View with images and charts

Routing Protocol of Wireless Sensor Network

Chapter 1

Introduction

A wireless sensor network (WSN) consists of spatially distributed autonomous devices called sensors and one or more base stations to cooperatively monitor physical or environmental conditions such as temperature sound vibration pressure motion or pollutants at different locations The research in sensor networks received a big boost with a number of funding initiatives by US military NSF and DARPA SENSIT [17] Many novel sensor based applications have emerged in recent past We may classify the sensor applications into following classes [6]

Monitoring spaces This class refers to passive data gathering recognizing occurrence of some events or conditions The gathered data are typically inputs to a number of target applications These target applications includes habitat monitoring monitoring of crops (failure pest attack) climate control security surveillance intelligent alarms (fire flash flood volcanic eruption) etc

Monitoring things This class refers to gathering data to recognize occurrence of specific states of a system On occurrence of these states the system may execute a sequence of internal transitions to get into a desirable state The target applications could be structural monitoring (bridge health monitoring) equipment maintenance medical diagnostics etc

Monitoring complex interactions It involves monitoring of complex interactions of systems and things For example wild-life habitat monitoring disaster management emergency response smart environments surface and sea navigation health care manufacturing process flow etc would require complex interactions of systems with objects or things

Such wide-ranging applications requiring WSNs make them candidates for intense research The research spans hardwares systems networking and programming methodologies Considering ubiquity of applications one of the crucial design decisions for sensor nodes has been to settle for a small form factor The advantage of small form factor is that these miniature devices are inexpensive Thousands of sensor nodes can be deployed with a small cost Therefore the key to success of sensor based applications is to network sensors in an efficient way for gathering sensory data from their respective deployed environments

Every sensor has three basic units namely sensing radio and battery The major constraint being limited energy (small battery unit) Consequently a large number of sensor nodes can be networked to gather sensory data and each sensor performs two main responsibilities namely (i) sensing activities and (ii) routing the sensed data to the base station or a controller The base station is a master node which is generally fixed and assumed to have uninterrupted power supply or accessible for maintenance (such as replacement of battery) It acts as an interface of the WSN for complex interactions with other objects The main

responsibility of base station is to collect information from various sensor nodes and process it for further disseminationactions

Energy dissipation at sensor node is a major concern as in many applications sensor have to be deployed in inaccessible environments Sensing alone is not an energy consuming activity but networking and programming certainly are So the major problem lies in activities like routing addressing and support for different class of programming services In this thesis our focus will be on routing problems

The topology of a WSN determined by

1 a set of external parameters such as node mobility weather interference and noise and

2 Also by a set controllable parameters like transmission power antenna type and direction etc

Normally the sensor nodes are uniformly spread over the whole of physical space to get the best coverage The nodes are mostly static unless they are either mounted on moving vehicles robots or tagged to live animals

Therefore topology control is mainly possible through adjustment of transmission power and associated radio adjustments Thus careful design of routing protocols that combine well with topology control mechanisms is extremely important for working of WSNs

The design of routing protocols for WSN are influenced by many factors including hardware constraints network topology and power consumption The sensor networks are mainly of two types - event-driven and time-driven (or continuous dissemination networks) In the event driven networks the sensor nodes sense the data and transmit it only if the data is considered critical enough to be communicated In the time driven networks sensors sense the data and transmit it to the central controller periodically The periodicity of relaying data packets is application dependent

The two broad categorization of routing protocols are (i) proactive and (ii) reactive In a proactive protocol the routes are discovered and maintained using periodic messaging Each node is expected to send periodic packets to the neighboring nodes The packets contain the routing information of the sender node The recipient nodes configure their routing tables based on received packets or make necessary updates as required These periodic packets known as Hello packets indicate that the sender can still take part in routing process As topology information is exchanged on the regular basis it causes unwanted overhead On the one hand a route to base station will always be available on request but on the other hand the nodes unnecessarily update routing tables on periodic basis even for the routes which may not be required at all The reactive protocol discovers the route once and then maintains it reactively Every time a path is unable to deliver a message that path is rendered invalid and new path is discovered Though the reactive protocols are quick to respond to the nodes moving in and out of the network the latency for discovering a path could be long and may be inadmissible for certain applications Both proactive and reactive protocols have their pros and cons So a protocol can best work mid way between the two

The coverage area of a sensor node (or the reach ability of a node) depends on its transmission range If the transmission range of all the nodes is high enough to reach the base station then it is considered one hop network Such networks do not incur overhead of additional control packets for route discovery and maintenance However as the wireless is a shared medium one hop networks lead to densely connected networks and suffer from severe congestion In other words there is a trade off in selecting a suitable transmission range for the nodes and severity of congestion The range should be chosen optimally to eliminate congestion and to retain desired network connectivity

A network in which all the nodes have same transmission range as in the example depicted by Fig 11(b) is called a Symmetric Network In a symmetric network as indicated by Fig 11(b) if node A is in the transmission range of node B then B must also be within transmission range A

If the transmission ranges of nodes are configured at the network start-up time and all the nodes do not have same transmission range then the network is considered to be an Asymmetric Network Fig 11(a) provides an example where a pair of asymmetric links is shown to exist between nodes A and B Node B is in range of A but A is not in range of B

Figure11 a) Asymmetric Link b) Symmetric Link

11 Motivation

As stated earlier the sensor nodes due to their small form factor have limited power In order to prolong the life of the wireless sensor networks the routing protocols apart from being robust and scalable needs to be highly energy efficient A lot of research [7 8 23 30] has taken place in this direction and various routing protocols are proposed to achieve these objectives

The work reported in this thesis aims at designing of a multihop energy efficient reliable and fault-tolerant routing protocol In a fully connected network all nodes can directly access the base station However wireless being a broadcast medium the congestion [13] in such a network is very high Typically each node in a multihop WSN would discover a path to the base station and route its data through this path This causes the nodes near the base station to be used more frequently than the nodes away from the base station The reason is the former set of nodes not only send their own sensed data but are also responsible for forwarding the packets from the far off nodes in the network This results in a bottleneck around the base

station If the nodes around the base station go dead then the nodes away from base station will be unable to send the data unless they increase their transmission ranges

Figure 12 Example Network

Fig 12 shows an example of a typical sensor network The filled black node is the base station The lines depict the connectivity and the filled gray nodes are the normal sensing nodes In this example node-2 and node-3 are one hop nodes Node-2 is responsible not only for sending its own data but also for forwarding the data from nodes-4 5 6 9 and 10 Similarly node-3 is responsible for sending its own data and as well as forwarding data of nodes-7 8 11 and 12 Thus the nodes situated at a distance of one hop from base station are used more often than the other nodes It causes such nodes to dissipate energy at a substantially higher rate than the rest of the nodes in the network Consequently the network becomes dead very soon The residual energy in the nodes near the base station may be sufficient to sense but may not be sufficient communicate the sensed data to the base station This observation led us to think how a routing protocol may avoid the formation of transmission bottleneck around the base station The underlying idea is to ensure that the energy dissipation at each node is more or less same over the entire network The approach we devised is to adjust transmission range of individual nodes to provide connectivity of base station without involving the nodes near the base station as and when desired It led to dynamicity in the sensor network environment We can view the dynamicity as insertion of new nodes and deletion of nodes (node failure) at random time

Our second observation is that WSNs experience high packet losses So reliability is also important for WSN Transmission reliability can be achieved through an acknowledgment based protocol It helps in deciding the requirement of retransmissions if the packet loss occurs and thus increases the reliability

An usual trend in protocol design suggests that the location parameters play crucial role for the routing purposes [12] The determination of position parameters requires use of embedded GPS [4] receivers But with embedded GPS receivers will make the sensor nodes substantially expensive Additionally also the nodes will consume more power Therefore our aim is also to ensure come up with routing protocols which avoids use of location based information

12 Problem statement and the Challenges Involved

The investigation in this thesis is oriented towards the design a multi hop routing protocol for time driven WSN which achieves following objectives

Avoid the formation of the congestion bottleneck around the base station

Handle dynamic changes in topology caused by transmission power adjustments

Handle node failures on existing paths by a novel route repair procedure that leverages ability to adjust node transmission range

Ensure reliability through use of acknowledgements and limited retransmission of lost packets

Improve the network lifetime by considering residual node energy during route discovery

Analyze and compare the new protocol with the existing ones

The objectives listed at items 1 and 5 are achieved by better utilization of the resources and treating the network as a dynamically changing asymmetric network The topology change in the network is mainly determined by adjustment of transmission range from time to time The resulting change topology modifies path from every node to base station The routing protocol is therefore is a mix of both proactive and reactive protocol The tasks of route discovery and route maintenance become more challenging in an asymmetric network as Hello messages can not be used for such purposes in asymmetric links The latency of route discovery also has to be minimized

Balancing distribution of load for routing (forwarding packets) among the nodes shall ensure the uniform and better resource utilization The concept of asymmetric network shall allow the nodes to use appropriate transmission ranges which is more suitable for its longer lifetime depending on its position in the network

Fault tolerance in routing protocols for WSN is a novel idea It has been possible to tolerate node failure simply by eliminating failed nodes through a route maintenance step It involves the adjustment of transmission power at some nodes for establishing connectivity with the base station when the routes are found to be broken due to node failures We experimented with up to 5 node failures

The objective concerning reliability in transmission is achieved by acknowledgments for the data packets The protocol supports at most three retransmissions of the data packets to handle packet loss

The routing metric is chosen by optimizing both latency of routes and residual energies at the nodes The other desirable property of a protocol which is addressed is load sharing The load for routing data packets is distributed over the multiple alternate paths available at a node It ensures overall better utilization of energy resources and effectively extends lifetime of the network

Finally we performed simulation over OMNet++ [1] and compared our results with other notable time driven WSN routing protocols such as LEACH [12] and SPIN [11] Our results show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation

13 Organization of Thesis

The rest of the thesis is organized as followsChapter 2 discusses the related work and lays the background further discussion The important existing protocols and the improvements subsequently proposed upon them are also discussed It also indicates how these solutions differ from the new protocol proposed in this thesis

In Chapter 3 the new protocol is described We focus on robustness network lifetime routing latency to show how the proposed algorithm has potentials to outperform other existing protocols The protocol is discussed in details by dwelling on its features various phases of operation and implementation aspects

Chapter 4 throws light on the actual logic of the proposal by presenting the algorithms An analysis of performance parameters is also presented to provide an insight into its performance The chapter also describes simulation software and the simulation environment The details of the radio characteristics used for the purpose of empirical evaluation

The results duly supported by the relevant plots for performance characteristics and related analysis are presented in Chapter 5

Chapter 6 concludes the thesis with some directions for future work

Chapter 2

Related Work

As explained in chapter 1 the design of routing protocols for WSN is a subject of intense research as both quality and quantity of information delivered to the end-users is very important for the applications centered on WSN It has been observed that different protocols work better in different environmentsapplications The issue of the effective utilization of energy resource has also been addressed in extensively in the literature This chapter deals with related work and the underlying concepts which form the basis of energy aware routing protocols for WNSs

Section 21 explains how simple routing ideas could lead to unnecessary wastage of precious energy resources Section 22 deals with some initial thinking done by researchers on the ways to avoid broadcast storm which appear to be the main reason behind excessive energy wastage in routings ideas based on flooding and gossiping Section 23 describes a number of routing protocols namely LEACH [12] PEGASIS [19] TEEN [20] and APTEEN [21] which try to eliminate redundant data broadcast and conserve the crucial energy resource by aggregation in some way or other Later section 24 which are relevant in context to this thesis The subject of Discussion in section 25 is the quality and reliability of data delivered by the sensors

21 Flooding and Gossiping

The conventional mechanism for relaying data from individual sensors to base station is Flooding or Gossiping [10] The underlying idea of flooding is that each sensor on receiving a data packet broadcasts the same to its every neighbor This packet is further re-broadcast until it arrives at the destination or the maximum number of hops for the packet is reached Quite obviously flooding raises unmanageable broadcast storms as depicted in Fig 21

Figure 21 Broadcast storm

It may be appropriate for networks which experience quick topology changes WSN mainly consists of static sensors nodes So topology remains more or less static for a considerable period of time Of course it is possible to control topology by adjusting transmission power of sensor motes But such power adjustments can only be realized by program controls and hence predictable The other serious problems with broadcast storm are redundancies contentions and collisions which result in wasteful consumption of node energies Since sensor nodes have only limited energy reserves in the form of small batteries they can ill-afford such associated problems arising out of broadcast storms Therefore flooding does not seem to be an appropriate approach of design of routing protocols in WSNs The deficiencies of data reporting by flooding can be viewed under three categories [11] namely

Implosion it refers to nodes disseminating data to all of their neighbors regardless of whether the data has already been received by these neighbors It results in wastage of energy across several nodes since the same piece of data arrives at every node via multiple paths

Overlap it is not possible to have a precise deployment whereby sensor could cover disjoint spatial regions So nodes often have overlapping spatial coverage Data dissemination by flooding therefore results in data from overlapping regions to be routed by different alternative sensors and flooded over the network

Resource ignorance flooding does not require the nodes to modify their activities on the basis of energy threshold Consequently a node may run out of energy too soon to perform any meaningful data sensing or reporting For example nodes near the base station could run out of their batteries being over used while forwarding data

from distant nodes This may result in network going dead too soon even when many distant nodes have sufficient energy to carry out their reporting and sensing activities

Gossiping [9] is just a slightly enhanced version of flooding wherein the recipient node sends the data packet to a randomly selected neighbor That neighbor in turn picks another random neighbor to forward the packet to and so on Another usual variation of gossip protocol is that the recipient could rebroadcasts or discard the data with probabilities p and (11048576 p) respectively By controlling the fraction of executions where gossip die out relatively low and keeping the gossip probability low it is possible to save up to 35 message overhead compared to flooding [9] But gossip based protocol exhibit bimodal behavior in the sense that in almost all executions either most of the nodes receive a message or hardly any of them do Furthermore in many applications gossip delay may be unacceptable

22 SPIN and Directed Diffusion

The focus Sensor Protocols for Information via Negotiation (SPIN) [11] was to work around the problems of simple flood based protocol mentioned in the previous section

SPIN was successful in addressing all the problems existing in the flood based protocols It labeled the data with the high level data descriptors called metadata The metadata was used to negotiate between the nodes eliminating the redundant data from being transmitted to the nodes As the metadata was supposed to be the compressed form of the original data so the network congestion could be avoided So the SPIN mainly comprises of three phases

STEP 1 Advertisement of the metadata

STEP 2 Request for the data

STEP 3 Actual data transmission

Fig 22 depicts the stepwise process

Figure 22 SPIN

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

responsibility of base station is to collect information from various sensor nodes and process it for further disseminationactions

Energy dissipation at sensor node is a major concern as in many applications sensor have to be deployed in inaccessible environments Sensing alone is not an energy consuming activity but networking and programming certainly are So the major problem lies in activities like routing addressing and support for different class of programming services In this thesis our focus will be on routing problems

The topology of a WSN determined by

1 a set of external parameters such as node mobility weather interference and noise and

2 Also by a set controllable parameters like transmission power antenna type and direction etc

Normally the sensor nodes are uniformly spread over the whole of physical space to get the best coverage The nodes are mostly static unless they are either mounted on moving vehicles robots or tagged to live animals

Therefore topology control is mainly possible through adjustment of transmission power and associated radio adjustments Thus careful design of routing protocols that combine well with topology control mechanisms is extremely important for working of WSNs

The design of routing protocols for WSN are influenced by many factors including hardware constraints network topology and power consumption The sensor networks are mainly of two types - event-driven and time-driven (or continuous dissemination networks) In the event driven networks the sensor nodes sense the data and transmit it only if the data is considered critical enough to be communicated In the time driven networks sensors sense the data and transmit it to the central controller periodically The periodicity of relaying data packets is application dependent

The two broad categorization of routing protocols are (i) proactive and (ii) reactive In a proactive protocol the routes are discovered and maintained using periodic messaging Each node is expected to send periodic packets to the neighboring nodes The packets contain the routing information of the sender node The recipient nodes configure their routing tables based on received packets or make necessary updates as required These periodic packets known as Hello packets indicate that the sender can still take part in routing process As topology information is exchanged on the regular basis it causes unwanted overhead On the one hand a route to base station will always be available on request but on the other hand the nodes unnecessarily update routing tables on periodic basis even for the routes which may not be required at all The reactive protocol discovers the route once and then maintains it reactively Every time a path is unable to deliver a message that path is rendered invalid and new path is discovered Though the reactive protocols are quick to respond to the nodes moving in and out of the network the latency for discovering a path could be long and may be inadmissible for certain applications Both proactive and reactive protocols have their pros and cons So a protocol can best work mid way between the two

The coverage area of a sensor node (or the reach ability of a node) depends on its transmission range If the transmission range of all the nodes is high enough to reach the base station then it is considered one hop network Such networks do not incur overhead of additional control packets for route discovery and maintenance However as the wireless is a shared medium one hop networks lead to densely connected networks and suffer from severe congestion In other words there is a trade off in selecting a suitable transmission range for the nodes and severity of congestion The range should be chosen optimally to eliminate congestion and to retain desired network connectivity

A network in which all the nodes have same transmission range as in the example depicted by Fig 11(b) is called a Symmetric Network In a symmetric network as indicated by Fig 11(b) if node A is in the transmission range of node B then B must also be within transmission range A

If the transmission ranges of nodes are configured at the network start-up time and all the nodes do not have same transmission range then the network is considered to be an Asymmetric Network Fig 11(a) provides an example where a pair of asymmetric links is shown to exist between nodes A and B Node B is in range of A but A is not in range of B

