Sink localization and topology control in large scale heterogeneous wireless sensor networks

Sink localization and topology control in large scale heterogeneous wireless sensor networks

Material Information

Sink localization and topology control in large scale heterogeneous wireless sensor networks
Zhang, Rui
Place of Publication:
[Tampa, Fla.]
University of South Florida
Publication Date:


Subjects / Keywords:
Location service
Energy efficiency
Multiple moving sinks and targets
Data dissemination
Location-based routing
Network longevity
Dissertations, Academic -- Computer Science and Engineering -- Doctoral -- USF ( lcsh )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )


ABSTRACT: Wireless Sensor Networks (WSNs) continue to evolve as new applications emerge. In the recent past, WSNs were mostly single sink networks with a few number of homogeneous and static sensor nodes. Now, several applications require networks with multiple and moving sinks and targets as well as thousands of heterogeneous devices. However, the same constraints remain: sensor nodes continue to be very limited in resources, posing new challenges in the design of scalable and energy-efficient algorithms and communication protocols to support these new applications. This dissertation first addresses the problem of sink localization in large scale WSNs. A scalable and energy-efficient sink localization mechanism, called the Anchor Location Service (ALS), is introduced to support the use of location-based routing protocols. ALS avoids frequent and costly flooding procedures derived from the mobility of the sinks and targets, and utilizes face routing to guarantee the success of localization. The problem of topology control in heterogeneous environments is addressed next. A new topology control mechanism, the Residual Energy-Aware Dynamic (READ) algorithm, is devised to extend the lifetime of the network while maintaining connectivity. READ extends the lifetime of the network by assigning a more prominent role to more powerful devices. ALS and READ are evaluated and compared with other well-known protocols using analytical means and simulations. Results show that ALS provides a scalable sink location service and reduces the communication overhead in scenarios with multiple and moving sinks and targets. Results also show that READ increases both the network lifetime and the packet delivery rate.
Dissertation (Ph.D.)--University of South Florida, 2007.
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 111 pages.
General Note:
Includes vita.
Statement of Responsibility:
by Rui Zhang.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
001917510 ( ALEPH )
181742143 ( OCLC )
E14-SFE0002048 ( USFLDC DOI )
e14.2048 ( USFLDC Handle )

Postcard Information



This item has the following downloads:

Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001917510
003 fts
005 20071121124041.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 071121s2007 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002048
1 100
Zhang, Rui.
0 245
Sink localization and topology control in large scale heterogeneous wireless sensor networks
h [electronic resource] /
by Rui Zhang.
[Tampa, Fla.] :
b University of South Florida,
3 520
ABSTRACT: Wireless Sensor Networks (WSNs) continue to evolve as new applications emerge. In the recent past, WSNs were mostly single sink networks with a few number of homogeneous and static sensor nodes. Now, several applications require networks with multiple and moving sinks and targets as well as thousands of heterogeneous devices. However, the same constraints remain: sensor nodes continue to be very limited in resources, posing new challenges in the design of scalable and energy-efficient algorithms and communication protocols to support these new applications. This dissertation first addresses the problem of sink localization in large scale WSNs. A scalable and energy-efficient sink localization mechanism, called the Anchor Location Service (ALS), is introduced to support the use of location-based routing protocols. ALS avoids frequent and costly flooding procedures derived from the mobility of the sinks and targets, and utilizes face routing to guarantee the success of localization. The problem of topology control in heterogeneous environments is addressed next. A new topology control mechanism, the Residual Energy-Aware Dynamic (READ) algorithm, is devised to extend the lifetime of the network while maintaining connectivity. READ extends the lifetime of the network by assigning a more prominent role to more powerful devices. ALS and READ are evaluated and compared with other well-known protocols using analytical means and simulations. Results show that ALS provides a scalable sink location service and reduces the communication overhead in scenarios with multiple and moving sinks and targets. Results also show that READ increases both the network lifetime and the packet delivery rate.
Dissertation (Ph.D.)--University of South Florida, 2007.
Includes bibliographical references.
Text (Electronic dissertation) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 111 pages.
Includes vita.
Advisor: Miguel A. Labrador, Ph.D.
Location service.
Energy efficiency.
Multiple moving sinks and targets.
Data dissemination.
Location-based routing.
Network longevity.
Dissertations, Academic
x Computer Science and Engineering
t USF Electronic Theses and Dissertations.
4 856


SulkLocalIzatIOn and Topology Control m Large Scale Heterogeneous WIfeless Sensor Networks by Rui Zhang A dissertation submitted in partial fulfillmentofthe reqUIrements for the degree of Doctor of PhIlosophy Department of Computer SCIence andEngmeenngCollegeofEngineering UniversityofSouth FloridaMajorProfessor: Miguel A. Labrador, Ph.D. Kimon Valavams, Ph.D. Rafael Perez, Ph.D. TapasK.Das, Ph.D. Wilfrido A. Moreno, Ph.D. April 23, 2007 Keywords: LocatIOn ServIce, Energy EfficIency, MultipleMovmgSmks and Targets, Data DissemmatIOn, LocatIOn-Based Routmg, ConnectivIty, Network LongevItyCopyright 2007, Rui Zhang


