EISLAB A Sensor Network Architecture - DiVA portalltu.diva-portal.org/smash/get/diva2:990852/FULLTEXT01.pdf · This thesis investigates the feasibility of using a heterogeneous sensor

LICENTIATE T H E S I S

Department of Computer Science, Electrical and Space EngineeringEISLAB

A Sensor Network Architecture

for Mobile Users

Chen Zhong

ISSN: 1402-1757 ISBN 978-91-7439-263-0

Luleå University of Technology 2011

Chen Z

hong A Sensor N

etwork A

rchitecture for Mobile U

sers

ISSN: 1402-1757 ISBN 978-91-7439-XXX-X Se i listan och fyll i siffror där kryssen är

A Sensor Network Architecturefor Mobile Users

Chen Zhong

Dept. of Computer Science, Electrical and Space EngineeringLulea University of Technology

Lulea, Sweden

Supervisors:

Jens Eliasson & Jerker Delsing

Printed by Universitetstryckeriet, Luleå 2011

ISSN: 1402-1757 ISBN 978-91-7439-263-0

Luleå 2011

www.ltu.se

To Jing

iii

iv

Abstract

Recent developments in sensor technology have resulted in the feasibility of deployingsensors on a human user in order to sense physiological properties such as body temper-ature, pulse and posture. This enables monitoring of a user’s health condition and stresslevels.

Some professions, such as first responders, are often operating in dangerous situations,which causes increased stress levels. Since stress can lead to poor performance with injuryor even death as a result, it would be very beneficial to remotely monitor the personnelto find those with critical body status and evacuate them from the dangerous area. Thiswould increase work safety and also allow for more efficient training. Another categoryof potential users for remote monitoring of health conditions are patients and elderly,who could enjoy a higher quality of life with the increased safety of remote monitoring.As a result, it is highly motivated to find a technology which is applicable for monitoringboth a number of collaborative users, i.e., a team, and as well as a single user. Oneinteresting solution is to use a wireless sensor network (WSN). Another technology forremote monitoring is a personal area network (PAN). A PAN usually consists of consumerdevices, which enables it to communicate with existing infrastructure networks throughthe use of standardized protocols and technologies.

This thesis investigates the feasibility of using a heterogeneous sensor network archi-tecture for the remote monitoring of a group of highly mobile users. The architecture is acombination of a WSN and PANs, which benefits from the mesh networking capabilitiesin WSN and the support for standardized protocols in PAN. The usage of the WSNincreases the network reliability and provides mobility and additional service (e.g. local-ization) to each user. The wearable PANs provide infrastructure network access pointsto sensor nodes, and hence support the mobility of the entire sensor network (by reusinguser’s mobile phone). Experimental results show that the proposed network architectureis a competitive solution for the targeted application, and I believe that this researchwork has the potential to drastically improve the life quality of a very large number ofusers, such as elderly and patients.

v

vi

ContentsPart I 1

Chapter 1 – Thesis Introduction 31.1 Previous Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.3 Thesis Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61.4 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

Chapter 2 – Sensor Networks 72.1 Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . 82.2 Personal Area Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

Chapter 3 – A Heterogeneous Sensor Network 153.1 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163.2 Network Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173.3 Performance Evaluation Metrics . . . . . . . . . . . . . . . . . . . . . . . 20

Chapter 4 – Summary of Papers 234.1 Paper A - A cluster-based localization method using RSSI for heteroge-

neous wireless sensor networks . . . . . . . . . . . . . . . . . . . . . . . . 234.2 Paper B - A Heterogeneous Sensor Network Architecture for Highly Mobile

Users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234.3 Paper C - Evaluation of a Heterogeneous Sensor Network Architecture for

Highly Mobile Users . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

Chapter 5 – Conclusions and Future Work 255.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

References 27

Part II 31

Paper A 331 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 Application scenario and requirements . . . . . . . . . . . . . . . . . . . 384 RSSI-Distance Characterization . . . . . . . . . . . . . . . . . . . . . . . 40

vii

5 Testbed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 417 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

Paper B 471 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492 Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513 Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 586 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 587 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

Paper C 631 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 652 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 674 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 705 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 766 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 767 Acknowledgment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

viii

Acknowledgments

I would like to thank my supervisor, assistant professor Jens Eliasson for his guidance onmy research. Moreover, Jens Eliasson also helped me to get used to the life in Sweden.I would also like to express my thanks to associate professor Evgeny Osipov for beingmy assistant supervisor. His advices are very useful and helpful for my research. I spe-cially thank guest Ph.D student Fan Zhang and research engineer Henrik Makitaavola inEISLAB for their knowledge sharing. Furthermore, I would like to thank Johan Borg forhelping me to handle miscellaneous issues. I also appreciate the interesting discussionswith Andrey Kruglyak and Rumen Kyusakov. Their suggestions are very inspiring formy research.

Finally, I would like to thank my wife, my parents and my grand mother for theirsupport and understanding.

ix

x

Part I

1

2

Chapter 1

Thesis Introduction

Sensor networks [1] capable of collecting distributed information have attracted more andmore interests in academia and industry. A sensor network consists of a number of sensornodes which are equipped with radios to form networks, and a gateway which is usedto connect the sensor nodes with external infrastructure, such as the Internet. Sensornodes can be deployed in various locations, such as buildings, machines in a factory anda natural environment. A sensor network can also be deployed on a human body in orderto collect health information, such as heart beat rate, body temperature, posture andlocation [2]. Such information indicates, for example, a patient’s health, a sportsman’sperformance or a firefighter’s status. However, in this type of applications, an end user ofcollected information is usually far away from the person being monitored. For examplehealth information about a patient can be transmitted to a hospital, allowing the patientto stay at home while still being monitored. Thus sensor data need to be forwarded toan infrastructure network, such as the Internet.

Sensor networks using general purpose communication technologies for wireless com-munication to monitor a single user is called a personal area (sensor) network (PASN).Fig. 1.1 shows the setup of a PASN on a user. Besides distributed sensor nodes, a gate-way is used to forward sensor data from a sensor node to an infrastructure network. InFig. 1.1, the gateway is connected to a mobile phone access network. The mobile phoneaccess network provides wide coverage, making it suitable for highly mobile applications.This approach allows remote monitoring even in rural locations, such as forests.

1.1 Previous Research

Remote monitoring of a user using wireless sensors has attracted a great deal of attentionfrom both the academia and the industry in the last decades. For example, the westerncountries are seeing an aging population with increased health care costs as a result.To reduce the need for an elderly (or patient) to occupy hospital resources, a personalarea sensor network, sometimes referred to as a body area network (BAN), can be usedto continuously monitor the user’s health parameters when the user is located at home.

3

4 Thesis Introduction

Figure 1.1: Personal Area Sensor Network with a mobile phone operator service network, theInternet and a remote user

This topic has attracted many research efforts.In [3], Delsing et al. introduced an Embedded Internet System (EIS) architecture.

They analyzed potential methods to access the Internet and chose a Bluetooth equippedmobile phone as an Internet access point for sensor nodes which are used to capture user’shealth information. They presented the implementation of the software and hardwarefor the architecture, and showed the result of a heart rate monitoring application. TheEIS architecture is a basis for the sensor network architecture discussed in this thesis.

Jovanov et al. presented that the system architecture of a body area network could bedivided into three levels (sensor level, personal server level and medical services) [4]. Forthe first level, the category of sensor information and the requirements for sensor nodesare discussed. Personal server is implemented by using commercial products, such as apersonal digital assistant (PDA), a 3G mobile phone or a computer. The medical serviceslevel is on the top of the architecture, and mainly defines the process and managementservices of collected data of different users. The usage of a computer as a personal serveris only suitable when the activity range of a user is limited. For highly mobile users, thecomputer will prohibit their activities. Thus using a PDA and a 3G mobile phone as apersonal server will be more suitable for mobile users. Moreover, dedicated applicationsoftware is usually needed to run on a PDA, a 3G mobile phone or a computer.

The authors of [5] introduced a health monitoring system for mobile patients. Infras-tructure network which has good coverage, such as 2.5G and 3G networks is essential formonitoring highly mobile users, and the infrastructure network capability is usually theconstraint to the network performance of a mobile monitoring system. The authors ex-perimentally tested the performance of a Universal Mobile Telecommunications System(UMTS) network provided in a specific location, and further discussed the issues thatexisted when using an infrastructure network for mobile monitoring.

Another interesting category of applications of body area networks is remote moni-toring of a group of first responders, such as firefighters. Such professions usually require

1.2. Motivation 5

personnel enter a hazard environment as a result, the application situation is much com-plex than a patient monitoring. RUNES (Reconfigurable Ubiquitous Networked Embed-ded Systems) IST project [6] focused on the solutions to an emergency application. Themain scenario this project considered is road tunnel fire where firefighters will be sentafter an incidence occurs. Challenges for communications and networking in the scenario,and motivations and issues for health monitoring of users in presence are analyzed.

In [7], Yuce et al. presented a sensor network system for remote monitoring of multiplepatients. This system used the medical band for the communication between sensornodes and a base station which forwards sensor data to a computer via RS232 protocol.Further, this computer sends sensor data to a remote computer via the Internet or anetwork. In order to support the sharing of the communication channel for multiplepatients, a CSMA/CA media access control protocol is implemented. In this solution, acomputer is used as a gateway to access to the Internet or a network. It will significantlyrestrict the mobility of users. Therefore, their solution is suitable for patient monitoringbut not suitable for remote monitoring of a group of highly mobile users.

Many sensor node platforms have been developed for the real-world demonstration ofsensor networks. Shimmer [8] is a typical example which provides supports of both Blue-tooth and IEEE 802.15.4 radios. TCP/IP for IEEE 802.15.4 is supported by Shimmer,while TCP/IP over Bluetooth is not supported.

1.2 Motivation

Previous research has well studied both single user monitoring and multiple users mon-itoring. A sensor network can use a computer as a gateway to access infrastructurenetworks (e.g. the Internet). Dedicated software program need to be implemented andrun on the computer. The obvious disadvantage of the usage of a computer as a gatewayis that the mobility of a user is highly restricted. In order to improve the users’ mobility,a PDA or a mobile phone is suggested to be used as a gateway to access the infras-tructure networks. As a result, all devices can be carried by the user and the mobilityof the user is much higher. However, in most current research, the dedicated softwareprogram is still inevitable. For example, the user may need to run a program on a PDA.Alternatively, if a mobile phone is used as the gateway, different programs may need tobe implemented for different mobile phones and even worse, not all of the mobile phonesare capable of running of a program. An ideal solution to that issue is to make thesensor node intelligent to avoid the installation and running of the program on a PDAor a mobile phone. The sensor node will do the smart work to access the infrastructurenetworks, while the user’s device (PDA and mobile phone) keep non-intelligent. As aresult, the mobility of the sensor network can be supported. Another issue motivates thisresearch is the mobility of an individual user and the reliability of the sensor network.There are multiple users in the targeted applications. Some user’s infrastructure net-work connectivity might fail due to an unexpected reason (e.g. mobile phone runs out ofpower) during their operation. It is highly demanded that the sensor node deployed onthis user is capable of forwarding sensor data to another access point so that the health

6 Thesis Introduction

information of the user can be resumed through another user’s mobile phone. Further-more, the network architecture needs to support each user’s mobility since the monitoredusers are highly mobile during an operation. Thus it is highly motivated to find out anetwork architecture and an corresponding wearable platform to meet the requirementsof (access points and users’) mobility, reliability and interoperability.

1.3 Thesis Objective

The general research question this thesis addresses is formulated as:Is it feasible to use a sensor network for remote monitoring of a group ofhighly mobile users?More specific, the general research question contains these two aspects:

• Q1: Is it possible to support the mobility for the monitored group ofusers?There are two issues of users’ mobility to be addressed. First, the sensor networkshould support the mobility of the entire group. Second, the sensor network shouldsupport the mobility of each user in the group. As a result, the first mobility issuerequires that all sensor nodes are wearable and at least one sensor node is capable ofcommunicating with user’s mobile phone. The second mobility issue requires thatsensor nodes deployed on different users should be able to form mesh networkingrapidly and easily even the users are highly mobile.

• Q2: Can the reliability of the network be increased?

Reliability is a common issue for many sensor networks as sensor nodes using radiosfor communication. Signal propagation can be easily affected by environment, e.g.walls, the ground and human bodies. Another issue contributing to the reliabilitychallenge is the mobility of sensor network users. Though the mobile phone accessnetwork can cover a very large area, a group of users may reach an area where thequality of mobile phone operator service is poor or even completely not available.

1.4 Thesis Outline

Chapter 2 introduces different types of sensor networks. Chapter 3 presents the re-quirements on the sensor network, and the corresponding design of a sensor networkarchitecture. Chapter 4 summarizes the papers. Chapter 5 presents the conclusions andfuture work. Finally, three papers are appended.

Chapter 2

Sensor Networks

Figure 2.1: A Wireless Sensor Network

There various sensor networks exist, such as wireless sensor networks, personal area sensornetworks, Bluetooth sensor networks, body area networks and mobile sensor networks.Sensor networks usually use wireless communication technology to collect distributedinformation, and different types of sensor network are designed to suit different classes ofapplications. This thesis focus on the discussion of wireless sensor networks and personalarea (sensor) networks since the proposed network architecture is based on these twonetwork types.

7

8 Sensor networks

Figure 2.2: The basic components of a sensor node

2.1 Wireless Sensor Networks

A wireless sensor network (WSN) consists of a large number of low power, low cost sensornodes. As shown in Fig. 2.2, the central component of a sensor node is a microcontroller(MCU). It is equipped with a broadcast radio capable of forming a multi-hop meshnetwork, memories, a power unit and sensors. WSNs can be used for many applications,such as environmental monitoring [9], and military applications [10]. WSNs are usuallyconnected to an existing network infrastructure, such as the Internet, to centralize andpresent sensor data to remote users. Fig. 2.1 illustrates a general application topologyof a wireless sensor network, including a database server on the Internet which can beaccessed by end users.

Figure 2.3: Routing in a Wireless Sensor Network

2.1. Wireless Sensor Networks 9

2.1.1 Research areas

Although much progress has been achieved on the research of wireless sensor networks,there still a large number of challenges that need to be addressed [11].

