Wsn Survey(08)

Embed Size (px)

Citation preview

  • 7/31/2019 Wsn Survey(08)

    1/42

    UNCO

    RREC

    TED

    PR

    OOFWireless sensor network surveyJennifer Yick, Biswanath Mukherjee, Dipak Ghosal *

    Department of Computer Science, University of California, Davis, CA 95616, United States

    a r t i c l e i n f o

    Article history:

    Received 20 March 2007

    Received in revised form 3 April 2008

    Accepted 7 April 2008

    Available online xxxx

    Responsible Editor: E. Ekici

    Keywords:

    Wireless sensor network

    Protocols

    Sensor network services

    Sensor network deployment

    Survey

    a b s t r a c t

    22A wireless sensor network (WSN) has important applications such as remote environmental

    23monitoring and target tracking. This has been enabled by the availability, particularly in24recent years, of sensors that are smaller, cheaper, and intelligent. These sensors are25equipped with wireless interfaces with which they can communicate with one another to26form a network. The design of a WSN depends significantly on the application, and it must27consider factors such as the environment, the applications design objectives, cost, hard-28ware, and system constraints. The goal of our survey is to present a comprehensive review29of the recent literature since the publication of [I.F. Akyildiz, W. Su, Y. Sankarasubramaniam,30E. Cayirci, A survey on sensor networks, IEEE Communications Magazine, 2002]. Following a31top-down approach, we give an overview of several new applications and then review the32literature on various aspects of WSNs. We classify the problems into three different catego-33ries: (1) internal platform and underlying operating system, (2) communication protocol34stack, and (3) network services, provisioning, and deployment. We review the major35development in these three categories and outline new challenges.36 2008 Published by Elsevier B.V.

    37

    1. Introduction

    Wireless sensor networks (WSNs) have gained world-

    wide attention in recent years, particularly with the prolif-

    eration in Micro-Electro-Mechanical Systems (MEMS)

    technology which has facilitated the development of smart

    sensors. These sensors are small, with limited processing

    and computing resources, and they are inexpensive com-

    pared to traditional sensors. These sensor nodes can sense,

    measure, and gather information from the environment

    and, based on some local decision process, they can trans-

    mit the sensed data to the user.Smart sensor nodes are low power devices equipped

    with one or more sensors, a processor, memory, a power

    supply, a radio, and an actuator.1 A variety of mechanical,

    thermal, biological, chemical, optical, and magnetic sensors

    54may be attached to the sensor node to measure properties55of the environment. Since the sensor nodes have limited56memory and are typically deployed in difficult-to-access

    57locations, a radio is implemented for wireless communica-58tion to transfer the data to a base station (e.g., a laptop, a59personal handheld device, or an access point to a fixed infra-60structure). Battery is the main power source in a sensor61node. Secondary power supply that harvests power from62the environment such as solar panels may be added to the63node depending on the appropriateness of the environment64where the sensor will be deployed. Depending on the appli-

    65cation and the type of sensors used, actuators may be incor-66porated in the sensors.

    67A WSN typically has little or no infrastructure. It con-68sists of a number of sensor nodes (few tens to thousands)69working together to monitor a region to obtain data about

    70the environment. There are two types of WSNs: structured

    71and unstructured. An unstructured WSN is one that con-72tains a dense collection of sensor nodes. Sensor nodes73may be deployed in an ad hoc manner2 into the field. Once

    1389-1286/$ - see front matter 2008 Published by Elsevier B.V.doi:10.1016/j.comnet.2008.04.002

    * Corresponding author. Tel.: +1 530 754 9251; fax: +1 530 752 4767.

    E-mail addresses: [email protected] (J. Yick), [email protected].

    edu (B. Mukherjee), [email protected] (D. Ghosal).1 An actuator is an electro-mechanical device that can be used to control

    different components in a system. In a sensor node, actuators can actuate

    different sensing devices, adjust sensor parameters, move the sensor, or

    monitor power in the sensor node.

    2 In ad hoc deployment, sensor nodes may be randomly placed into the

    field.

    Computer Networks xxx (2008) xxxxxx

    Contents lists available at ScienceDirect

    Computer Networks

    j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m n e t

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

    mailto:[email protected]:[email protected]:[email protected].%20edumailto:[email protected].%20edumailto:[email protected]://www.sciencedirect.com/science/journal/13891286http://www.elsevier.com/locate/comnethttp://www.elsevier.com/locate/comnethttp://www.sciencedirect.com/science/journal/13891286mailto:[email protected]:[email protected].%20edumailto:[email protected].%20edumailto:[email protected]
  • 7/31/2019 Wsn Survey(08)

    2/42

    UNCO

    RREC

    TED

    PR

    OOF

    4 deployed, the network is left unattended to perform moni-5 toring and reporting functions. In an unstructured WSN, net-6 work maintenance such as managing connectivity and

    7 detecting failures is difficult since there are so many nodes.8 In a structured WSN, all or some of the sensor nodes are de-9 ployed in a pre-planned manner.3 The advantage of a struc-0 tured network is that fewer nodes can be deployed with1 lower network maintenance and management cost. Fewer2 nodes can be deployed now since nodes are placed at spe-3 cific locations to provide coverage while ad hoc deployment4 can have uncovered regions.

    5 WSNs have great potential for many applications in sce-6 narios such as military target tracking and surveillance7 [2,3], natural disaster relief[4], biomedical health monitor-

    8 ing [5,6], and hazardous environment exploration and9 seismic sensing [7]. In military target tracking and surveil-0 lance, a WSN can assist in intrusion detection and identifi-1 cation. Specific examples include spatially-correlated and2 coordinated troop and tank movements. With natural3 disasters, sensor nodes can sense and detect the environ-

    4 ment to forecast disasters before they occur. In biomedical

    5 applications, surgical implants of sensors can help monitor6 a patients health. For seismic sensing, ad hoc deployment7 of sensors along the volcanic area can detect the develop-

    8 ment of earthquakes and eruptions.9 Unlike traditional networks, a WSN has its own design

    0 and resource constraints. Resource constraints include a

    1 limited amount of energy, short communication range,2 low bandwidth, and limited processing and storage in each3 node. Design constraints are application dependent and are

    4 based on the monitored environment. The environment

    5 plays a key role in determining the size of the network,6 the deployment scheme, and the network topology. The

    7 size of the network varies with the monitored environ-8 ment. For indoor environments, fewer nodes are required9 to form a network in a limited space whereas outdoor envi-

    0 ronments may require more nodes to cover a larger area.

    1 An ad hoc deployment is preferred over pre-planned2 deployment when the environment is inaccessible by hu-3 mans or when the network is composed of hundreds to4 thousands of nodes. Obstructions in the environment can5 also limit communication between nodes, which in turn af-

    6 fects the network connectivity (or topology).

    7 Research in WSNs aims to meet the above constraints8 by introducing new design concepts, creating or improving9 existing protocols, building new applications, and develop-0 ing new algorithms. In this study, we present a top-down1 approach to survey different protocols and algorithms pro-2 posed in recent years. Our work differs from other surveys

    3 as follows:

    4 While our survey is similar to [1], our focus has been to5 survey the more recent literature.6 We address the issues in a WSN both at the individual7 sensor node level as well as a group level.

    8 We survey the current provisioning, management and

    9 control issues in WSNs. These include issues such as

    130localization, coverage, synchronization, network secu-

    131rity, and data aggregation and compression.

    132 We compare and contrast the various types of wireless133sensor networks.134 Finally, we provide a summary of the current sensor135technologies.136

    137The remainder of this paper is organized as follows:

    138Section 2 gives an overview of the key issues in a WSN.139Section 3 compares the different types of sensor networks.140Section 4 discusses several applications of WSNs. Section 5

    141presents issues in operating system support, supporting142standards, storage, and physical testbed. Section 6 summa-143rizes the control and management issues. Section 7 classi-

    144fies and compares the proposed physical layer, data-link145layer, network layer, and transport layer protocols. Section1468 concludes this paper. Appendix A compares the existing147types of WSNs. Appendix B summarizes the sensor tech-148nologies. Appendix C compares sensor applications with149the protocol stack.