ACKNOWLEDGEMENTSpeople that have supported and guided me these past four years.Iwould especially like to acknowledge my mentor and advisor Dr. Miguel Labrador, who has made considerablec()IltrIJ)lltI()Ilst()J)()tliIllYaca(leIllIcall(lpers()Ilal lIte. Tlie c()llIltless IIlsIglits all(lIIly allla1jleadvice that he has providedmeover the years havediredlyhelped in the completionofthisIttakes more than words to express my gratitude to my family for their constant and uncondItIonal support and love.Myparents are everythmg to me.Iwould also lIke to take this opportunity to thank all my friends for their continuous encouragement and support during those long nights.


TABLE OF CONTENTSABSTRACT viii1.1Wireless Sensor Networks1.3Topology Control in WSNs1.5Organizationofthe Dissertation CHAPTER 2 LITERATURE REVIEW CHAPTER 3 ANCHOR LOCATION SERVICE PROTOCOL3.1The Anchor Location Service Protocol691020 203.1.2 3.1.3 3.1.4 3.2.1 Anchor Selection Process Query and Data Dissemination Processes Sink and Target Mobility and Agent Chain Maintenance Scenario and Notation232526273.2.2.1 Process #1: Establishmentofthe Global Grid and Anchor SystemProcess#4:QueryingtheAnchorSystemProcess #5: Data Packet Transmission2830333.3 Performance Evaluation 3.3.2 Impactofthe NumberofSinks and Sourcesi3540


3.3.4 3.3.5 3.3.6 ImpactofSensor Density ImpactofNetwork Area ImpactofSink Mobility 45 47 50 CHAPTER 4 RESIDUAL ENERGY AWARENESS DYNAMIC TOPOLOGY CONTROL 56 4.1 Network Model 56 4.1.1 Maxpower Graph 57 4.2Tl1ec:elltraIIze(J;R.esI(Jllal}\\Vare[)YllaIIlIc Top()l()gyC:()lltr()lAlgorithm 63 4.2.1 Centralized Residual Energy Awareness Dynamic Algorithm 63 Results and Analysis with Packet Transmission 69 4.3 The DIstrIbuted ResIdual Energy Aware DynamIc Topology Control 4.3.1 4.3.2 4.3.3 [)Istfll)llte(J;R.esI(Jllal}\\Vare [)YllaIIlIc }\lg() rItI1IIlComplexityofthe DREAD Topology Control Algorithm Simulation Results and EvaluationofDREAD Simulation Setup 738182 82CHAPTER 5 CONCLUSIONS AND FUTURE WORK 90 AppendIx A Property Proof of ALS and READ 98 A. I Convergence Proof of ALS' Anchor Setup Process and Query Process 98 A.2 Connectivity and Symmetric Property ProofofREAD 109 ABOUT THE AUTHORiiEnd Page


Table 4.4 LIST OF TABLES Two-hop neighbor table.ill76



Figure 3.16 Anchor setup time vs. the numberofsources.vs. FIgure3.18LocatIOn tIme vs. the number of sources. Figure 3.19 Location overhead vs. the numberofsources. sources. 434445VS.sources. Figure 3.21 Average anchor setup time. vs. the network density.oveThe:ad.VS.46J

Figure 4.4 FIgure4.6Figure 4.7 Figure 4.9 FIgure4.11Figure 4.12 Figure 4.13 Figure 4.14 FIgure4.16Figure 4.17 R&M with link addition. READ wIth two-degree connectIvIty. READ with one-degree connectivity. Average link length. Successful delIvery rate m centralIzed ImplementatIOn. Numberofmilitary nodes alive in centralized implementation. Numberofrobots alive in centralized implementation. NumberofPDA nodes alive in centralized implementation. sensors EstablIshment of two-hop tables. Generation local minimal spanning tree.6869 69707173 7374 7778Figure 4.19 FIgure4.21Figure 4.22 Figure 4.23 Figure 4.24 FIgure4.26Figure 4.27 Figure 4.28 Numberofnodes alive in distributed implementation. Number of mIlItary nodes alIve m dIstnbuted ImplementatIOn. Numberofrobots alive in distributed implementation. NumberofPDA nodes alive in distributed implementation. Numberofsensors nodes alive in distributed implementation. Successful delIvery rate of CREAD and DREAD. Numberofmilitary nodes aliveofCREAD and DREAD. Numberofrobots aliveofCREAD and DREAD. vi 8587 87 87 87 8889 89