Figure11 a) Asymmetric Link b) Symmetric Link

11 Motivation

As stated earlier the sensor nodes due to their small form factor have limited power In order to prolong the life of the wireless sensor networks the routing protocols apart from being robust and scalable needs to be highly energy efficient A lot of research [7 8 23 30] has taken place in this direction and various routing protocols are proposed to achieve these objectives

The work reported in this thesis aims at designing of a multihop energy efficient reliable and fault-tolerant routing protocol In a fully connected network all nodes can directly access the base station However wireless being a broadcast medium the congestion [13] in such a network is very high Typically each node in a multihop WSN would discover a path to the base station and route its data through this path This causes the nodes near the base station to be used more frequently than the nodes away from the base station The reason is the former set of nodes not only send their own sensed data but are also responsible for forwarding the packets from the far off nodes in the network This results in a bottleneck around the base

station If the nodes around the base station go dead then the nodes away from base station will be unable to send the data unless they increase their transmission ranges

Figure 12 Example Network

Fig 12 shows an example of a typical sensor network The filled black node is the base station The lines depict the connectivity and the filled gray nodes are the normal sensing nodes In this example node-2 and node-3 are one hop nodes Node-2 is responsible not only for sending its own data but also for forwarding the data from nodes-4 5 6 9 and 10 Similarly node-3 is responsible for sending its own data and as well as forwarding data of nodes-7 8 11 and 12 Thus the nodes situated at a distance of one hop from base station are used more often than the other nodes It causes such nodes to dissipate energy at a substantially higher rate than the rest of the nodes in the network Consequently the network becomes dead very soon The residual energy in the nodes near the base station may be sufficient to sense but may not be sufficient communicate the sensed data to the base station This observation led us to think how a routing protocol may avoid the formation of transmission bottleneck around the base station The underlying idea is to ensure that the energy dissipation at each node is more or less same over the entire network The approach we devised is to adjust transmission range of individual nodes to provide connectivity of base station without involving the nodes near the base station as and when desired It led to dynamicity in the sensor network environment We can view the dynamicity as insertion of new nodes and deletion of nodes (node failure) at random time

Our second observation is that WSNs experience high packet losses So reliability is also important for WSN Transmission reliability can be achieved through an acknowledgment based protocol It helps in deciding the requirement of retransmissions if the packet loss occurs and thus increases the reliability

An usual trend in protocol design suggests that the location parameters play crucial role for the routing purposes [12] The determination of position parameters requires use of embedded GPS [4] receivers But with embedded GPS receivers will make the sensor nodes substantially expensive Additionally also the nodes will consume more power Therefore our aim is also to ensure come up with routing protocols which avoids use of location based information

12 Problem statement and the Challenges Involved

The investigation in this thesis is oriented towards the design a multi hop routing protocol for time driven WSN which achieves following objectives

Avoid the formation of the congestion bottleneck around the base station

Handle dynamic changes in topology caused by transmission power adjustments

Handle node failures on existing paths by a novel route repair procedure that leverages ability to adjust node transmission range

Ensure reliability through use of acknowledgements and limited retransmission of lost packets

Improve the network lifetime by considering residual node energy during route discovery

Analyze and compare the new protocol with the existing ones

The objectives listed at items 1 and 5 are achieved by better utilization of the resources and treating the network as a dynamically changing asymmetric network The topology change in the network is mainly determined by adjustment of transmission range from time to time The resulting change topology modifies path from every node to base station The routing protocol is therefore is a mix of both proactive and reactive protocol The tasks of route discovery and route maintenance become more challenging in an asymmetric network as Hello messages can not be used for such purposes in asymmetric links The latency of route discovery also has to be minimized

Balancing distribution of load for routing (forwarding packets) among the nodes shall ensure the uniform and better resource utilization The concept of asymmetric network shall allow the nodes to use appropriate transmission ranges which is more suitable for its longer lifetime depending on its position in the network

Fault tolerance in routing protocols for WSN is a novel idea It has been possible to tolerate node failure simply by eliminating failed nodes through a route maintenance step It involves the adjustment of transmission power at some nodes for establishing connectivity with the base station when the routes are found to be broken due to node failures We experimented with up to 5 node failures

The objective concerning reliability in transmission is achieved by acknowledgments for the data packets The protocol supports at most three retransmissions of the data packets to handle packet loss

The routing metric is chosen by optimizing both latency of routes and residual energies at the nodes The other desirable property of a protocol which is addressed is load sharing The load for routing data packets is distributed over the multiple alternate paths available at a node It ensures overall better utilization of energy resources and effectively extends lifetime of the network

Finally we performed simulation over OMNet++ [1] and compared our results with other notable time driven WSN routing protocols such as LEACH [12] and SPIN [11] Our results show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation

13 Organization of Thesis

The rest of the thesis is organized as followsChapter 2 discusses the related work and lays the background further discussion The important existing protocols and the improvements subsequently proposed upon them are also discussed It also indicates how these solutions differ from the new protocol proposed in this thesis

In Chapter 3 the new protocol is described We focus on robustness network lifetime routing latency to show how the proposed algorithm has potentials to outperform other existing protocols The protocol is discussed in details by dwelling on its features various phases of operation and implementation aspects

Chapter 4 throws light on the actual logic of the proposal by presenting the algorithms An analysis of performance parameters is also presented to provide an insight into its performance The chapter also describes simulation software and the simulation environment The details of the radio characteristics used for the purpose of empirical evaluation

The results duly supported by the relevant plots for performance characteristics and related analysis are presented in Chapter 5

Chapter 6 concludes the thesis with some directions for future work

Chapter 2

Related Work

As explained in chapter 1 the design of routing protocols for WSN is a subject of intense research as both quality and quantity of information delivered to the end-users is very important for the applications centered on WSN It has been observed that different protocols work better in different environmentsapplications The issue of the effective utilization of energy resource has also been addressed in extensively in the literature This chapter deals with related work and the underlying concepts which form the basis of energy aware routing protocols for WNSs

Section 21 explains how simple routing ideas could lead to unnecessary wastage of precious energy resources Section 22 deals with some initial thinking done by researchers on the ways to avoid broadcast storm which appear to be the main reason behind excessive energy wastage in routings ideas based on flooding and gossiping Section 23 describes a number of routing protocols namely LEACH [12] PEGASIS [19] TEEN [20] and APTEEN [21] which try to eliminate redundant data broadcast and conserve the crucial energy resource by aggregation in some way or other Later section 24 which are relevant in context to this thesis The subject of Discussion in section 25 is the quality and reliability of data delivered by the sensors

21 Flooding and Gossiping

The conventional mechanism for relaying data from individual sensors to base station is Flooding or Gossiping [10] The underlying idea of flooding is that each sensor on receiving a data packet broadcasts the same to its every neighbor This packet is further re-broadcast until it arrives at the destination or the maximum number of hops for the packet is reached Quite obviously flooding raises unmanageable broadcast storms as depicted in Fig 21

Figure 21 Broadcast storm

It may be appropriate for networks which experience quick topology changes WSN mainly consists of static sensors nodes So topology remains more or less static for a considerable period of time Of course it is possible to control topology by adjusting transmission power of sensor motes But such power adjustments can only be realized by program controls and hence predictable The other serious problems with broadcast storm are redundancies contentions and collisions which result in wasteful consumption of node energies Since sensor nodes have only limited energy reserves in the form of small batteries they can ill-afford such associated problems arising out of broadcast storms Therefore flooding does not seem to be an appropriate approach of design of routing protocols in WSNs The deficiencies of data reporting by flooding can be viewed under three categories [11] namely

Implosion it refers to nodes disseminating data to all of their neighbors regardless of whether the data has already been received by these neighbors It results in wastage of energy across several nodes since the same piece of data arrives at every node via multiple paths

Overlap it is not possible to have a precise deployment whereby sensor could cover disjoint spatial regions So nodes often have overlapping spatial coverage Data dissemination by flooding therefore results in data from overlapping regions to be routed by different alternative sensors and flooded over the network

Resource ignorance flooding does not require the nodes to modify their activities on the basis of energy threshold Consequently a node may run out of energy too soon to perform any meaningful data sensing or reporting For example nodes near the base station could run out of their batteries being over used while forwarding data

from distant nodes This may result in network going dead too soon even when many distant nodes have sufficient energy to carry out their reporting and sensing activities

Gossiping [9] is just a slightly enhanced version of flooding wherein the recipient node sends the data packet to a randomly selected neighbor That neighbor in turn picks another random neighbor to forward the packet to and so on Another usual variation of gossip protocol is that the recipient could rebroadcasts or discard the data with probabilities p and (11048576 p) respectively By controlling the fraction of executions where gossip die out relatively low and keeping the gossip probability low it is possible to save up to 35 message overhead compared to flooding [9] But gossip based protocol exhibit bimodal behavior in the sense that in almost all executions either most of the nodes receive a message or hardly any of them do Furthermore in many applications gossip delay may be unacceptable

22 SPIN and Directed Diffusion

The focus Sensor Protocols for Information via Negotiation (SPIN) [11] was to work around the problems of simple flood based protocol mentioned in the previous section

SPIN was successful in addressing all the problems existing in the flood based protocols It labeled the data with the high level data descriptors called metadata The metadata was used to negotiate between the nodes eliminating the redundant data from being transmitted to the nodes As the metadata was supposed to be the compressed form of the original data so the network congestion could be avoided So the SPIN mainly comprises of three phases

STEP 1 Advertisement of the metadata

STEP 2 Request for the data

STEP 3 Actual data transmission

Fig 22 depicts the stepwise process

Figure 22 SPIN

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The coverage area of a sensor node (or the reach ability of a node) depends on its transmission range If the transmission range of all the nodes is high enough to reach the base station then it is considered one hop network Such networks do not incur overhead of additional control packets for route discovery and maintenance However as the wireless is a shared medium one hop networks lead to densely connected networks and suffer from severe congestion In other words there is a trade off in selecting a suitable transmission range for the nodes and severity of congestion The range should be chosen optimally to eliminate congestion and to retain desired network connectivity

A network in which all the nodes have same transmission range as in the example depicted by Fig 11(b) is called a Symmetric Network In a symmetric network as indicated by Fig 11(b) if node A is in the transmission range of node B then B must also be within transmission range A

If the transmission ranges of nodes are configured at the network start-up time and all the nodes do not have same transmission range then the network is considered to be an Asymmetric Network Fig 11(a) provides an example where a pair of asymmetric links is shown to exist between nodes A and B Node B is in range of A but A is not in range of B

Figure11 a) Asymmetric Link b) Symmetric Link

11 Motivation

As stated earlier the sensor nodes due to their small form factor have limited power In order to prolong the life of the wireless sensor networks the routing protocols apart from being robust and scalable needs to be highly energy efficient A lot of research [7 8 23 30] has taken place in this direction and various routing protocols are proposed to achieve these objectives

The work reported in this thesis aims at designing of a multihop energy efficient reliable and fault-tolerant routing protocol In a fully connected network all nodes can directly access the base station However wireless being a broadcast medium the congestion [13] in such a network is very high Typically each node in a multihop WSN would discover a path to the base station and route its data through this path This causes the nodes near the base station to be used more frequently than the nodes away from the base station The reason is the former set of nodes not only send their own sensed data but are also responsible for forwarding the packets from the far off nodes in the network This results in a bottleneck around the base

station If the nodes around the base station go dead then the nodes away from base station will be unable to send the data unless they increase their transmission ranges

Figure 12 Example Network

Fig 12 shows an example of a typical sensor network The filled black node is the base station The lines depict the connectivity and the filled gray nodes are the normal sensing nodes In this example node-2 and node-3 are one hop nodes Node-2 is responsible not only for sending its own data but also for forwarding the data from nodes-4 5 6 9 and 10 Similarly node-3 is responsible for sending its own data and as well as forwarding data of nodes-7 8 11 and 12 Thus the nodes situated at a distance of one hop from base station are used more often than the other nodes It causes such nodes to dissipate energy at a substantially higher rate than the rest of the nodes in the network Consequently the network becomes dead very soon The residual energy in the nodes near the base station may be sufficient to sense but may not be sufficient communicate the sensed data to the base station This observation led us to think how a routing protocol may avoid the formation of transmission bottleneck around the base station The underlying idea is to ensure that the energy dissipation at each node is more or less same over the entire network The approach we devised is to adjust transmission range of individual nodes to provide connectivity of base station without involving the nodes near the base station as and when desired It led to dynamicity in the sensor network environment We can view the dynamicity as insertion of new nodes and deletion of nodes (node failure) at random time

Our second observation is that WSNs experience high packet losses So reliability is also important for WSN Transmission reliability can be achieved through an acknowledgment based protocol It helps in deciding the requirement of retransmissions if the packet loss occurs and thus increases the reliability

An usual trend in protocol design suggests that the location parameters play crucial role for the routing purposes [12] The determination of position parameters requires use of embedded GPS [4] receivers But with embedded GPS receivers will make the sensor nodes substantially expensive Additionally also the nodes will consume more power Therefore our aim is also to ensure come up with routing protocols which avoids use of location based information

12 Problem statement and the Challenges Involved

The investigation in this thesis is oriented towards the design a multi hop routing protocol for time driven WSN which achieves following objectives

Avoid the formation of the congestion bottleneck around the base station

Handle dynamic changes in topology caused by transmission power adjustments

Handle node failures on existing paths by a novel route repair procedure that leverages ability to adjust node transmission range

Ensure reliability through use of acknowledgements and limited retransmission of lost packets

Improve the network lifetime by considering residual node energy during route discovery

Analyze and compare the new protocol with the existing ones

The objectives listed at items 1 and 5 are achieved by better utilization of the resources and treating the network as a dynamically changing asymmetric network The topology change in the network is mainly determined by adjustment of transmission range from time to time The resulting change topology modifies path from every node to base station The routing protocol is therefore is a mix of both proactive and reactive protocol The tasks of route discovery and route maintenance become more challenging in an asymmetric network as Hello messages can not be used for such purposes in asymmetric links The latency of route discovery also has to be minimized

Balancing distribution of load for routing (forwarding packets) among the nodes shall ensure the uniform and better resource utilization The concept of asymmetric network shall allow the nodes to use appropriate transmission ranges which is more suitable for its longer lifetime depending on its position in the network

Fault tolerance in routing protocols for WSN is a novel idea It has been possible to tolerate node failure simply by eliminating failed nodes through a route maintenance step It involves the adjustment of transmission power at some nodes for establishing connectivity with the base station when the routes are found to be broken due to node failures We experimented with up to 5 node failures

The objective concerning reliability in transmission is achieved by acknowledgments for the data packets The protocol supports at most three retransmissions of the data packets to handle packet loss

The routing metric is chosen by optimizing both latency of routes and residual energies at the nodes The other desirable property of a protocol which is addressed is load sharing The load for routing data packets is distributed over the multiple alternate paths available at a node It ensures overall better utilization of energy resources and effectively extends lifetime of the network

Finally we performed simulation over OMNet++ [1] and compared our results with other notable time driven WSN routing protocols such as LEACH [12] and SPIN [11] Our results show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation

13 Organization of Thesis

The rest of the thesis is organized as followsChapter 2 discusses the related work and lays the background further discussion The important existing protocols and the improvements subsequently proposed upon them are also discussed It also indicates how these solutions differ from the new protocol proposed in this thesis

In Chapter 3 the new protocol is described We focus on robustness network lifetime routing latency to show how the proposed algorithm has potentials to outperform other existing protocols The protocol is discussed in details by dwelling on its features various phases of operation and implementation aspects

Chapter 4 throws light on the actual logic of the proposal by presenting the algorithms An analysis of performance parameters is also presented to provide an insight into its performance The chapter also describes simulation software and the simulation environment The details of the radio characteristics used for the purpose of empirical evaluation

The results duly supported by the relevant plots for performance characteristics and related analysis are presented in Chapter 5

Chapter 6 concludes the thesis with some directions for future work

Chapter 2

Related Work

As explained in chapter 1 the design of routing protocols for WSN is a subject of intense research as both quality and quantity of information delivered to the end-users is very important for the applications centered on WSN It has been observed that different protocols work better in different environmentsapplications The issue of the effective utilization of energy resource has also been addressed in extensively in the literature This chapter deals with related work and the underlying concepts which form the basis of energy aware routing protocols for WNSs

Section 21 explains how simple routing ideas could lead to unnecessary wastage of precious energy resources Section 22 deals with some initial thinking done by researchers on the ways to avoid broadcast storm which appear to be the main reason behind excessive energy wastage in routings ideas based on flooding and gossiping Section 23 describes a number of routing protocols namely LEACH [12] PEGASIS [19] TEEN [20] and APTEEN [21] which try to eliminate redundant data broadcast and conserve the crucial energy resource by aggregation in some way or other Later section 24 which are relevant in context to this thesis The subject of Discussion in section 25 is the quality and reliability of data delivered by the sensors

21 Flooding and Gossiping

The conventional mechanism for relaying data from individual sensors to base station is Flooding or Gossiping [10] The underlying idea of flooding is that each sensor on receiving a data packet broadcasts the same to its every neighbor This packet is further re-broadcast until it arrives at the destination or the maximum number of hops for the packet is reached Quite obviously flooding raises unmanageable broadcast storms as depicted in Fig 21