    1502. Overview of key issues

    151Current state-of-the-art sensor technology provides a

    152solution to design and develop many types of wireless sen-153sor applications. A summary of existing sensor technolo-154gies is provided in Appendix A. Available sensors in the155market include generic (multi-purpose) nodes and gate-156way (bridge) nodes. A generic (multi-purpose) sensor157nodes task is to take measurements from the monitored

    158environment. It may be equipped with a variety of devices159which can measure various physical attributes such as160light, temperature, humidity, barometric pressure, veloc-161ity, acceleration, acoustics, magnetic field, etc. Gateway162(bridge) nodes gather data from generic sensors and relay163them to the base station. Gateway nodes have higher pro-

    164cessing capability, battery power, and transmission (radio)

    165range. A combination of generic and gateway nodes is typ-166ically deployed to form a WSN.167To enablewireless sensor applications using sensortech-168nologies, the range of tasks can be broadly classified169into three groups as shown in Fig. 1. The first group is the

    170system. Each sensor node is an individual system. In order

    171to support different application software on a sensor sys-172tem, development of new platforms, operating systems,173and storage schemes are needed. The second group is com-174munication protocols, which enable communication be-175tween the application and sensors. They also enable

    176communication between the sensor nodes. The last group

    177is services which are developed to enhance the application178and to improve system performance andnetwork efficiency.179From application requirements and network manage-

    180ment perspectives, it is important that sensor nodes are181capable of self-organizing themselves. That is, the sensor

    182nodes can organize themselves into a network and subse-

    183quently are able to control and manage themselves effi-184ciently. As sensor nodes are limited in power, processing185capacity, and storage, new communication protocols and

    186management services are needed to fulfil these187requirements.

    3

    In pre-planned deployment, sensor nodes are pre-determined to beplaced at fixed locations.

    2 J. Yick et al. / Computer Networks xxx (2008) xxxxxx

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

    http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-
  • 7/31/2019 Wsn Survey(08)

    3/42

    UNCO

    RREC

    TED

    PR

    OOF

    The communication protocol consists of five standard

    protocol layers for packet switching: application layer,

    transport layer, network layer, data-link layer, and physical

    layer. In this survey, we study how protocols at different

    layers address network dynamics and energy efficiency.

    Functions such as localization, coverage, storage, synchro-

    nization, security, and data aggregation and compression

    are explored as sensor network services.

    Implementation of protocols at different layers in the

    protocol stack can significantly affect energy consumption,

    end-to-end delay, and system efficiency. It is important to

    optimize communication and minimize energy usage. Tra-

    ditional networking protocols do not work well in a WSN

    since they are not designed to meet these requirements.

    Hence, new energy-efficient protocols have been proposed

    for all layers of the protocol stack. These protocols employ

    cross-layer optimization by supporting interactions across

    the protocol layers. Specifically, protocol state information

    at a particular layer is shared across all the layers to meet

    the specific requirements of the WSN.

    As sensor nodes operate on limited battery power, en-

    ergy usage is a very important concern in a WSN; and there

    has been significant research focus that revolves around

    harvesting and minimizing energy. When a sensor node

    is depleted of energy, it will die and disconnect from the

    213network which can significantly impact the performance

    214of the application. Sensor network lifetime depends on

    215the number of active nodes and connectivity of the net-216work, so energy must be used efficiently in order to maxi-217mize the network lifetime.218Energy harvesting involves nodes replenishing its en-219ergy from an energy source. Potential energy sources in-

    220clude solar cells [8,9], vibration [10], fuel cells, acoustic

    221noise, and a mobile supplier [11]. In terms of harvesting222energy from the environment [12], solar cell is the current223mature technique that harvest energy from light. There is

    224also work in using a mobile energy supplier such as a robot225to replenish energy. The robots would be responsible in226charging themselves with energy and then delivering en-

    227ergy to the nodes.228Energy conservation in a WSN maximizes network life-229time and is addressed through efficient reliable wireless230communication, intelligent sensor placement to achieve231adequate coverage, security and efficient storage manage-232ment, and through data aggregation and data compression.

    233The above approaches aim to satisfy both the energy con-

    234straint and provide quality of service (QoS)4 for the applica-235tion. For reliable communication, services such as congestion

    236control, active buffer monitoring, acknowledgements, and237packet-loss recovery are necessary to guarantee reliable238packet delivery. Communication strength is dependent on239the placement of sensor nodes. Sparse sensor placement240may result in long-range transmission and higher energy241usage while dense sensor placement may result in short-242range transmission and less energy consumption. Coverage243is interrelated to sensor placement. The total number of sen-244sors in the network and their placement determine the de-245gree of network coverage. Depending on the application, a246higher degree of coverage may be required to increase the247accuracy of the sensed data. In this survey, we review new248protocols and algorithms developed in these areas.

    2493. Types of sensor networks

    250Current WSNs are deployed on land, underground, and251underwater. Depending on the environment, a sensor net-

    252work faces different challenges and constraints. There are253five types of WSNs: terrestrial WSN, underground WSN,254underwater WSN, multi-media WSN, and mobile WSN

    255(see Appendix B).

    256Terrestrial WSNs [1] typically consist of hundreds to

    257thousands of inexpensive wireless sensor nodes deployed258in a given area, either in an ad hoc or in a pre-planned259manner. In ad hoc deployment, sensor nodes can be260dropped from a plane and randomly placed into the target

    261area. In pre-planned deployment, there is grid placement,

    262optimal placement [13], 2-d and 3-d [14,15] placement263models.264In a terrestrial WSN, reliable communication in a dense265environment is very important. Terrestrial sensor nodes266must be able to effectively communicate data back to the

    267base station. While battery power is limited and may not

    ServicesLocalizationCoverageSecurity

    SynchronizationData Aggregation

    Cross-layerOptimization

    CommunicationProtocol

    Transport LayerNetwork Layer

    Data Link Layer

    System

    PlatformOperating System

    SupportPerform Evaluation

    Storage

    Applications

    Sensor Technology

    Fig. 1. Broad classification of various issues in a WSN.

    4

    QoS defines parameters such as end-to-end delay which must beguaranteed to an application/user.

    J. Yick et al. / Computer Networks xxx ( 2008) xxxxxx 3

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    4/42

    UNCO

    RREC

    TED

    PR

    OOF

    8 be rechargeable, terrestrial sensor nodes however can be

    9 equipped with a secondary power source such as solar

    0 cells. In any case, it is important for sensor nodes to1 conserve energy. For a terrestrial WSN, energy can be con-2 served with multi-hop optimal routing, short transmission3 range, in-network data aggregation, eliminating data4 redundancy, minimizing delays, and using low duty-cycle

    5 operations.

    6 Underground WSNs [16,17] consist of a number of sen-7 sor nodes buried underground or in a cave or mine used to8 monitor underground conditions. Additional sink nodes

    9 are located above ground to relay information from the0 sensor nodes to the base station. An underground WSN is1 more expensive than a terrestrial WSN in terms of equip-

    2 ment, deployment, and maintenance. Underground sensor3 nodes are expensive because appropriate equipment parts4 must be selected to ensure reliable communication5 through soil, rocks, water, and other mineral contents.6 The underground environment makes wireless communi-7 cation a challenge due to signal losses and high levels of

    8 attenuation. Unlike terrestrial WSNs, the deployment of

    9 an underground WSN requires careful planning and energy0 and cost considerations. Energy is an important concern in1 underground WSNs. Like terrestrial WSN, underground

    2 sensor nodes are equipped with a limited battery power3 and once deployed into the ground, it is difficult to re-

    4 charge or replace a sensor nodes battery. As before, a

    5 key objective is to conserve energy in order to increase6 the lifetime of network which can be achieved by imple-7 menting efficient communication protocol.

    8 Underwater WSNs [18,19] consist of a number of sensor

    9 nodes and vehicles deployed underwater. As opposite to0 terrestrial WSNs, underwater sensor nodes are more

    1 expensive and fewer sensor nodes are deployed. Autono-2 mous underwater vehicles are used for exploration or3 gathering data from sensor nodes. Compared to a dense

    4 deployment of sensor nodes in a terrestrial WSN, a sparse

    5 deployment of sensor nodes is placed underwater. Typical6 underwater wireless communications are established7 through transmission of acoustic waves. A challenge in8 underwater acoustic communication is the limited band-9 width, long propagation delay, and signal fading issue.