• Routing Wireless sensor networks mainly target large scale applications, thus, aWSN can comprise thousands up to millions sensor nodes. Finding an efficient andreliable route to forward sensor data from a far away sensor node to a gatewayis a challenging problem [12]. Fig. 2.3 shows an example to explain the routingproblem. After sensor nodes have been deployed in an area of interest, they mustdiscover suitable paths, or routes, to gateways. Gateways are used to forward sensordata from the WSN to an external infrastructure network, usually the Internet [13].Sensor nodes, (indicated by dashed circle) might need to use multi-hop transmissionto reach a network gateway (the solid black node) as the sensor node is out of singlehop transmission range of the gateway. In order to find a particular path betweena sensor node and a gateway, routing algorithm need to be run in a WSN. Fig.2.3 depicts two paths to forward sensor data from the sensor nodes, s1 and s2 tothe gateway which is located on the top of the figure. The computation of routesdepends on routing protocols [14] which can be classified into different categories,such as flat or hierarchical. It should be noticed that many routing protocols thatperform well in simulations also may fail in the real-world applications [11]. Thisis because simulators can not capture the influence of the environment and thecomplexity of a real-world application.

Figure 2.4: Localization

• Power Consumption Power consumption is another very challenging problem forWSN [15]. Sensor nodes are usually powered by batteries. Once they are deployed,it is not practical to charge or replace the batteries, while many applications wouldrequire a long term operation. For example, environment monitoring [16] mayneed to collect data for a period which could be in the range of months up to years.Sensor nodes which frequently forward packet for other sensor nodes would be likelyto run out of power earlier comparing with idle nodes. For example, as depicted inFig. 2.3, there are two routes for source node s1 to send data packet to a gateway

10 Sensor networks

which is located on network edge, while there is one route for source node s2 tosend data packet. If both s1 and s2 use route g - h - i -j - k - l - m, these nodes(g,h,i,j,k,l and m) will consume more power than other sensor nodes. To prolongthe life time of nodes g,h,i,j,k,l and m, routing protocol can be designed in a waythat when source node s1 realizes route g - h - i -j - k - l - m has traffic load for nodes2, s1 will switch to route a - b - c - d - e -f. In such way, the power consumptionis distributed and allocated among different sensor node averagely. Therefore, thewhole network’s life time can be extended. Moreover, power consumption can alsobe the focus of routing protocol design [17]. Another potential solution to prolongthe life time of a wireless sensor network is energy scavenging [18]. Other energies(e.g. solar and vibration) can be converted into electricity, and hence the batteryof the sensor nodes can be charged. This solution relies on the availability of theenergy sources and the energy conversion efficiency.

• Localization Localization for WSNs [19] has attracted more and more researchefforts since the position information of sensor nodes is very useful for some ap-plications. For example, in environment monitoring, sensor node detects certainevent, such as presence of pollutant, can issue an alarm message to a user. It is notsufficient for the user only get notified. The location of this event is also importantto know as responding actions can be carried out to handle the event. Sensor nodesare deployed either in a controlled order or randomly. For former approach, thelocation of sensor nodes is preassigned, and thus location information is known inadvance, while for the latter approach, sensor nodes’ location is derived by a certainmethod. Global position system (GPS) is a mature technology and GPS sensorscan be equipped to sensor nodes to get their positions. However, there are severalrestrictions [20]. Another type of method to acquire sensor nodes’ positions arecomputing positions based on packet transmission among sensor nodes. Assumea packet is received by a sensor node in a free space, the received signal strength(RSS) which implies the distance between a sending node and a receiving node canbe derived. Theoretically, the RSS is decreasing when the transmission distanceis increasing. Thus, the distance can be estimated based on RSS in an appropri-ate mathematical model [21]. Furthermore, localization algorithms can use thesedistances as input to calculate the specific position for any sensor nodes. Fig. 2.4shows a localization scenario. Node a, b, c and d are deployed as anchors, andthus their coordinates are known in advance. To acquire the location of node t, itis programmed to broadcast several messages. The anchors will derive RSSs locallyonce they receive these messages, and centralize all RSSs to a computer. With theusage of RSSs, the computer will use a Distance - RSS model to estimate the dis-tances between each anchor and the node t. Finally, the coordinates of t is derivedby running a localization algorithm [22] in the computer, which use the distancesas inputs. Such method is applicable for both 2D localization (targeted node andall anchor nodes are always in the same plane) and 3D localization (targeted nodeand all anchor nodes are deployed in a 3D space). The advantage of this approachto acquire position is it does not require additional hardware, and the RSS can be

2.1. Wireless Sensor Networks 11

derived from any packets transmitted without requiring dedicated packet transmis-sion. The drawback of such method is the positioning estimation is not accuratedue to the unpredictable influence of environment to signal propagation [23]. Therealso other localization methods exist, such as the usage of ultrasound. They requirededicated hardware implementation and consume the power of sensor nodes. Theadvantage of these methods is that the localization result is usually more precisecomparing with the method using the RSS of the received packets.

Figure 2.5: Real-time networking

• Real-time This challenge for WSNs becomes more and more important [24]. Forsome application in which sensor node are static, sensor data are not required toreach user in a limited time period. For example, in environment and biologyresearch, sensor data might do not need to be presented to a user in a short periodas it is not urgent. However, for some other kind of applications, especially securitymonitoring, the event/alarm message should reach a user as fast as possible sincethe emergency might need to be handled immediately. Sensor data usually cannot reach the gateway in a single hop range since sensor nodes can be deployedin a wide area (large network topology). As a result, the delay increases duringpacket’s multihop transmission. Besides, the choice of forwarding path (networklayer) can also affect the packet delay. For example, in Fig. 2.5, there are tworoutes for sensor node s1 to send data to the gateway. Route a - b - c - d is onehop shorter than route e - f - g. If sensor node s2 is idle, s1 can use route e - f -g which is capable of forwarding packet to the gateway in less time (ignore otherinfluences, such as the interference and distance). Now, assume node s2 starts totransmit packets. If s1 still uses route e - f - g which is a short route for nodes2, both s1 and s2 may fail to reach the gateway in an acceptable time intervaldue to the sharing of route e - f - g. One solution to such failure is to make s1use the route a - b - c - d. Although this route is one hop longer than e - f - g,the competition between s1 and s2 to use the same forwarding path is avoided.More important, the packet delivery delay for both s1 and s2 may be reduced.Such intelligence is very useful to improve the network performance. It can be

12 Sensor networks

achieved if a sensor node can detect its link quality (e.g. packet delay) with thenext hop sensor node so that the sensor node can choose a best relay node from allavailable neighboring nodes [25]. However, the difficulties of designing the routingprotocol are also increased. Another issue should be considered is how to balancethe network performance between real-time and power consumption. Transmittingsensor data as fast as possible may cause high power consumption. For example,if a packet only carries one sample, though the sample can reach its destinationwith a small delay since the packet can be sent once it is ready, each sample causesone transmission in such method. Alternatively, if several samples are aggregatedinto one packet, the power consumption will be reduced since the number of packettransmissions is reduced [26]. However, the latency for a sample will be larger sincea packet needs to assemble several samples. Thus, the tradeoff between real-timeand power consumption needs to be made according to the requirements of thetargeted application.

Figure 2.6: A Personal Area Network

2.2 Personal Area Networks

Personal area network (PAN) is another type of wireless networks. It consists of fewdevices which are usually off-the-shelf products, such as mobile phone, personal digi-tal assistant and laptop. These devices have more powerful processors comparing withWSN’s sensor nodes which usually use a microcontroller as central component. Largerbattery is used to power PAN’s device, and even though the working time of PAN’s de-vice is still much shorter than WSN’s sensor node. Therefore, the battery of PAN deviceneeds to be recharged frequently. On the other hand, a PAN can provide much higherdata rate. For example, PAN using Bluetooth technology [27] for communication canprovide a data rate of up to 2 Mbps, while WSN using IEEE 802.15.4 standard [28] can

2.2. Personal Area Networks 13

provide a data rate of 250Kbps if the 2.4 GHz physical layer is used. The performanceof Bluetooth used for PAN has been well studied [29].

Fig. 2.6 depicts the network setup of a PAN and its connection with infrastructurenetworks. A user can use a mobile phone to connect to a laptop to exchange files or thelaptop can access the Internet via the mobile phone if the mobile phone can connect toa 3G network (telephone operator service). As an extension of PAN, sensors can alsobe included in PAN to constitute a personal area sensor network (PASN). For example,acceleration sensors integrated in user’s mobile phones can sense user’s motion. Personalarea sensor network is a suitable solution to monitor single human body, such a patient[30]. Specially designed platform which uses the same communication protocol with theoff-the-shelf products can also be used as networking device in a PASN. Such platform canbe equipped with various sensors which can capture different human body information.

14 Sensor networks

Chapter 3

A Heterogeneous Sensor Network

This thesis focuses on the research of remote monitoring of a highly mobile group of users.The motivation of such research is to ensure users’ safety during their mission operationin a hazardous environment. For example, a firefighter team may enter a fire scene, and aremote command center wants to monitor the firefighters’ health information (e.g. pulse,body temperature, remaining oxygen and posture) during their operation. One solutionto achieve this purpose is to use wireless sensor networks. Fig. 3.1 shows the scenariowhere many sensor nodes are deployed on all users. Health information can be capturedby the sensors on the nodes, and sensor data are sent to a remote command center via aninfrastructure network, for example, GPRS, 3G networks and the Internet. Some otherapplication examples which have similar scenarios include:

• policeman team

• assault team

• military group

A sensor network can also be deployed to remotely monitor a single user, such as a patientand an elderly, and the solution of multiple users monitoring can also be applicable forsingle user monitoring. This thesis focuses on the research of remote monitoring of agroup of highly mobile users. For this type of applications, their characteristics aresummarized as:

• Large scale A large number of sensor nodes need to be deployed to cover allmonitored persons. Thus the network scale is very large comparing with a singlebody area network.

• Hazardous working environment The monitored targets (persons) may enterhazard environment, such as a fire scene. The users might be injured, and inorder to report such incidence, a corresponding alarm message needs to reach thecommand center within a small delay.

15

16 A Heterogeneous Sensor Network

Figure 3.1: Remote monitoring of a group

• Highly mobile users Many sensor networks deployed in real-world is aimed tomonitor environments which are static. However, the monitored target of thisresearch is a group of users on an operation. A team is sent out to accomplish theoperation, and the operation location is arbitrary. Thus, the network and the usersare mobile.

In the following content of this chapter, the networking requirements for the targeted ap-plications and network architecture to address corresponding requirements are presentedrespectively.

3.1 Requirements

Due to the purpose and conditions of the targeted applications, different requirementsare placed on the design of the sensor network architecture. Below are the most criticalrequirements:

• Support highly mobile usersAs mentioned before, a monitored team is highly mobile. Therefore the networkdevices which are deployed on their bodies should not prohibit users’ high mobility.Furthermore, the sensor network needs to be able to connect to an infrastructurenetwork, such as a mobile phone network to gather data to a remote commandcenter.

3.2. Network Architecture 17

• Manage to access to infrastructure networksDue to the mobility of users, sensor nodes usually can not reach users’ commandcenter directly. One solution addressing this problem is to use an existing infrastruc-ture network, such as a mobile phone network and the Internet. An ideal solutionis to use mobile phone network to collect data and store the data in database onthe Internet.

• Reliable performanceFor environmental monitoring, the loss of sensor data is not critical. But for emer-gency monitoring, such as the posture of a person when he/she has fallen down, theloss of an alarm message is not acceptable. So the network performance especiallyfor message delivery service needs to be guaranteed.

3.2 Network Architecture

A straightforward solution to achieve remote monitoring of multiple users is to use awireless sensor network to collect massive distributed information, and a dedicated per-sonal server to forward data to a mobile phone network. Fig. 3.2 depicts the networkarchitecture of such solution.

Figure 3.2: WSN with a personal server

3.2.1 Background

The main challenge of this WSN which consists of homogeneous sensor nodes is to find anefficient route for a packet to reach a sink (the personal server) which could be deployedin an opposite side of the network. Furthermore since the monitored targets are highlymobile, the network topology will keep changing constantly. This increases the difficulties


to send each packet to the sink as a new forwarding path might need to be established afterthe change of the network topology. One solution to address the routing issue is to usecluster based routing protocols, and the hardware for a cluster head node can be differentfrom a regular sensor node (node heterogeneity), as more tasks can be accomplished on acluster head, such as sensor data aggregation and multihop transmissions between clusterheads to a sink [31].

Figure 3.3: Sensor network connects to users’ mobile phones

Users’ mobile phones can also be reused in the applications of multiple users monitor-ing. Fig. 3.3 illustrates a network architecture where users’ mobile phones are integratedinto a sensor network, and connected with sensor nodes via a Bluetooth module. Sensornodes are deployed in a manner of clusters. Each cluster corresponds to a user. Onesensor node in a cluster is equipped with a Bluetooth module and it connects to user’smobile phone. Thus it can act as a gateway node for other sensor nodes accessing theinfrastructure network.

However, users’ mobile phones could fail (e.g. break or run out of battery) as theymay enter hazardous environment. Fig. 3.4 shows a scenario where a user’s mobile phoneservice is not available during mission operation.

3.2.2 A heterogeneous sensor network architecture

In order to continuously receive this user’s information, sensor nodes on this user need tobe able to send packets to another gateway node that connects to a usable mobile phone.One solution is to enable mesh networking between nodes (including sensor nodes andgateway nodes), so that the packets can reach other clusters. The IEEE 802.15.4 radio isa suitable technology for such purpose as it supports the mesh networking topology andthe mobility of the users As a result, a gateway node is equipped with both a Bluetoothmodule and an IEEE 802.15.4 module. In each cluster, the gateway node and user’s

3.2. Network Architecture 19

Figure 3.4: A sensor network using heterogeneous radios

mobile phone constitute a personal area network, while all the sensor nodes and gatewaynodes in all of the clusters form a wireless sensor network which supports multihoptransmission in case a packet needs to be forwarded to another cluster. The advantageof this architecture is that all sensor data can reach the infrastructure network (e.g. 3Gnetwork) via local PAN without the need of multihop transmission among sensor node,and in the case of local gateway failure, user’s data can reach another gateway to resumethe accessibility of an infrastructure network. As a result, the reliability of sensor networkis increased.

Benefiting from the mesh networking among the sensor nodes, the indoor localizationservice for all users becomes feasible. The sensor nodes deployed on each user form a clus-ter which is regarded as a basic element in the localization since the desired informationfor the command center is each user’s location, other than a single sensor node. This isan important feature as the distance between two users can be estimated via the packettransmissions between different sender nodes and receiver nodes which are deployed onthese two users respectively. If a certain number of users (at least 3 for 2D localizationand 4 for 3D localization) located at the boundary of a group are treated as anchors,the positions of other users in this group can be calculated by running a localizationalgorithm. The distances between tracked users and anchor users are used as inputs forthe algorithm. As mentioned before, the localization information is very important forthe targeted applications, such as firefighter monitoring since injured firefighter needs tobe located for fast rescue. The advantage of such localization method is that it doesnot require additional hardware (free service) and only relies on the mesh networkingbetween sensor nodes.