Figure 21 Broadcast storm

It may be appropriate for networks which experience quick topology changes WSN mainly consists of static sensors nodes So topology remains more or less static for a considerable period of time Of course it is possible to control topology by adjusting transmission power of sensor motes But such power adjustments can only be realized by program controls and hence predictable The other serious problems with broadcast storm are redundancies contentions and collisions which result in wasteful consumption of node energies Since sensor nodes have only limited energy reserves in the form of small batteries they can ill-afford such associated problems arising out of broadcast storms Therefore flooding does not seem to be an appropriate approach of design of routing protocols in WSNs The deficiencies of data reporting by flooding can be viewed under three categories [11] namely

Implosion it refers to nodes disseminating data to all of their neighbors regardless of whether the data has already been received by these neighbors It results in wastage of energy across several nodes since the same piece of data arrives at every node via multiple paths

Overlap it is not possible to have a precise deployment whereby sensor could cover disjoint spatial regions So nodes often have overlapping spatial coverage Data dissemination by flooding therefore results in data from overlapping regions to be routed by different alternative sensors and flooded over the network

Resource ignorance flooding does not require the nodes to modify their activities on the basis of energy threshold Consequently a node may run out of energy too soon to perform any meaningful data sensing or reporting For example nodes near the base station could run out of their batteries being over used while forwarding data

from distant nodes This may result in network going dead too soon even when many distant nodes have sufficient energy to carry out their reporting and sensing activities

Gossiping [9] is just a slightly enhanced version of flooding wherein the recipient node sends the data packet to a randomly selected neighbor That neighbor in turn picks another random neighbor to forward the packet to and so on Another usual variation of gossip protocol is that the recipient could rebroadcasts or discard the data with probabilities p and (11048576 p) respectively By controlling the fraction of executions where gossip die out relatively low and keeping the gossip probability low it is possible to save up to 35 message overhead compared to flooding [9] But gossip based protocol exhibit bimodal behavior in the sense that in almost all executions either most of the nodes receive a message or hardly any of them do Furthermore in many applications gossip delay may be unacceptable

22 SPIN and Directed Diffusion

The focus Sensor Protocols for Information via Negotiation (SPIN) [11] was to work around the problems of simple flood based protocol mentioned in the previous section

SPIN was successful in addressing all the problems existing in the flood based protocols It labeled the data with the high level data descriptors called metadata The metadata was used to negotiate between the nodes eliminating the redundant data from being transmitted to the nodes As the metadata was supposed to be the compressed form of the original data so the network congestion could be avoided So the SPIN mainly comprises of three phases

STEP 1 Advertisement of the metadata

STEP 2 Request for the data

STEP 3 Actual data transmission

Fig 22 depicts the stepwise process

Figure 22 SPIN

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

station If the nodes around the base station go dead then the nodes away from base station will be unable to send the data unless they increase their transmission ranges

Figure 12 Example Network

Fig 12 shows an example of a typical sensor network The filled black node is the base station The lines depict the connectivity and the filled gray nodes are the normal sensing nodes In this example node-2 and node-3 are one hop nodes Node-2 is responsible not only for sending its own data but also for forwarding the data from nodes-4 5 6 9 and 10 Similarly node-3 is responsible for sending its own data and as well as forwarding data of nodes-7 8 11 and 12 Thus the nodes situated at a distance of one hop from base station are used more often than the other nodes It causes such nodes to dissipate energy at a substantially higher rate than the rest of the nodes in the network Consequently the network becomes dead very soon The residual energy in the nodes near the base station may be sufficient to sense but may not be sufficient communicate the sensed data to the base station This observation led us to think how a routing protocol may avoid the formation of transmission bottleneck around the base station The underlying idea is to ensure that the energy dissipation at each node is more or less same over the entire network The approach we devised is to adjust transmission range of individual nodes to provide connectivity of base station without involving the nodes near the base station as and when desired It led to dynamicity in the sensor network environment We can view the dynamicity as insertion of new nodes and deletion of nodes (node failure) at random time

Our second observation is that WSNs experience high packet losses So reliability is also important for WSN Transmission reliability can be achieved through an acknowledgment based protocol It helps in deciding the requirement of retransmissions if the packet loss occurs and thus increases the reliability

An usual trend in protocol design suggests that the location parameters play crucial role for the routing purposes [12] The determination of position parameters requires use of embedded GPS [4] receivers But with embedded GPS receivers will make the sensor nodes substantially expensive Additionally also the nodes will consume more power Therefore our aim is also to ensure come up with routing protocols which avoids use of location based information

12 Problem statement and the Challenges Involved

The investigation in this thesis is oriented towards the design a multi hop routing protocol for time driven WSN which achieves following objectives

Avoid the formation of the congestion bottleneck around the base station

Handle dynamic changes in topology caused by transmission power adjustments

Handle node failures on existing paths by a novel route repair procedure that leverages ability to adjust node transmission range

Ensure reliability through use of acknowledgements and limited retransmission of lost packets

Improve the network lifetime by considering residual node energy during route discovery

Analyze and compare the new protocol with the existing ones

The objectives listed at items 1 and 5 are achieved by better utilization of the resources and treating the network as a dynamically changing asymmetric network The topology change in the network is mainly determined by adjustment of transmission range from time to time The resulting change topology modifies path from every node to base station The routing protocol is therefore is a mix of both proactive and reactive protocol The tasks of route discovery and route maintenance become more challenging in an asymmetric network as Hello messages can not be used for such purposes in asymmetric links The latency of route discovery also has to be minimized

Balancing distribution of load for routing (forwarding packets) among the nodes shall ensure the uniform and better resource utilization The concept of asymmetric network shall allow the nodes to use appropriate transmission ranges which is more suitable for its longer lifetime depending on its position in the network

Fault tolerance in routing protocols for WSN is a novel idea It has been possible to tolerate node failure simply by eliminating failed nodes through a route maintenance step It involves the adjustment of transmission power at some nodes for establishing connectivity with the base station when the routes are found to be broken due to node failures We experimented with up to 5 node failures

The objective concerning reliability in transmission is achieved by acknowledgments for the data packets The protocol supports at most three retransmissions of the data packets to handle packet loss

The routing metric is chosen by optimizing both latency of routes and residual energies at the nodes The other desirable property of a protocol which is addressed is load sharing The load for routing data packets is distributed over the multiple alternate paths available at a node It ensures overall better utilization of energy resources and effectively extends lifetime of the network

Finally we performed simulation over OMNet++ [1] and compared our results with other notable time driven WSN routing protocols such as LEACH [12] and SPIN [11] Our results show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation

13 Organization of Thesis

The rest of the thesis is organized as followsChapter 2 discusses the related work and lays the background further discussion The important existing protocols and the improvements subsequently proposed upon them are also discussed It also indicates how these solutions differ from the new protocol proposed in this thesis

In Chapter 3 the new protocol is described We focus on robustness network lifetime routing latency to show how the proposed algorithm has potentials to outperform other existing protocols The protocol is discussed in details by dwelling on its features various phases of operation and implementation aspects

Chapter 4 throws light on the actual logic of the proposal by presenting the algorithms An analysis of performance parameters is also presented to provide an insight into its performance The chapter also describes simulation software and the simulation environment The details of the radio characteristics used for the purpose of empirical evaluation

The results duly supported by the relevant plots for performance characteristics and related analysis are presented in Chapter 5

Chapter 6 concludes the thesis with some directions for future work

Chapter 2

Related Work

As explained in chapter 1 the design of routing protocols for WSN is a subject of intense research as both quality and quantity of information delivered to the end-users is very important for the applications centered on WSN It has been observed that different protocols work better in different environmentsapplications The issue of the effective utilization of energy resource has also been addressed in extensively in the literature This chapter deals with related work and the underlying concepts which form the basis of energy aware routing protocols for WNSs

Section 21 explains how simple routing ideas could lead to unnecessary wastage of precious energy resources Section 22 deals with some initial thinking done by researchers on the ways to avoid broadcast storm which appear to be the main reason behind excessive energy wastage in routings ideas based on flooding and gossiping Section 23 describes a number of routing protocols namely LEACH [12] PEGASIS [19] TEEN [20] and APTEEN [21] which try to eliminate redundant data broadcast and conserve the crucial energy resource by aggregation in some way or other Later section 24 which are relevant in context to this thesis The subject of Discussion in section 25 is the quality and reliability of data delivered by the sensors

21 Flooding and Gossiping

The conventional mechanism for relaying data from individual sensors to base station is Flooding or Gossiping [10] The underlying idea of flooding is that each sensor on receiving a data packet broadcasts the same to its every neighbor This packet is further re-broadcast until it arrives at the destination or the maximum number of hops for the packet is reached Quite obviously flooding raises unmanageable broadcast storms as depicted in Fig 21

Figure 21 Broadcast storm

It may be appropriate for networks which experience quick topology changes WSN mainly consists of static sensors nodes So topology remains more or less static for a considerable period of time Of course it is possible to control topology by adjusting transmission power of sensor motes But such power adjustments can only be realized by program controls and hence predictable The other serious problems with broadcast storm are redundancies contentions and collisions which result in wasteful consumption of node energies Since sensor nodes have only limited energy reserves in the form of small batteries they can ill-afford such associated problems arising out of broadcast storms Therefore flooding does not seem to be an appropriate approach of design of routing protocols in WSNs The deficiencies of data reporting by flooding can be viewed under three categories [11] namely

Implosion it refers to nodes disseminating data to all of their neighbors regardless of whether the data has already been received by these neighbors It results in wastage of energy across several nodes since the same piece of data arrives at every node via multiple paths

Overlap it is not possible to have a precise deployment whereby sensor could cover disjoint spatial regions So nodes often have overlapping spatial coverage Data dissemination by flooding therefore results in data from overlapping regions to be routed by different alternative sensors and flooded over the network

Resource ignorance flooding does not require the nodes to modify their activities on the basis of energy threshold Consequently a node may run out of energy too soon to perform any meaningful data sensing or reporting For example nodes near the base station could run out of their batteries being over used while forwarding data

from distant nodes This may result in network going dead too soon even when many distant nodes have sufficient energy to carry out their reporting and sensing activities

Gossiping [9] is just a slightly enhanced version of flooding wherein the recipient node sends the data packet to a randomly selected neighbor That neighbor in turn picks another random neighbor to forward the packet to and so on Another usual variation of gossip protocol is that the recipient could rebroadcasts or discard the data with probabilities p and (11048576 p) respectively By controlling the fraction of executions where gossip die out relatively low and keeping the gossip probability low it is possible to save up to 35 message overhead compared to flooding [9] But gossip based protocol exhibit bimodal behavior in the sense that in almost all executions either most of the nodes receive a message or hardly any of them do Furthermore in many applications gossip delay may be unacceptable

22 SPIN and Directed Diffusion

The focus Sensor Protocols for Information via Negotiation (SPIN) [11] was to work around the problems of simple flood based protocol mentioned in the previous section

SPIN was successful in addressing all the problems existing in the flood based protocols It labeled the data with the high level data descriptors called metadata The metadata was used to negotiate between the nodes eliminating the redundant data from being transmitted to the nodes As the metadata was supposed to be the compressed form of the original data so the network congestion could be avoided So the SPIN mainly comprises of three phases

STEP 1 Advertisement of the metadata

STEP 2 Request for the data

STEP 3 Actual data transmission

Fig 22 depicts the stepwise process

Figure 22 SPIN

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

12 Problem statement and the Challenges Involved

The investigation in this thesis is oriented towards the design a multi hop routing protocol for time driven WSN which achieves following objectives

Avoid the formation of the congestion bottleneck around the base station

Handle dynamic changes in topology caused by transmission power adjustments

Handle node failures on existing paths by a novel route repair procedure that leverages ability to adjust node transmission range

Ensure reliability through use of acknowledgements and limited retransmission of lost packets

Improve the network lifetime by considering residual node energy during route discovery

Analyze and compare the new protocol with the existing ones

The objectives listed at items 1 and 5 are achieved by better utilization of the resources and treating the network as a dynamically changing asymmetric network The topology change in the network is mainly determined by adjustment of transmission range from time to time The resulting change topology modifies path from every node to base station The routing protocol is therefore is a mix of both proactive and reactive protocol The tasks of route discovery and route maintenance become more challenging in an asymmetric network as Hello messages can not be used for such purposes in asymmetric links The latency of route discovery also has to be minimized

Balancing distribution of load for routing (forwarding packets) among the nodes shall ensure the uniform and better resource utilization The concept of asymmetric network shall allow the nodes to use appropriate transmission ranges which is more suitable for its longer lifetime depending on its position in the network

Fault tolerance in routing protocols for WSN is a novel idea It has been possible to tolerate node failure simply by eliminating failed nodes through a route maintenance step It involves the adjustment of transmission power at some nodes for establishing connectivity with the base station when the routes are found to be broken due to node failures We experimented with up to 5 node failures

The objective concerning reliability in transmission is achieved by acknowledgments for the data packets The protocol supports at most three retransmissions of the data packets to handle packet loss

The routing metric is chosen by optimizing both latency of routes and residual energies at the nodes The other desirable property of a protocol which is addressed is load sharing The load for routing data packets is distributed over the multiple alternate paths available at a node It ensures overall better utilization of energy resources and effectively extends lifetime of the network

Finally we performed simulation over OMNet++ [1] and compared our results with other notable time driven WSN routing protocols such as LEACH [12] and SPIN [11] Our results show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation

13 Organization of Thesis

The rest of the thesis is organized as followsChapter 2 discusses the related work and lays the background further discussion The important existing protocols and the improvements subsequently proposed upon them are also discussed It also indicates how these solutions differ from the new protocol proposed in this thesis

In Chapter 3 the new protocol is described We focus on robustness network lifetime routing latency to show how the proposed algorithm has potentials to outperform other existing protocols The protocol is discussed in details by dwelling on its features various phases of operation and implementation aspects

Chapter 4 throws light on the actual logic of the proposal by presenting the algorithms An analysis of performance parameters is also presented to provide an insight into its performance The chapter also describes simulation software and the simulation environment The details of the radio characteristics used for the purpose of empirical evaluation

The results duly supported by the relevant plots for performance characteristics and related analysis are presented in Chapter 5

Chapter 6 concludes the thesis with some directions for future work

Chapter 2

Related Work

As explained in chapter 1 the design of routing protocols for WSN is a subject of intense research as both quality and quantity of information delivered to the end-users is very important for the applications centered on WSN It has been observed that different protocols work better in different environmentsapplications The issue of the effective utilization of energy resource has also been addressed in extensively in the literature This chapter deals with related work and the underlying concepts which form the basis of energy aware routing protocols for WNSs

Section 21 explains how simple routing ideas could lead to unnecessary wastage of precious energy resources Section 22 deals with some initial thinking done by researchers on the ways to avoid broadcast storm which appear to be the main reason behind excessive energy wastage in routings ideas based on flooding and gossiping Section 23 describes a number of routing protocols namely LEACH [12] PEGASIS [19] TEEN [20] and APTEEN [21] which try to eliminate redundant data broadcast and conserve the crucial energy resource by aggregation in some way or other Later section 24 which are relevant in context to this thesis The subject of Discussion in section 25 is the quality and reliability of data delivered by the sensors

21 Flooding and Gossiping

The conventional mechanism for relaying data from individual sensors to base station is Flooding or Gossiping [10] The underlying idea of flooding is that each sensor on receiving a data packet broadcasts the same to its every neighbor This packet is further re-broadcast until it arrives at the destination or the maximum number of hops for the packet is reached Quite obviously flooding raises unmanageable broadcast storms as depicted in Fig 21

Figure 21 Broadcast storm

It may be appropriate for networks which experience quick topology changes WSN mainly consists of static sensors nodes So topology remains more or less static for a considerable period of time Of course it is possible to control topology by adjusting transmission power of sensor motes But such power adjustments can only be realized by program controls and hence predictable The other serious problems with broadcast storm are redundancies contentions and collisions which result in wasteful consumption of node energies Since sensor nodes have only limited energy reserves in the form of small batteries they can ill-afford such associated problems arising out of broadcast storms Therefore flooding does not seem to be an appropriate approach of design of routing protocols in WSNs The deficiencies of data reporting by flooding can be viewed under three categories [11] namely

Implosion it refers to nodes disseminating data to all of their neighbors regardless of whether the data has already been received by these neighbors It results in wastage of energy across several nodes since the same piece of data arrives at every node via multiple paths

Overlap it is not possible to have a precise deployment whereby sensor could cover disjoint spatial regions So nodes often have overlapping spatial coverage Data dissemination by flooding therefore results in data from overlapping regions to be routed by different alternative sensors and flooded over the network

Resource ignorance flooding does not require the nodes to modify their activities on the basis of energy threshold Consequently a node may run out of energy too soon to perform any meaningful data sensing or reporting For example nodes near the base station could run out of their batteries being over used while forwarding data

from distant nodes This may result in network going dead too soon even when many distant nodes have sufficient energy to carry out their reporting and sensing activities

Gossiping [9] is just a slightly enhanced version of flooding wherein the recipient node sends the data packet to a randomly selected neighbor That neighbor in turn picks another random neighbor to forward the packet to and so on Another usual variation of gossip protocol is that the recipient could rebroadcasts or discard the data with probabilities p and (11048576 p) respectively By controlling the fraction of executions where gossip die out relatively low and keeping the gossip probability low it is possible to save up to 35 message overhead compared to flooding [9] But gossip based protocol exhibit bimodal behavior in the sense that in almost all executions either most of the nodes receive a message or hardly any of them do Furthermore in many applications gossip delay may be unacceptable

22 SPIN and Directed Diffusion

The focus Sensor Protocols for Information via Negotiation (SPIN) [11] was to work around the problems of simple flood based protocol mentioned in the previous section