    0 Another challenge is sensor node failure due to environ-

    1 mental conditions. Underwater sensor nodes must be able2 to self-configure and adapt to harsh ocean environment.3 Underwater sensor nodes are equipped with a limited bat-4 tery which cannot be replaced or recharged. The issue of5 energy conservation for underwater WSNs involves devel-6 oping efficient underwater communication and network-

    7 ing techniques.8 Multi-media WSNs [20] have been proposed to enable9 monitoring and tracking of events in the form of multi-0 media such as video, audio, and imaging. Multi-media1 WSNs consist of a number of low cost sensor nodes2 equipped with cameras and microphones. These sensor

    3 nodes interconnect with each other over a wireless con-

    4 nection for data retrieval, process, correlation, and com-5 pression. Multi-media sensor nodes are deployed in a6 pre-planned manner into the environment to guarantee7 coverage. Challenges in multi-media WSN include high8 bandwidth demand, high energy consumption, quality of

    329service (QoS) provisioning, data processing and compress-

    330ing techniques, and cross-layer design. Multi-media con-

    331tent such as a video stream requires a large amount of332bandwidth in order for the content to be delivered. As a re-333sult, high data rate leads to high energy consumption.334Transmission techniques that support high bandwidth335and low energy consumption have to be developed. QoS

    336provisioning is a challenging task in a multi-media WSN

    337due to the variable delay and variable channel capacity. It338is important that a certain level of QoS must be achieved339for reliable content delivery. In-network processing, filter-

    340ing, and compression can significantly improve network341performance in terms of filtering and extracting redundant342information and merging contents. Similarly, cross-layer

    343interaction among the layers can improve the processing344and the delivery process.345Mobile WSNs consist of a collection of sensor nodes that346can move on their own and interact with the physical envi-347ronment. Mobile nodes have the ability sense, compute,348and communicate like static nodes. A key difference is mo-

    349bile nodes have the ability to reposition and organize itself

    350in the network. A mobile WSN can start off with some ini-351tial deployment and nodes can then spread out to gather352information. Information gather by a mobile node can be

    353communicated to another mobile node when they are354within range of each other. Another key difference is data

    355distribution. In a static WSN, data can be distributed using

    356fixed routing or flooding while dynamic routing is used in a357mobile WSN. Challenges in mobile WSN include deploy-358ment, localization, self-organization, navigation and con-

    359trol, coverage, energy, maintenance, and data process.

    360Mobile WSN applications include but are not limited to361environment monitoring, target tracking, search and res-

    362cue, and real-time monitoring of hazardous material. For363environmental monitoring in disaster areas, manual364deployment might not be possible. With mobile sensor

    365nodes, they can move to areas of events after deployment

    366to provide the required coverage. In military surveillance367and tracking, mobile sensor nodes can collaborate and368make decision based of the target. Mobile sensor nodes369can achieve a higher degree of coverage and connectivity370compared to static sensor nodes. In the presence of obsta-

    371cles in the field, mobile sensor nodes can plan ahead and

    372move appropriately to obstructed regions to increase373target exposure.

    3744. Applications

    375WSN applications can be classified into two categories:

    376monitoring and tracking (see Fig. 2). Monitoring applica-377tions include indoor/outdoor environmental monitoring,378health and wellness monitoring, power monitoring, inven-

    379tory location monitoring, factory and process automation,380and seismic and structural monitoring. Tracking applica-

    381tions include tracking objects, animals, humans, and vehi-

    382cles. While there are many different applications, below383we describe a few example applications that have been de-384ployed and tested in the real environment.

    385PinPtr [2] is an experimental counter-sniper system386developed to detect and locate shooters. The system

    4 J. Yick et al. / Computer Networks xxx (2008) xxxxxx

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    5/42

    UNCO

    RREC

    TED

    PR

    OOF

    utilizes a dense deployment of sensors to detect and mea-

    sure the time of arrival of muzzle blasts and shock waves

    from a shot. Sensors route their measurements to a base

    station (e.g., a laptop or PDA) to compute the shooters

    location.

    Sensors in the PinPtr system are second-generation

    Mica2 motes connected to a multi-purpose acoustic sensor

    board. Each multi-purpose acoustic sensor board is de-

    signed with three acoustic channels and a Xilinx Spartan

    II FPGA. Mica2 motes run on a TinyOS [21] operating sys-

    tem platform that handles task scheduling, radio commu-

    nication, time, I/O processing, etc. Middleware services

    developed on TinyOS that are exploited in this application

    include time synchronization, message routing with data

    aggregation, and localization.

    Macroscope of redwood [22] is a case study of a WSN

    that monitors and records the redwood trees in Sonoma,

    California. Each sensor node measures air temperature, rel-

    ative humidity, and photo-synthetically-active solar radia-

    tion. Sensor nodes are placed at different heights of the

    tree. Plant biologists track changes of spatial gradients in

    the microclimate around a redwood tree and validate their

    biological theories.

    Semiconductor plants and oil tanker application reported

    in [23] focus on preventive equipment maintenance using

    vibration signatures gathered by sensors to predict equip-

    ment failure. Based on application requirements and site

    survey, the architecture of the network is developed to

    meet application data needs. Two experiments were car-

    ried out: the first was in a semiconductor fabrication plant

    and the second on an onboard oil tanker in the North Sea.

    418The goal was to reliably validate the requirements for

    419industrial environments and evaluate the effect of the sen-420sor network architecture. The study also analyzed the im-

    421pact of platform characteristics on the architecture and

    422performance of real deployment.423Underwater monitoring study in [24] developed a plat-424form for underwater sensor networks to be used for long-

    425term monitoring of coral reefs and fisheries. The sensor426network consists of static and mobile underwater sensor427nodes. The nodes communicate via point-to-point links

    428using high speed optical communications. Nodes broadcast429using an acoustic protocol integrated in the TinyOS proto-430col stack. They have a variety of sensing devices, including

    431temperature and pressure sensing devices and cameras.

    432Mobile nodes can locate and move above the static nodes433to collect data and perform network maintenance func-

    434tions for deployment, re-location, and recovery. The chal-435lenges of deploying sensors in an underwater436environment were some key lessons from this study.

    437MAX [25] is a system for human-centric search of the

    438physical world. MAX allows people to search and locate439physical objects when they are needed. It provides location440information reference to identifiable landmarks rather441than precise coordinates. MAX was designed with the442objectives of privacy, efficient search of a tagged object,

    443and human-centric operation. MAX uses a hierarchical

    444architecture that requires objects to be tagged, sub-sta-445tions as landmarks, and base-station computers to locate446the object. Tags on objects can be marked as private or447public which is searchable by the public or owner only.448MAX is designed for low energy and minimal-delay que-

    Sensor

    Network

    Monitoring

    Military

    Enemy Tracking

    Habitat

    Animal Tracking

    Public/Industrial

    Traffic Tracking

    Car/Bus Tracking

    Environment

    Environmental Monitoring

    (weather, temperature, pressure)

    Military

    Security Detection

    Habitat

    Animal Monitoring

    (Zebra, birds, Cane toad)

    Health

    Patient monitoring

    Tracking

    Business

    Human tracking

    Public/Industrial

    Structural Monitoring

    Factory Monitoring

    Inventory Monitoring

    Machine Monitoring

    Chemical Monitoring

    Business

    Inventory Monitoring

    Fig. 2. Overview of sensor applications.

    J. Yick et al. / Computer Networks xxx ( 2008) xxxxxx 5

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    6/42

    UNCO

    RREC

    TED

    PR

    OOF

    9 ries. The implementation of MAX was demonstrated using

    0 Crossbow motes where trials were conducted in a room of

    1 physical objects.2 Connection-less sensor-based tracking system using3 witness (CenWits) [26] is a search-and-rescue system de-4 signed, implemented, and evaluated using Berkeley Mica25 sensor motes. The system uses several small radio frequen-

    6 cies (RF)-based sensors and a small number of storage and

    7 processing devices. CenWits is not a continuously-con-8 nected network. It is designed for intermittent network9 connectivity. It is comprised of mobile sensors worn by

    0 subjects (people), access points that collect information1 from these sensors and GPS receivers, and location points2 to provide location information to the sensors. A subject

    3 will use the GPS receivers and location points to determine4 its current location. The key concept is the use of witnesses5 to convey a subjects movement and location information6 to the outside world. The goal of CenWits is to determine7 an approximate small area where search-and-rescue ef-8 forts can be concentrated.

    9 Cyclops [27] is a small camera device that bridges the

    0 gap between computationally-constrained sensor nodes1 and complimentary metal-oxide semiconductor (CMOS)2 imagers. This work provides sensor technology with CMOS

    3 imaging. With CMOS imaging, humans can (1) exploit a4 different perspective of the physical world which cannot

    5 be seen by human vision, and (2) identify their importance.

    6 Cyclops attempts to interface between a camera module7 and a lightweight sensor node. Cyclops contains program-8 mable logic and memory circuits with high speed data

    9 transfer. It contains a micro-controller to interface with

    0 the outside world. Cyclops is useful in a number of applica-1 tions that require high speed processing or high resolution

    2 images.3 WSN in a petroleum facility [28] can reduce cost and im-4 prove efficiency. The design of this network focused on the

    5 data rate and latency requirement of the plant. The net-

    6 work consists of four sensor node and an actuator node.7 The sensor nodes are based on T-mote sky devices [29].8 Two AGN1200 pre-802.11N Series MIMO access points9 [30] are used to create an 802.11b 2.4 GHz wireless local0 area network. In this multi-hop WSN, the T-mote sky de-

    1 vices send their radio packets to the base station which

    2 is forwarded to a crossbow stargate gateway. The crossbow3 stargate gateway translates the radio packets and sends it4 along the Ethernet MIMO to a single board TS-3300 com-5 puter [31]. The single board TS-3300 computer outputs6 the sensor data to the distributed control system. The dis-7 tributed control system can also submit changes to the

    8 actuator. In this study, results of network performance,9 RSSI and LQI measurement and noise were gathered. Re-0 sults show that the effect of latency and environmental1 noise can significantly affect the performance of a WSN2 placed in an industrial environment.3 Volcanic monitoring [32] with WSN can help accelerate

    4 the deployment, installation, and maintenance process.