Sensor networks can be heterogeneous in terms of, for example, energy heterogeneity


and radio heterogeneity [32]. The proposed sensor network architecture can also beclassified as a heterogeneous sensor network but in terms of radios.

The key feature of this architecture is it reuses available resources to create local sinksfor many sensor nodes, and it also supports mesh networking to increase the reliability ofa network and provide indoor localization service. Moreover, there is no specific softwareprogram needs to be implemented and run on the mobile phones in this architecturesince the wearable gateway node is intelligent and hence, the mobile phones do not needto perform any intelligent work. In traditional solutions, a dedicated software programusually needs to be implemented and run on a computer or a PDA.

3.3 Performance Evaluation Metrics

In order to guarantee the quality of service of the proposed network architecture, theperformance needs to be evaluated. First, evaluation metrics need to be selected accord-ing to the requirements of the targeted applications. The performance of the proposedsensor network architecture can be evaluated in the following metrics:

• End-to-end delayThe targeted applications in this thesis require sending alarm message. Thus oneof the most important metrics to evaluate the network performance is end-to-enddelay which includes the time from packet sent by a sensor node to packet storedin a server on the Internet.

• Packet reception ratioPacket reception ratio (PRR) is a very typical metric to evaluate the link qualitybetween a transmitter and a receiver. Packet can be lost during the transmissionbetween a sensor node and a gateway node in different kinds of sensor networks.

• Network scalabilityNetwork scalability indicates the number of sensor nodes that can work simultane-ously in a sensor network. The element of proposed heterogeneous sensor networkarchitecture is a cluster which consists of a number of sensor nodes and a gatewaynode. It is interesting to know how many sensor nodes can be supported by agateway node in a cluster.

Since an emergency message is supposed to reach a command center in a reason-able delay, this thesis measured the end-to-end delay with different experiment setups.Network scalability is also investigated as one gateway node is supposed to support anumber of sensor nodes in a cluster. This thesis simulated the traffic load generated bysensor nodes in Matlab, and simulation results show that a gateway node is capable ofsupporting a reasonable number of sensor nodes in a cluster. Paper C appended to thisthesis presented the evaluation results of the network architecture. The derived measure-ment shows that the one hop (in the WSN) end-to-end (from a sensor to a database onthe Internet) delay of a packet with minimal payload length (5 bytes) is around 300 ms.

3.3. Performance Evaluation Metrics 21

This value increases to around 500 ms when the payload length is 114 bytes (the maximalavailable length in the IEEE 802.15.4 MAC layer). If the number of transmission hopsin the WSN is increased by one, the end-to-end delay for packets with both minimal andmaximal payload lengths will slightly increase. The simulation results shows that if everysensor node generates samples in a pattern of normal distribution (mean 30 and standarddeviation 8) and the payload length is 5 bytes, a gateway node can support 4 and 5 sensornodes on a user with nearly no packet loss and very few packet losses respectively. Theseresults imply that the for the proposed network architecture, the packet delay will bemainly attributed to the resource constrained gateway node since the number of trans-mission hops in mesh network is expected to be small and the delay for an additionalhop is also very small according to the results of the two hop transmission (in the WSN)experiments. The number of sensor nodes deployed on a user is limited by the capabilityof the gateway node as well (only consider the circumstance of one hop transmission inthe WSN) since the bandwidth of the WSN exceeds the upload capability of the gatewaynode. Resolutions to the bottle neck of the network architecture could be the increasingof the gateway node capability and the implementation of intelligent functionalities (e.g.data aggregation).


Chapter 4

Summary of Papers

4.1 Paper A - A cluster-based localization method

using RSSI for heterogeneous wireless sensor net-

works

Authors: Chen Zhong, Jens Eliasson, Henrik Makitaavola and Fan Zhang

Published: presented at 6th International Conference on Wireless CommunicationsNetworking and Mobile Computing (WiCOM)), Chengdu, China, September 2010

Summary: This paper is dedicated to the research of a cluster-based in-door localizationsystem for a specific heterogeneous sensor network. The received signal strength of apacket transmitted using the IEEE 802.15.4 standard is used in this method. In order tointerpret the received signal strength to a distance between a transmitter and a receiver, amodel that reflects the relationship between the received signal strength and the distanceis established based on experimental measurements. The localization target in this paperis a cluster which comprises of a number of sensor nodes. The number of sensor nodesthat act as anchors in a cluster affect the accuracy of the localization. The influence ofthe number of packets that are sent for localization to the accuracy of positioning is alsoinvestigated. Moreover, this paper experimentally demonstrates that the sensor nodewhich is in the highest position of a cluster achieves least estimation error.

4.2 Paper B - A Heterogeneous Sensor Network Ar-

chitecture for Highly Mobile Users

Authors: Jens Eliasson, Chen Zhong and Jerker Delsing

Published: presented at Sixth IEEE Conference on Wireless Communication and SensorNetwork (WCSN-2010), Allahabad, India, December 2010

23

24 Summary of Papers

Summary: In this paper, a heterogeneous sensor networks architecture consisting ofWSNs and PANs is presented. The requirements of monitoring a group of mobile usersinclude accessibility to the Internet, mobility for the users and reliability. The proposedarchitecture and its relevant hardware and software are introduced in this paper. Theend-to-end delay for sending a packet from a sensor node to a database on the Internet ismeasured. Finally, the memory usage of the gateway which combines WSNs and PANsfor the proposed heterogeneous sensor network architecture is measured in this paper.

4.3 Paper C - Evaluation of a Heterogeneous Sensor

Network Architecture for Highly Mobile Users

Authors: Chen Zhong, Jens Eliasson, Jerker Delsing and Rumen KyusakovPublished: Journal paper, accepted in Communications and Network, Scientific Re-search PublishingSummary: In this paper, experimental evaluation of a heterogeneous sensor network ar-chitecture is presented. The experiment setup which includes hardware, an infrastructurenetwork, software and test cases are introduced. The performance of the infrastructurenetwork which is a 3G network is evaluated by measuring round trip time between alaptop and a web server. The end-to-end delay of the data collection chain is measuredwhen using different experiment setups: one hop to send data from sensor node to gate-way node and two hops to send data from sensor node to gateway node. The influence ofpayload length in a transmitted packet to the total end-to-end delay is also investigated.Network scalability is studied to investigate the maximum number of sensor nodes thatcan be supported by a gateway node. Finally, the memory usage of a gateway node ispresented.

Chapter 5

Conclusions and Future Work

5.1 Conclusions

A sensor network architecture using multiple radios is proposed as a feasible solutionto the targeted applications. Bluetooth radio is used to connect a gateway node and auser’s mobile phone, so that sensor data can reach a 3G network as long as the user islocated in an area covered with the 3G network. Therefore, the mobility of the entiresensor network can be supported, and this is the answer to the group mobility issue of theresearch question Q1 presented in Chapter 1. The IEEE 802.15.4 standard is used for thecommunication between sensor nodes and gateway nodes. Thus mesh networking betweensensor nodes is supported by the proposed architecture, and therefore, the mobility ofeach user in a team can be supported and the reliability of the sensor network is increased.This feature is the answer to the user mobility issue of the research question Q1 and theanswer to the research question Q2 as well.

Performed experiments show that the proposed architecture can achieve mobility,interoperability and reliability for remote monitoring of highly mobile users. Measure-ment results are derived based on specific hardware and infrastructure network choice,but the result can be generalized to the sensor networks which use comparable resourcesconstrained hardware and a typical 3G network since all communication protocols usedin this architecture are standardized (the IEEE 802.15.4 and Bluetooth) and there isno particular requirements on the hardware. Therefore, the proposed sensor networkarchitecture is a general and feasible solution to the target applications.

5.2 Future work

The future work of this thesis consists of four parts. First, the scalability of the proposedsensor network architecture needs to be investigated since packet forwarding capabilityfrom the gateway node to a mobile phone network is limited. Second, in order to preventmalicious users from interference of the deployed sensor network, security features need

25

26 Conclusions and Future Work

to be investigated, and they would add even more requirements to the nodes whichalready have many existing tasks. Third, in order to increase the reliability, routingmechanism for mobile networks needs to be investigated. Finally, it is interesting to usethe 6LoWPAN with IPv6 on the sensor network since IP-all-the-way will further increasethe interoperability of the sensor network.

References

[1] C.-Y. Chong and S. P. Kumar, “Sensor networks: evolution, opportunities, andchallenges,” in Proceedings of the IEEE, vol. 91, Aug. 2003, pp. 1247 – 1256.

[2] M. A. Hanson, H. C. Powell, A. T. Barth, K. Ringgenberg, B. H. Calhoun, J. H.Aylor, and J. Lach, “Body area sensor networks: Challenges and opportunities,” inComputer, vol. 42, Jan. 2009, pp. 58 – 65.

[3] J. Delsing, P. Lindgren, and A. Ostmark, “Mobile Internet enabled sensors usingmobile phones as access network,” in Electronic Journal of Information Technologyin Construction. 9, Special Issue Mobile Computing in Construction, 2004, pp. 381– 388.

[4] E. Jovanov, A. Milenkovic, C. Otto, and P. C. de Groen, “A wireless body areanetwork of intelligent motion sensors for computer assisted physical rehabilitation,”in Journal of NeuroEngineering and Rehabilitation, vol. 2, Mar. 2005.

[5] A. van Halteren, R. Bults, K. Wac, D. Konstantas, I. Widya, N. Dokovsky, G. Ko-prinkov, V. Jones, and R. Herzog, “Mobile Patient Monitoring: The MobiHealthSystem,” in The Journal on Information Technology in Healthcare, vol. 2, 2004, pp.365 – 373.

[6] RUNES, “Reconfigurable ubiquitous networked embedded systems,”http://www.ist-runes.org/, 2011.

[7] M. Yuce, P. Ng, C. Lee, J. Y. Khan, and W. Liu, “A wireless medical monitoringover a heterogeneous sensor network,” in 29th Annual International Conference ofthe IEEE Engineering in Medicine and Biology Society, 2007., Lyon, Aug. 2007, pp.5894 – 5898.

[8] “Wireless Sensor Platform for Wearable Applications,” http://www.shimmer-research.com/, 2011.

[9] A. Mainwaring, J. Polastre, R. Szewczyk, D. Culler, and J. Anderson, “Wireless sen-sor networks for habitat monitoring,” in ACM International Workshop on WirelessSensor Networks and Applications (WSNA’02), Atlanta, GA, USA, Sep. 2002, pp.88 – 97.

27

28 References

[10] N. Alsharabi, L. R. Fa, F. Zing, and M. Ghurab, “Wireless sensor networks of battle-fields hotspot: Challenges and solutions,” in 6th International Symposium on Mod-eling and Optimization in Mobile, Ad Hoc, and Wireless Networks and Workshops,2008. WiOPT 2008., Berlin, Apr. 2008, pp. 192 – 196.

[11] J. A. Stankovic, “Research challenges for wireless sensor networks,” in ACMSIGBED Review - Special issue on embedded sensor networks and wireless com-puting, vol. 1, New York, USA, Jul. 2004, pp. 9 – 12.

[12] P. R. Pereira, A. Grilo, F. Rocha, M. S. Nunes, A. Casaca, C. Chaudet, P. Almstrom,and M. Johansson, “End-to-end reliability in wireless sensor networks: survey andresearch challenges,” in EuroFGI Workshop on IP QoS and Traffic Control, 2007.

[13] D. Christin, A. Reinhardt, P. S. Mogre, and R. Steinmetz, “Wireless Sensor Networksand the Internet of Things: Selected Challenges,” in Proceedings of the 8th GI/ITGKuVS Fachgesprach Drahtlose Sensornetze, 2009.

[14] J. Al-Karaki and A. Kamal, “Routing techniques in wireless sensor networks: asurvey,” in Wireless communications, IEEE, vol. 11, Dec. 2004, pp. 6 – 28.

[15] N. Zaman and A. B. Abdullah, “Energy efficient routing in wireless sensor network:Research issues and challenges,” in IEEE International Conference on Intelligenceand Information Technology, Aug. 2010.

[16] G. Barrenetxea, F. Ingelrest, G. Schaefer, and M. Vetterli, “Wireless Sensor Net-works for Environmental Monitoring: The SensorScope Experience,” in 2008 IEEEInternational Zurich Seminar on Communications, Mar. 2008, pp. 98 – 101.

[17] C. Schurgers and M. B. Srivastava, “Energy efficient routing in wireless sensor net-works,” in Military Communications Conference, 2001. MILCOM 2001. Commu-nications for Network-Centric Operations: Creating the Information Force. IEEE,vol. 1, 2001, pp. 357 – 361.

[18] E. M. Yeatman, “Energy scavenging for wireless sensor nodes,” in 2nd InternationalWorkshop on Advances in Sensors and Interface, 2007. IWASI 2007, Bari, Jun.2007, pp. 1 – 4.

[19] G. Mao, B. Fidan, and B. D. Anderson, “Wireless sensor network localization tech-niques,” in Computer Networks, vol. 51, Jul. 2007, pp. 2529 – 2553.

[20] F. Benbadis, T. Friedman, M. de Amorim, and S. Fdida, “GPS-free-free positioningsystem for wireless sensor networks,” in Second IFIP International Conference onWireless and Optical Communications Networks, Piscataway, NJ, Mar. 2005, pp.541 – 545.

References 29

[21] C. Zhong, J. Eliasson, H. Makitaavola, and F. Zhang, “A cluster-based localiza-tion method using rssi for heterogeneous wireless sensor networks,” in 6th Interna-tional Conference on Wireless Communications Networking and Mobile Computing(WiCOM), Chengdu, China, 2010, pp. 1 – 6.

[22] A. Savvides, C.-C. Hand, and M. B. Strivastava, “Dynamic fine-grained localizationin ad-hoc networks of sensors,” in Proc. 7th international Conference on MobileComputing and Networking, Italy, 2001, pp. 166 – 179.

[23] E. Elnahraway, X. Li, and R. P. Martin, “The limits of localization using RSS,” inSenSys ’04 Proceedings of the 2nd international conference on Embedded networkedsensor systems, Baltimore, MD, USA, 2004, pp. 283 – 284.

[24] Y. Li, C. Chen, Y. Song, and Z. Wang, “Real-time QoS support in wireless sensornetworks: A survey,” in 7th IFAC International Conference on Fieldbuses and Net-works in Industrial and Embedded Systems (FeT’07), Toulouse, France, Nov. 2007.