SPIN was successful in addressing all the problems existing in the flood based protocols It labeled the data with the high level data descriptors called metadata The metadata was used to negotiate between the nodes eliminating the redundant data from being transmitted to the nodes As the metadata was supposed to be the compressed form of the original data so the network congestion could be avoided So the SPIN mainly comprises of three phases

STEP 1 Advertisement of the metadata

STEP 2 Request for the data

STEP 3 Actual data transmission

Fig 22 depicts the stepwise process

Figure 22 SPIN

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Finally we performed simulation over OMNet++ [1] and compared our results with other notable time driven WSN routing protocols such as LEACH [12] and SPIN [11] Our results show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation

13 Organization of Thesis

The rest of the thesis is organized as followsChapter 2 discusses the related work and lays the background further discussion The important existing protocols and the improvements subsequently proposed upon them are also discussed It also indicates how these solutions differ from the new protocol proposed in this thesis

In Chapter 3 the new protocol is described We focus on robustness network lifetime routing latency to show how the proposed algorithm has potentials to outperform other existing protocols The protocol is discussed in details by dwelling on its features various phases of operation and implementation aspects

Chapter 4 throws light on the actual logic of the proposal by presenting the algorithms An analysis of performance parameters is also presented to provide an insight into its performance The chapter also describes simulation software and the simulation environment The details of the radio characteristics used for the purpose of empirical evaluation

The results duly supported by the relevant plots for performance characteristics and related analysis are presented in Chapter 5

Chapter 6 concludes the thesis with some directions for future work

Chapter 2

Related Work

As explained in chapter 1 the design of routing protocols for WSN is a subject of intense research as both quality and quantity of information delivered to the end-users is very important for the applications centered on WSN It has been observed that different protocols work better in different environmentsapplications The issue of the effective utilization of energy resource has also been addressed in extensively in the literature This chapter deals with related work and the underlying concepts which form the basis of energy aware routing protocols for WNSs

Section 21 explains how simple routing ideas could lead to unnecessary wastage of precious energy resources Section 22 deals with some initial thinking done by researchers on the ways to avoid broadcast storm which appear to be the main reason behind excessive energy wastage in routings ideas based on flooding and gossiping Section 23 describes a number of routing protocols namely LEACH [12] PEGASIS [19] TEEN [20] and APTEEN [21] which try to eliminate redundant data broadcast and conserve the crucial energy resource by aggregation in some way or other Later section 24 which are relevant in context to this thesis The subject of Discussion in section 25 is the quality and reliability of data delivered by the sensors

21 Flooding and Gossiping

The conventional mechanism for relaying data from individual sensors to base station is Flooding or Gossiping [10] The underlying idea of flooding is that each sensor on receiving a data packet broadcasts the same to its every neighbor This packet is further re-broadcast until it arrives at the destination or the maximum number of hops for the packet is reached Quite obviously flooding raises unmanageable broadcast storms as depicted in Fig 21

Figure 21 Broadcast storm

It may be appropriate for networks which experience quick topology changes WSN mainly consists of static sensors nodes So topology remains more or less static for a considerable period of time Of course it is possible to control topology by adjusting transmission power of sensor motes But such power adjustments can only be realized by program controls and hence predictable The other serious problems with broadcast storm are redundancies contentions and collisions which result in wasteful consumption of node energies Since sensor nodes have only limited energy reserves in the form of small batteries they can ill-afford such associated problems arising out of broadcast storms Therefore flooding does not seem to be an appropriate approach of design of routing protocols in WSNs The deficiencies of data reporting by flooding can be viewed under three categories [11] namely

Implosion it refers to nodes disseminating data to all of their neighbors regardless of whether the data has already been received by these neighbors It results in wastage of energy across several nodes since the same piece of data arrives at every node via multiple paths

Overlap it is not possible to have a precise deployment whereby sensor could cover disjoint spatial regions So nodes often have overlapping spatial coverage Data dissemination by flooding therefore results in data from overlapping regions to be routed by different alternative sensors and flooded over the network

Resource ignorance flooding does not require the nodes to modify their activities on the basis of energy threshold Consequently a node may run out of energy too soon to perform any meaningful data sensing or reporting For example nodes near the base station could run out of their batteries being over used while forwarding data

from distant nodes This may result in network going dead too soon even when many distant nodes have sufficient energy to carry out their reporting and sensing activities

Gossiping [9] is just a slightly enhanced version of flooding wherein the recipient node sends the data packet to a randomly selected neighbor That neighbor in turn picks another random neighbor to forward the packet to and so on Another usual variation of gossip protocol is that the recipient could rebroadcasts or discard the data with probabilities p and (11048576 p) respectively By controlling the fraction of executions where gossip die out relatively low and keeping the gossip probability low it is possible to save up to 35 message overhead compared to flooding [9] But gossip based protocol exhibit bimodal behavior in the sense that in almost all executions either most of the nodes receive a message or hardly any of them do Furthermore in many applications gossip delay may be unacceptable

22 SPIN and Directed Diffusion

The focus Sensor Protocols for Information via Negotiation (SPIN) [11] was to work around the problems of simple flood based protocol mentioned in the previous section

SPIN was successful in addressing all the problems existing in the flood based protocols It labeled the data with the high level data descriptors called metadata The metadata was used to negotiate between the nodes eliminating the redundant data from being transmitted to the nodes As the metadata was supposed to be the compressed form of the original data so the network congestion could be avoided So the SPIN mainly comprises of three phases

STEP 1 Advertisement of the metadata

STEP 2 Request for the data

STEP 3 Actual data transmission

Fig 22 depicts the stepwise process

Figure 22 SPIN

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

21 Flooding and Gossiping

The conventional mechanism for relaying data from individual sensors to base station is Flooding or Gossiping [10] The underlying idea of flooding is that each sensor on receiving a data packet broadcasts the same to its every neighbor This packet is further re-broadcast until it arrives at the destination or the maximum number of hops for the packet is reached Quite obviously flooding raises unmanageable broadcast storms as depicted in Fig 21

Figure 21 Broadcast storm

It may be appropriate for networks which experience quick topology changes WSN mainly consists of static sensors nodes So topology remains more or less static for a considerable period of time Of course it is possible to control topology by adjusting transmission power of sensor motes But such power adjustments can only be realized by program controls and hence predictable The other serious problems with broadcast storm are redundancies contentions and collisions which result in wasteful consumption of node energies Since sensor nodes have only limited energy reserves in the form of small batteries they can ill-afford such associated problems arising out of broadcast storms Therefore flooding does not seem to be an appropriate approach of design of routing protocols in WSNs The deficiencies of data reporting by flooding can be viewed under three categories [11] namely

Implosion it refers to nodes disseminating data to all of their neighbors regardless of whether the data has already been received by these neighbors It results in wastage of energy across several nodes since the same piece of data arrives at every node via multiple paths

Overlap it is not possible to have a precise deployment whereby sensor could cover disjoint spatial regions So nodes often have overlapping spatial coverage Data dissemination by flooding therefore results in data from overlapping regions to be routed by different alternative sensors and flooded over the network

Resource ignorance flooding does not require the nodes to modify their activities on the basis of energy threshold Consequently a node may run out of energy too soon to perform any meaningful data sensing or reporting For example nodes near the base station could run out of their batteries being over used while forwarding data

from distant nodes This may result in network going dead too soon even when many distant nodes have sufficient energy to carry out their reporting and sensing activities

Gossiping [9] is just a slightly enhanced version of flooding wherein the recipient node sends the data packet to a randomly selected neighbor That neighbor in turn picks another random neighbor to forward the packet to and so on Another usual variation of gossip protocol is that the recipient could rebroadcasts or discard the data with probabilities p and (11048576 p) respectively By controlling the fraction of executions where gossip die out relatively low and keeping the gossip probability low it is possible to save up to 35 message overhead compared to flooding [9] But gossip based protocol exhibit bimodal behavior in the sense that in almost all executions either most of the nodes receive a message or hardly any of them do Furthermore in many applications gossip delay may be unacceptable

22 SPIN and Directed Diffusion

The focus Sensor Protocols for Information via Negotiation (SPIN) [11] was to work around the problems of simple flood based protocol mentioned in the previous section

SPIN was successful in addressing all the problems existing in the flood based protocols It labeled the data with the high level data descriptors called metadata The metadata was used to negotiate between the nodes eliminating the redundant data from being transmitted to the nodes As the metadata was supposed to be the compressed form of the original data so the network congestion could be avoided So the SPIN mainly comprises of three phases

STEP 1 Advertisement of the metadata

STEP 2 Request for the data

STEP 3 Actual data transmission

Fig 22 depicts the stepwise process

Figure 22 SPIN

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

from distant nodes This may result in network going dead too soon even when many distant nodes have sufficient energy to carry out their reporting and sensing activities

Gossiping [9] is just a slightly enhanced version of flooding wherein the recipient node sends the data packet to a randomly selected neighbor That neighbor in turn picks another random neighbor to forward the packet to and so on Another usual variation of gossip protocol is that the recipient could rebroadcasts or discard the data with probabilities p and (11048576 p) respectively By controlling the fraction of executions where gossip die out relatively low and keeping the gossip probability low it is possible to save up to 35 message overhead compared to flooding [9] But gossip based protocol exhibit bimodal behavior in the sense that in almost all executions either most of the nodes receive a message or hardly any of them do Furthermore in many applications gossip delay may be unacceptable

22 SPIN and Directed Diffusion

The focus Sensor Protocols for Information via Negotiation (SPIN) [11] was to work around the problems of simple flood based protocol mentioned in the previous section

SPIN was successful in addressing all the problems existing in the flood based protocols It labeled the data with the high level data descriptors called metadata The metadata was used to negotiate between the nodes eliminating the redundant data from being transmitted to the nodes As the metadata was supposed to be the compressed form of the original data so the network congestion could be avoided So the SPIN mainly comprises of three phases

STEP 1 Advertisement of the metadata

STEP 2 Request for the data

STEP 3 Actual data transmission

Fig 22 depicts the stepwise process

Figure 22 SPIN

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

There is also an energy efficient version of the SPIN in which a node enters into this three way handshake only if it has sufficient energy to complete it till the end SPIN not only achieved a considerable improvement over flooding but also instrumental to open up the existing gaps in the research on routing protocols It is observed that the data actually traverses along multiple paths but gets finally accepted from just one path and discarded from remaining others SPINrsquos implementation did not talk about the reliability which may be one of the important issues in reliability considering that the wireless networks are highly error-prone

Subsequently a new protocol called Directed Diffusion [14] was proposed for event driven Applications Directed diffusion concentrated on reducing the number of multiple paths along which data traverses It also incorporates some novel features - data-centric disseminationReinforcement based adaptation to the best path in-network data aggregation and caching A central controller called Sink injects its interest in the network by normal flooding with a large update interval Sensors report data if they match with the interest received from the sink node A sensor sends to the interested node through multiple paths The neighboring node establishes a gradient towards each other based on the direction from which they have received interest This way the interested data finds its path to the sink Apart from being unsuitable for continuous data delivery it incurs extra overhead for data matching and interest injection

23 Aggregation Protocols

Considering that the sensors are resource constraint and the existing inherent redundancies which characterize data reporting in SPIN as well as directed diffusion a lot of scope for improvements was there LEACH [12] is another protocol in line to throw some light on the same issues and propose some more improvements The basic idea employed in LEACH is cut down the overhead by reduce volume of data reportage LEACH is an energy efficient aggregation protocol It includes cluster formation and local processing to reduce global communication by collecting the packets of all the nodes in the cluster in one place and aggregating the information contained The randomized rotation of the cluster-heads helps in implementing the load sharing property

The LEACH protocol involves the following steps

STEP 1 Status collection of all nodes by the base station

STEP 2 Selection of the cluster heads by the base station

STEP 3 Choosing of the best cluster head by the node

STEP 4 Data dissemination to the cluster head

STEP 5 Aggregated data dissemination to the base station by the cluster heads

The problem with LEACH is that it considers all the nodes to have accessibility to the base station when it becomes the cluster head or when sending the status message to the base station According to the MICAz [2] specifications the maximum outdoor transmission range which is allowed for a node is 100m The indoor transmission range is even less it is 30m at the maximum That means to provide one hop connectivity the maximum distance of a node

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

from the base station can be at most 100mIf the base station is placed at edge of the network area then the maximum area which can be allowed in LEACH is (_ _ 502=2) m2

In LEACH each node decides its cluster head based on which of the heads is closest to it It means that the location parameter plays an important role here The nodes need to know the position coordinates of all the cluster heads chosen for that round LEACH needs GPS enabled node which requires considerable amount of energy apart from making the sensor motes quite expensive

The original LEACH proposal provisions for 5 of the total nodes to be selected as the cluster head With 100 nodes the number of clusters formed will be 5 Given that the maximum area which can be covered is (_ _ 502=2) m2 then each cluster will cover about 785m2 Therefore even after using so many nodes in such a small area the actual data transferred to the base station is not much If the volume of data to be transferred to the control-center is considerable then either the number of nodes must be increased or the percentage of the total nodes selected as the cluster heads should be increased

The energy dissipation at nodes is proportional to the square of the distance from the destination therefore the node which dies first is always the farthest nodes

LEACH is not an acknowledgement based protocol So the nodes receive no information of data being lost or corrupted on the way to the destination The protocol therefore lacks reliability The wireless networks are full of noisy interferences and LEACH does allow any redundancy in data as the aggregation removes redundancy so non-reliable nature can be considered as one of the major drawbacks

Although the LEACH does not talks directly about asymmetric links nor does introduce the concept it is implicit from the explanation that the nodes adjust their transmission range every time they need to communicate based on their distance from the node with which they want to communicate In a sense the network can be considered to be asymmetric The asymmetric nature of the network here does help in avoiding the wastage of energy and congestion but as the protocol is not implemented in a distributed way the asymmetric characteristic is under emphasized Rather it is considered a network with nodes having adjustable transmission range

Environmental energy harvesting has also been considered for improving the sustainable lifetime of the resource constraint network Numerous harvesting modalities have been suggested including solar vibration biochemical and light A successful demonstration of the solar energy harvesting has been shown in [28] The SOLAR-aware LEACH [28] aims at extending the lifetime of the sensor network by choosing solar-aware nodes as the preferable cluster heads

While the harvesting energy provides the ability to extract energy from the environment it must be efficiently integrated into an embedded system to translate that harvested energy into increased application performance and the system lifetime This shows that harvesting itself is very complex The other drawback being that the solar energy harvesting requires outdoor system deployment thus does not work for the indoor networks

Lindsay and Raghavan [19] proposed Power Efficient Gathering in Sensor Information Systems (PEGASIS) which achieved about 100-300 improvement over LEACH over a

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

range Of percentages of nodes dying out in different network sizes In PEGASIS sensor nodes form a chain to transmit and receive the data Each node transmits or receives data from a neighbor and only one node from the chain is selected to send the data to the base station Each node on receiving the data aggregates it to its own data and then transmits further The aggregated data is eventually sent to the base station The chain construction is done in a greedy way

Figure 23 PEGASIS Fig 23 illustrates that node-0 transmits data to node-1 Node-1 aggregates it to its own data and transmits it to the node-2 Similarly node-4 transmits its data to node-3 Node-3 aggregates it to its own data and transmits it to the node-2 The node to finally aggregates all the received data along with its own and relays it to the base station

Unlike LEACH PEGASIS uses multi hop routing and only one node sends the data to the base station Although PEGASIS shows 100 1048576 300 improvements over LEACH for different network and topologies there are many issues which cannot be neglected when analyzing a protocol There is only one packet which is reported to the base station every second no matter how large a network may be The nodes are supposed to be power adjustable do that if they are eventually selected to the send the final packet to the base station then many must be able to do it in one hop This again brings to us the constraint of maximum power which a node can possess Given that the maximum outdoor range is 100m restricts the area which can be covered by PEGASIS As PEGASIS reduces the redundancy to zero in the network there is expected to be some scheme to ensure reliability which is again absent

Threshold sensitive Energy Efficient sensor Network (TEEN) proposed by Manjeshwar and Agrawal [20] is actually the modified version of LEACH The modification proposed was to convey two attributes to the nodes from the cluster heads namely (i) Hard Threshold and (ii) Soft Threshold Hard threshold was the absolute value of the sensed parameter beyond which the value must be t ransmitted to the cluster head Soft threshold was the value of the sensed parameter beyond which the node must activate its transmitter

As the data was expected to be transmitted less frequently than being sensed so the protocol is much more energy efficient then LEACHES But there are drawbacks which cannot be neglected If the threshold is not reached then data is not communicated to the base station There were no messages to inform the base station that whether the node has gone dead or the data is not crucial enough to be reported Thus the reliability of data reportage becomes an important drawback

Adaptive Periodic Threshold sensitive Energy Efficient sensor Network APTEEN [21] was designed to remove the drawbacks of TEEN APTEEN adds an extra attribute to the packet sent by the cluster heads to the cluster members It not only includes the thresholds but also includes the maximum interval between two packets This modification makes the protocol

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

usable even by time driven networks As the data get sent periodically it is easy for base station to know that the node is dead or has failed to send data