    5 WSN equipments are smaller, lighter, and consume less6 power. The challenges of a WSN application for volcanic7 data collection include reliable event detection, efficient8 data collection, high data rates, and sparse deployment of9 nodes. Given these challenges, a network consists of 16

    510sensor nodes was deployed on Volcn Reventador in north-

    511ern Ecuador. Each sensor node is a T-mote sky device [29]

    512equipped with an external omni-directional antenna, a513seismometer, a microphone, and a custom hardware inter-514face board. Of the 16 sensor nodes, 14 sensor nodes are515equipped with a single axis Geospace Industrial GS-11516Geophone with corner frequency of 4.5 Hz while the other

    517two sensor nodes carried triaxial Geospace Industries GS-1

    518seismometers with corner frequencies of 1 Hz. The custom519hardware interface board was designed with four Texas520Instruments AD7710 analog-to-digital converters to inte-

    521grate with the T-mote sky devices. Each sensor node draws522power from a pair of alkaline D cell batteries. Each sensor523node are placed approximately 200400 m apart from each

    524other. Nodes relay data via multi-hop routing to a gateway525node. The gateway node connected to a long-distance Free-526Wave radio modem transmits the collected data to the527base station. During network operation, each sensor node528samples two or four channels of seismoacoustic data at529100 Hz. The data is stored in local flash memory. When

    530an interesting event occurs, the node will route a message

    531to the base station. If multiple nodes report the same532event, then data is collected from the nodes in a round-533robin fashion. When data collection is completed, the

    534nodes will return to sampling and storing sensor data535locally.

    536In the 19 days of deployment, the network observed

    537230 eruptions and other volcanic events. About 61% of538the data was retrieved from the network due to short out-539ages in the network from software component failure and

    540power outage. Overall, the system performed well in this

    541study.542Health monitoring applications [33] using WSN can im-

    543prove the existing health care and patient monitoring. Five544prototype designs have been developed for applications545such as infant monitoring, alerting the deaf, blood pressure

    546monitoring and tracking, and fire-fighter vital sign moni-

    547toring. The prototypes used two types of motes: T-mote548sky devices [29] and SHIMMER (Intel Digital Health549Groups Sensing Health with Intelligence, Modularity,550Mobility, and Experimental Re-usability).551Because many infant die from sudden infant death syn-

    552drome (SIDS) each year, Sleep Safe is designed for monitor-

    553ing an infant while they sleep. It detects the sleeping554position of an infant and alerts the parent when the infant555is lying on its stomach. Sleep Safe consists of two sensor556motes. One SHIMMER mote is attached to an infants cloth-557ing while a T-mote is connected to base station computer.558The SHIMMER node has a three-axis accelerometer sensing

    559the infants position relative to gravity. The SHIMMER node560periodically sends packets to the base station for process-561ing. Based on the size of the sensing window and the562threshold set by the user, the data is processed to deter-563mine if the infant is on their back.564Baby Glove prototype is designed to monitor vitals.

    565Baby Glove is a swaddling baby wrap with sensors that

    566can monitor an infants temperature, hydration, and pulse567rate. A SHIMMER mote is connected to the swaddling wrap568to transmit the data to the T-mote connected to the base569station. Like Sleep Safe, an alert is sent to the parent if570the analyzed data exceeds the health settings.

    6 J. Yick et al. / Computer Networks xxx (2008) xxxxxx

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    7/42

    UNCO

    RREC

    TED

    PR

    OOF

    FireLine is a wireless heart rate sensing system. It is

    used to monitor a fire fighters heart rate in real-time to de-

    tect any abnormality and stress. FireLine consist of a T-

    mote, a custom made heart rate sensor board, and three

    re-usable electrodes. All these components are embedded

    into a shirt that a fire fighter will wear underneath all his

    protective gears. The readings are taken from the T-mote

    is then transfer to another T-mote connected to the base

    station. If the fire fighters heart rate is increasing too high,

    an alert is sent.

    Heart@Home is a wireless blood pressure monitor and

    tracking system. Heart@Home uses a SHIMMER mote lo-

    cated inside a wrist cuff which is connected to a pressure

    sensor. A users blood pressure and heart rate is computed

    using the oscillometric method. The SHIMMER mote re-

    cords the reading and sends it to the T-mote connected

    to the users computer. A software application processes

    the data and provides a graph of the users blood pressure

    and heart rate over time.

    LISTSENse enables the hearing impaired to be informed

    of the audible information in their environment. A user

    carries the base station T-mote with him. The base station

    T-mote consists of a vibrator and LEDs. Transmitter motes

    are place near objects (e.g., smoke alarm and doorbell) that

    can be heard. Transmitter motes consist of an omni-direc-

    tional condenser microphone. They periodically sample the

    microphone signal at a rate of 20 Hz. If the signal is greater

    than the reference signal, an encrypted activation message

    is sent to the user. The base station T-mote receiving the

    message actives the vibrator and its LED lights to warn

    the user. The user must press the acknowledge button to

    deactivate the alert.

    ZebraNet[9] system is a mobile wireless sensor network

    used to track animal migrations. ZebraNet is composed of

    sensor nodes built into the zebras collar. The node consists

    of a 16-bitTI microcontroller,4 Mbits of off-chip flashmem-

    ory,a 900 MHz radio, and a GPS unit. Positionalreadings are

    taking using the GPS and send multi-hop across zebras to

    the base station. The goal is to accurately log each zebras

    position and usethem for analysis. A total of 610 zebra col-

    lars were deployed at the Sweetwaters game reserve in cen-

    tral Kenya to study the effects and reliability of the collar

    and to collect movement data. After deployment, the biolo-

    gists observed that the collared zebras were affected by the

    collars. They observed additional head shakes from those

    zebra in the first week. After the first week, the collared ze-

    bra show no difference than the uncollared zebra. A collec-

    tion of movement data was also collected during this study.

    From the data, the biologists can better understand the ze-

    bra movements during the day and night.

    Open research issues

    The enabling applications provide some a key attributes

    that determine the driving force behind WSN research.

    Existing applications such as environmental monitoring,

    health monitoring, industrial monitoring, and military

    tracking have application-specific characteristics and

    requirements. These application-specific characteristics

    and requirements coupled with todays technology lead to

    different hardware platforms and software development.

    A variety of hardware platforms and technology have been

    631developed over the years; however, more experimental

    632work is necessary to make these applications more reliable

    633and robust in the real world. Appendix C compares the634application with the protocol stack.635WSNs have the potential to enhance and change the636way people interact with technology and the world. The637direction of future WSNs lies in identifying real business

    638and industry needs. Interactions between research and

    639development are necessary to bridge the gap between640existing technology and the development of business solu-641tions. Applying sensor technology to industrial applica-

    642tions will improve business processes as well as open up643more problems for researchers.

    6445. Internal sensor system

    645For a sensor to operate in a wireless sensor system,

    646there are several internal system issues that need to be ad-

    647dressed through the system platform and operating system648(OS) support. In addition, supporting standards, storage,

    649and physical testbeds are reviewed in the following650subsections.

    6515.1. System platform and OS support

    652Current WSN platforms are built to support a wide653range of sensors. Products that offer sensors and sensor654nodes have different radio components, processors, and655storage. It is a challenge to integrate multiple sensors on656a WSN platform since sensor hardware is different and

    657processing raw data can be a problem with limited re-658sources in the sensor node. System software such as the659OS must be designed to support these sensor platforms.660Research in this area involves designing platforms that661support automatic management, optimizing network lon-662gevity, and distributed programming. Below we discuss

    663two platforms: a Bluetooth-based sensor system [34] and

    664a detection-and-classification system [35].665Bluetooth-based sensor networks [20] reported a study to666determine if a Bluetooth-based sensor node is viable for a667WSN. Typical radio components used in a WSN are based668on fixed frequencies where sensor nodes within communi-

    669cation range compete for a shared channel to transmit

    670data. But Bluetooth is based on spread-spectrum transmis-671sion where separate channels are used to transmit data.672The Bluetooth-based devices used in the experiments673are BTnodes developed by ETH Zurich [36]. A stripped-674down version of the Bluetooth stack for TinyOS was de-

    675signed and ported into the BTnodes. In order to support a

    676multi-hop network, each BTnode is equipped with two677radios: one configured to operate as a master and the other678as a slave. The master radio can support up to seven con-

    679nections while the slave radio looks for another node to680connect to. Because Bluetooth is connection oriented, a

    681master and slave connection must be established before

    682data is exchanged. When a new node joins the network,683its slave radio is first enabled. The new node tries to connect684itself with the rest of the network. When the new node