[25] A. A. Ahmed and N. Fisal, “A real-time routing protocol with load distributionin wireless sensor networks,” in Computer Communications, vol. 31, Sep. 2008, pp.3190 – 3203.

[26] L. Krishnamachari, D. Estrin, and S. Wicker, “The impact of data aggregation inwireless sensor networks,” in 22nd International Conference on Distributed Comput-ing Systems Workshops, Washington, DC, USA, Nov. 2002, pp. 575 – 578.

[27] Bluetooth SIG, “Bluetooth special interest group,” http://www.bluetooth.org, 2011.

[28] “IEEE standard for information technology–telecommunications and informa-tion exchange between systems–local and metropolitan area networks specificrequirements–,” http://standards.ieee.org/getieee802/download/802.15.4-2003.pdf.

[29] R. A. Rashid and R. Yusoff, “Bluetooth Performance Analysis in Personal AreaNetwork (PAN),” in RF and Microwave Conference, Putra Jaya, Sep. 2006, pp. 393– 397.

[30] E. Jovanov, D. Raskovic, J. Price, J. Chapman, A. Moore, and A. Krishnamurthy,“Patient monitoring using personal area networks of wireless intelligent sensors,” inBiomedical Sciences, vol. 37, 2001, pp. 373 – 378.

[31] X. Du and F. Lin, “Improving routing in sensor networks with heterogeneous sensornodes,” in Vehicular Technology Conference, 2005. VTC 2005-Spring., vol. 4, 2005,pp. 2528 – 2532.

[32] M. Yarvis, N. Kushalnagar, H. Singh, A. Rangarajan, Y. Liu, and S. Singh, “Exploit-ing heterogeneity in sensor networks,” in INFOCOM 2005. 24th Annual Joint Con-ference of the IEEE Computer and Communications Societies. Proceedings IEEE,vol. 2, Mar. 2005, pp. 878 – 890.

30 References

Part II

31

32

Paper A

A Cluster-based LocalizationMethod using RSSI for

Heterogeneous Wireless SensorNetworks

Authors:Chen Zhong, Jens Eliasson, Henrik Makitaavola and Fan Zhang

Reformatted version of paper originally published in:6th International Conference on Wireless Communications Networking and Mobile Com-puting (WiCOM)). IEEE Communications Society, 2010

c© 2010, WiCOM 2010, Reprinted with permission.

33

34

A cluster-based localization method using RSSI for

heterogeneous wireless sensor networks

Chen Zhong, Jens Eliasson, Henrik Makitaavola

Abstract

In this paper, we investigate the performance of cluster-based localization using receivedsignal strength indicator (RSSI). The proposed solution is designed to meet the require-ments of monitoring of firefighters or similar applications. The empirical relationshipbetween signal strength and distance is determined using experiment data. One of themost popular localization algorithms found today, Min-Max, is used for our testbeds.Our solution is implemented in TinyOS and experimentally evaluated on a Mulle v5.2IEEE 802.15.4 platform. The aim of our research is to develop a heterogeneous wirelesssensor network consisting of inter connected body area networks, or clusters. Using lo-calization, the network’s robustness and reliability, as well as the safety of its users, canbe improved.

1 Introduction

Localization is one of the most important services in the field of wireless sensor networks(WSNs). Firstly, localization is necessary when implementing some applications, such aswater quality monitoring and indoor air quality monitoring [1], as it might be inadequateto only collect sensor data without the knowledge of its corresponding location. Secondly,knowing the relative position of sensor nodes allows the use of geography-based routingprotocols, which can improve the performance of a network in terms of reliability andlow power consumption [2]. The Global Positioning System is a very popular locationestimation system, however it only works outdoors where it can receive the satellites’signals. However, many WSN applications are running indoors which prohibits the us-age of GPS. Alternatively, methods using other technologies, e.g. ultrasound [3], canachieve high accuracy in positioning, but each device adds to the sensor node size, cost,and energy consumption (which is one of the most critical issues in WSNs). Comparingwith mentioned localization approaches, the advantages of using received signal strengthindicator (RSSI) based localization are quite apparent. Technically, it can be used bothinside of a building or outdoors. Furthermore, no additional hardware is required sincewireless sensor nodes are equipped with radio transceivers. The challenges of localiza-tion based on RSSI are mainly due to its unpredictable propagation, especially in indoorenvironment. The reflections of a signal against walls, floors and ceilings may result insevere multi-path interference at the receiving antenna, as shown by Zanca et al. [4],

35

36 Paper A

with increased errors as a result.

Figure 1: Body Area Network overview

The target application is monitoring of firefighters in order to improve their safetyin hazardous working environments. The goal of our research is to develop an architec-ture for heterogeneous wireless sensor network consisting of a number of inter connectedsmall body area (sensor) networks (Fig. 1). The proposed localization approach will beused for improving overall network connectivity as well as locating firefighters in need ofassistance.

The rest of the paper is organized as follows: Section 2 discusses related work. Theapplication scenario and requirements are presented in Section 3. Experiments and resultsare described in Section 6. In the last section, conclusions are drawn and future workare outlined.

Figure 2: Cluster-based localization in a heterogeneous wireless sensor network

2. Related Work 37

2 Related Work

Considerable research has been carried out in the area of localization based on RSSI. Stoy-anova et al. [5] performed a theoretical analysis and experimental evaluation of impactfactors on received signal strength (RSS) accuracy, and Sugano et al. [6] implemented anindoor localization system on a Ubiquitous Device. Besides the environment where thesensor nodes are deployed, radio characteristics, such as antenna type and size, channelfrequency, modulation and transmit power, can influence the signal propagation drasti-cally as well. For instance, the Ubiquitous Device uses a relatively large antenna whichimproves the radio performance, but make the deployment of sensor nodes inconvenientin the real world because of its large size.

In [7] Elnahraway et al. tried to experimentally demonstrate the limits of a varietyof RSS-based localization algorithms. They used commodity IEEE 802.11 technologyand concluded that the limitations are fundamental. Comparing with that, Zanca et al.[4] investigated the real-world performance of some well known localization algorithmsusing the IEEE 802.15.4 standard. According to their findings, RSSI-based localization inindoor environments presents severe limitations, but they also suggested that an accurateradio channel model might alleviate the problems mainly introduced by the environment.

Generally, current research suggests that RSS-based localization is a promising tech-nique but is currently suffering from high inaccuracy. However, this situation can beimproved by analyzing a specific application scenario, and adjusting a localization algo-rithm towards its specific requirements. Using this approach it is possible to devise alocalization method which takes advantage of specific characteristics that general meth-ods can not.

One of the most popular localization algorithms is the Min-Max [8]. Compared toother algorithms, it is not complicated to implement. Complex functionalities runningon a sensor node, such as localization, can provide a significant contribution to higherlevel applications. However, a sensor node typically has very limited resources in termsof computation power, memory, and energy. Therefore, it is crucial that any functioncan be implemented using small resources as possible. In [9], localization based on themaximum likelihood (ML) method has been presented. The ML estimation of a target isobtained by using the minimum mean square error (MMSE). The main idea of MMSE isthat if the distance between an anchor and the target is known, then the target must layon a circle centered at the anchor and with a radius equaling to the distance. At leastthree distances with different anchors are needed.

The localization algorithm investigated in this paper is the Min-Max due to its rela-tively good performance and low resource requirements. For each anchor, it will constructa bounding square box. The center is the anchor itself and the edges is the double of theestimated distance with the target. Assume there are three anchors which are distributedin 2D manner. As a result, three corresponding boxes and the intersection of these boxescan be obtained. The position of a target is defined as the center of intersection of theboxes. In [4], Zanca et al. compared the 2D Min-Max with other algorithm, such as ML.Their research was however general and thus lacking application characteristics. In [10],

38 Paper A

Parker et al. used the Min-Max localization for vehicle positioning based on the IEEE802.11p standard. They used the algorithm in pure 2D experiments.

3 Application scenario and requirements

The goal of this project is to increase safety for personnel, e.g. police officers, assaultteams, and firefighters, working in hazardous environments. In order to achieve this goal,it is necessary to be able to remotely monitor the personnel participating in an operation.One main issue to deal with is that this types of operations are highly mobile, e.g. takingplace in forests, on roads, in buildings, and in tunnels. Therefore, it is important thatthe communication infrastructure is wireless and has extremely good coverage. Today,the only means of achieving this is by using commonly available infrastructures, such asthe mobile telephone network (3G, GPRS) and the Internet. One approach is to equipevery sensor node with a GPRS module. This is however an expensive solution. Anotherissue is that it is desired that each user can be equipped several sensors in order tocollect different types of information. The use of cables is prohibited due to the fact thatcables can conduct heat and might interfere with the user’s movement. The approachadopted here therefore to use several sensor nodes communicating wirelessly with eachother and the gateway using low cost, short range radios. The gateway can be either aPDA, mobile phone, or some specialized device. All nodes located on a user form a bodyarea network (BAN, cluster), which collect and forward sensor data to the gateway. Fig.1 shows one example of a BAN consisting of three sensor nodes (one node also acts ascluster head) and mobile phone gateway. Another advantage when using a mobile phoneis that a user’s standard mobile phone can be reused as infrastructure access point. Thisapproach eliminates the need to carry additional hardware and reduces the cost. Also,the possibility when the mobile phone connection in a cluster is lost is considered. In suchcircumstance, the BAN will try to route its sensor data to another BAN in its vicinity.Then, the network topology will become a mesh network. Localization can assist suchrouting when mobile phone failure occur and thus improve the network reliability.

Sensor nodes within a BAN can acquire information about its user, such as whena user falls down, and position, temperature, pulse, posture, etc. When an anomaly isdetected, an alarm message must be sent to the network’s infrastructure and collectedat a command center. It might be vital for the command center to know the locationfor each person. If an emergency would happen to a person, actions to assist this personshould be taken as soon as possible. As a result, using localization can accelerate thesearch procedure and save time for following measures. Thus, it is motivated to explorethe indoor localization. One significant characteristic for this application is that a clusterwhich contains several sensor nodes should be located. Fig. 2 shows an example of awireless sensor network consisting of five body area networks (clusters). Every sensornode within the same cluster can estimate the distance between itself and neighboringclusters. However, usually not all sensor nodes in a cluster can estimate the distanceprecisely due to the degradation of the signal propagation due to the environment. It iseven worse that the nodes might report quite different distances although actually they

3. Application scenario and requirements 39

are near each other. The contribution to this issue can be the propagation of signal andthe orientation of the antenna. In the experiments considered in this paper it was foundthat the latter contribution was one of the most critical challenges. For instance in Fig.2, if the orientation of anchor nodes, a, b, c is different, the RSSIs between the Targetnode in cluster 4 and the anchors, a, b, c in cluster 1, can be rather dissimilar. The focusis not on the relationship between the antenna orientation and received signal strengthsince the users are mobile and antenna orientation will constantly change. Therefore, thehighest RSSI out of all sensor nodes’ RSSIs in a cluster was chosen to reflect the distancebetween the target node and other clusters. The performance of the localization methodusing different configurations, such as, the number of anchors in a cluster, and the usedamount of RSSI(s) from different anchor(s) in the same reference point (cluster) werestudied.

Figure 3: 10 cm above the ground

Figure 4: 80 cm above the ground

40 Paper A

4 RSSI-Distance Characterization

Before performing localization experiments, it was essential to find a suitable model forthe relationship between the measured RSSI value and the distance between nodes. Inorder to achieve this, a number of packets were sent when the target and anchor nodeswere separated with different distances. The average of all RSSI values were taken asthe measured RSSI value for a certain distance. These experiments are performed indifferent heights above the ground, 0.1 m and 0.8 m. Results are shown in Fig. 3 andFig. 4 respectively.

Based on obtained experimental data and using logarithmic curve fitting, the rela-tionships between measured signal strength (dBm) and distance (m) can be expressed inthe following way:

10 cm above the ground

s = −9.73681 ln d − 43.3088 (1)

80 cm above the ground

s = −10.14731 ln d − 42.5258 (2)

The formulas are plotted as a straight line in logarithmic coordinates. As it can beseen in Fig. 3 and Fig. 4, the RSS values distribute quite close to the fitting line in theoccasion of 0.1 m above the ground, while in comparison, the RSS values spread slightlyaround their fitting line in the 0.8 m case. In the following sections, these relationshipsare used to estimate the distances from the target to all anchor nodes.

Figure 5: Test bed

5 Testbed

The experiments have been performed in a specially designed testbed. The testbed isbased on a coordinate system consisting of four reference points. The x, y coordinates

6. Experimental results 41

for each test position and anchor are depicted in Fig. 5. Near each reference point, weuniformly deployed three sensor nodes as anchors. Nearby anchors constitute a cluster.The fourteen vertical and six horizontal small squares stand for the positions where thetarget node is placed and at each of these location, it broadcasts ten localization requestpackets. On the receiver side, the anchors might receive all these ten packets, and thusretrieve the RSSIs for the calculation of corresponding location. In order to investigatethe influence of aptitude above the ground in positioning, each time, the target was placedin several different heights above the ground. Three setup positions (heights) of targetnode on a body were emulated, head, belly and knee. The heights (Z axis) assumed forthese three positions are 2 m, 1m and 0.5 m respectively.

Figure 6: Mulle v5.2, IEEE 802.15.4 sensor node

6 Experimental results

Our localization solution is evaluated on the Mulle v5.2 IEEE 802.15.4 platform fromEistec AB [11], which uses an Atmel AT86RF230 [12] transceiver. The experimentalresults are obtained based on running Min-Max algorithm. The influence of RSS-Distancemodel’s parameters, the number of localization request packets and the number of anchorsin a cluster on positioning accuracy were analyzed respectively. To evaluate the over-all performance among different configurations, some estimation error means were alsocalculated.

• Estimation Error - RSS-Distance Model First of all, the effect of RSS-Distancemodel’s parameters to estimation error is studied. As mentioned, two models tocharacterize the relationship between RSS and distance at two different heightsabove ground (0.1 m and 0.8 m) were derived. However, in Fig. 7, it is shownthat the performance of localization seems to not rely on the model. For differentnumber (one, two and three) of anchor nodes in each reference point (cluster) used,the error distributions are nearly the same for different RSS-Distance models. Thusthe localization system is almost independent of models and the 0.8 m above groundRSS-Distance model will be adopted for further analysis.