24 Other Related Works

Parameters Flooding Gossiping SPIN

LEACH

PEGASIS TEEN APTEEN AROS

Energy Efficient

No NO YES YES YES YES YES YES

Redundancy YES YES NO NO NO NO NO NOReliability IMPLICI

TIMPLICT NO NO NO NO NO NO

Reported Data

HIGH HIGH HIGH

Medium LOW LOW LOW Medium

Message Complexity

HIGH HIGH LOW

Medium LOW LOW LOW Medium

BS Responsibility

LOW LOW LOW

Medium Medium Medium Medium HIGH

The problem of bottleneck around the base station was previously addressed in [24] It proposed to increase the density of nodes as we move towards the base station So more nodes are available near the base station to share the responsibility of relaying the data packets The proposal recommends certain nodes as relay nodes which act only as forwarding nodes and not as sensors One of the problems faced with such an approach is that if sensor nodes are dropped from a height with possible error in location so having certain nodes in predetermined coordinates is not possible Furthermore interferences also increase with increase in density of the nodes An energy efficient probabilistic scheme has also been proposed in [27]

Table 21 Summary of Some Well known Routing Protocols for Wireless Sensor Networks

The author the view that always using lowest energy path may not be optimal for the lifetime and the long-term connectivity of the network So they propose to use probabilistic forwarding The paper concentrates on the importance of energy but only in a symmetric network The weight age to be given to the residual energy factor is also not clearly justified The paper compares the proposed protocol over existing directed diffusion protocol Not much work has been done on the sensor networks with asymmetric links Asymmetric communication and Routing in Sensor networks (AROS) protocol [22] proposes improvements over the existing LEACH protocol In AROS all nodes are not at one hop from the base station The cluster heads which does not have direct reach ability to the base station uses a path through other cluster heads AROS was compared to LEACH and showed 97 improvement in the lifetime of the network AROS loads base station with most of the responsibility The path discovery from a particular cluster head to the base station is actually considered to be a centralized process to be handled by the base station This way it becomes very difficult to account for any dynamicity in the network ie addition of new nodes and the failure of any existing one are difficult to handle

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The protocol which overloads the base station with high responsibilities is not always suited for the applications where the base station may also run on battery eg a laptop The bridge monitoring application can be considered as one such example

25 Quality and Reliability in Sensor Networks

As the wireless sensor networks have found their applications in many crucial domains like medical defense and navigation system it is important to lay due emphasis on the reliability and quality of the data transferred The traditional protocols paid least attention towards the reliability issue The delivery of the data packets was expected to be positive as there was much redundancy in the network Later the protocols [11 12 and 14] reduced redundancy to conserve energy but the reliability issue was left untouched The middleware quality attributes [26] has always been an area of interest But the researchers were of the view that to achieve these quality attributes there has to be trade-off in performance The major attributes which attracted attention includes Scalability Modifiability Availability (fault tolerance) and Maintainability Currently there is lot of interest on ensuring quality attributes with high performance guarantee

The work in this thesis is motivated by all the properties which are desirable considering the sensor network applications and are found lacking in existing protocols In particular we try to address the following issues

Load distribution among the nodes which are near and far-off from the base station

Dynamic enough to handle the insertion and deletion of the nodes in a distributed way

Route discovery and maintenance also through a distributed algorithm

The redundancy of the data packets is reduced to zero and the protocol is made reliable through the presence of the acknowledgements

Energy is considered as one of the most crucial resources and invested with care

The aim of this work is to propose a design with all the above properties and performs better compared to the other exiting protocols

Chapter 3

Proposed Methodology

Wireless Sensor Networks (WSNs) like mobile ad hoc Networks (MANETs) are self-configuring networks But unlike MANETs they are statically deployed and each WSN has a central authority known as base station Each sensor performs the sensing task independent of others but the routing of the sensed data to the base station needs communication among the nodes Sensor nodes do not communicate except for forwarding data to the base station or helping in disseminating data from the base station Therefore when determining the routes in a WSN single destination shortest paths are needed Sensing requires very little energy as compared to communication

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Typically the major physical layer property which is assumed to be fixed in design of a routing protocol for WSNs is the node transmission power The network with nodes having uniformly fixed transmission power is referred to symmetric networks The traditional protocol stack independently models the physical layer and the higher layers The abstraction of one la1yer is added to the other layer to provide power of multi-layering Power control is very important in designing of any protocol for wireless sensor networks as the nodes are battery dependent To control the energy consumption at the nodes the transmission range could be varied But the variation should be done considering the fact that the nodes have at least one path to the base station and at the same time do not increase their respective transmission ranges so as to give rise to a densely connected graph The configuration of the physical layer properties at the startup time needs hardware instructions The transmission range is adjusted thereby to get the best results at the network layer

When working with asymmetric networks the physical layer properties come into picture very often as transmission range is required to be adjusted from time to time The current chapter of this thesis deals with a proposal for a new routing protocol based on the above principle of cross layer optimization [18] Extensive simulations were carried out to compare the proposed protocol with some existing protocols The simulated protocol includes SPIN [11] LEACH [12] and the two versions of symmetric protocols The symmetric protocol versions are based on DSDV [25] so minimum hop count has been used as routing metric

The rest of the chapter organized as follows Section 31 gives a brief statement of the problems discussed in the chapter Section 33 presents the various technicalities of the newly proposed protocol describing each phase and its features and the reason for the same In the end section 34 concludes the chapter highlighting the properties of the proposed design

31 Energy-aware Protocols

The research on energy-aware protocols for WSN has so far focused on two major strategies to optimize energy usage viz (i) reducing transmission of redundant data and (ii) confining communication to local sub-regions (clusters) by partitioning areas of sensing environment Two well known protocols namely SPIN and LEACH which we chose to simulate are the representative cases of the above two strategies We also simulated two versions of symmetric protocols The characteristics of symmetric protocol versions are as follows

The basic protocol is a modification over Destination Sequenced Distance Vector (DSDV) [25] protocol for ad hoc networks The routing metric used is minimum hop count The routing in the first version does not have any energy optimization criterion

The second version attempts to make the first version energy efficient by introducing residual energy consideration

The newly proposed asymmetric routing protocol has been simulated with the purpose of comparing it with the existing work We looked at the following three versions of the proposed protocol

The first version is just a simple (no energy consideration) routing protocol that works for asymmetric network It uses minimum hop count as the routing metric

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The second version proposes to make the above protocol energy efficient by introducing the consideration of residual energy while relaying the data

The third version adds an aggregation protocol to combine residual energy consideration with the idea of confining communication among nodes locally

32 Routing Protocols for Symmetric Networks

Since WSNs are resource constraint the most important thing to be kept in view when designing a routing protocol for WSN is that it should be (i) resource aware and (ii) should not be too complex to implement The symmetric routing protocol has been designed with these two constraints in mind The underlying idea is to avoid flooding or injecting redundant data packets while the data is being transmitted a route discovery is initiated or a route maintenance event occurs

321 Multihop Minimum Hopcount Protocol

Our first attempt is to find the shortest path between the nodes and the base station and to be able to maintain it in a distributed way We have employed acknowledgements for the data packets to ensure data transmission reliable

Route Discovery

Normally route discovery in symmetric networks is performed in top-down fashion The base station initiates a route discovery process by sending a broadcast message to all the neighbor nodes Each node caches a routing table which contains the preferred neighbor and the hop count of the path Fig 31 illustrates the fields of routing table The receiving node updates its routing table and rebroadcasts the request packet including its own details downstream towards destination The updates are performed only if the hops reported by the route information packets are less than the hop count of the path cached

Figure 31 Routing Table

The route discovery process is a one time task which is performed at the network startup

Route Maintenance

Symmetric network uses periodic Hello packets to maintain network topology A Hello packet tells a node about the status its neighbors These packets contain routing table information of the neighbors When relaying any data packet along the preferred neighbor the node checks whether it has received Hello message from that neighbor or not If no hello packet is received from that neighbor for a long time then that neighbor is considered to be dead or out of service for the time being

If the frequency of these packets is low then it takes long time to determine a non-functioning route Consequently all the packets routed through existing routes may either be

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

dropped or have to be retransmitted Therefore it necessitates frequency of Hello packets sending to be substantial in the case of time-driven networks

Reliability

The wireless medium is not a reliable medium as it experiences interferences and noises from various sources The reliability has to achieve by designing a robust protocol Thus our design is an acknowledgement based protocol The acknowledgement is sent by base station on receiving a data packet The sender node caches the send packet till it receives an acknowledgement A data packet may be retransmitted three times in the case of subsequent failures If failure occurs even after the three successive retransmission then the packet is discarded and the cached neighbor is invalidated The node is then left with no valid path to relay the data packets But as it keeps on receiving the periodic Hello packets from its neighbors it may cache the sender of the next Hello packet as the preferred neighbor Later it can update its cache if any other Hello packet offers a better path in terms of the hop count

Evaluation of the Protocol

Although the above routing protocol perform better than the traditional protocols like flooding gossiping SPIN still there are certain problems In the routing tree the nodes near the base station are loaded with the responsibility of relaying the data packets of the far-off nodes as well Therefore the nodes near the base station exhaust their energy much before the far-off nodes With this even if the far-off nodes have sufficient energy to sense they are unable to send data to the base station because of the absence of a valid routing path As the links are required to be symmetric all nodes need to maintain same transmission power Each node i select the respective minimum level li at which they have at least one path to the base station The transmission range is the maximum of the power level selected by each node ie y = max l1 l2 ln So a node has to maintain higher power level even when it can possibly work well with lower power As already explained above the Hello packets are required to be more frequent than the data packets in case of time-driven networks

322 Energy Efficient Multihop Minimum Hopcount Protocol

This protocol present a modification over the simple protocol mentioned in the previous subsection The routing table is enlarged to cache three preferred neighbors all with the minimum hop count Fig 32 shows the structure of the routing table It contains three fields the new field is introduced for caching the residual energy parameter This version aims at providing load balancing by caching alternate paths and selecting the best out of them for each turn

Energy Consideration

The energy updates in symmetric network are sent along with the periodic Hello packets As such no extra energy updates are required to be sent

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Figure 32 Extended Routing Table

Every node caches three preferred neighbors (all of them lead to the minimum hop count path to the base station) One of them is chosen to send the data to the base station The choosing of the path has to be done in a way to utilize the resources best and balance the load If the same neighbor is used often then it will exhaust its energy before other node does and the area being covered by that node could be left unsensed Furthermore it might even leave certain other nodes disconnected So it is advisable to balance the load The best path is selected comparing the residual energy E of the three cached neighbor nodes The node with the maximum E is selected as the neighbor for relaying the data for that turn This accounts for the change in the routing tree every turn and thus exploits the resources best

33 Routing Protocols for Asymmetric Networks

This work as already stated aims at proposing a multihop energy efficient reliable routing protocol for wireless sensor networks The idea is to retain the positive characteristics of the existing protocols and overcome the drawbacks The protocol also targets to be fault tolerant as well as scalable

331 Energy Efficiency and Reliability

Our first target is to eliminate the bottleneck around base station by sharing the responsibilities of data transmission equitably among the near as well as far-off nodes It involves adjustment of per node transmission power The periodic variation in the transmission power changes the network topology and gives rise to existence of asymmetric links in the network It maintains multiple optimal paths for communication so that an alternate path is available if one being used fails

Maintaining Discrete Levels

Our protocol maintains discrete levels of the transmission range The transmission ranges changes in steps Fig 33 depicts the changes in network topology when the transmission range of the nodes changes in step of 5 It may be observed that by adjusting node transmission power the number of nodes at one hop distance from the base station changes

Initially as Fig 33 (a) shows the transmission range is 10m and the there are 2 nodes namely node-2 and node-3 at distance one hop to the base station So the data transmission load from all the nodes gets distributed among these two nodes

Fig 33 (b) depicts the change in topology when the transmission range is 15m There are four nodes viz node-2 node-3 node-4 and node-5 at one hop distance to the base station With this new configuration the load of data transmission can now be distributed among four nodes implying lesser loads on node-2 and node-3

Fig 33 (c) gives the topology when the transmission range is raised 20m Now five nodes viz node-2 node-3 node-4 node-5 node-6 are at one hop distance from the base station As a result the load of data transmission can be distributed among five nodes

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Figure 33 Topology variation a) Transmission Range=10m b) Transmission Range=15m c) Transmission Range=20m

Thus by regulating between the three transmissions ranges periodically we can avoid the bottleneck around the base station

The lowest level transmission range of different nodes may be different It depends on the position of a node in the network so that the node remains connected with the rest of the network where connectivity means existence of a route to the base station At a particular time instant two nodes can be at a different level of transmission range The maximum level of transmission range is fixed based on two conditions

It should be large enough so that no node should be left disconnected with this Transmission range when considering a symmetric network

It should not be very large so as to make the graph very densely connected and lead to wastage of energy resources

The transmission range of the node increases in levels But there is no point in increasing the transmission range of a node further once it has reached the level when it is at one hop distance away from the base station This way the maximum transmission range level attained by different nodes is different

Transmission Range Period

The period for which a particular level should be maintained is same for all nodes This period is decided based on number of nodes in the network the application and the area to be covered The reason is sufficient time should be allowed for the newly formed topology to converge The time required to converge the topology depends on the RTT (round trip time) since the route discovery takes time proportional to RTT (when done in bottom up fashion) The round trip time is different for different nodes It depends on the number of hops (counted on the basis of initial transmission range) between the node and the base station As the number of

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

hops increases the round trip time increases The period of change of transmission range is required to be large enough to allow the topology convergence

Suppose the time taken to cover one hop distance is x seconds For N nodes the worst topology possible is linear topology and so in worst case the hop count of a node to reach the destination (base station) is N In this scenario the round trip time to discover the route is 2N x seconds Therefore to provide enough time to a topology to converge and use the discovered path the transmission range period _ should be greater than 2N x seconds ie _ gt 2N x

Route Discovery

The periodic changes in transmission ranges of nodes create asymmetric networks The process of route discovery in such networks is different from that in symmetric networks In an asymmetric network a path which may be valid from the base station to a node may not be valid when reversed

The reason is some node-A on the one way may be in range of other node-B but node-B might not be in the range of node-A Fig 34 illustrates an example where there is a valid path 1-2-5-9 from base station to node-9 but this path is not valid in the reverse direction In other words node-9 can not reach 1 via the reverse path 9-5-2-1 The data packets from node-9 can be transmitted along the path 9-4-2-1 Such a path can be discovered only by bottom-up approach and not in top-down manner The node which needs to discover the route initiates a route request which is broadcast to all the neighbor nodes A node receiving the route request packet checks for the staleness of the packet and on finding it fresh rebroadcasts it further till the request reaches the base station Assuming that the base station is a powerful node (connected to stable power supply) so it can reach all the nodes through one hop The base station sends back the reply which includes the preferred neighbor of the requesting node The base station avoids sending the complete path to the requesting node It has three advantages

Figure 34 Asymmetric Network

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The sending complete path would unnecessarily increase the size of the reply packet If the number of nodes is N assuming the worst case topology (linear topology) the maximum number of hop-counts to reach the destination (base Station) can be as large as N So the size of the reply packet may increase by a factor of N

The requesting node caches only the preferred neighborrsquos id If the complete path was to be cached then memory (an important resource) would be wasted for no fruitful reasons

If complete path were cached then the nodes caching the three paths to the base station has an option of using one out of three alternate paths available to a node On the other hand by caching preferred neighbors potentially more alternate paths can be stored In Fig 35 the number in curly braces is the cached ids of the preferred neighbors and arrows represent the connectivity Consider for example node-11 if the complete paths were cached then the number of alternate paths available is 3 ie 11-5-2-1 11-6-3-1 11-7-3-1 Whereas if only the neighbors are cached then the available number of alternates increase to 5 namely 11-5-2-1 11-5-3-1 11-6-2-1 11-6-3-1 11-7-3-1 Actually if all the three neighbors lead to three valid paths each then the potential number of number of alternate paths could be as many as 3k where k is the number of hops from node to the base station

To avoid the same request packet from being rebroadcast by the same node route request packets are numbered This sequence number is of the form lt node_id request_no gt If node receives a request packet with a repeated sequence number then that packet is discarded This helps in avoiding the loop formation in the paths also removes unnecessary congestion in the network

Figure 35 Wireless Sensor Network

Acknowledgement Based Protocol

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The protocol retains the reliable nature of symmetric protocols by inheriting their acknowledgement characteristic The base station sends the acknowledgements for the received data packets Like the previous case there can be at most three retransmissions in the case of delivery failures If the failure occurs for the fourth time then the cached neighbor is declared as invalid After invalidating the neighbor if no cached neighbor is available then reactive route discovery takes place

Route Maintenance

In asymmetric networks hello packets can serve no purpose Suppose node-A and node-B share an asymmetric link Even it a node A has direct access to another node B B may not have direct access A A could cache B as its neighbor for routing the data packets to the destination Since the hello packets from B can not reach A the use of these packets does not work for maintaining routes in asymmetric networks

Consequently discovered routes in asymmetric network are maintained in reactive manner The initial route discovery is done in a proactive way at the start of each transmission period Our design combines the advantages of both proactive and reactive protocols A proactive protocol has low latency but discovers and rediscovers routes which are not required whereas a reactive protocol discovers only the required routes but has higher latency In the proposed approach the proactive discovery of the routes is done when the node changes its transmission range (change in transmission range is periodic) The maintenance of routes is done in reactive way In case the acknowledgement is not received for any data packet even after three retransmissions the path invalidated and a reactive route discovery takes place

332 Energy-Aware Routing

The second version of our proposed protocol is designed to uncover the importance of residual node energy in data transmission It provides an integrated approach to cache the residual energy of the preferred neighbors and choose the best alternate available for relaying the data packets