    685finds a node to connect to as its slave, it turns on the master686radio to accept connections from nodes that are not yet

    J. Yick et al. / Computer Networks xxx ( 2008) xxxxxx 7

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    8/42

    UNCO

    RREC

    TED

    PR

    OOF

    7 connected to the network. If the new node fails to connect

    8 to other nodes in its vicinity due to the maximum number

    9 of connections being reached at the other nodes, it re-con-0 nects to the first node it had contacted in the network. With1 the second request, the master radio in that node will drop2 one of its slave node connections and accept the connection3 from the new node. The disconnected node will find an-

    4 other node in its vicinity to connect. The network topology

    5 formed by this procedure is a connected tree.6 Experimental results indicate that Bluetooth-based sen-7 sor networks using BTnodes are suitable for applications

    8 that are active over a limited time period with a few unpre-9 dictable traffic bursts. BTnodes can achieve high through-0 put; however, they consume a lot of energy even when

    1 idle. Connection maintenance is expensive and dual radios2 are needed to support multi-hop routing. Hence, Bluetooth3 can only serve as an alternative to broadcast radios.4 Detection-and-classification system developed in VigilNet5 [35] can detect and classify vehicles, persons, and persons6 carrying ferrous objects. It targets objects with a maximum

    7 velocity error of 15%. The VigilNet surveillance system con-

    8 sists of 200 sensor nodes which are deployed in a pre-9 planned manner into the environment. Their locations0 are assigned at the time they are deployed. Each sensor

    1 node is equipped with a magnetometer, a motion sensor,2 and a microphone.

    3 A hierarchical architecture was designed for this system

    4 in order to distribute sensing and computation tasks to dif-5 ferent levels of the system. The hierarchical architecture is6 comprised of four tiers: sensor-level, node-level, group-le-

    7 vel, and base-level. The lowest level, the sensor-level, deals

    8 with the individual sensor and its sensing algorithm to de-9 tect and classify objects. Once the sensing algorithm has

    0 processed the sensor data, the classification result is sent1 to the next level, namely the node-level. At the node-level,2 classification deals with the fusion of various sensor data

    3 obtained by the individual nodes. The node-level sensing

    4 algorithm relays the sensor data from each sensor and5 forms node-level classification results. Both the sensor-le-6 vel and node-level classification functions reside on the7 node itself. The next level is the group-level. This level of8 classification is performed by a group of nodes. A set of

    9 nodes is organized in a group, and a group leader is elected

    0 to perform group-level classification. The input to the1 group-level classification is the node-level classification re-2 sults of the aggregated attributes. At group-level classifica-3 tion, group leaders can accomplish more advanced tasks4 and gain better knowledge of the location of the targets.5 The highest level is the base-level classification. At this le-

    6 vel, the results from the group-level classification are7 transmitted via multi-hop to the base station. The base-le-8 vel classification algorithm finalizes the results collected9 and reduces false positives among the reported results.0 VigilNet was deployed and tested in an outdoor site. The1 system was able to accurately detect targets and reduce

    2 false negatives with a dense deployment of sensor nodes.

    3 5.2. Standards

    4 Wireless sensor standards have been developed with5 the key design requirement for low power consumption.

    746The standard defines the functions and protocols necessary

    747for sensor nodes to interface with a variety of networks.

    748Some of these standards include IEEE 802.15.4 [37], ZigBee749[38,39], WirelessHART [40,41], ISA100.11 [42], IETF 6LoW-750PAN [4345], IEEE 802.15.3 [46], Wibree [47]. Below751describes these standards in more detail.752IEEE 802.15.4: IEEE 802.15.4 [37] is the proposed

    753standard for low rate wireless personal area networks

    754(LR-WPANs). IEEE 802.15.4 focuses on low cost of deploy-755ment, low complexity, and low power consumption. IEEE756802.15.4 is designed for wireless sensor applications that

    757require short data range communication to maximizing758battery life. The standard allows the formation of the star759and peer-to-peer topology for communication between

    760network devices. Devices in the star topology communi-761cate with a central controller while in the peer-to-peer762topology ad hoc and self-configuring networks can be763formed. IEEE 802.15.4 devices are designed to support764the physical and data-link layer. The physical layer sup-765ports 868/915 MHz low bands and 2.4 GHz high bands.

    766The MAC layer controls access to the radio channel using

    767the CSMA-CA mechanism. The MAC layer is also responsi-768ble for validating frames, frame delivery, network inter-769face, network synchronization, device association, and

    770secure services. Wireless sensor applications using IEEE771802.15.4 include residential, industrial, and environment

    772monitoring, control and automation.

    773ZigBee [38,39] defines the higher layer communication774protocols built on the IEEE 802.15.4 standards for LR-PANs.775ZigBee is a simple, low cost, and low power wireless com-

    776munication technology used in embedded applications.

    777ZigBee devices can form mesh networks connecting hun-778dreds to thousands of device together. ZigBee devices use

    779very little power and can operate on a cell battery for many780years. There are three types of ZigBee devices: ZigBee coor-781dinator, ZigBee router, and ZigBee end device. ZigBee coor-

    782dinator initiates network formation, stores information,

    783and can bridge networks together. ZigBee routers link784groups of devices together and provide multi-hop commu-785nication across devices. ZigBee end device consists of the786sensors, actuators, and controllers that collects data and787communicates only with the router or coordinator. The

    788ZigBee standard was publicly available as of June 2005.

    789WirelessHART: The WirelessHART [40,41] standard pro-790vides a wireless network communication protocol for pro-791cess measurement and control applications. The standard792is based on IEEE 802.15.4 for low power 2.4 GHz operation.793WirelessHART is compatible with all existing devices,794tools, and systems. WirelessHART is reliable, secure, and

    795energy efficient. It supports mesh networking, channel796hopping, and time-synchronized messaging. Network com-797munication is secure with encryption, verification, authen-798tication, and key management. Power management799options enable the wireless devices to be more energy effi-800cient. WirelessHART is designed to support mesh, star, and

    801combined network topologies. A WirelessHART network

    802consists of wireless field devices, gateways, process auto-803mation controller, host applications, and network man-804ager. Wireless field devices are connected to process or805plant equipment. Gateways enable the communication be-806tween the wireless field devices and the host applications.

    8 J. Yick et al. / Computer Networks xxx (2008) xxxxxx

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    9/42

    UNCO

    RREC

    TED

    PR

    OOF

    The process automation controller serves as a single con-

    troller for continuous process. The network manager con-

    figures the network and schedule communication

    between devices. It also manages the routing and network

    traffic. The network manager can be integrated into the

    gateway, host application, or process automation control-

    ler. In short, WirelessHART standards have been released

    to the industry in September 2007 and will soon be avail-

    able in commercial products.

    ISA100.11a: ISA100.11a [42] standard is designed for

    low data rate wireless monitoring and process automation

    applications. It defines the specifications for the OSI layer,

    security, and system management. The standard focuses

    on low energy consumption, scalability, infrastructure,

    robustness, and interoperability with other wireless de-

    vices. ISA100.11a networks use only 2.4 GHz radio and

    channel hopping to increase reliability and minimize inter-

    ference. It offers both meshing and star network topolo-

    gies. ISA100.11a also provides simple, flexible, and

    scaleable security functionality.

    6LoWPAN: IPv6-based Low power Wireless Personal

    Area Networks [4345] enables IPv6 packets communica-

    tion over an IEEE 802.15.4 based network. Low power de-

    vice can communicate directly with IP devices using IP-

    based protocols. Using 6LoWPAN, low power devices have

    all the benefits of IP communication and management.

    6LoWPAN standard provides an adaptation layer, new

    packet format, and address management. Because IPv6

    packet sizes are much larger than the frame size of IEEE

    802.15.4, an adaptation layer is used. The adaptation layer

    carries out the functionality for header compression. With

    header compression, smaller packets are created to fit into

    an IEEE 802.15.4 frame size. Address management mecha-

    nism handles the forming of device addresses for commu-

    nication. 6LoWPAN is designed for applications with low

    data rate devices that required Internet communication.

    IEEE 802.15.3: IEEE 802.15.3 [46] is a physical and MAC

    layer standard for high data rate WPAN. It is designed to

    support real-time multi-media streaming of video and mu-

    sic. IEEE 802.15.3 operates on a 2.4 GHz radio and has data

    rates starting from 11 Mbps to 55 Mbps. The standard uses

    time division multiple access (TDMA) to ensure quality of

    service. It supports both synchronous and asynchronous

    data transfer and addresses power consumption, data rate

    scalability, and frequency performance. The standard is

    used in devices such as wireless speakers, portable video

    electronics, and wireless connectivity for gaming, cordless

    phones, printers, and televisions.