• Estimation Error - Localization Request Packet Used Amount Secondly,the influence of the number of RSSI measurement packets used for localization

42 Paper A

Figure 7: Five localization request packets and different number of anchors used in each referencepoint (cluster) with different signal propagation model

Figure 8: Three anchors in each reference point, different number of RSSI measurement packetsused in each localization experiment

is investigated. Originally, there are ten packets broadcasted by a target. Dueto environmental interference to signal propagation, some packets are lost, and ina certain test, only six packets can be received by a anchor node. As a result,a maximum of six RSSI measurement packets are available to compute the meanvalue of estimated distance. Therefore, the influence of packet number to estimationerror can be depicted by six curves as in Fig. 8. In this plot, it is assumed thatthree anchor nodes participate in the positioning calculation. As it can be seen,with more localization request packets used, the errors can be smaller in certaintest positions, such as # 60. But in contrast, in some other positions, more packetsused in calculation leads to worse performance, e.g. test position # 50. Basically,the general positioning performance when different number of packets are used iscomparable.

6. Experimental results 43

To further evaluate the over-all performance, the estimation error means based ondifferent number of used RSSI measurement packets are depicted in Fig. 9. Threeanchor nodes in each reference point participate calculation. As it can been seen,the best performance is achieved when three localization request packets are used.Even though the differences when different number of packets are used are rathersmall. More packets used in the calculation do not improve the result. Thus,localization system can achieve acceptable results even without many localizationrequest packets broadcasted. In the practical leverage, the target might merelyneed to broadcast three to five localization request packets. This is an obviousadvantage since the traffic load can be reduced in networking.

Figure 9: Mean values of error using different amount of RSSI measurement packets when threeanchor nodes in each reference point participate calculation

Figure 10: Errors at different height when using 6 RSSI measurement packets and 1 anchor ineach reference point

• Estimation Error - Number of Involved Anchor Nodes in Each Cluster(Reference Point) During the experiment, we realized that the RSSI differ fromeach other even the receiver Mulles are deployed quite close to each other. The

44 Paper A

Figure 11: Errors at different height when using 6 RSSI measurement packets and 2 anchorsin each reference point

Figure 12: Errors at different height when using 6 RSSI measurement packets and 3 anchorsin each reference point

main contribution to such behavior can be the orientation of antennas betweentransceivers. Thus we also explore if it is necessary to deploy several anchor nodes(three anchors in our experiment) with different antenna orientation. Fig. 10, 11and 12 illustrate the performance of positioning assuming one, two and three an-chor node(s) is or are retrieving RSSI in each reference point (cluster) respectively.According to the target’s real-world amplitudes, the corresponding positioning er-rors can be sorted into three groups, red (target is 0.5 m above ground), green (1m) and blue (2 m). Therefore, the influence of target’s height (Z axis) on a bodyis shown in these pictures as well. Basically, positioning is slightly improved byadding anchor nodes into each reference point (cluster). And the position estima-tion is better when the target is set on head comparing with on knee and belly. Benoticed, the number of RSSI measurement packets used in these calculations is six.

Fig. 13 illustrates the estimation error means for the three error groups against thenumber of anchor nodes that participate localization in each reference point. Forthe estimation group when the target is deployed on head, the results are much

7. Conclusion 45

Figure 13: Error means at different height and using different number of anchors in eachreference point with six RSSI measurement packets

better than other groups. Generally, more Mulles participating positioning willimprove the result slightly.

7 Conclusion

In this paper, a cluster-based localization method has been presented. The method wasevaluated using a new IEEE 802.15.4 based platform, the Mulle v5.2, in a large scalespace: 40 m ∗ 20 m ∗ 6.5 m. The proposed approach is designed to draw from theadvantages of a cluster-based sensor network where the signal strength is obtained byhaving several sensor nodes within a body area sensor network (cluster) collaborate. Thenetwork topology consists of a number of inter connected clusters, where each cluster isa body area network. Each BAN is worn by a firefighter in order to remotely monitorhis health state and position. By using the cluster approach, a number of devices withdifferently aligned antennas are equipped on each firefighter. This enables mitigation ofthe effects of multi-path and obstacles. The proposed localization method is based on acluster-based version of the Min-Max algorithm, which also has the advantage of elimi-nating the need to transmit a large number of localization request packets. Experimentshave shown that different postures of target user will lead to different estimation errors.Potentially, by adding a posture detecting functionality and combining the knowledge ofposture with RSSI values, it is expected that the performance of localization method canbe further increased. The next step is to investigate if the approach can be used in 3Dapplication scenarios. Tests have shown that the approach can achieve an average errorof around 5 meters, which for the present application is sufficiently accurate.

References

[1] N. Patwari and A. O. H. III, “Using proximity and quantized rss for sensor localiza-tion in wireless networks,” in Proceedings of the 2nd ACM international conference

46

on Wireless sensor networks and applications, September 2003.

[2] J. N. Al-Karaki and A. E. Kamal, “Routing techniques in wireless sensor networks:A survey,” in Wireless Communications, IEEE, vol. 11, December 2004, pp. 6–28.

[3] N. B. Priyantha, A. Chakraborty, and H. Balakrishnan, “The cricket location-support system,” in Proceedings of the 6th annual international conference on Mobilecomputing and networking, 2000, pp. 32–43.

[4] G. Zanca, F. Zorzi, A. Zanella, and M. Zorzi, “Experimental comparison of rssi-based localization algorithms for indoor wireless sensor networks,” in Proceedings ofthe workshop on Real-world wireless sensor networks, Glasgow, Scotland, 2008, pp.1–5.

[5] T. Stoyanova, F. Kerasiotis, A. Prayati, and G. Papadopoulos, “Evaluation of impactfactors on rss accuracy for localization and tracking applications,” in Proceedings ofthe 5th ACM international workshop on Mobility management and wireless access,New York, 2007, pp. 9–16.

[6] M. Sugano, T. Kawazoe, Y. Ohta, and M. Murata, “Indoor localization systemusing rssi measurement of wireless sensor network based on zigbee standard,” inThe IASTED International Conference on Wireless Sensor Networks (WSN 2006),Banff (Canada), July 2006.

[7] E. Elnahraway, X. Li, and R. P. Martin, “The limits of localization using rss,”in Proceedings of the 2nd international conference on Embedded networked sensorsystems, Baltimore, MD, USA, 2004, pp. 283–284.

[8] X. Nguyen and T. Rattentbury, “Localization algorithms for sensor networks usingrf signal strength,” in University of California at Berkeley, Tech. Rep., May 2003.

[9] A. Savvides, C.-C. Hand, and M. B. Strivastava, “Dynamic fine-grained localizationin ad-hoc networks of sensors,” in Proc. 7th international Conference on MobileComputing and Networking, Italy, 2001, pp. 166–179.

[10] R. Parker and S. Valaee, “Robust min-max localization algorithm,” in Proc. of theIEEE ITSC 2006, Toronto, Canada, September 17-20 2006.

[11] “Eistec AB.” http://www.eistec.se/, 2010.

[12] “AT86RF230.” http://www.atmel.com/, 2009.

Paper B

A Heterogeneous Sensor NetworkArchitecture for Highly Mobile

Users

Authors:Jens Eliasson, Chen Zhong and Jerker Delsing

Reformatted version of paper originally published in:Sixth IEEE Conference on Wireless Communication and Sensor Networks (WCSN-2010)

c© 2010, WCSN 2010, Reprinted with permission.

47

48

A Heterogeneous Sensor Network Architecture for

Highly Mobile Users

Jens Eliasson, Chen Zhong and Jerker Delsing

Abstract

Wireless sensor networks and personal area networks are two relatively new and emerg-ing technologies, capable of addressing a large number of applications such as homeautomation, medical monitoring, and sports monitoring. This article presents a new sen-sor network architecture suitable for emerging applications that require a highly mobilegroup of persons, such as fire fighters or assault teams, to be monitored in real-time. Thearchitecture combines features from WSN, such as mesh routing and efficient communi-cation, with the benefits of using a PAN architecture communicating with standardizedBluetooth profiles and the TCP/IP protocol suite. The use of Bluetooth enables theproposed architecture to use a user’s mobile phone in order to achieve Internet connec-tivity, and IEEE 802.15.4 to create an internal mesh network. This dual-radio approachis necessary since today’s mobile phones lack support for radio technologies traditionallyutilized by WSNs. Performed tests indicate that the approach can support dependabledelivery of sensor data to the Internet from a group of highly mobile users.

1 Introduction

Personal area networks (PANs) and wireless sensor networks (WSNs) are two promisingwireless communication technologies for present and future applications. A PAN is com-prised of a small number of battery powered devices connected together using suitablewireless technologies. Traditional PAN devices, which are usually standard consumerdevices such as: laptops, smart phones, and PDAs, have relatively high processing capa-bilities and can therefore communicate using general-purpose technologies and protocols.A personal area (sensor) network (PASN) is an extension to PANs where small sensornodes are introduced. PASN nodes must communicate using the same protocols andtechnologies as the PAN they connect to. One kind of PASN that is normally used forapplications such as patient monitoring and sports monitoring is a body area network(BAN) [1].

Wireless sensor networks normally consist of a large number of low cost, low powersensor nodes, or motes, which are more specialized than PASN devices. WSN sensornodes are usually very resource constrained; i.e. they have low computation power, lowstorage capacity, and limited energy storage capabilities [2]. The use of broadcast radiosenables formation of multi-hop (mesh) networks, as shown in Fig. 1. As a result, sensordata can be forwarded from one node to another until it reaches dedicated gateways, or

49

50 Paper B

sinks, located at the network’s edge. A WSN can cover a large area with a large number ofnodes, which makes it an efficient solution to a number of applications such as: industrialcontrol, environmental monitoring, and battlefield surveillance [3, 4]. However, the multi-hop communication approach used in WSNs also exhibits a major challenge: efficientrouting. Mobile sensor networks makes the routing issue an even greater challenge.

Previous research on the BAN area has mainly targeted applications with singleusers, such as medical monitoring and sports monitoring. For such applications, theBAN network topology has a low complexity with a limited number of devices. However,some applications contain a large number of highly mobile users, e.g. firefighters, assaultteams, or soldiers. In order to monitor the activities of an entire team, using only onesingle BAN is not a viable solution. Therefore, one straightforward solution for suchtype of applications is to equip every individual team member with its own BAN. Onecommon issue with the types of applications described above is that missions are carriedout in hazardous environments, and often in rural locations. This introduces additionalrequirements on the gateway technology used since it must guarantee coverage even in themost rural places. Using the mobile telephone access network is one attractive solutionsince it has the best coverage today. The use of the Bluetooth technology [5] combinedwith standard mobile phones as gateway is a technique that was suggested by Ostmarket al. [6], and gives several advantages such as low cost since user’s own mobile phone canbe re-used as gateway and thereby eliminating the need of carrying additional devices.

However, today’s mobile phones lack support for radio technologies typically used byWSNs, e.g. IEEE 802.15.4. This precludes them from acting as gateways directly. Inthis article, a heterogeneous sensor network architecture combining the best from PANand WSN technologies is presented. The architecture is based on using a sensor nodeequipped with dual radios as a gateway: a Bluetooth module is utilized to communicatewith the Internet using a standard mobile phone, and a broadcast-enabled radio is usedto form a mesh network to efficiently route data from one sensor node to another whenthe access network is unavailable.

Figure 1: Wireless sensor network

2. Related work 51

2 Related work

Single personal/body area networks monitoring a person’s physical state, used in appli-cations such as remote medical care, sports monitoring has attracted interest both fromacademia and the industry in recent years. Quach et al. showed in [7] an implemen-tation of a prototype wireless health monitoring system. They used the IEEE 802.15.4standard to transmit human physiological condition information to a base station node,which was USB-cable connected to a PC or PDA. Sensor information is later stored in aSQL database. In [8], a multi-hop protocol named Wireless Autonomous Spanning treeProtocol (WASP) was presented for BANs as the conclusion was made that direct com-munication between all nodes has a high energy cost (more transmission power needed)or may even be impossible due to the interference from a human body. WASP uses pro-tocol stack cross-layer technology and manipulates medium access control and routinglayers to achieve low end-to-end packet transmission delay and a high packet deliveryratio. Nevertheless, their multi-hop protocol design was mainly targeted to be used witha single body area network. BANs and PANs usually use a star topology which makesthem vulnerable for occasions when the star’s center node fails to operate. In [9], Hansonet al. described the challenges and advantages of a new type of network topology, calledstar-mesh hybrid, where several coordinators cooperate with each other. As a result,a star-mesh hybrid topology increased the reliability of a single BAN by adding morecoordinators and designing relevant failure detection and handling mechanisms.

Several research projects have been targeted fire fighter and police monitoring aswell. In the Fire Information and Rescue Equipment (FIRE) project [10], Wilson et al.developed an architecture to support fire fighters working in hazardous environments.The architecture includes a fixed wireless sensor network situated in a building combinedwith mobile nodes carried by users. Users are also equipped with a small headAmounteddisplay that can show for example floor layouts to guide its user or warning messages toalert its user of danger. In [11], a simulation of an energy-efficient architecture based onad hoc sensors is presented. The architecture, which can be used as a life line to guidefire fighters, uses sensors deployed in a building in an either ad-hoc manner or duringbuilding construction. In [12], Kemp et al. presented a body are network capable ofmonitoring bomb disposal units. The BAN can be used by squad leaders to abort bombdisposal operations in the case where an operator is overheated or shows signs of fatigue.

However, all these architectures above assumes that there is an existing infrastructure,either already integrated in the building or created in an ad hoc manner by the firefightersthemselves. This approach has some drawbacks as the network’s range will be limitedby the number of nodes deployed, with limited mobility as a result.

Dedicated hardware platforms have also been developed in previous research projectsto be used in real-world BAN applications. SHIMMER [13] and BTnode [14] are twowell-known platforms. SHIMMER is a dual radio platform which is capable of commu-nicating with a Bluetooth and IEEE 802.15.4 transceiver. Several sensors are includedon this platform such as accelerometer, electrocardiogram, and a passive infrared sensor.The BTnode also contains dual radios and is equipped with a low-power radio and a

52 Paper B

Bluetooth transceiver. However, both the SHIMMER and the BTnode only supports alimited set of the Bluetooth specification. This prohibits the SHIMMER and BTnodefrom communicating with the Internet without using specialized gateways or middlewareapplications. This is a drawback for mobile sensor networks since users must carry addi-tional hardware or install middleware applications on existing gateway devices in orderto send sensor data to the Internet.

3 Requirements

As stated in Section 1, the architecture presented in the article aims at addressing chal-lenges when monitoring multiple highly mobile persons in hazardous working environ-ments. Examples of such applications can be monitoring of firefighters or assault teams.However, the presented architecture can be used in other scenarios as well, such as mon-itoring of elderly and in security and safety applications.