Residual Energy Consideration

In the protocol for symmetric networks the energy updates were provided by the Hello packets However in the light of the discussions in section 331 Hello packets cannot help Therefore each node sends a periodic energy update to the neighbors informing the maximum level of the transmission power with which it can operate If the sender of the Energy Update packet is cached as the preferred neighbor at the receiver the recipient also stores the updated residual energy value Since energy update packets have no other purpose like route maintenance they can be sent less frequent than route update packets (Hello packets) in symmetric networks

333 Role of Aggregation Protocol

We also integrated an aggregation protocol for data transmission in the proposed energy aware routing protocol the asymmetric network There are many applications where the precise location of data is not important In other words the spatiality of data can be extended to include data from any point within a sub-region So it we may aggregate data from the cluster members by either finding the mean median or mode then transmit a single value to

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

base station It helps to eliminate spatial redundancy in the sensor data An example application can be temperature monitoring in an area

Cluster Formation

Typically the cluster formation all the clustering protocols is performed by the base station But our design this unwanted overhead on the base station has been eliminated by the distributed formation of the clusters

Fig 36 explains the clustering process The nodes sharing the same preferred neighbor for a particular round forms a cluster For example as node-2 3 and 4 all have node-1 as their preferred neighbor so these nodes belong to the same cluster Similarly node-5 and 6 cache node-2 as their preferred neighbor implying that 5 6 form a cluster In all seven clusters can be identified in Fig 36 These are (234) (56) (78) (910) (1112) (1314)

Figure 36 Aggregation in Asymmetric Network

34 Evaluation of the Proposed Protocol

Before carrying out empirical evaluation of the proposed protocol through simulation it is instructive to identify the basic parameters on which an evaluation should be based and to qualitatively argue out the reasons behind expecting improvements compared to existing routing protocols for WSNs In this section we discuss about these parameters

341 Improved Lifetime

The proposed protocol eliminates bottleneck around the base station by periodic change of topology through per node power adjustment The protocol is resource-aware as the preferred path is selected on the basis of the maximum residual energy Therefore the proposed protocol when compared with the existing protocols should results in better network lifetime

342 Resource-awareness

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The design is resource-aware as is chooses the path with maximum residual energy from among the existing alternatives for relaying the data packets to the base station

343 Reduced Message Complexity and Zero Redundancy

There exists no redundancy in transmission of the data packets The redundancy in the route request packets has been avoided by using the request ids in the respective packets

Every node caches at most three neighbors to be used as the preferred next hops for relaying the data Once three neighbors have been discovered further discovery is considered redundant The base station uses a REQ_CAN packet which contains the id of the requesting node and the sequence number of the request It is broadcast by the base station once three neighbors for any node are discovered Any node on receiving this packet discards the route request quoted by the REQ_CAN packet Thus this packet is used as a death certificate for the route request packets from a particular node and a specific sequence number

344 Reliability

As the redundancy is reduced to the minimum level the overhead due to reliability is required to be considered There are acknowledgements for the data packets The protocol design supports at max three retransmissions of a data packet There is trade-off concerning reliability versus redundancy But reliability in a way helps to eliminate requirement for redundancy

345 Scalability

The protocol puts no constraint of the deployment area New nodes can be injected to the network A new node performs a reactive route discovery for a path to the base station It can be used as a forwarding node only after the next proactive discovery takes place or if there is any reactive discovery before the next proactive schedule

346 Fault Tolerance

The protocol is dynamic enough to handle the node failures If a node fails the packets relayed along that path will result in no acknowledgements so the paths will be invalidated after three retransmissions If on invalidating the path a node is left with no valid path it reactively starts a new route discovery

347 Load Balancing

Each node maintains three alternate preferred neighbors to be used for relaying the data to the base station As the best is chosen based on the maximum residual energy the load is balanced among the three neighbors

348 Requirements for Specialized Support Parameters

Some routing protocols like LAR [15] or GAF [29] make use of position vectors of the nodes to eliminate flooding during route discovery However our proposed protocol does not require any such additional support The protocol works independent of the location of the

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

individual nodes Therefore suited both for indoor and outdoor networks The routing algorithm is simple to implement

Chapter 4

Implementation Details

Considering applications for which WSN are usually deployed some of the routine maintenance requirements like changing battery could be difficult to perform So in chapter 3 we discussed about various routing protocols for WSN with approaches to make these resource aware The underlying motivation is to extend the lifetime of WSN as much as possible Our approach was to use cross layer optimization to make the node aware of their own energy level and operate cooperatively to transmit data in way that per-node energy dissipation is more or less uniform The purpose of this chapter is to find out if our approach is feasible in terms of real implementation So in this chapter we present the nuts and bolts for implementing the proposed routing protocol for asymmetric WSNs In chapter 3 we presented an overview of the proposed protocol outlining the ideas behind function of the proposed protocol in terms of route discovery route maintenance reliability of data transmission etc In this chapter our aim is to examine how these ideas can indeed be realized algorithmically and programmed for simulation using standard simulation tools such OMNeT++ TOSSIM ns-2 etc We also discuss about the reasons for our choice of OMNeT++ for simulation of the protocol

The chapter is organized as follows Section 41 presents the algorithms specifying the actions of individual sensor nodes on receiving a particular packet or when a timeout occurs We emphasize on the fact that the protocol is as easy to implement as to understand Section 42 deals with the analysis of the various important parameters which might help us in judging the performance of the protocol It also describes simulation environment used for evaluating the performance of the proposed protocol Section 43 talks about the network configuration used It specifies radio characteristics of sensor nodes and also describes other parameters in some details Section 44 concentrates on the simulation tool used for the purpose of creating the required network prototype

41 Algorithms

We have two classes of entities in any routing protocol for WSN namely (i) ordinary sensor nodes and (ii) a base station Broadly the task of every sensor node is identical Each node senses some data and transmits the sensed data to base station The job of the base station mainly passive except for acknowledging the receipt of the sensor data and assisting sensor nodes to find and maintain routes to it So we can specify the protocol in terms of the actions at these two entities

Lets first examine the tasks performed by a sensor node It has to

1 Participate in route discovery

2 Take various actions on certain timeout events The timeout actions are as follow

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

(a) Acknowledgement timeout Since the proposed protocol is an acknowledgement protocol the retransmission of a data item should be initiated if an acknowledgement is not received from the base station within a finite time interval

(b) Route request timeout The base station is expected to send the reply once it receives the route request packet from a sensor node The initiating sensor node would try to increase its transmission power if possible to establish a route in case there is no reply within a finite interval

(c) Transmission period timeout Though increase of transmission range could help to eliminate congestion near the base station there are occasions when a route becomes invalid due to low energy levels at intermediate nodes on the path to base station In this situation a reactive discovery may be necessitated to establish a new path

3 Participate in aggregation protocol for reducing the volume of application specific data transmission requirements

Algorithm 41 Algorithm executed at sensor node on receiving a packet ON Receiving Route Request Packet by sensor node if RouteRequest Packet thenif ReqNotCancelled AND noloop thenRebroadcast the Route RequestelseDiscard the Route Request Data Packet Received by a sensor node if Data Packet thenif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor ON Receiving Route Reply Packet if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborelseif hops = hop count thenif no duplicates thenCache new neighborelseDiscard the ReplyelseDiscard the Replyif buffer is not empty thenTransmit the data packets with the interval of 1 sec eachStart the data packet timer

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

On Receiving Request cancel packet and Acknowledgment packet if Request Cancel Packet thencache the node id and the request id included in the request cancel packetif Acknowledgment Packet thenDelete the packet for which acknowledgment has been received On Receiving Energy Update Packet if Energy Update Packet thenif source node is cached as a neighbor thenupdate the cached neighbors Residual Energy E

Algorithm 42 Algorithm executed at sensor node on Timeout Acknowledgment Timeout if Acknowledgment Timeout thenif Retransmission_Count lt 3 thenIncrement Retransmission_CountRetransmit packet along the same neighborelseDiscard the data packetInvalidate the neighbor used for the transmissionif neighbor cache is empty thenInitiate Route RequestStart the route request timer Route Request Timeout if Route Request Timeout thenif Transmission_Range == Min_Transmission_Range thenIncrement the transmission power T by the step size sSet the Min_Transmission_Range to Current_Transmission_RangeelseIncrement the transmission power T by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInitiate the route requestStart the route request timer Transmission Period Timeout if Transmission Period Timeout thenif hop_count == 1 thenSet Transmission_Range to Min_Transmission_RangeelseIncrement the transmission power by the step size sif T gt Max_Transmission_Range thenSet Transmission_Range to Min_Transmission_RangeInvalidate all the neighborsStart the route request timer

Algorithms 41 and 42 present the actions of sensor node for first two tasks listed above The base station is primarily passive It reacts when it either receives a route request from a sensor node or it receives a sensed data packet from sensor node On receiving a sensed data item it needs to send an acknowledgement to the sender In case of route discovery it needs to send

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

id of preferred neighbor to the requesting node Algorithm 43 describes the actions of the base station

Algorithm 43 Algorithm executed at Base Station on receiving a packet ON Receiving Route Request Packet by base station if RouteRequest Packet thenif hop_count gt hops thenInvalidate all the replies sendSet the reply_count to ZeroCache the new reply and the new hop_countSend the replyIncrement the reply_countelseif hop_count = hops thenif reply_count lt 3 AND no duplicates thenCache the ReplySend the replyIncrement the reply_countif reply_count = 3 thenSend the REQ_CAN messageelseDiscard the messageelseDiscard the Reply Data Packet received by Base Station if Data Packet thenSend the AckThe remaining part of the protocol is concerned with aggregation protocol Though aggregation protocol is executed by a sensor node the node may have to play dual role of sensing its own data and collecting sensed data from its neighbors Algorithm 44 described the actions of a sensor node when it executes aggregation

Algorithm 44 Algorithm for the Aggregation ProtocolInitialize data_counter neighbor_list to ZeroPeriodically send data packet (DATA_PKT) to the preferred neighborSend no data signal (NO_DATA_PKT) to the other cached neighbors ON Receiving Route Reply Packet by Sensor Node if RouteReply Packet thenif hops lt hop_count thenInvalidate the cached NeighborsSet hop_count to the new valueCache the new neighborSend the Neighbor information packet (NEIGH_INFO) to the cached neighborelseif hops = hop_count thenif no duplicates thenCache new neighborSend the information to the cached neighbor

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

elseDiscard the ReplyelseDiscard the Reply Data Packet Received by a sensor node if Data Packet thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighborelseCollect the data packet ON Receiving No data signal if NO_DATA_PKT thenDecrement the data_counterif data_counter == 0 thenAggregate the collected data packetsif no neighbor node available thenInitiate route requestStart the route request timerPut the data packet in the bufferelseSelect the neighbor with the maximum Residual Energy EForward the data packet to the selected neighbor

ON Receiving Neighbor Information Packet by Sensor Node if NEIGH_INFO Packet thenIncrement the neighbor_listInitialize data_counter to neighbor_list TIMEOUT for data packets if data packet timeout thenif data_counter lt nei ghbour_l ist thenAggregate the collected dataSend the data to the preferred neighborAlgorithmAlgorithm for the Aggregation Protocol Contd

42 Analysis

The protocol analysis involves the time to converge the topology total buffer required percentage of packet loss and the delay parameter

The time to converge the topology is the maximum RTT of the network

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The maximum buffer space requirement is 5x y where x is the RTT and y is packets send per second The protocol involves maintenance of two buffers

1 Buffer to cache the packets send but waiting for acknowledgments The space requirement for this buffer is computed as follows Let RTT be x seconds packets sent per second be y So packets sent per RTT = x y Since there can be at most 3 re-transmissions of a packet before it is discarded the buffer size to cache waiting packets is 3x y + x y = 4x y 2 Buffer to cache the packets which are delayed due to non-availability of path and will be transmitted on receiving the route reply Route discovery takes one RTT time (assuming the case when a path is always available) Therefore the maximum local buffer size needed is x y

The worst case loss percentage is 20 We again base our computation on the same parameter values for RTT and packets sent per second Since the packets lost due in 5x time could be x y packets x y5x ie at most 15 th packets are lost per second

The Maximum delay in sending a packet is 3 RTT + k (otherwise the packet will be dropped) where k is a parameter introduced due to congestion

43 Simulation Details

Now let us discuss about the simulation tools we used to carry out the experiments There are two important issues to be settled even before we can discuss about the appropriateness of the choice of simulation tool These are the radio model of individual node and the network configuration

431 Radio Model

For the purpose of comparisons among different protocols it is very important to adhere to specific radio characteristics In this work we assume a simple model where radio dissipates Ee = 50nJbit to run the transmitter or receiver circuitry and Eamp = 100pJ=bi t=m2 for the transmitter amplification

The formulas used are as follows

The energy Required to transmit a data packet of size l bits from a node i to node j is given by

Tij= lEe + lEamp(dij)2 where dij is the distance between the node i and j

The energy required to receive a l bit packet for any node i is given by Ri = Eel

The energy required to aggregate i data packets is given by is given by Ai= Eamp i

432 Network Configuration

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The simulation environment constructs a prototype area of (125X190)m2 The nodes are uniformly randomly distributed over this entire area The number of nodes taken is 80 Although certain experiments have also been done to see the effect of change in the network density over the lifetime

The prototype created strongly adheres to the details provided by the MICAz motes The MICAz motes is a 24 GHz IEEE 802154 complaint Mote module used for enabling low power wireless sensor networks It clearly specifies that the indoor range of for the sensor networks is 20m

As specified in the IEEE 802154 the standard data packet size taken is 128 bytes The simulation assumes the base station to be fixed at the origin and is a powerful node

44 Simulation Tools

WSNs have the potential to become significant subsystems of engineering applications The simulation software helps in accurately modeling the real world scenario The practical testing of the design in wireless sensor networks requires large quantity of hardware and involves high cost Using simulators is an intelligent way to determine whether testing of the proposed methodology can be fruitful or not In summary simulation software provide us a low cost way to test the scalability dynamicity and ease of implementation of the proposed design

There are many simulation tools for WSNs These include Network Simulator-2 (ns-2) [3] OMNeT++ [1] OPNET [5] GloMoSim [31] etc

NS-2 [3] being the most popular among them It uses C++ and OTcl and follows an object oriented approach But this simulator does not scale well So ns-2 is not advisable for simulation of large networks There is also lack of customization

GloMoSim [31] was developed in 1998 for mobile ad hoc networks It is written in Parsec which is an extension of C for parallel programming To run GloMoSim we need to have the latest Parsec compiler So familiarity with Parsec is one of the basic requirements for developing a protocol in GloMoSim Parsec code is used extensively in the GloMoSim kernel Therefore to understand the GloMoSim kernel extensive knowledge of Parsec is needed

OPNET [5] is another discrete event object oriented general purpose network simulator It uses a hierarchical model to define each aspect of the system OPNET suffers from the same object-oriented scalability problems as ns-2 It is not as popular as ns-2 or GloMoSim at least in research being made publicly available and thus does not have the high number of protocols available to those simulators Additionally OPNET is only available in commercial form

TOSSIM [16] was developed to provide a scalable high fidelity simulation of a complete TinyOS (operating system) for sensor network Instead of compiling a TinyOS application for a mote users can compile it into the TOSSIM framework which runs on computer system This allows users to debug test and analyze algorithms in a controlled and repeatable environment The TOSSIM and TinyViz the GUI for TOSSIM simulation capabilities are anyhow constrained to TinyOS based applications (protocols and modules

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

already implemented in TinyOS) More-over they can be seen more as emulators rather than simulators TOSSIMrsquos debugging aid tools are inadequate Therefore only way to detect bugs is to inspect large log files which is tedious and time consuming Simulations can be carried out for hours instead of minutes TOSSIM still has lot undetected bugs as of date

OMNeT++ [1] was designed with a view to provide the researchers with a better discrete event simulation environment Its primary application area is the simulation of communication networks but because of its generic and flexible architecture is successfully used in other areas like the simulation of complex IT systems queuing networks or hardware architectures as well

OMNeT++ allows the design of modular simulation models which can be combined and reused flexibly

It is possible to compose models with any granular hierarchy

The object-oriented approach of OMNeT++ allows the flexible extension of the base classes provided in the simulation kernel of the Simulator

OMNeT++ offers an extensive simulation library that includes support for inputoutput statistics data collection graphical presentation of simulation data random number generators and data structures

OMNeT++ simulation kernel uses C++ which makes it possible to be embedded in larger applications

OMNeT++ models are built with NED and omnetppini and do not use scripts which makes it easier for various simulations to be configured

441 Reasons for Choosing OMNeT++

In view of the preceding discussion on simulation tools the choice of simulation was between ns-2 and OMNeT++ Both are open source tools and available for unrestricted download We chose OMNeT++ OMNeT++ offers following advantages over ns-2

Flexibility OMNeT++ is a flexible and generic simulation framework It is possible to simulate anything that can be mapped to active components that communicate by passing messages Some good example systems include queuing networks multiprocessor systems hardware architectures (routers optical switches file servers etc) or business processes There are model frameworks available for different problem domains some popular ones include INET Fw Mobility Fw OverSim NesCT MACSimulator etc

Ns-2 is basically a (TCPIP) network simulator and it is generally found difficult to simulate things other than packet-switching networks and protocols with it It has highly hard coded concepts about the defined components nodes agents protocols links packet representation and network addresses etc which may be found good to work in traditional ways but makes it very hard if one wants to do things a little differently In summary ns-2 are more appropriate simulation experiments over wire network for which it is originally designed

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Programming Model

Omnet++ is an object-oriented event-driven simulator written in C++ Topology descriptions are either written as text files (NED language) or can be dynamically created at run-time This makes it scalable and robust for testing the design for different configurations There is also a graphical interface (GNED) for creating and editing the topologies which automatically creates the topology file