    Wibree: Wibree [47] is a wireless communication tech-

    nology designed for low power consumption, short-range

    communication, and low cost devices. Wibree allows the

    communication between small battery-powered devices

    and Bluetooth devices. Small battery powered devices in-

    clude watches, wireless keyboard, and sports sensors

    which connect to host devices such as personal computer

    or cellular phones. Wibree operates on 2.4 GHz and has a

    data rate of 1 Mbps. The linking distance between the de-

    vices is 510 m. Wibree is designed to work with Blue-

    tooth. Bluetooth with Wibree makes the devices smaller

    and more energy-efficient. BluetoothWibree utilizes the

    existing Bluetooth RF and enables ultra-low power

    868consumption. Wibree was released publicly in October

    8692006.

    8705.3. Storage

    871Conventional approaches in WSNs require that data be872transferred from sensor nodes to a centralized base station

    873because storage is limited in sensor nodes. Techniques

    874such as aggregation and compression reduce the amount875of data transferred, thereby reducing communication and876energy cost. These techniques are important for real-time

    877or event-based applications, but they may not suffice.878Applications that operate on a query-and-collect approach879will selectively decide which data are important to collect.

    880Optimizing sensor storage becomes important in this case881when massive data is stored over time.882Given that storage space is limited and communication883is expensive, a storage model is necessary to satisfy storage884constraints and query requirements. In this subsection, we885evaluate several storage methods in terms of design goals,

    886assumptions, operation models, and performance.

    887GEM: Graph EMbedding (GEM) [48] provides an infra-888structure for routing and data-centric storage for sensor889networks. The idea of graph embedding works in two

    890steps. The first step is choosing a labelled guest graph for891routing and data-centric storage. The second step is to

    892embed the guest graph onto the actual sensor topology.

    893Each sensor node in this network is given an identifier894and a label encoded with its position. Each sensor node895needs only to know the labels of its neighbors. To support

    896data-centric storage in GEM, each data item has a name

    897that can be mapped to a label and stored at different nodes.898When a client requests data, it sends a query with the da-

    899tas name into the network. The node that has the data will900route the data back to the requested. GEM enables node-901to-node routing by using a lookup mechanism to find a

    902nodes current label. If two nodes need to communicate,

    903the sender node must first retrieve the label of the receiv-904ing node. A lookup request message is sent by the sender to905the receiver. Upon receiving the lookup request, the recei-906ver retrieves the label in a distributed hash table. Once the907sender node has the receivers label, it can send messages

    908to the receiver.

    909To demonstrate how GEM is applied to a sensor net-910work, the virtual polar coordinate space (VPCS) was devel-911oped in this study. In VPCS, a ring-tree graph is embedded912into the network topology. Each sensor node is assigned a913level which is the number of hops from the root node. Each914node is also assigned a virtual angle range which identifies

    915the node within that level. The virtual angle range is a sub-916set of its parents virtual angle range. Children of a node917may not have overlapping angle ranges. The virtual polar918coordinate routing (VPCR) algorithm is built on top of VPCS919to route a message from a node to another. VPCR utilizes920polar coordinates for efficient routing. Each node has a la-

    921bel defined by a space in a VPCS. VPCR is greedy because it

    922forwards packets closer to the destination angle range.923Packet forwarding is accomplished by checking for nearby9242-hop neighbor nodes which have an angle range that is925closer to the destination angle than the current nodes an-926gle range. If so, VPCR forwards the packet to that node.

    J. Yick et al. / Computer Networks xxx ( 2008) xxxxxx 9

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    10/42

    UNCO

    RREC

    TED

    PR

    OOF

    7 Each node is required to store state information about its

    8 neighbors. VPCR makes routing more efficient by routing

    9 with cross-links in the ringed tree. Experimental results0 show that VPCR is efficient in both energy usage and1 routing.2 TSAR: Two-tier sensor storage architecture (TSAR) [49]3 uses interval skip graphs to employ a multi-resolution or-

    4 dered distributed index structure for efficient support of

    5 spatio-temporal and value queries. Sensor nodes send con-6 cise identifying information (or metadata) to a nearby7 proxy. Proxies interact with one another to construct a dis-

    8 tributed index of the metadata reported by the sensors and9 an index of the associated data stored at the sensors. The0 index provides a logical view of the distributed data. The

    1 index is used to pinpoint all data from the corresponding2 sensors. Actual data remains in the sensor nodes. TSAR re-3 duces energy overhead at sensor nodes by using the prox-4 ies for queries and low cost transmission of metadata to5 the proxies. There are four main contributions: (1) novel6 distributed index structure based on interval skip graphs,

    7 (2) each sensors local archive to store data in flash mem-

    8 ory, (3) a prototype of TSAR on a multi-tier test bed, and9 (4) a detailed evaluation of TSAR. Experimental results0 show feasibility and low energy latency of the distributed

    1 storage architecture in a multi-tier sensor network.2 Multi-resolution storage: Multi-resolution storage sys-

    3 tem [50] provides storage and long-term querying of the

    4 data for data-intensive applications. Multi-resolution stor-5 age uses in-network wavelet-based summaries to store6 data in a spatially- and hierarchically-decomposed distrib-

    7 uted storage structure. The storage system architecture is

    8 divided into three parts: (1) wavelet process to construct9 multi-resolution summaries, (2) drill-down query process

    0 to reduce search cost, and (3) a data-aging scheme to dis-1 card summaries. In the first part, the wavelet process uses2 a summarizing technique that provides data compression

    3 for spatio-temporal data sets. Wavelet construction has

    4 two phases: temporal summarization phase and spatial5 summarization phase. The first phase requires each node6 to compress the time-series data by exploiting temporal7 redundancy in the signal. The second phase constructs a8 hierarchical grid-based overlay. At each level, data is com-

    9 pressed more in a spatial scale. At the highest level, one or

    0 a few nodes contain an overall summary of all the data in1 the network.2 The second part of the system architecture is the drill-3 down query process to reduce the cost of search. Drill-4 down queries are inserted at the highest level of the hier-5 archy and use a coarse summary as a hint to indicate which

    6 region in the network will most likely contain the response7 to the query. The query is forwarded to nodes that store8 summaries of these regions. The query is routed from9 one sub-region to the next till it reaches the lowest level0 of the hierarchy or when there are enough results in the1 intermediate nodes. The drill-down query process is very

    2 efficient in that it can obtain query results in a few steps.

    3 Lastly, old data must be discarded in order to create4 space to store new data. To determine how old is the data5 in the network, each data is given an age that specifies the6 amount of time that the summary has been stored. Two7 data-aging schemes are proposed: a training-based algo-

    988rithm and a greedy algorithm. The training algorithm oper-

    989ates on a limited training set of data. During the training

    990period, aging parameters are extracted from a training991set. The training set is typically data sensed during system992deployment. A weighted cumulative error is computed993from different queries. The cumulative error is fed into994an optimization function to evaluate aging parameters

    995for different summaries. For the greedy algorithm, there

    996are no prior data sets to determine the aging parameters.997It assigns weights to summaries according to expected998importance of each resolution toward drill-down queries.

    999The goal of the aging schemes is to provide data manage-1000ment and enhance the query process. Results show that1001both schemes perform within 2% of the optimal scheme,

    1002but the training scheme performed better than the greedy1003scheme.

    10045.4. Testbeds

    1005A WSN testbed is consists of sensor nodes deployed in a

    1006control environment. It is designed to support experimen-

    1007tal research in a real-world setting. It provides researchers1008a way to test their protocols, algorithms, network issues1009and applications. Experiments can easily be configured,

    1010run, and monitor remotely. Experiments can also be re-1011peated produce the same results for analysis. Below de-

    1012scribes several WSN testbeds in more detail.

    1013ORBIT: Open access research testbed for next-genera-1014tion wireless networks (ORBIT) [51] consists of 64 remo-1015tely accessible sensor nodes placed indoor with $1 m

    1016spacing apart. Each ORBIT radio node consists of a 1-GHz

    1017VIA C3 processor, two wireless PCI 802.11a/b/g interface,1018two ethernet ports, and an integrated chassis manager.

    1019Users can log on remotely to set up their experiment. OR-1020BIT can be used test new applications, measure system1021performance, run cross-layer experiments, and test new

    1022protocols and algorithms.