Figure 2: Mobile Personal area (sensor) network

These type of applications can be identified with the following characteristics:

• Accessibility Usually, a command center for a firefighter team or assault teamcan be located far away from the personnel in the field. As shown in Fig. 2,for a remote command center to access sensor data collected by a highly mobileBAN can be an issue since a typical sensor node’s radio transceiver have limitedcommunication range, up to a few hundred meters. One solution to this issue isto connect BANs to an external infrastructure network where information can beretrieved remotely, e.g. mobile telephone network and the Internet. The mobiletelephone access network has the best coverage. Therefore, the authors suggestthat the available mobile phone access network should be used to enable sensordata to be received by the command center, regardless of the fire fighter’s position.By using the Internet as backbone network, the command center can be mobileitself, and even be located in another country or continent.

4. Architecture 53

• Mobility A sensor network used in the type of applications mentioned earlier hastwo types of mobility. Firstly, the entire network is mobile, i.e. it will be used indifferent locations. Secondly, each user within the network is mobile, so the networktopology will constantly change over time. This type of mobility will require anetwork coverage both in cities and in rural areas. The only available network typethat can handle this today is the mobile phone telephone access network. And byusing the mobile phone access network, standard commercial-off-the-shelf (COTS)devices such as mobile phones and GPRS/3G modems can be used to access theInternet. This approach precludes the need to develop or use some specialized longrange radio technology, which in turn will reduce cost and shorten developmentcycle time. The use of mobile phones as gateways also enables other types ofapplications to benefit from this architecture.

• Reliability Besides the mobility and accessibility challenge, another importantissue that exists in the targeted applications is reliability. As described above,with the usage of mobile phones, each BAN can independently access availableinfrastructure in order to deliver sensor data, i.e. status and alarm messages, tothe Internet. However, the connection between a BAN and the infrastructure canfail due to various reasons. For example, a fireman equipped with a PAN and amobile phone entering a fire scene may encounter unstable GPRS or 3G servicein such harsh environments. Another risk arises from infrastructure access deviceitself since this device can fail as well. Therefore, it is not possible to guarantee thatevery BAN will be constantly connected to the infrastructure network. Therefore, itis vital that backup mechanisms can take over and mitigate the loss in performance,and hence safety, when the connection to external infrastructure is lost. This canbe achieved by using a mesh network and allow other PAN’s gateways to alsoforward sensor data for each other. That way, several gateways may malfunction,but the network will still be able to deliver alarms and status information betweena command center and the firefighters in the field.

Mobility and reliability are two of the most critical issues in the targeted applications.Other issues that should be considered during the design phase is that the sensor nodesshould have a relatively low cost and low power consumption. The two latter requirementsare also found in other types of applications, such as health care, and security and safety.In the next section, a new architecture is proposed that can address all challenges andissues outlined above.

4 Architecture

To achieve mobility and accessibility for each BAN, the approach used in this workto connect a BAN with an infrastructure gateway is to use the Bluetooth technology.Bluetooth-equipped networked sensor nodes can achieve good interoperability with con-sumer devices, and have lower power consumption and cost than Wi-Fi. Bluetooth isalso by far the most wide-spread technology supported by existing consumer devices (e.g.

54 Paper B

computers, mobile phones, and PDAs), which further makes it an interesting technologyto use for personal area (sensor) networks. The Mulle platform from Eistec AB, is theresult from extensive research performed by Lulea University of Technology. The Mullev3.2 platform features a Mitsumi WML-C46 AHR Bluetooth 2.0 module.

To address the reliability challenge, one straightforward solution is to equip everysensor node with a GPRS module. This enables sensor nodes to access mobile phonenetwork independently. Thus, the reliability of a BAN is increased. However, this solutionhas a high cost and has a relatively high energy consumption. A low cost and lowpower consumption solution proposed in this paper to address the second challenge isenabling IEEE 802.15.4 standard [15] based mesh networking among BANs (Fig. 3).Theoretically, both Bluetooth and IEEE 802.15.4 can be used to constitute BANs [16].However, if Bluetooth is adopted, BANs are generally isolated with each other sinceBluetooth technology has very limited support for multi-hop mesh networks. Comparingwith that, IEEE 802.15.4 standard is specially designed for ad-hoc networking, whichmeans that when a BAN’s sink node fail, the IEEE 802.15.4 capable sensor nodes in thisBAN can detect such failure and forward its data to another sensor node in a BAN inthe vicinity, as illustrated in Fig. 3. In such way, a BAN can resume its connection toa command center after the failure of its infrastructure connection. Table 1 summarizesthe characteristics and differences between Bluetooth and IEEE 802.15.4.

Figure 3: Body area mesh network

For the targeted applications in this paper, hierarchical routing [17] is a suitablemethodology since sensor nodes are regularly organized and distributed in a BAN (orcluster manner). Hierarchical routing is a promising approach for point-to point routingwith a minimal routing state, which suits the proposed usage scenario well. A BAN inthe network can be identified as a cluster, which is the bottom level and basic unit ofhierarchical routing. Since the presented platform can communicate both with sensornodes and the infrastructure network, it is also assigned to take the role as cluster head

4. Architecture 55

(BAN coordinator).

Table 1: Bluetooth and IEEE 802.15.4Feature Bluetooth IEEE 802.15.4

Scalability Few devices Many DevicesTransmission Range 10 m to 100 m 200 to 1000 m

Data Rate 2 Mb/s 20 kb/s, 40 kb/s, 250 kb/sPower Consumption Low Ultra Low

Battery Life Short Several YearsNetwork Topology Star Star, Mesh

4.1 Hardware

The base for the dual-radio Mulle is the Mulle v3.1 platform. The Mulle architectureis based on the 16-bit Renesas M16C/62P micro controller. It has 31kB of RAM and384kB flash memory. The CPU runs at 10 MHz, but can be clocked at a maximum of20 MHz using a PLL. However, even with this powerful micro controller, the Mulle onlyconsumes 4 μA in sleep mode. This enables a Mulle to operating for months or evenyears depending on radio and micro controller duty cycles. All Bluetooth-based Mullesuse a Mitsumi WML-C46 Bluetooth 2.0 module.

The dual-radio Mulle’s IEEE 802.15.4 module was chosen to be an Atmel’s AT86RF230,same as on Mulle v5.2. It operates in the 2.4 GHz frequency band and has features forboth the physical layer and MAC layer of the IEEE 802.15.4 specification [15].

4.2 Software

The dual-radio Mulle system’s software consists of a hardware abstraction layer (HAL),IP- and Bluetooth stacks, and an IEEE 802.15.4 radio driver with support for TinyOS-compliant messages.

The lwBT Bluetooth stack used in the proposed architecture is depicted in Fig. 4.lwBT supports the most commonly used protocols, BCSP, BCCMD, HCI, L2CAP, SDP,RFCOMM, BNEP, and PPP. The following profiles are supported: SPP, LAP, DUN,PAN-U, PAN-GN, and PAN-NAP.

Atmel AT86RF230 is an IEEE 802.15.4 compliant radio chip. A low level softwaredriver was developed in order to provide integration with the lwIP [18] stack from SICSwhich is used on the Bluetooth-based Mulle platform. The driver was implemented usingthe standard C programming language. The software design for the driver is depicted inFig. 5.

As described in Section 3, the new Mulle platform must be able to be used as a clusterhead for WSN nodes, and gateway to the Internet for all nodes. As cluster members, thesingle-chip IEEE 802.15.4 platform Mulle v5.2 from Eistec AB [19] is used. The TinyOS

56 Paper B

Figure 4: Light weight Bluetooth stack

Figure 5: Software design for IEEE 802.15.4 driver

[20] operating system, which is a well-known widely used embedded operation system,has also been ported to the Mulle platform. TinyOS is written in nesC [21] which is basedon structural components. Although the software system for the cluster head/gateway isin standard C with lwIP as base, cluster members, i.e. Mulle v5.2, are running TinyOS.This required a driver for TinyOS’s ActiveMessage format to be developed as well andintegrated with lwIP. The resulting architecture enables a Bluetooth-based Mulle to usean IEEE 802.15.4 transceiver simultaneously and thus transfer sensor data to and from amesh network to the Internet via a Bluetooth network with PAN and TCP/IP. Anotherapproach would have been to use IPv6 over Bluetooth and 6LoWPAN over the 802.15.4network. However, the work needed to develop a driver for 6LoWPAN was estimated torequire much more work than implementing ActiveMessage, and the latter approach wastherefore chosen.

4.3 Communication performance

To measure the total communication delay, and thereby the performance, packets weregenerated from a sensor node running TinyOS. The packets were transmitted using thenode’s IEEE 802.15.4 radio to a dual-radio Mulle gateway node. The gateway node copiedthe packet into a TCP send buffer and transmitted the packet over it’s Bluetooth link viaa SonyEricsson K810 mobile phone with 3G support. The packet was finally received by

4. Architecture 57

a server software running on a laptop connected to the Internet via 100 MBit/s Ethernet.At the same time as the TinyOS node transmitted the packet, it also sent one byte overits UART, which also was connected to the laptop. When the byte was received by theserver software, it generated a time stamp which was used as the packet’s transmit time.When the packet was received by TCP, another timestamp was generated and the totaltransmission delay time could be calculated. The UART link baud rate was set to 57600,which would take a single byte less than 0.2 ms to be transmitted. Figure 6 shows thatan average delay for a packet to be transmitted was around 340 ms. One interesting issuehere is that the long transmission delay actually makes the gateway node the network’sbottle neck, since it’s uplink speed in slower that the mesh network. This will changethe standard view that a WSN’s gateways are more powerful than it’s sensor nodes, andput new requirements on routing protocols.

Figure 6: Total packet transmission delay

4.4 Memory usage

Running three stacks: IP, Bluetooth, and IEEE 802.15.4, simultaneously will leave arelatively large footprint in terms of memory usage. Table II shows the memory usageof a Mulle v3.2 node. The bare Bluetooth and TCP/IP system requires 112 kB ofcode space, leaving more than 271 kB of program memory available. When the driverfor the Atmel chip is added, the code memory consumption increases to 122 kB. Eventhough Atmel’s AT86RF230 IEEE 802.15.4 chip performs both PHY and MAC layerfunctionality in hardware, the code footprint is less than 10% of the necessary spacefor lwIP and lwBT. The RAM usage is around 19 kB out of the available 31 kB withBluetooth and IP stacks only, and increases to 21 kB when the 802.15.4 stack is added.This still leaves more than 10 kB of RAM memory to be used for other purposes ona Mulle v3.2. Next generation Mulles currently being evaluated are equipped with a

58 Paper B

Table 2: Memory footprintBluetooth + TCP/IP IEEE 802.15.4 Total

RAM (kB) 19.2 1.4 20.6ROM (kB) 112.6 9.7 122.3

Renesas M16C/65 MCU which has 47 kB of RAM, and 512 kB of ROM. As a comparison,the SHIMMER platform, which only supports the Bluetooth Serial Port Profile, uses aMCU with only 48 kB of code memory, and 10 kB of RAM. It is apparently that runninga complex IP stack combined with full Bluetooth PAN requires much more memory.However, the use of TCP/IP and PAN enables direct connection with the Internet withoutthe need of middleware software. When using a mobile phone as gateway it if feasible torun a middleware application. However, when using a standard Bluetooth access point,running a middleware application is not normally possible.

5 Results

The architecture presented in this paper has been experimentally verified using real-worldtests. Two different types of network technologies, one using standard Bluetooth proto-cols and profiles, and the other using IEEE 802.15.4 with TinyOS messages, have beensuccessfully integrated. Memory consumption measurements indicate that the proposedarchitecture is feasible to use even with resource-constrained sensor nodes. Performancetests with up to three sensor nodes and two gateways in a mobile network have beensuccessfully performed. ActiveMessage packets broadcast over the IEEE 802.15.4 net-work can be received, and encapsulated in an TCP/IP packet and sent to a server onthe Internet using a standard Bluetooth-equipped mobile phone. This indicates thatthe proposed architecture is feasible, and can be used when developing mobile sensornetworks. However, the architecture has only been verified with a small wireless sensornetwork. More research is needed to address the scalability issue when a larger networkis used.

6 Future work

Support for 6LoWPAN in the gateway node could further enable mobile IPv6-basedwireless sensor networks to use the proposed architecture as a gateway to the Internet.The authors will therefore continue by investigating the possibilities to include supportfor 6LoWPAN, and hence IPv6, together with the IP and Bluetooth stacks used today.Another issue that must be investigated further is how a robust, fault tolerant, andmobile large-scale mesh sensor network architecture could be integrated into the proposedarchitecture.

7. Conclusion 59

7 Conclusion

This paper has presented an architecture that combines PAN and WSN technologies asthe solution to monitor a group of highly mobile users. The architecture is designed toaddress the mobility and dependability problems inherent in the targeted applications.The use of Bluetooth, with standardized services such as TCP/IP over PAN, enablesthe proposed architecture to use a sensor node as an Internet-gateway with standardconsumer devices such as laptops, PDAs, and mobile phones. The proposed architecturehas been successfully implemented using a Mulle sensor platform. Interoperability withstandard consumer devices using Bluetooth and other sensor nodes running TinyOS hasbeen successfully verified. Performance tests using several sensor- and gateway nodesshow that ActiveMessage packets sent over an IEEE 802.15.4 network can be forwardedto a Internet server through the use of a mobile phone, thus enabling true mobility.

References

[1] M. A. Hanson, H. C. Powell Jr., A. T. Barth, K. Ringgenberg, B. H. Calhoun, J. H.Aylor, and J. Lach, “Body area sensor networks: Challenges and opportunities,”Computer, vol. 42, no. 1, pp. 58–65, 2009.

[2] W. Zhao, M. Ammar, and E. Zegura, “The energy-limited capacity of wireless net-works,” in Sensor and Ad Hoc Communications and Networks, 2004. IEEE SECON2004. 2004 First Annual IEEE Communications Society Conference on, 4-7 2004,pp. 279 – 288.

[3] A. Cerpa, J. Elson, D. Estrin, L. Girod, M. Hamilton, and J. Zhao, “Habitat moni-toring: Application driver for wireless communications technology,” in 2001 ACMSIGCOMM Workshop on Data Communications in Latin America and the Caribbean,April 2001., 2001. [Online]. Available: citeseer.ist.psu.edu/cerpa01habitat.html

[4] N. Alsharabi, L. R. Fa, F. Zing, and M. Ghurab, “Wireless sensor networks ofbattlefields hotspot: Challenges and solutions,” in Modeling and Optimization inMobile, Ad Hoc, and Wireless Networks and Workshops, 2008. WiOPT 2008. 6thInternational Symposium on, 1-3 2008, pp. 192 –196.