In ns2 there is mixed-mode OTcl (Object-Tcl) with underlying C++ classes OTcl is also used for creating and configuring networks recording results etc

Hierarchical Models (Reusability)

The OMNeT++ simulation kernel is a class library it can be easily linked to the user defined components (simple modules) dynamically at the execution time No need to modify OMNeT++ sources anywhere This enforces reusability A complex model can be formed from self-contained building blocks (ie simple modules and compound modules) which are reusable in other simulation

In contrast ns-2 tends to be a bit monolithic Implementing a complex protocol as a composition of several independent procedures (that appear as one unit) is not possible in ns-2

Scalability

OMNeT++ can simulate very large scale network topologies The limit is the virtual memory capacity of the computer used

In ns-2 there are scalability problems on simulating large network topologies It has been observed that with the increase in number of nodes the run-time of the network increases exponentially

Experiment Design

In OMNeT++ the parameters required by a simulation experiment are mentioned in the configuration file omnetppini which enforces the concept of separating model from experiments

Whereas both models and experiments are usually interwoven in ns-2 topology parameters model customizations result collection etc usually in the same Tcl script

Although ns-2 seems to be more popular among researcher OMNeT++ is also catching up due to fore mentioned advantages over ns-2 OMNeT++ has been used with success in a wide range of research areas from various network simulations (optical wireless ad-hoc IP etc) to performance analysis of complex software systems (at Siemens) and hardware design (at networking equipment vendor companies)

Chapter 5

Results of Simulation Experiments

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

In this chapter we present the results of experiments carried out for evaluating the performance of the proposed energy efficient routing protocol For an evaluation to be meaningful the performance of the proposed protocol should be compared with the performances of certain well known existing energy aware protocols We base our evaluation primarily by comparing with the performance of two other protocols namely SPIN [11] and LEACH [12] The choice of these two protocols for performance comparison is guided by two important reasons SPINrsquos approach is based on eliminating data redundancies due to implosion and overlaps exhibited by plain flooding On the other hand LEACHrsquos approach is to form random clusters in each round so that the load of data transmission borne by the cluster heads can be spread uniformly over all nodes So LEACH cuts back overheads due communication by localizing most of the communication through data aggregation Since the proposed protocol leverages cross layer optimization for energy efficiency in WSN routing it should independently show performance enhancement over symmetric routing protocols and over both SPIN and LEACH even without data aggregation The results are discussed through plots accompanied by explanation as needed

The simulation environment consists of 80 nodes uniformly randomly distributed over a space of (125X190)m2 Certain experiments have also been done to see the effect of change in the network density over the lifetime

The node attributes chosen to strongly resemble MICAz motes A MICAz mote is a 24GHz IEEE 802154 complaint chip The mote module used for enabling low power wireless interface Since the indoor range of for the sensor networks is 20m-30m the transmission range variations are applied within this range As specified in the IEEE 802154 we have used the standard data packet size of 128b y tes The simulation assumes the base station to be fixed at the origin and is a powerful node

The chapter is organized as follows Section 51 gives the prototype of the node distribution which is used for the purpose of evaluation Section 52 compares the Lifetime of the various protocols with the proposed asymmetric protocol Section 53 analyses the effect various parameters on the lifetime of the network in proposed protocol Section 54 evaluates the introduction of clustering in the proposed protocol comparing the improvement in lifetime number of delivered packets and the area covered by a single cluster

51 Node Placement

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Figure 51 Node Placement

The plot in Fig 51 shows the distribution of the nodes The base station or the control center is placed at the origin ie (00) The position of the other nodes is randomly generated by the simulating software The distribution followed is the random uniform distribution A black dot displays the position of a node the id of the node is written alongside the dot The base station id is 1 and the 80 sensor nodes are allocated ids from 2-81

52 Network Lifetime

As already mentioned one of the major objective of this work has been to improve the lifetime of the sensor networks So we compare the lifetime in the proposed methodology with the existing protocols Typically there are three different measures for the lifetime of a WSN namely

1 The time till the first node exhaust its energy

2 The time till the last node exhaust its energy and

3 The time till half the nodes in the network exhausts their energy In this work The first parameter has been used for the purpose of comparison of different protocols

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Figure 52 Lifetime

The plot in Fig 52 shows the improvement in the lifetime obtained with the proposed methodology over the existing symmetric protocol semantics The five protocols compared in this section are

SPIN The transmission range of the nodes is taken to be 29m as this is the minimum range when all nodes get at least one path to the base station

Symmetric Protocol The transmission range used is same as above

Symmetric protocol with Residual Energy Consideration The transmission ranged used is same as above

Asymmetric Network In this all nodes change their transmission range between a fixed range in steps

Parameter values

ndash Transmission Range 20m ndash 32mndash Transmission Period 12Nsecndash Step Size 3m

Asymmetric protocol with Residual Energy Consideration All the parameters are kept unchanged as above The Energy update packets are broadcast to the neighbors every 5secs

53 Effects of Different Parameters on the Lifetime

In this section the effect of various parameters has been evaluated on the asymmetric protocol with residual energy consideration

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

531 Variation in Number of Nodes

The plot in Fig 53 provides comparison of the lifetime of the network with change in network density

It is observed that with the increase in network density the lifetime of the network decreases This happens because as the density increase the network becomes densely connected The more densely connected network is victim to congestion as wireless is a broadcast medium

Figure 53 Lifetime Comparison Varying Number of nodes

532 Variation in Transmission Range Period

Figure 54 Lifetime Comparison Varying Transmission Range Period

The next plot in Fig 54 compares the lifetime of the network with change in transmission range period As already mentioned in chapter 3 the transmission range period has to be

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

greater than 2N x where N is the number of nodes and x is the time take to cover one hop distance by a node

The transmission range period changes over 4N ndash 20 N

The transmission range varies over 20m ndash 32m with step size of 3m

It is observed that the lifetime increases up to a certain level and then starts decreasing It exhibits that there is one optimal value of the period for a particular network configuration

533 Variation in Node Transmission Power

Figure 55 Lifetime Comparison Varying Range of the Transmission PowerThe plot of Fig 55 compares the lifetime of the network with varying range of transmission power

The period of the transmission range is 12Nsec

The transmission power is varied over the range 20m ndash 30m 20m ndash 32m and 20m ndash 35m with step size of 3m

The lifetime is observed to be maximum for the node ranges between 20m ndash 32m this range is found to provide better balancing of load between the nodes near to the base station and the nodes away from the base station But the lifetime decreases for 20m ndash 35 because 35m is a high enough range to create additional congestion in the network when network becomes very densely connected

534 Variation in Transmission Range Steps

The graph shown in Fig 56 compares the lifetime of the network with varying transmission range step size The specific parameter values for experiment are

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m ndash 32m

The steps in which transmission range is varied 2m 3m 4m and6m

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Figure 56 Lifetime Comparison Varying Transmission Range Step size

It is observed that the maximum lifetime is attained for step size 3m With much larger step size the needed granularity is lost The minimum transmission range is set by the node at the execution time If the step size is large then for some nodes the minimum value set is very large which leads to wastage of energy and increases congestion as well If the step Size is small then it may happen that the two different transmission ranges may lead to the same topology In that case it just leads to doubling of transmission period and consequently leading to decrease in lifetime

535 Effect of Node Failure

The plots in this subsection are used to observe the effect of node failure on the lifetime of the network when the nodes near the base station fail The network configuration used for the purpose of evaluation is as follows

The period of the transmission range 12Nsec

The range between which transmission power is varied 20m - 32m

The steps in which transmission range is varied 6m

The variation in percentage of node failures 1-5

The plot in Fig 57 shows the effect on the lifetime of the network when nodes near the base station fails It was observed that the lifetime in this case decreases by about 6 with 5 increase in the failure percentage This can be explained as follows If the nodes near the

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Figure 57 Effect of node failure on Lifetime

base station fails then the total load of data packet in the network is required to be distributed among reduced number of nodes As such the near nodes exhaust their energy much faster leading to decrease in the lifetime

Figure 58 Effect of node failure on Lifetime

Our second graph in Fig 58 presents the effect on the lifetime of the network when nodes lying in the middle of the region being sensed fail It was observed that the increase in the percentage of the dying nodes in the central region does not exhibit any specific pattern The explanation for this observation is as follow There is a trade-off between the following

1 The total energy drained out of the network because of the packets generated by centrally located nodes and

2 The total energy saved by the far-off nodes when they use those centrally located for routing data packets

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

The plot in Fig 59 shows the effect on the lifetime of the network when nodes far-off from the base station fail If the far-off nodes die then the lifetime shows an improvement of about 7 with 5 nodes going dead It happens because the load due to the route discovery and the data packet of the far-off nodes reduces The nodes near the base station are now less loaded and last longer

Figure 59 Effect of node failure on Lifetime

54 Effect of Aggregation

The aggregation version of the proposed protocol has been compared with the well known aggregation protocol LEACH [12] The lifetime obtained with proposed approach was 150400sec which shows an improvement of more than 400 is observed over LEACH (27099sec) As already stated in chapter 2 LEACH is not suited for such large area But still if we remove the constraint of the transmission range and assume to use nodes with very high transmission ranges that are maximum of approximately 250m the proposed protocol outperforms LEACH

541 Clusters in LEACH

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Figure 510 a) Clusters in LEACH b)Clusters in Asymmetric Network with Residual Energy Consideration

Although as mentioned in the chapter 2 LEACH can work in this configuration with the assumption that the nodes are able to increase their transmission range to around 250m the number of clusters to be formed are taken to be 20 of the number of nodes If the number is higher then the performance goes down further Since the total number of nodes is 80 the highest number of clusters formed at any particular instant of the simulation was 16 The total area covered by each cluster is approximately 1500m2 which is large enough

542 Effects of Aggregation in Proposed Protocol

The graph in Fig 510 b) shows that by introducing aggregation in the proposed asymmetric protocol around 40 clusters are formed at any particular instant of the simulation Therefore the number of packets delivered is much more than in the case LEACH as depicted in Fig 510

a) The total area covered by each cluster in case of the asymmetric network is approximately 600m2 which leads to better localization of communication

Chapter 6

Application of Wireless Sensor Network

61 Structural Health Monitoring

Smart Structures

Sensors embedded into machines and structures enable condition-based maintenance of these assets [32] Typically structures or machines are inspected at regular time intervals and components may be repaired or replaced based on their hours in service rather than on their working conditions This method is expensive if the components are in good working order and in some cases scheduled maintenance will not protect the asset if it was damaged in between the inspection intervals Wireless sensing will allow assets to be inspected when the sensors indicate that there may be a problem reducing the cost of maintenance and preventing catastrophic failure in the event that damage is detected Additionally the use of wireless reduces the initial deployment costs as the cost of installing long cable runs is often prohibitive

In some cases wireless sensing applications demand the elimination of not only lead wires but the elimination of batteries as well due to the inherent nature of the machine structure or materials under test These applications include sensors mounted on continuously rotating parts [33] within concrete and composite materials [34] and within medical implants [35 36]

Industrial Automation

In addition to being expensive lead wires can be constraining especially when moving parts are involved The use of wireless sensors allows for rapid installation of sensing equipment and allows access to locations that would not be practical if cables were attached An example of such an application on a production line is shown in Figure 2261 In this application typically ten or more sensors are used to measure gaps where rubber seals are to

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

be placed Previously the use of wired sensors was too cumbersome to be implemented in a production line environment The use of wireless sensors in this application is enabling allowing a measurement to be made that was not previously practical [37]

Figure 61 Industrial application of wireless sensors

Other applications include energy control systems security wind turbine health monitoring environmental monitoring location-based services for logistics and health care

62 Application Highlight ndash Civil Structure Monitoring

One of the most recent applications of todayrsquos smarter energy-aware sensor networks is structural health monitoring of large civil structures such as the Ben Franklin Bridge (Figure 2262) which spans the Delaware River linking Philadelphia and Camden NJ [38 39] The bridge carries automobile train and pedestrian traffic Bridge officials wanted to monitor the strains on the structure as high-speed commuter trains crossed over the bridge

A star network of ten strain sensors were deployed on the tracks of the commuter rail train The wireless sensing nodes were packaged in environmentally sealed NEMA rated enclosures The strain gauges were also suitably sealed from the environment and were spot welded to the surface of the bridge steel support structure Transmission range of the sensors on this star network was approximately 100 meters

The sensors operate in a low-power sampling mode where they check for presence of a train by sampling the strain sensors at a low sampling rate of approximately 6 Hz When a train is present the strain increases on the rail which is detected by the sensors Once detected the system starts sampling at a much higher sample rate The strain wave form is logged into local Flash memory on the wireless sensor nodes Periodically the waveforms are downloaded from the wireless sensors to the base station The base station has a cell phone attached to it which allows for the collected data to be transferred via the cell network to the engineersrsquo office for data analysis

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

Figure 62 Ben Franklin Bridge

This low-power event-driven data collection method reduces the power required for continuous operation from 30 mA if the sensors were on all the time to less than 1 mA continuous This enables a lithium battery to provide more than a year of continuous operation

Resolution of the collected strain data was typically less then 1 micro strains A typical waveform downloaded from the node is shown in Figure 2263 Other performance specifications for these wireless strain sensing nodes have been provided in an earlier work [40]

Figure 63 Bridge strain data

Chapter 7

Limitations

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

A number of security issues exist in WSN and need to be analyzed in detail in order to design appropriate security mechanisms and overcome security problems that arise in the sensor environment However designing new security protocols and mechanisms is constrained by the capabilities of the sensor nodes This section discusses the limitations that complicate the security design and deployment in sensor networks It is important to understand the constrained capabilities of sensor nodes if you wish to develop proper security that balances demanding security performance against sensor nodes limitations [41]

71 Hostile Environment

Sensor networks can be deployed in remote or hostile environments such as battlefields In these cases the nodes cannot be protected from physical attacks since anyone could have access to the location where they are deployed An adversary could capture a sensor node or even introduce his own malicious nodes inside the network If the latter is the case the adversaryrsquos aim is to trick the network into accepting his nodes as legitimates

In either case the adversary can compromise sensitive information which is either stored on the compromised nodes or is forwarded through the adversaryrsquos nodes to the next hop the sensitive information that is collected could be used for illegal purposes The challenge here for researchers and developers is to design resilient security protocols and solutions offering security even if a subset of sensor nodes are compromised It is important to ensure that if a node is compromised sensitive information stored on the node cannot be taken off with ease

72 Random topology

Most of the time deploying a sensor network in a hostile environment is done by random distribution ie from an airplane Therefore it is difficult to know the topology of sensor networks a priori In these situations it is hard to store various encryption keys on nodes in order to establish encryption among a group of neighbors since the neighborhood cannot be known a priori

The challenge is to design key agreement protocols that do not require certain nodes to be neighbors of some other nodes and also do not require encryption keys to be stored on sensors before deployment Appropriate key distribution algorithms must be designed along a flexible WSN architecture to securely provide encryption keys in real time

73 Power restrictions

The power restrictions of sensor nodes are raised due to their small physical size and lack of wires Since the absence of wires results in lack of a constant power supply not many power options exist Sensor nodes are typically battery-driven However because a sensor network contains hundreds to thousands of nodes and because often WSN are deployed in remote or hostile environments it is difficult to replace or recharge batteries The power is used for various operations in each node such as running the sensors processing the information gathered and data communication

Keep in mind that communication between sensor nodes consumes most of the available power much more than sensing and computation Power limitations greatly affect security since encryption algorithms introduce a communication overhead between the nodes more messages must be exchanged ie for key management purposes but also messages become larger as authentication initialization and encryption data must be included

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

74 Limited Computational power

In the case of computational power computations are linked with the available amount of power As you may understand since there is a limited amount of power computations are constrained also Although it is acknowledged that sensors are not expected to have the computing power of workstations or even mobile handheld devices researchers and developers are greatly concerned with the issue

As I mentioned in the previous section more power is used for communication than computations Therefore since the power for computations is even more constrained than the total quantity of power complex security solutions are prohibited The limitation of computational power limits the adoption of strong cryptographic algorithms such as the RSA public key algorithm which is computationally expensive

Instead symmetric encryption algorithms are used to secure sensor nodes communication since symmetric encryption doesnrsquot have as demanding computational requirements as asymmetric encryption However with asymmetric encryption features like digital signatures are not supported Therefore another challenge for researchers and developers is to design appropriate algorithms to establish and verify trust among the nodes participating in a communication Furthermore other security solutions must be adopted to cover the weaknesses of symmetric encryption when an adversary compromises a node he could retrieve the shared key used to encrypt the messages and then compromise the entire communication of the sensor network

75 Storage Restrictions

The limited capability for storage affects the storage of cryptographic keys as well According to the encryption scheme used each sensor node may need to know a number of keys for each other node in the network to secure communication and thus store the keys in the nodesrsquo storage space However the large number of sensor nodes requires a lot of memory which may not be provided As I mentioned previously having a single encryption key common to all nodes allows an adversary to compromise the whole network by compromising only a single node The challenge of storage restriction is for researchers to design security protocols in a way that a minimum number of encryption keys must be used to provide adequate protection to the network [41]

Chapter 8

Conclusion and Future Work

81 Conclusion

The work in this thesis is focused towards designing implementation and evaluation of energy efficient routing protocols for wireless sensor networks The elimination of bottleneck around the base station is a matter of major concern The effort was directed towards uniform distribution of data transmission and dissemination load among the nodes across the network We realized that only way to achieve this is to develop a methodology for distributed topology control We studied the specification of MICAz motes and came to conclusion that by per-node transmission power adjustment it is possible to control topology and thus eliminate the bottle- neck around the base station It resulted in increase in the lifetime of the network Through a series of simulation experiments using different network configurations we observed that the approach is justified and result in substantial performance enhancements