    1023MoteLab: MoteLab [52] is a web-based WSN test consist1024of a set of MicaZ motes [53] connected to a central server.1025The central server handles the scheduling, re-program-1026ming and data logging of the motes. A user can log onto1027a web interface to create and schedule experiments. The

    1028goal of MoteLab is to allow users to evaluate WSN applica-

    1029tions without manually re-programming and re-deploying1030the nodes into the physical environment. The users can re-1031trieve data through the web interface and interact with1032individual nodes. MoteLab consists of the following soft-1033ware components: a SQL database, web interface, DB log-1034ger, and job daemon. The SQL database stores all the

    1035information needed for the test-bed operation. The web1036interface uses PHP to generate the web contents for the1037users to access. The DB logger is connected to each node1038to receive messages and store them in the SQL database.1039The job daemon is responsible for re-programming each1040node, and starting and stopping system components. Mot-

    1041eLab have been used to study newly developed protocols,

    1042signal strength analysis, and cluster analysis.1043Emulab: Emulab [54] is a remotely accessible mobile1044and wireless sensor testbed. The testbed consists of Acro-1045name robots carrying an XScale based Startgate small com-1046puter and 900 Hz Mica2 mote [53]. The robots operate on

    10 J. Yick et al. / Computer Networks xxx (2008) xxxxxx

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    11/42

    UNCO

    RREC

    TED

    PR

    OOF

    battery power which last up to 3 h and uses 802.11 b for

    communication. The radios are set to 900 MHz. The robots

    motion and steering comes from two drive wheels at a

    maximum rate of 2 m/s. There are six infrared proximity

    sensors on all sides of the robot to detect obstructions.

    Users can create experiments through a web interface

    and schedule events to control the robots movement.

    Emulab can be used to study network topologies, mobility

    effects on protocols, test algorithms, and mobile

    applications.

    5.5. Diagnostics and debugging support

    In order to guarantee the success of the sensor network

    in the real environment, it is important to have a diagnos-

    tic and debugging system that can measure and monitor

    the sensor node performance of the overall network. Stud-

    ies that deal with handling various types of hardware and

    software failures help extend the life of each sensor which

    in turn help increase the sensor network lifetime. In addi-

    tion to failures, addressing methods to enhance communi-

    cation performance can make the system more efficient. In

    the following subsections, we first describe a tool call Sym-

    pathy [26] that detects and localizes failures. We then dis-

    cuss the study reported in [55] which analyzes packet

    delivery performance at the physical and the medium ac-

    cess control (MAC) layers.

    Sympathy: Sympathy [26] is a diagnosis tool for detect-

    ing and debugging failures in sensor networks. It is specif-

    ically designed for data-collection applications where

    nodes periodically send data back to a centralized base sta-

    tion or sink. Sympathy detects failures in a system by

    selecting metrics such as connectivity, data flow, nodes

    neighbor and next hops. Connectivity metrics provide con-

    nectivity information from every node in the network.

    Sympathy collects every nodes current routing table with

    information for next hop and path quality. Flow metrics

    provide the networks traffic load as well as its connectiv-

    ity. Sympathy collects packet level information transmit-

    ted and received from each node. In addition, Sympathy

    also maintains information for packets transmitted from

    the sink to the nodes. Based on these metrics, Sympathy

    detects when nodes are not delivering sufficient data to

    the sink and locates the cause of the failure.

    Sympathy can identify three types of failures: self, path,

    and sink. In self failure, the node itself has failed due to a

    crash, re-boot, bug in software code, or connectivity issue.

    In path failure, a node along the path fails, causing other

    nodes to fail or there are collisions along the path. In sink

    (i.e., base station) failure, the whole network appears to

    be failing when it is the sink that has failed. Failure at

    the sink may be due to bad sink placement, changes in

    the environment after deployment, and connectivity

    issues.

    In Sympathy, the sink/base station runs the necessary

    software to detect and localize the failure. Localizing a fail-

    ure is a four-stage process. In the first stage, the sink col-

    lects metrics from the sensor nodes in the system. Upon

    receiving a packet, Sympathy looks for failures by analyz-

    ing the received metrics and running tests to determine

    the cause. Common causes include a node crashing or re-

    1106booting, no route to the base station/sink, or the request

    1107never reaching the node. In these cases, Sympathy identi-

    1108fies the type of failure and reports it to the user. Hence, col-1109lecting information about each node allows Sympathy to1110detect failures more quickly.1111Analysis of data packet delivery: the work in [55] studied1112packet delivery performance of a sensor network at the

    1113physical and MAC layers. At the physical layer, the work

    1114in [55] studies the performance of packet delivery under1115different transmit powers and physical-layer encoding. At1116the MAC layer, different MAC layer mechanisms such as

    1117carrier sensing and link-layer re-transmission are used to1118measure the efficiency of packet delivery. Up to 60 Mica1119motes were used to measure packet delivery under three

    1120different environmental settings: an office building, a hab-1121itat with moderate foliage, and an open parking lot. Under1122these settings, results show that both physical and MAC1123layers contribute to the packet-delivery performance,1124which is defined as the fraction of packets not successfully1125received by the receiver within a time window.

    1126At the physical layer, traffic is generated by one node at

    1127one end of the line transmitting one packet per second.1128Packet-delivery performance is measured with the MAC1129layer disabled under different environments, coding

    1130schemes, and transmission settings. Results show that at1131least 20% of the nodes had at least 10% packet loss and at

    1132least 10% of the nodes had greater than 30% packet loss.

    1133Spatial characteristics show the existence of a gray area1134for some nodes. Nodes that are a certain distance from1135the sender have uniformly high packet reception rate. Be-

    1136yond this distance is a gray area in which the reception

    1137rate changes dramatically. Receiving nodes in this gray1138area are likely to experience either 90% successful recep-

    1139tion or less than 50% reception rate. The gray area defined1140for an office building and open parking lot is one-third of1141the total communication range while for habitat setting,

    1142it is one-fifth.

    1143At the MAC layer, experiments vary in topology, envi-1144ronment, and traffic pattern. Packet losses in this case are1145largely due to lost transmissions. Under light load, nearly114650% of the links have an efficiency of 70% or higher. Under1147heavy load, nearly 50% of the links have efficiency less than

    114820%. Depending on the load, between 50% and 80% of the

    1149communication energy is used for repairing lost transmis-1150sions. Packet-delivery performance can be greatly im-1151proved by adding a simple set of mechanisms such as1152topology control to discard neighbors with asymmetric1153links.

    1154Open research issues

    1155The design of a WSN platform must deal with chal-1156lenges in energy efficiency, cost, and application require-1157ments. It requires the optimization of both the hardware1158and software to make a WSN efficient. Hardware includes1159using low cost tiny sensor nodes while software addresses

    1160issues such as network lifetime, robustness, self-organiza-1161tion, security, fault tolerance, and middleware. Application1162requirements vary in terms of computation, storage, and

    1163user interface and consequently there is no single platform1164that can be applied to all applications. Existing platforms

    1165discussed here include a Bluetooth-based sensor system

    J. Yick et al. / Computer Networks xxx ( 2008) xxxxxx 11

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    12/42

    UNCO

    RREC

    TED

    PR

    OOF

    6 [34] and a detection-and-classification system [35]. Future

    7 work in this area entails examining a more practical plat-

    8 form solution for problems in new applications.9 Storage capacity in low end sensor nodes is limited.0 Rather than sending large amounts of raw data to the base1 station, a local sensor nodes storage space is used as a dis-2 tributed database to which queries can send to retrieve

    3 data. Existing approaches [4850] present data structures

    4 that can efficiently manage and store the data. Neverthe-5 less, energy-efficient storage data structure is still an open6 area of research that requires optimizing various types of

    7 database queries both with respect to performance and en-8 ergy efficiency.9 Performance studies provide valuable information for

    0 developing tools and solutions to improve system perfor-1 mance. Critical factors that influence system performance2 include scalability, communication, protocols at different3 layers, failures, and network management. Scalability is-4 sues can degrade system performance. Communication5 protocols are still trying to achieve a reasonable through-

    6 put when the size of the network increases. Optimizing

    7 and analyzing protocols at different layers can improve8 system performance and determine their benefits and lim-9 itations. Sensor nodes can fail at any time due to hardware,

    0 software, or communication reasons. It is important that1 there are services to handle these failures before and after

    2 they occur. Development of network management tools

    3 enables monitoring of system performance and configur-4 ing of sensor nodes.

    5 6. Network services

    6 Sensor provisioning, management, and control services

    7 are developed to coordinate and manage sensor nodes.8 They enhance the overall performance of the network in9 terms of power, task distribution, and resource usage. Pro-0 visioning properly allocates resources such as power and1 bandwidth to maximize utilization. In provisioning, there

    2 is coverage and localization. Coverage in a WSN needs to

    3 guarantee that the monitored region is completely covered4 with a high degree of reliability. Coverage is important be-5 cause it affects the number of sensors to be deployed, the

    6 placement of these sensors, connectivity, and energy.7 Localization is the process by which a sensor node tries

    8 to determine its own location after deployment. Manage-

    9 ment and control services play a key role in WSNs as they

    0 provide support to middleware services such as security,1 synchronization, data compression and aggregation,

    2 cross-layer optimization, etc. In this section, we study pro-3 visioning, control, and management services based on their4 objectives. A brief summary of each plane is described in

    5 each of the sections below.