[5] “Bluetooth Special Interest Group,” Bluetooth SIG, 2010,http://www.bluetooth.org.

[6] A. Ostmark, L. Svensson, P. Lindgren, and J. Delsing, “Mobile Medical ApplicationsMade Feasible Through the Use of EIS Platforms,” in Proceedings of the 20th IEEEInstrumentation and Measurement Technology Conference, 2003. IMTC ’03, vol. 1,May 2003, pp. 292–295.

[7] B. Quach, M. Balakrishnan, D. Benhaddou, and X. Yuan, “Implementation of in-tegrated wireless health monitoring network,” in WiMD ’09: Proceedings of the 1st

60 Paper B

ACM international workshop on Medical-grade wireless networks. New York, NY,USA: ACM, 2009, pp. 63–68.

[8] B. Braem, B. Latre, C. Blondia, I. Moerman, and P. Demeester, “The wirelessautonomous spanning tree protocol for multihop wireless body area networks,” pp.1–8, 2006. [Online]. Available: http://hdl.handle.net/1854/6937

[9] M. Hanson, H. Powell, A. Barth, K. Ringgenberg, B. Calhoun, J. Aylor, and J. Lach,“Body area sensor networks: Challenges and opportunities,” Computer, vol. 42,no. 1, pp. 58 –65, jan. 2009.

[10] J. Wilson, V. Bhargava, A. Redfern, and P. Wright, “A wireless sensor network andincident command interface for urban firefighting,” in Mobile and Ubiquitous Sys-tems: Networking Services, 2007. MobiQuitous 2007. Fourth Annual InternationalConference on, 6-10 2007, pp. 1 –7.

[11] H. Salam, S. Rizvi, S. Ainsworth, and S. Olariu, “A durable sensor enabled lifelinesupport for firefighters,” in INFOCOM Workshops 2008, IEEE, 13-18 2008, pp. 1–6.

[12] J. Kemp, E. I. Gaura, J. Brusey, and C. D. Thake, “Using body sensor networksfor increased safety in bomb disposal missions,” Sensor Networks, Ubiquitous, andTrustworthy Computing, International Conference on, vol. 0, pp. 81–89, 2008.

[13] “SHIMMER - Sensing Health with Intelligence, Modular-ity, Mobility, and Experimental Reusability,” Intel, 2008,http://docs.tinyos.net/index.php/Intel SHIMMER.

[14] J. Beutel, “Fast-prototyping using the btnode platform,” in Design, Automation andTest in Europe, 2006. DATE ’06. Proceedings, vol. 1, 6-10 2006, pp. 1 –6.

[15] IEEE Standard for Information technology – Telecommunications and informationexchange between systems– Local and metropolitan area networks– Specific require-ments Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer(PHY) Specifications for Low Rate Wireless Personal Area Networks (LR-WPANs).,IEEE 802.15.4 2003, http://standards.ieee.org/getieee802/download/802.15.4-2003.pdf.

[16] L. Yan, L. Zhong, and N. K. Jha, “Energy comparison and optimization of wirelessbody-area network technologies,” in BodyNets ’07: Proceedings of the ICST 2ndinternational conference on Body area networks. ICST, Brussels, Belgium, Belgium:ICST (Institute for Computer Sciences, Social-Informatics and TelecommunicationsEngineering), 2007, pp. 1–8.

[17] K. Iwanicki and M. van Steen, “On hierarchical routing in wireless sensor networks,”in IPSN ’09: Proceedings of the 2009 International Conference on Information Pro-cessing in Sensor Networks. Washington, DC, USA: IEEE Computer Society, 2009,pp. 133–144.

61

[18] A. Dunkels, “lwIP - A lightweight TCP/IP stack,” http://www.sics.se/adam/lwip,2002.

[19] “Eistec AB,” 2010, http://www.eistec.se/.

[20] “TinyOS web site,” 2010, http://www.tinyos.net/.

[21] D. Gay, P. Levis, R. von Behren, M. Welsh, E. Brewer, and D. Culler, “The nesclanguage: A holistic approach to networked embedded systems,” in PLDI ’03: Pro-ceedings of the ACM SIGPLAN 2003 conference on Programming language designand implementation. New York, NY, USA: ACM, 2003, pp. 1–11.

62

Paper C

Evaluation of a HeterogeneousSensor Network Architecture for

Highly Mobile Users

Authors:Chen Zhong, Jens Eliasson, Jerker Delsing and Rumen Kyusakov

Reformatted version of paper accepted for publication in:Communications and Network

c© 2011, Scientific Research Publishing, Reprinted with permission.

63

64

Evaluation of a Heterogeneous Sensor Network

Architecture for Highly Mobile Users

Chen Zhong, Jens Eliasson, Jerker Delsing and Rumen Kyusakov

Abstract

This paper presents experimental results of a heterogeneous sensor network architecture,which is a combination of a wireless sensor network and a personal area network. Theproposed architecture uses the IEEE 802.15.4 standard to transmit sensor data to a sen-sor node which in turn forwards the data using TCP/IP to a database on the Internetvia a Bluetooth-equipped mobile phone and the mobile telephone access network. Theperformance of the entire communication chain is evaluated. First, a 3G network’s per-formance is evaluated by measuring its round trip time for packet transmission. Second,the real-world end-to-end delay between a sensor node and a database server on the In-ternet is measured using two different experimental setups: single-hop transmission andtwo hops transmission. Finally, the proposed architecture’s scalability is estimated in aMatlab simulation using the results of the experiments as a base. The results show thatthe proposed architecture is applicable for small-scale sensor networks used by highlymobile users.

1 Introduction

Sensor networks are a promising technology to gather distributed information about thephysical world and transmit it to the virtual domain. Sensor networks can be dividedinto two main categories: wireless sensor networks (WSNs) and personal area (sensor)networks (PASNs). Wireless sensor networks normally consist of a large number ofdistributed low-cost, low-power sensor nodes, or motes. Motes use low power microcon-trollers and are equipped with broadcast-enabled radios which enables formation of multihop mesh networks. WSNs are usually used in large-scale, low data rate applications thatrequires monitoring over long periods of time, such as environmental monitoring, defenseand industrial monitoring [1][2]. For dynamic and mobile applications, such as sportsmonitoring, a personal area network (PAN) with additional sensor nodes (PASN) is amore suitable solution. Traditional PANs consist of a small number of consumer deviceswhich have higher processing capabilities compared to a commonly used sensor nodes.For example, mobile phones, personal digital assistants, and computers communicatewith each other using standardized protocols, such as Wi-Fi, Bluetooth and TCP/IP. APASN placed on a human user is sometimes referred to as a Body area network (BAN).

The targeted application class investigated in this paper is the monitoring of a groupof highly mobile users, for example, fire fighters or assault teams. In many cases, these

65

66 Paper C

teams carry out operations in very hazardous environments. Therefore it is beneficialto monitor the health status of the teams during an operation. Sensor nodes can bedeployed on a human body in order to sense health status, such as heart beat rate, stresslevel, body temperature, activity and pulse and transmit sensor data to a commandcenter at some remote location. The second characteristic of the targeted applications isthe network scale. The scale can be considered large compared to traditional body areanetworks [3], but relatively small compared to traditional WSN installations. In order togather sufficient information, several sensor nodes must be deployed on each user. A teamalso consists of several users. The resulting sensor network will therefore consist of severalsmaller body area networks, i.e. a network of networks. Totally, there could be more thana hundred sensor nodes deployed. The third noticeable characteristic is the influence ofthe environment to the wireless communication. The wireless transmission could fail dueto unpredictable reasons in a complex environment. Another issue requiring considerationis the operational time. Fire rescue operations usually do not last for extensive periodsof time, and a system life time in the range of hours up to a week is therefore consideredto be sufficient.

A WSN can address the scalability and life time issues. However it cannot addressthe mobility issue in a trivial way since WSN need to connect to an infrastructure net-work, such as Wi-Fi, mobile phone access network service or satellites [4] to send sensordata to the remote command center. As a contrast, a PAN which contains off-the-shelfdevices, such as mobile phones, can easily access available infrastructure networks andgain access to the Internet. However, a PAN is only applicable for single user monitoringand the device’s life time can be very short comparing with a WSN. Therefore, in [5],a network architecture which combines a WSN and PANs is proposed as a solution forthe targeted applications. In this architecture, all sensor nodes deployed on users usebroadcast radio (IEEE 802.15.4 [6]) in order to form a mesh network (WSN). At the sametime, a dual-radio (Bluetooth [7] and IEEE 802.15.4) gateway/sensor node connects toa user’s Bluetooth equipped mobile phone, which constitutes a PAN. This gateway nodeforwards the sensor data from the WSN to the user’s mobile phone which links to a 3Gnetwork, and thus can access database servers on the Internet. Furthermore, cloud com-puting and web services can be used to perform resource intensive computations and addintelligence to the sensor nodes without increasing their processing power. This conceptis widely adopted in the smart phones today and can be applied on Internet connectedsensor nodes as well. This requires that the web service protocol stack is implemented ontop of TCP/IP and deployed on the motes. Although challenging, new implementationand data encoding techniques have been developed that shows promising results in thisdirection [8].

Benefiting from the combination of WSN and PANs, the scalability, mobility, andinteroperability of a mobile sensor network can be achieved. However, the network per-formance which reflects the feasibility of the proposed architecture is not investigated. Itis very important and challenging to guarantee that critical sensor data, such as alarms,successfully reach the infrastructure networks and are stored on a server in an acceptabledelay, typically in the range of hundreds of milliseconds. Furthermore, it is necessary to

2. Related Work 67

investigate the scalability of the proposed architecture since the number of sensor nodesthat can be supported by a resource constrained gateway node is limited compared totraditional gateways that are less restricted in resources.

This paper is organized in the following way: Section 2 discusses related work of thispaper. Experimental design for the system test is presented in Section 3, and in Section4, the experiments’ results presented. Finally, future work and conclusion of the researchare presented in Sections 5 and 6, respectively.

2 Related Work

In [9], a mobile patient monitoring system, named MobiHealth was introduced by Alterenet al. This system used the general packet radio service (GPRS) and universal mobiletelecommunications system (UMTS) as infrastructure networks to remotely monitor mo-bile patients. An iPAQ H3870 was used as the infrastructure gateway of this system andalso to graphically visualize sensor data in real time. The system was tested in differentEuropean countries with different telephone network operators. Milenkovic et al. imple-mented a personal health monitoring system based on wireless sensor networks [10]. Theirapproach required a personal server program to reside on a personal digital assistant, amobile phone or a computer during operation. Sensor data were forwarded from thenetwork coordinator to the personal server via an USB connection. For the performanceevaluation of the system, the investigated metric was the power consumption.

The IEEE 802.15.4 standard is also well studied in many simulations. Zheng et al.implemented an IEEE 802.15.4 patch for NS2. They presented the performance com-parison between IEEE 802.15.4 and IEEE 802.11, and studied the association efficiency,orphaning, collision and data transmission methods of the IEEE 802.15.4 standard [11].In [12], Rousselot et al. investigated the possibility of accurate simulation of the IEEE802.15.4 standard in Omnet++. They compared the simulation and experiment resultsin terms of timeliness and transmission success rate and analyzed the attributions of thedifferences between the results.

In this article, the investigated network’s scalability is simulated using Matlab. Thisis because the gateway node uses a bandwidth limited Bluetooth link to communicatewith consumer devices. The attribution to such slow Bluetooth data rate is due to thefact of low UART speed (57.6 Kbps) between the gateway node’s microcontroller andthe Bluetooth module. There are current no simulation tools available that can simulatean IEEE 802.15.4 network with such a constraint.

3 Experiments

For evaluating the performance of the data collection chain, experiments were designedand carried out based on the following hardware and software resources:

68 Paper C

Figure 1: Resource-constrained gateway node

3.1 Gateway node

The gateway node is a dual-chip, resource-constrained platform. Fig. 1 illustrates theprototype of the gateway node. It consists of a Mulle v3.1 [13] sensor node with an extraradio transceiver. The Mulle uses a Renesas M16C/62P microcontroller as its centralcomponent running at 10 MHz. The M16C MCU features 31 kB of RAM and 384 kB offlash memory. Besides an IEEE 802.15.4 module, the gateway node is also equipped witha Mitsumi’s WML-C46 AHR Bluetooth module, which can provide a 2 Mbps link withother device. However, the bandwidth to the Bluetooth 2.0 module is UART limited to57.6 kbps in the current version. This limits the system’s bandwidth since Bluetooth 2.0can support up to 2.1 Mbit/s transfer rate.

Figure 2: Multi-hop experiment hardware

3.2 Sensor node

The sensor node platform used for the experiments is the Mulle v5.2 which also uses aRenesas M16C at 10 MHz. The radio module is an Atmel AT86RF230 [14] which is anIEEE 802.15.4 standard compliant module. It is used for internal communication betweensensor nodes and gateway nodes. Moreover, the IEEE 802.15.4 module is configured to

3. Experiments 69

utilize the 2.4 GHz physical layer, which can provide up to 250 kbps of data rate. Fig. 2shows two sensor nodes (on the right hand side) transmitting sensor data to a gatewaynode in a two hop manner. The right hand side node is a data source node and it sendsunicast packets to the node in the middle. The middle node forwards the packets to thegateway node on the left hand side. This is the experimental setup which is a part of theproposed heterogeneous sensor network architecture.

3.3 User devices

Currently, the system has been verified to work with several off-the-shelf consumer de-vices. Supported mobile phone brands include iPhone, Nokia, Sony Erisson, and Mo-torola. It should be noticed that no application software is needed to be installed on anyof these mobile phones. A laptop with the Linux operating system can also connect tothe gateway node. Another popular consumer device used for mobile sensor networks ispersonal digital assistant (PDA). Such a device has not been tested with the gatewaynode, but is planned as future work. For performance evaluation described in this paper,the user device used was a Sony Ericsson K810i 3G mobile phone.

3.4 3G network

The location where the experiments were carried out is Lulea University of Technology(LTU) in northern Sweden. A Turbo 3G network, which provides 1 Mbps maximum uplink data rate and 14.4 Mbps maximum down link data rate is available at the university,and was therefore used for the experiments. The operator chosen was TeliaSonera.

3.5 Software

The gateway node utilizes the lwIP and lwBT stacks for communication, which arewritten in standard C code. The ActiveMessage packet format in TinyOS was usedbetween the gateway and sensor nodes. The network’s sensor nodes were programmedusing TinyOS [15]. As the gateway node is highly resource constrained, its memory usageis also presented in this paper to show the performance impact by running three stacks:IP, Bluetooth and 802.15.4, in parallel.