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

over existing energy-aware protocols We addressed the following issues through our experiments

Energy efficiency The lifetime of the network increased by about 155 in comparison to the symmetric network

Scalability The protocol is dynamic enough to handle the failures and new insertions

Low latency As the protocol is a mixture of both proactive and reactive protocols the total delay factor is 3RT T + k where k is a parameter introduced due to congestion

Reliability It is an acknowledgment based protocol If acknowledgment is not received for a packet till acknowledgment timeout retransmissions take place

Ease of implementation The implementation details presented clearly shows that the protocol is easy to implement

The proposed protocol when compared to the symmetric multihop protocol was found to attain 155 improvement in the lifetime The aggregation version of the proposed protocol showed an improvement of 400 over the existing aggregation protocol LEACH [12]

82 Future Work

There are several opportunities for future work It might be interesting to study the pattern of the nodes dying due to exhaustion of their energy This pattern might give some insight into the proposal and may actually lead to performance enhancements In fact the distribution pattern of routing load could be important starting point of applicability of cross layer optimization

The study of the percentage of packet loss may also lead to some important conclusions like effect of interferences and noise brought about by range adjustment

Extending the proposed protocol for the real life event driven networks might be fruitful The protocol is expected to perform better for event driven networks with some minor modifications

REFERENCES

1 Documentation and tutorials for omnet++ httpwwwomnetpporg

2 Micaz 24 ghz data sheet httpwwwopenautomationnetpageproductosid22titleMICAz-24-GHz

3 The network simulator ns-2 httpwwwisiedunsnamns

4 Piyush Agrawal R K Ghosh and Sajal K Das Localization of wireless sensor nodes using proximity information In Proceedings of the 16th International Conference on Computer Communications and Networks IEEE ICCCN pages 485ndash490 2007

5 Xinjie Chang Network simulations with opnet Proceedings of the 31st conference on Winter simulation Simulationmdasha bridge to the future 1307ndash314 1999

6 D Culler D Estrin and M Srivastava Overview of sensor networks Computer pages 41ndash49 August 2004

7 B Deb S Bhatnagar and B Nath Reinform Reliable information forwarding using multiple paths in sensor networks In 28th Annual IEEE International Conference on Local Computer Networks pages 406ndash415 October 2003

8 S Dulaman T Nieberg J Wu and P Havinga Trade-off between traffic overhead and reliability in multi path routing for wireless sensor networks In IEEE Wireless Communication and Networking volume 3 pages 1918ndash1922 March 2003

9 Zygmunt Haas Joseph Y Halpern and Li L Gossip-based ad hoc routing IEEEACM Trans Netw 14(3)479ndash491 2006

10 S Hedetniemi and A Liestman A survey of gossiping and broadcasting in communication networks Networks 18(4)319ndash349 1988

11 W Heinzelman J Kulik and H Balakrishnan Adaptive protocols for information dissemination in wireless sensor networks 1999

12 Wendi Heinzelman Anantha Chandrakasan and Hari Balakrishnan Energy-efficient communication protocols for wireless microsensor networks In Proceedings of the 33rd Annual Hawaii International Conference on System Sciences 2000 pages 10 pp vol2ndash Jan 4-7 2000

13 B Hull K Jamieson and H Routing protocols overview and design issues for self-organized networks In International Conference on Communication Technology (WCC-ICCT) volume 2 pages 1298ndash1303 2000

14 Chalermek Intanagonwiwat Ramesh Govindan and Deborah Estrin Directed diffusion a scalable and robust communication paradigm for sensor networks In MobiCom rsquo00 Proceedings of the 6th annual international conference on Mobile computing and networking pages 56ndash67 New York NY USA 2000 ACM Press

15 Young-Bae Ko and Nitin H Vaidya Location-aided routing (LAR) in mobile ad hoc networks Volume 6 pages 307ndash321 2000

16 Philip Levis Nelson Lee Matt Welsh and David Culler Tossim accurate and scalable simulation of entire tinyos applications In SenSys rsquo03 Proceedings of the 1st international conference on Embedded networked sensor systems pages 126ndash137 New York NY USA 2003 ACM

17 F L Lewis Wireless sensor networks In D J Cook and S K Das editors Smart Environments Technologies Protocols and Applications pages 12ndash13 John Wiley New York USA 2004

18 X Lin N B Shroff and R Srikant A tutorial on cross-layer optimization in wireless networks IEEE Journal on Selected Areas in Communications 24(8)1452ndash1463 2006

19 Stephanie Lindsey and Cauligi S Raghavendra Pegasis Power-efficient gathering in sensor information systems 2002

20 Arati Manjeshwar and Dharma P Agrawal Teen A routing protocol for enhanced efficiency in wireless sensor networks 15th International Parallel and Distributed Processing Symposium (IPDPSrsquo01) 03 2001

21 Arati Manjeshwar and Dharma P Agrawal Apteen A hybrid protocol for efficient routing and comprehensive information retrieval in wireless sensor networks In IPDPS rsquo02 Proceedings of the 16th International Parallel and Distributed Processing Symposium page 48 2002

22 J Neander E Hansen M Nolin and M Bjoumlrkman Asymmetric multi hop communication in large sensor networks In Wireless Pervasive Computing 2006 1st International Symposium on pages 7 ppndash 16-18 Jan 2006

23 J Neha D K Madathil and D P Agrawal Exploiting multi-path routing to achieve service differentiation in sensor networks In 11th IEEE International Conference on Networks (ICON) pages 681ndash686 October 2003

24 Kumar Padmanabh and Rajarshi Roy Bottleneck around base station in wireless sensor network and its solution In Mobile and Ubiquitous Systems - Workshops 2006 3rd Annual International Conference on pages 1ndash5 17-21 July 2006

25 Charles Perkins and Pravin Bhagwat Highly dynamic destination-sequenced distance-vector routing (DSDV) for mobile computers In ACM SIGCOMMrsquo94 Conference on Communications Architectures Protocols and Applications pages 234ndash244 1994

26 Sharmila Ravula Ji Eun Kim Brad Petrus and Christoph Stoermer Quality attributes in wireless sensor networks Third IEEE Workshop on Software Technologies for Future Embedded and Ubiquitous Systems (SEUSrsquo05) 0030ndash32 2005

27 Rahul C Shah and Jan Rabaey Energy aware routing for low energy ad hoc sensor networks In Wireless Communications and Networking Conference 2002 WCNC2002 2002 IEEE volume 1 pages 350ndash355 17-21 Mar 2002

28 T Voigt A Dunkels J Alonso H Ritter and J Schiller Solar-aware clustering in wireless sensor networks In Proceedings of the Ninth International Symposium on Computers and Communications (ISCC04) volume 2 pages 238ndash243 2004

29 Ya Xu John Heidemann and Deborah Estrin Geography-informed energy conservation for ad hoc routing In MobiCom rsquo01 Proceedings of the 7th annual international conference on Mobile computing and networking pages 70ndash84 New York NY USA 2001 ACM

30 C Ying Q Lu andM Shi Routing protocols overview and design issues for self-organized networks In International Conference on Communication Technology (WCC-ICCT) volume 2 pages 1298ndash1303 2000

31 X Zeng R Bagrodia and M Gerla Glomosim A library for parallel simulation of large-scale wireless networks In Workshop on Parallel and Distributed Simulation pages 154-161 1998

32 A Tiwari A Lewis FL Shuzhi S-G ldquoDesign amp Implementation of Wireless Sensor Network for Machine Condition Based Maintenancerdquo Intrsquol Conf Control Automation Robotics amp Vision (ICARV) Kunming China 6ndash9 Dec 2004

33 Arms SA Townsend CP ldquoWireless Strain Measurement Systems ndash Applications amp Solutionsrdquo Proceedings of NSF-ESF Joint Conference on Structural Health Monitoring Strasbourg France Oct 3ndash5 2003

34 Arms SW Townsend CP Hamel MJ ldquoValidation of Remotely Powered and Interrogated Sensing Networks for Composite Cure Monitoringrdquo paper presented at the 8th InternationalConference on Composites Engineering (ICCE8) Tenerife Spain August 7ndash11 2001

35 Townsend CP and Arms SW Hamel MJ ldquoRemotely Powered Multichannel Micropro-cessor-Based Telemetry systems for Smart Implantable Devices and Smart Structuresrdquo SPIErsquos 6th Annual Intrsquol Conference on Smart Structures and Materials Newport Beach CA Mar 1ndash5 1999

36 Morris BA DrsquoLima DD Slamin J Kovacevic N Townsend CP Arms SW Colwell CW e-Knee The Evolution of the Electronic Knee Prosthesis Telemetry Technology Development Supplement to Am Journal of Bone amp Joint Surgery January 2002

37 Kohlstrand KM Danowski C Schmadel I Arms SW ldquoMind The Gap Using Wireless Sensors to Measure Gaps Effi cientlyrdquo Sensors Magazine October 2003

38 Galbreath JH Townsend CP Mundell SW Hamel MJ Esser B Huston D Arms SW (2003) Civil Structure Strain Monitoring with Power-Effi cient High-Speed Wireless SensorNetworks Proceedings International Workshop for Structural Health Monitoring Stanford CA

39 Arms SW Newhard AT Galbreath JH Townsend CP ldquoRemotely Reprogrammable Wireless Sensor Networks for Structural Health Monitoring Applicationsrdquo ICCES International Conference on Computational and Experimental Engineering and Sciences Medeira Portugal July 2004

40 Arms SW Townsend CP Galbreath JH Newhard AT ldquoWireless Strain Sensing Networksrdquo Proceedings 2nd European Workshop on Structural Health Monitoring Munich Germany July 7ndash9 2004

41 Limitations 1 httpwebhostingdevshedcomcaWeb-Hosting-ArticlesWireless-Sensor-Networks-part-2-Limitations

2 httpwebhostingdevshedcomcaWeb-Hosting-ArticlesWireless- Sensor-Networks-part-2-Limitations1

21 Arati Manjeshwar and Dharma P Agrawal Apteen A hybrid protocol for efficient routing and comprehensive information retrieval in wireless sensor networks In IPDPS rsquo02 Proceedings of the 16th International Parallel and Distributed Processing Symposium page 48 2002

22 J Neander E Hansen M Nolin and M Bjoumlrkman Asymmetric multi hop communication in large sensor networks In Wireless Pervasive Computing 2006 1st International Symposium on pages 7 ppndash 16-18 Jan 2006

23 J Neha D K Madathil and D P Agrawal Exploiting multi-path routing to achieve service differentiation in sensor networks In 11th IEEE International Conference on Networks (ICON) pages 681ndash686 October 2003

24 Kumar Padmanabh and Rajarshi Roy Bottleneck around base station in wireless sensor network and its solution In Mobile and Ubiquitous Systems - Workshops 2006 3rd Annual International Conference on pages 1ndash5 17-21 July 2006

25 Charles Perkins and Pravin Bhagwat Highly dynamic destination-sequenced distance-vector routing (DSDV) for mobile computers In ACM SIGCOMMrsquo94 Conference on Communications Architectures Protocols and Applications pages 234ndash244 1994

26 Sharmila Ravula Ji Eun Kim Brad Petrus and Christoph Stoermer Quality attributes in wireless sensor networks Third IEEE Workshop on Software Technologies for Future Embedded and Ubiquitous Systems (SEUSrsquo05) 0030ndash32 2005

27 Rahul C Shah and Jan Rabaey Energy aware routing for low energy ad hoc sensor networks In Wireless Communications and Networking Conference 2002 WCNC2002 2002 IEEE volume 1 pages 350ndash355 17-21 Mar 2002

28 T Voigt A Dunkels J Alonso H Ritter and J Schiller Solar-aware clustering in wireless sensor networks In Proceedings of the Ninth International Symposium on Computers and Communications (ISCC04) volume 2 pages 238ndash243 2004

29 Ya Xu John Heidemann and Deborah Estrin Geography-informed energy conservation for ad hoc routing In MobiCom rsquo01 Proceedings of the 7th annual international conference on Mobile computing and networking pages 70ndash84 New York NY USA 2001 ACM

30 C Ying Q Lu andM Shi Routing protocols overview and design issues for self-organized networks In International Conference on Communication Technology (WCC-ICCT) volume 2 pages 1298ndash1303 2000

31 X Zeng R Bagrodia and M Gerla Glomosim A library for parallel simulation of large-scale wireless networks In Workshop on Parallel and Distributed Simulation pages 154-161 1998

32 A Tiwari A Lewis FL Shuzhi S-G ldquoDesign amp Implementation of Wireless Sensor Network for Machine Condition Based Maintenancerdquo Intrsquol Conf Control Automation Robotics amp Vision (ICARV) Kunming China 6ndash9 Dec 2004

33 Arms SA Townsend CP ldquoWireless Strain Measurement Systems ndash Applications amp Solutionsrdquo Proceedings of NSF-ESF Joint Conference on Structural Health Monitoring Strasbourg France Oct 3ndash5 2003

34 Arms SW Townsend CP Hamel MJ ldquoValidation of Remotely Powered and Interrogated Sensing Networks for Composite Cure Monitoringrdquo paper presented at the 8th InternationalConference on Composites Engineering (ICCE8) Tenerife Spain August 7ndash11 2001

35 Townsend CP and Arms SW Hamel MJ ldquoRemotely Powered Multichannel Micropro-cessor-Based Telemetry systems for Smart Implantable Devices and Smart Structuresrdquo SPIErsquos 6th Annual Intrsquol Conference on Smart Structures and Materials Newport Beach CA Mar 1ndash5 1999

36 Morris BA DrsquoLima DD Slamin J Kovacevic N Townsend CP Arms SW Colwell CW e-Knee The Evolution of the Electronic Knee Prosthesis Telemetry Technology Development Supplement to Am Journal of Bone amp Joint Surgery January 2002

37 Kohlstrand KM Danowski C Schmadel I Arms SW ldquoMind The Gap Using Wireless Sensors to Measure Gaps Effi cientlyrdquo Sensors Magazine October 2003

38 Galbreath JH Townsend CP Mundell SW Hamel MJ Esser B Huston D Arms SW (2003) Civil Structure Strain Monitoring with Power-Effi cient High-Speed Wireless SensorNetworks Proceedings International Workshop for Structural Health Monitoring Stanford CA

39 Arms SW Newhard AT Galbreath JH Townsend CP ldquoRemotely Reprogrammable Wireless Sensor Networks for Structural Health Monitoring Applicationsrdquo ICCES International Conference on Computational and Experimental Engineering and Sciences Medeira Portugal July 2004

40 Arms SW Townsend CP Galbreath JH Newhard AT ldquoWireless Strain Sensing Networksrdquo Proceedings 2nd European Workshop on Structural Health Monitoring Munich Germany July 7ndash9 2004

41 Limitations 1 httpwebhostingdevshedcomcaWeb-Hosting-ArticlesWireless-Sensor-Networks-part-2-Limitations

2 httpwebhostingdevshedcomcaWeb-Hosting-ArticlesWireless- Sensor-Networks-part-2-Limitations1

33 Arms SA Townsend CP ldquoWireless Strain Measurement Systems ndash Applications amp Solutionsrdquo Proceedings of NSF-ESF Joint Conference on Structural Health Monitoring Strasbourg France Oct 3ndash5 2003

34 Arms SW Townsend CP Hamel MJ ldquoValidation of Remotely Powered and Interrogated Sensing Networks for Composite Cure Monitoringrdquo paper presented at the 8th InternationalConference on Composites Engineering (ICCE8) Tenerife Spain August 7ndash11 2001

35 Townsend CP and Arms SW Hamel MJ ldquoRemotely Powered Multichannel Micropro-cessor-Based Telemetry systems for Smart Implantable Devices and Smart Structuresrdquo SPIErsquos 6th Annual Intrsquol Conference on Smart Structures and Materials Newport Beach CA Mar 1ndash5 1999

36 Morris BA DrsquoLima DD Slamin J Kovacevic N Townsend CP Arms SW Colwell CW e-Knee The Evolution of the Electronic Knee Prosthesis Telemetry Technology Development Supplement to Am Journal of Bone amp Joint Surgery January 2002

37 Kohlstrand KM Danowski C Schmadel I Arms SW ldquoMind The Gap Using Wireless Sensors to Measure Gaps Effi cientlyrdquo Sensors Magazine October 2003

38 Galbreath JH Townsend CP Mundell SW Hamel MJ Esser B Huston D Arms SW (2003) Civil Structure Strain Monitoring with Power-Effi cient High-Speed Wireless SensorNetworks Proceedings International Workshop for Structural Health Monitoring Stanford CA

39 Arms SW Newhard AT Galbreath JH Townsend CP ldquoRemotely Reprogrammable Wireless Sensor Networks for Structural Health Monitoring Applicationsrdquo ICCES International Conference on Computational and Experimental Engineering and Sciences Medeira Portugal July 2004

40 Arms SW Townsend CP Galbreath JH Newhard AT ldquoWireless Strain Sensing Networksrdquo Proceedings 2nd European Workshop on Structural Health Monitoring Munich Germany July 7ndash9 2004

41 Limitations 1 httpwebhostingdevshedcomcaWeb-Hosting-ArticlesWireless-Sensor-Networks-part-2-Limitations

2 httpwebhostingdevshedcomcaWeb-Hosting-ArticlesWireless- Sensor-Networks-part-2-Limitations1