    6 6.1. Localization

    7 In WSNs, sensor nodes that are deployed into the envi-8 ronment in an ad hoc manner do not have prior knowledge9 of their location. The problem of determining the nodes

    0 location (position) is referred to as localization. Existing1 localization methods include global positioning system

    1222(GPS), beacon (or anchor) nodes, and proximity-based

    1223localization. Equipping the sensor nodes with a GPS recei-

    1224ver is a simple solution to the problem. However, such a1225GPS-based system may not work when the sensors are de-1226ployed in an environment with obstructions such as dense1227foliage areas. The beacon (anchor) method makes use of1228beacon (anchor) nodes, which know their own position,

    1229to help sensors determine their position. This method has

    1230its shortcoming. It does not scale well in large networks1231and problems may arise due to environmental conditions.1232Proximity-based localization makes use of neighbor nodes

    1233to determine their position and then act as beacons for1234other nodes. Below we review some of the key localization1235techniques that differ from the above methods.

    1236Moores algorithm: Ref. [56] presents a distributed local-1237ization algorithm for location estimation without the use1238of GPS or fixed beacon (anchor) nodes. A key feature of this1239algorithm is the use of a robust quadrilateral. A robust1240quadrilateral is a fully-connected quadrilateral whose four1241sub-triangles are robust. Localization based on robust

    1242quadrilateral can be adjusted to support noisy measure-

    1243ments and it correctly localizes each node with a high1244probability.1245This algorithm has three phases: cluster localization

    1246phase, cluster optimization phase, and cluster transforma-1247tion phase. In the first phase, each node becomes the cen-

    1248ter of a cluster and measures the distance of its one-hop

    1249neighbors. The information gathered is broadcasted. For1250each cluster, each node computes the complete set of ro-1251bust quadrilaterals and finds the largest sub-graph of over-

    1252lapping robust quadrilaterals. Position estimations for a

    1253local coordinate system are computed for as many nodes1254as possible using the overlap graph using a breadth-first

    1255search. The second phase is an optimization phase that1256can be omitted. Position estimations are refined using1257numerical optimization such as spring relaxation or the

    1258NewtonRaphson method. The last phase computes the

    1259transformation between local coordinate system of con-1260nected clusters. The transformation computes the rotation,1261translation, and possible reflection that best aligns the1262nodes of two local coordinate systems.1263There is, however, one drawback to this system. Under

    1264conditions of low node connectivity and high measure-

    1265ment noise, the algorithm may not be able to localize some1266nodes.1267RIPS: The work in [57] proposes a localization system1268called Radio Interferometric Positioning System (RIPS)1269which utilizes two radio transmitters to create an interfer-1270ence signal. Two radio transmitters are placed at different

    1271locations and set at slightly different radio frequencies to1272provide ranging information for localization. At least two1273receivers are needed to calculate the phase offset of the ob-1274served signals. The relative phase offset is a function of the1275relative positions between the two transmitters and the1276receivers, and the carrier frequency. By measuring the rel-

    1277ative phase offset, one can analyze and determine the rel-

    1278ative locations of the two receivers or the location of the1279radio source if the receiver locations are known.1280Spotlight: Spotlight [58] is a system that achieves high1281accuracy of localization without the use of expensive hard-1282ware like other localization systems. Spotlight uses an

    12 J. Yick et al. / Computer Networks xxx (2008) xxxxxx

    COMPNW 3774 No. of Pages 42, Model 3G

    5 May 2008 Disk UsedARTICLE IN PRESS

    Please cite this article in press as: J. Yick et al., Wireless sensor network survey, Comput.Netw. (2008), doi:10.1016/

    j.comnet.2008.04.002

  • 7/31/2019 Wsn Survey(08)

    13/42

    UNCO

    RREC

    TED

    PR

    OOF

    asymmetric architecture where computation resides on a

    single Spotlight device. The Spotlight device uses a steer-

    able laser light source which illuminates the sensor nodes

    that are placed in a known terrain. The main idea of the

    Spotlight localization system is to generate controlled

    events in the field where the sensor nodes are deployed.

    An event can be defined as a lighted sensor area. Using

    time events perceived by a sensor node and spatio-tempo-

    ral properties of the generated events, spatial information

    regarding the sensor node can be inferred. Results show

    that Spotlight is more accurate than other range-based

    localization schemes and much more effective for long-

    range localization problems. The cost of localization is

    low since only one single device is necessary to localize

    the network.

    Secure localization: Secure localization [59] focuses on

    securing the localization process. The goal is to prevent

    malicious beacon nodes from providing false location to

    sensors. Sensors rely on beacon information to compute

    their position. To prevent the localization process from

    being compromised, the following security requirement

    must be satisfied. Sensors must only accept information

    from authenticated beacon nodes. Sensors should only

    use information that has not been tampered. Sensors

    should be able request location information at anytime.

    Upon a location request, information exchange must take

    place immediately and not at a later time. Neither a

    sources or sensors location should be disclosed at any

    time to prevent malicious nodes from taking over a loca-

    tion in the network. If any one of these requirements is

    breached, the localization process is compromised.

    Some of the existing secure location techniques include

    SeRloc [60], Beacon Suite [61], DRBTS [62], SPINE [63], and

    ROPE [64]. SeRloc uses a set of locator nodes equipped with

    directional antennas to provide sensors with location

    information. Each locator transmits a different beacon at

    each antenna sector. An attacker would have to imperson-

    ate several locators to compromise the localization pro-

    cess. While SeRloc prevents attackers comprising the

    localization process, beacon suite identifies the malicious

    beacon nodes. Beacon nodes serve two purposes: (1) pro-

    vide location information to sensor nodes, and (2) detect

    malicious beacon signals. To detect malicious beacon sig-

    nals, a beacon can request location information from an-

    other beacon in order to observe its behaviour. When a

    beacon node determines that the beacon that its observing

    is misbehaving, it reports the beacon to the base station. A

    similar approach called distributed reputation and trust-

    based security (DRBTS) protocol identifies malicious infor-

    mation by enabling beacon node monitoring. Beacon nodes

    monitor each other and provide information to the sensor

    nodes. Sensor nodes can choose to accept a beacons infor-

    mation based on votes from their common neighbors.

    Using this voting approach, sensor nodes can determine

    the trustworthy beacons within their range. It is demon-

    strated through simulation the robustness and effective-

    ness of DRBTS in large networks.

    A centralized approach, secure positioning in sensor

    network (SPINE) is based on verifiable multi-lateration.

    SPINE bounds each sensor to at least three reference points

    within its range in order to compute its position. SPINE

    1344effectively prevents against nodes from lying about its po-

    1345sition. Like SeRLoc, ROPE uses a set of locators to provide

    1346location information to the sensor nodes. Each sensor1347shares a pairwise key with every locator. Prior to data col-1348lection, ROPE provides a location verification mechanism1349to verify the locations of the sensors.1350MAL: Mobile-assisted localization (MAL) [65] utilizes a

    1351mobile user (a human or robot) to assist in collecting dis-

    1352tance information between itself and static sensor nodes1353for node localization. In node localization, a minimum1354number of distance samples must be collected before a

    1355nodes coordinates can be computed. The goal is to re-con-1356struct the position of the nodes given a graph with mea-1357sured distance edges. In MAL, a mobile user explores the

    1358sensor region and incrementally builds a localization graph1359between the mobiles various positions and the static sen-1360sor nodes. The number of measurements required by the1361mobile is linear to the number of static sensor nodes.1362When the required number measurement to build a rigid1363graph is obtained, an anchor-free localization (AFL) algo-

    1364rithm is run to compute the nodes coordinate. AFL first

    1365computes the initial coordinate assignment of all the nodes1366using only node connectivity information. AFL then uses a1367non-linear optimization procedure to reduce the sum of

    1368squared distance errors between the nodes actual distance1369and the distance of the current coordinate assignment.

    1370Simulation results show that MAL performs better in large

    1371mobile coverage areas. The estimated distance error de-1372creases with the increasing number of nodes.

    13736.2. Synchronization

    1374Time synchronization in a wireless sensor network is

    1375important for routing and power conservation. The lack1376of time accuracy can significantly reduce the networks1377lifetime. Global time synchronization allows the nodes to

    1378cooperate and transmit data in a scheduled manner. En-

    1379ergy is conserved when there are less collisions and re-1380transmissions. In addition, energy is saved when nodes1381are duty-cycled.5 Existing time synchronization protocols1382aim to accurately estimate time uncertainty and synchro-1383nize each nodes local clock in the network. In the following1384subsection, we briefly review a few of these protocols.

    1385Uncertainty-driven approach: Ref. [66] proposes an1386uncertainty-driven approach to duty-cycling by modelling1387long-term clock drifts between nodes to minimize the1388duty-cycling overheads. This approach uses long-term1389empirical measurements to evaluate and analyze three1390key parameters that influence long-term synchronization.

    1391The parameters are synchronizing rate, history