3.6 Test Cases

Test cases were designed in order to analyze the architecture’s performance in terms ofend-to-end delay and the number of sensor node manageable in the system (networkscalability). Therefore, the test cases are:

• 3G network round trip time statistics

• 1 hop end-to-end delay with three different payload sizes

• 2 hops end-to-end delay with three different payload sizes

70 Paper C

• simulation of packet generation rate for sensor nodes and scalability estimation

The purpose of the first test case is to investigate the 3G network’s real-world perfor-mance with a regular application. Such information can be used to identify the boundaryof the system’s performance. The second and third test cases aim on investigating theinfluence of packet size and network topology to the end-to-end delay. The last test caseuses Matlab to simulate the packet generation rate of sensor node. Based on that, thenumber of sensor nodes that can be supported by a bandwidth constrained gateway, e.g.the network scalability, can be estimated.

4 Results

4.1 Infrastructure network (3G) statistics

The first test case measures the infrastructure network (3G) performance. In this casea Bluetooth equipped laptop accessed the Internet via a 3G mobile phone as shown inFig. 4, and has been continuously transmitting ICMP PINGs to Lulea University ofTechnology’s web server. Fig. 3 depicts the performance of the TeliaSonera 3G networkat different time periods during the same day, and on different days. To make theillustration comparable, y axle’s scale is limited from 0 ms to 1000 ms. Thus, the RTTinstances which exceed 1000 ms are not shown in Fig. 3. As it can be seen, the round triptime (RTT) between the laptop and the server is mainly in the range from 400 ms to 500ms independent of the time period. Table 1 summarizes the performance statistics forthese three trials. It includes all RTT instances. It is important to know the 3G networkperformance as the up link data transmission rate and high delay is one bottle neck forthe proposed architecture. For example, according to the average RTT in Table 1, packetdelivery delay from a sensor node to a database server on the Internet should exceed 432/2ms, which is 220 ms if TCP/IP payload length in a gateway node is the same within thelaptop. In some cases the RTT measured was very large (around 5070 ms in the thirdtrial) comparing with common RTTs. Such behavior explains some severe situations formobile sensor network’s end-to-end delay which is shown in following subsections. Thishigh delay in combination with the limited memory and processing resources of the sensornode acting as gateway will introduce severe bandwidth limitations.

Another point to be noticed is that the 3G network performance can vary dependingon the time of the day, the location, and the number of simultaneous users. Though thetime when measurements are taken in this paper is only spread over two days, and thelocation is a specific city, it is still predictable that the 3G network performance will notdrastically vary at different times and locations under normal network operation.

4.2 Single and multi hop data delivery delay

After collecting the infrastructure performance analysis, the sensor nodes and the gatewaynode are added to the existing 3G network. Fig. 5 illustrates a single hop and two hops

4. Results 71

Figure 3: Statistics of round trip time for 3G networks

Figure 4: 3G network performance test setup

heterogeneous sensor network architecture. Data delivery delay is measured for these twosetups respectively.

Fig. 6 shows packet delivery delay for the first experiment setup, single hop transmis-sion. The delay is represented as a time difference between packets issued from a sensornode, and received by a database server on the Internet. The effect of the IEEE 802.15.4packet payload length to the delay is investigated in three experiments with 5 bytes, 50bytes and 114 bytes data length. Packet payload needs to contain two bytes sensor nodeID, two bytes serial number and at least one byte sensor data. Thus, the minimal datalength used in the experiments is 5 bytes. The maximal data payload length is 114 bytesaccording to the IEEE 802.15.4 standard specification. The 50 bytes is chosen as themedium of minimal and maximal data size. Generally, the end-to-end delay decreaseswhen the data length is shortened. The maximal delay is around 950 ms when the datalength is 114 bytes and the minimal delay is approximately 300 ms when the data length

72 Paper C

Table 1: 3G Network StatisticsMin. Avg. Max. Mdev

First Trial (ms) 367 435 609 29Second Trial (ms) 370 432 3493 111Third Trial (ms) 388 444 5071 147

Figure 5: Single hop and two hops data collection chain

is 5 bytes.

Fig. 7 shows the measurements of end-to-end delay from sensor node to a databaseserver on the Internet when there are two hops between a sensor node and a gatewaynode. As it can be seen, in most cases, the delay becomes smaller when the data lengthdecreases. During the experiment for 50 bytes transmission, some delays were very large:around 3300 ms, 1800 ms and 2200 ms (3 spikes in Fig. 7). This could happen due tounpredictable reasons, such as 3G network fluctuation or sensor node transmission prob-lem. For example, during the infrastructure performance evaluation in 4.1, the instability(large end-to-end delay) of 3G network is captured. The RTT for pure 3G network canbe more than 5000 ms. Comparing with single hop setting, the two hops end-to-enddelay does not increase drastically for all three data lengths used. To further evaluatethe overall performance, the delay means are summarized in Table 2. For smallest datasize (5 bytes), the delay mean value difference between single hop transmission and twohops transmission is 33 ms, and for largest data size (114 bytes), this difference is 45ms. For medium data size (50 bytes), the delay mean value difference between singlehop transmission and two hops transmission is 134 ms. This value is relatively large dueto the spikes as mentioned before. However, for regular cases in Fig. 7, the green curvelocates at similar height between red and blue as in Fig. 6. Therefore, ideally, for 50bytes data transmission, the difference of delay mean value between single hop and twohops should also be between 30 ms and 45 ms. One delay mean that needs to be noticedis that for one hop and very small data size configuration, the delay is 335 ms. Thisindicates that a gateway node can forward around 3 packets per second to a server onthe Internet. This result is used for the network scalability evaluation.

4. Results 73

Figure 6: 1 hop transmission delay from packet issued to stored in server

Figure 7: 2 hops transmission delay from packet issued to stored in server

4.3 Network scalability

In section 4.2, the experiment results of end-to-end delay are presented. If the sensordata length is set to 5 bytes and only one hop is needed to send this data, the end-to-enddelay is 335 ms on average (in Table 2). Therefore, the number of packets with 5 bytespayload can be forwarded by the gateway node to a server on the Internet via a mobilephone’s 3G network in one minute is

60s ÷ 334.6ms ≈ 180

The number of sensor nodes which can be supported by a single gateway node is calculatedas

180 ÷ packetGenerationRateOfSensorNode

74 Paper C

Table 2: Transmission delay mean values5 Bytes data 50 Bytes data 114 Bytes data

One hop (ms) 335 377 478Two hops (ms) 368 511 523

Figure 8: Packet generation rate distribution simulation for one sensor node

If the packet generation rates for all sensor nodes are constant and equal, for example,30 packets per minute, the network system can support 6 sensor nodes with very highpacket reception rate. However, in real world, the sensor events are usually random.Thus, instead of generating constant data rate, a sensor node is more likely to generaterandom rate of sensor data. In this paper, the normal distribution is used to simulate thepacket generation rate (number of packet per minute) for a sensor node. Fig. 8 depictsa Matlab simulation result for two hundred minutes. As it can be seen in Fig. 8, in mostof the time units (minutes), a sensor node will transmit 30 packets while in some raresituations, a sensor node will issue few (close to 0) or many (close to 60) packets perminute.

By using the approach of packet generation rate simulation for one sensor node, thetotal network traffic generated by all sensor nodes can be simulated. Fig. 9 plots thesimulation results of the entire number of packets in each minute generated by differentnumber of sensor nodes. This experiment simulates three packet traffic scenarios where4, 5 and 6 sensor nodes are issuing packets respectively. Each of them lasts for twothousand minutes. The red horizontal line indicates the throughput capacity of onegateway node (180 packets/minute). As shown, in the simulation of four sensor nodesgenerating packets, no time instance (one minute time period) exists when the totaltraffic exceeds gateway throughput limit. For five sensor node traffic simulation, thereare few instances when the required packet rate is larger than the throughput limit. Itdepends on the application whether such packet lose rate is acceptable or not, e.g. status

4. Results 75

Figure 9: Packet generation rate simulation for different sensor node numbers

packets can be dropped while alarm messages cannot. In the last experiment where thereare six sensor nodes sending packets, a packet could be lost in nearly 50% of the time.This shows that a gateway should use compression, aggregation and other bandwidthincreasing techniques for it to support larger networks.

4.4 Memory usage

The link between the sensor network and the infrastructure network is the gatewaynode which is resource constrained in terms of energy, bandwidth and memory. Table3 summarizes the memory usage of the gateway node. Memory is consumed by theBluetooth and TCP/IP stacks, and the IEEE 802.15.4 stack. The required static randomaccess memory (RAM) is 20.6 kB. The required read only memory (ROM) is 122.3 kB.

Table 3: Memory footprintBluetooth + TCP/IP IEEE 802.15.4 Total

RAM (kB) 19.2 1.4 20.6ROM (kB) 112.6 9.7 122.3

76 Paper C

5 Future Work

Future work consists of four parts. First, data aggregation at the gateway node to increasepacket throughput should be investigated. Currently, when a gateway node receives oneIEEE 802.15.4 packet, it copies the packet’s payload to its TCP/IP packet’s payload, andimmediately sends out this TCP/IP packet to the 3G network. Such behavior achievesthe lowest end-to-end delay. However, since the 802.15.4 packet’s payload length ismuch smaller than the maximal payload of a TCP/IP packet bandwidth is wasted. Inthe experiments, the maximum payload size was set to 129 bytes. This value couldbe increased to several hundred bytes, allowing the gateway node to aggregate several802.15.4 packets in one TCP/IP packet. For example, under the current settings, aTCP/IP packet can carry 25 802.15.4 packets if the 802.15.4 packet’s payload length is5 bytes, as used in the experiments. As a result, theoretically, the packet throughputcould be increased by around 25 times.

Second, enabling buffering capability for the gateway node should be considered.Although the gateway node is memory constrained, there is still free RAM available tosupport packet buffering inside the gateway node. As the sensor nodes’ instant packet rateis always changing according to the packet generation rate simulation, packet bufferingis beneficial. This means the demand of gateway up link throughput is changing as well.Therefore, the gateway node could buffer incoming 802.15.4 packets when it reachesuplink limit, and forward these packets later to decrease packet loss rate.

It is necessary to verify the simulation result for scalability investigation in anothersimulator, for example NS2, since it can simulate the networking performance in a moredetailed way.

Finally, it is important to investigate the possibility of adopting 6LoWPAN in theproposed architecture to enable IP at WSN level as well. This approach would increaseinteroperability. Another way to improve interoperability would be to use a standardizedway to transmit sensor data. The use of a service-oriented architecture (SOA) wouldbe very beneficial since it can enable automatic device and service discovery, and allowsensor data to be transmitted using standardized communication protocols, e.g. SOAPand DPWS.

6 Conclusion

The real world performance of a heterogeneous sensor network architecture which com-bines features from both WSN and PAN architectures has been experimentally studied.In order to investigate the characteristics of the used 3G network, its performance wasmeasured at different time periods during two days. The performance of the proposedsensor network architecture was evaluated by measuring the end-to-end delay when dif-ferent packet size and network topologies were used. The number of sensor nodes thatcan be supported by a single gateway node was simulated in Matlab. The key device ofthe proposed architecture is the gateway node which enables reuse of existing infrastruc-ture network access points. The derived measurement results were based on a specific

7. Acknowledgment 77

hardware and network architecture, but clearly indicate that the proposed heterogeneoussensor network architecture, using a resource-constrained sensor node as gateway to theInternet is a feasible solution for small-scale wireless mobile sensor networks.

7 Acknowledgment

The authors would like to thank the NSS and AESOP projects for sponsoring, and HenrikMakitaavola for his support.

References

[1] A. Mainwaring, J. Polastre, R. Szewczyk, D.Culler, and J. Anderson, “Wireless sen-sor networks for habitat monitoring,” in ACM International Workshop on WirelessSensor Networks and Applications (WSNA’02), Atlanta, GA, USA, September 2002.

[2] N. Alsharabi, L. R. Fa, F. Zing, and M. Ghurab, “Wireless sensor networks ofbattlefields hotspot: Challenges and solutions,” in Modeling and Optimization inMobile, Ad Hoc, and Wireless Networks and Workshops, 2008. WiOPT 2008. 6thInternational Symposium on, 1 - 3 2008, pp. 192–196.

[3] M. Hanson, H. P. Jr., A. Barth, K. Ringgenberg, B. Calhoun, J. Aylor, and J. Lach,“Body area sensor networks: Challenges and opportunities,” in IEEE Computer,January 2009, pp. 58–65.

[4] L. Steenkamp, S. Kaplan, and R. Wilkinson, “Wireless sensor network gateway,” inAFRICON, 2009, 2009.

[5] J. Eliasson, C. Zhong, and J. Delsing, “A heterogeneous sensor networks architecturefor highly mobile users,” in Sixth IEEE Conference on Wireless Communication andSensor Network (WCSN-2010), 2010.

[6] “IEEE standard for information technology–telecommunications and informa-tion exchange between systems–local and metropolitan area networks specificrequirements–,” http://standards.ieee.org/getieee802/download/802.15.4-2003.pdf.

[7] B. SIG, “Bluetooth special interest group,” http://www.bluetooth.org, 2010.

[8] Z. Shelby, “Embedded web services,” in Wireless Communications, IEEE, vol. 17,2010, p. 52–57.

[9] A. V. Halteren, R. Bults, K. Wac, D. Konstantas, I. Widya, N. Dokovsky, G. Ko-prinkov, V. Jones, and R. Herzog, “Mobile patient monitoring: The mobihealthsystem,” in The Journal on Information Technology in Healthcare, 2004, pp. 365–373.

78 Paper C

[10] A. Milenkovic, C. Otto, and E. Jovanov, “Wireless sensor networks for personalhealth monitoring: Issues and an implementation,” in Computer Communications,vol. 29, August 2006, pp. 2521–2533.

[11] J. Zheng and M. Lee, “A comprehensive performance study of ieee 802.15.4,” inSensor Network Operations, 2006, p. 218–237.

[12] J. Rousselot, M. Aoun, and R. S. Oliver, “Accurate timeliness simulations for real-time wireless sensor networks,” in Third UKSim European Symposium on ComputerModeling and Simulation, 2009, p. 476–481.

[13] “The Mulle platform,” http://www.eistec.se, 2011.

[14] “AT86RF230,” http://www.atmel.com/, 2011.

Documents

EISLAB A Sensor Network Architecture - DiVA portalltu.diva-portal.org/smash/get/diva2:990852/FULLTEXT01.pdf · This thesis investigates the feasibility of using a heterogeneous sensor