Figure 4.29 NumberofPDA nodes aliveofCREAD and DREAD. 89 FIgure 4.30 Number of sensors nodes ahve of CREAD and DREAD. 89 FigureA.lA graphical representationofreal-grid-nodes (solid nodes), void-gridnodes (hollow nodes), real-edges, vOId-edges, andvOIdareas. 100 Figure A.2 A graphical representationoffour Real Polygon examples.101Figure A.3 A graphical representationofvoid areas and their envelops.101case Figure A.5 Figure A.6 FIgure A.7 FIgure A.8 Base case1with 2 void-grid-points. Base case 2 with 2 void-grid-points. Step wIth I vOld-gnd-node as far neIghbor. Step wIth I vOld-gnd-node as neIghbor. sIm,ple max powerne1tw()rkexclmr)le.vii 104 106 106 108


SINK LOCALIZATION AND TOPOLOGY CONTROL IN LARGE SCALE HETEROGENEOUS WIRELESS SENSOR NETWORKS RuiZhang ABSTRACTas new aPr)lIcatu)ns em,erge. the recent past, WSNs were mostly single sink networks with a few numberofhomogeneous and statIc sensor nodes. Now, several applIcatIons reqUIre networksWIthmultIple and movmg smks and targets as well as thousands of heterogeneous deVIces. However, sameCOllstraiJntsrennarn: sensor very limited resources, posing tion protocols to support these new applications.ThISdIssertatIOn first addresses the problem of smk localIzatIOn m large scale WSNs. A scalable and energy-efficIent smk localIzatIOn mechamsm, called the Anchor LocatIOn Service (ALS), is introduced to support the useoflocation-based routing protocols. ALS targets, and utilizes face routing to guarantee the successoflocalization. The problemoftopology control in heterogeneous environments is addressed next. A new topology control mechamsm, the ReSIdual Energy-Aware DynamIC (READ) algorithm, is devised to extend the lifetimeofthe network while maintaining connectivity. viii assIgnmg a moreprC)milneJntmore


ALS and READ are evaluated and compared with other well-known protocols using analytIcal means and sImulatIons. Results show that ALS provIdes a scalable sulk locatIon service and reduces the communication overhead in scenarios with multiple and movingix


<:HAPTERI INTRODUCTION1.1Wireless Sensor NetworksSensors mtegrated wIth microelectromc technologIes emerged decades ago. Early sen sors were used individually in applications to monitor smoke inside residences, collect indoor or outdoor temperature, or collect sound stimulus in stairways. Since then, sensors sensors wirelessly and collaborate with each other. The proliferationofsensing and wireless com mum cation technologIes mconjUnctIOnwIth the development of microelectromcs has made a new breedofmore powerful wireless sensor devices available. Their application scenar ios have also expanded from simple cases to a countless numberofmore complicated ones. new areen'viSlonleasmall wireless devices spread over very large areas to monitor the environment, perform mtrusIOn detectIOn, collectseIsmICmformatIon, etc. One proposed concrete applIcation could be a vast heterogeneous sensor network deployment along the border between the United States and Mexico, from San Diego, California, to Brownsville, Texas, to protect tfa'verses aofterrains, which makes it very difficult,ifnot impossible, to monitor illegal intruders by human power. Wireless Sensor Networks (WSNs) make this very difficult mission much eaSIer to achieve.1


Classic wireless sensor node and network architectures are shown in Figure1.1.All sensors have the followmg hardware components: sensmg umt, processmg umt, wIth stor age, transmission unit and power supply.Intermsofthe network architecture, sensors auto one anc)tht:rareh-''-Ul.,wi""r!CBS),which may connect the sensor network to the Internet or any other publicorprivate network. Although sensors differentiate from each other in termsofsensing functionality, physdinaerlsicm or evensome commonch;aractE:ris:tics.most sensor nodes have very limited computational capabilities, storage capacity, and en ergy resources. Therefore, it is very important to design simple and energy efficient com1111111IcatI()1lpr()t()c()IsallClalg() rItllIl1sf()r'YI\Ts.2


DataDisseminationOthersRoutingProtocolsLocalizationTopologyControlSchedulingInPhysicalLayerTimeSynchronizationFigure 1.2 Mechanism and protocols that support the operationofWSNs.UUIOUlghul1rp>I,p>,",,",sensor consumep>n,"'rovdlJnrlg tasks, communication is by far the most energy consuming task. For example, the RFMTRlOOOradIOtransceIver mcluded m the Berkeley motes consumesIp,}to transmIt one bIt and0.5ILJto receive one, while it takes around8nJofenergy per instruction. This results in a communication to processing ratioofabout 190. Other transceivers, such as Rockwell's elusions can be drawn from these power consumption figures. First, communication costs must be mimmized. Second, ItISworth spendmg addItional processmg cyclesIfthey can contnbute to commumcatIOn savmgs.ThISISknown as "m-network processmg".WSNs are supported by many mechanisms and protocols. A sampleofthe most im-nrvrt,"",tonesarewith constraintsofWSNs in mind. For instanlce,Protocolsutilize either location information or local neighbor tables to route packets from Source to Destination.Schedul-3


ing in PhysicalLayeris a strategy employed to turn nodes on and off in order to reduce energy consumptIOn. SImIlarly, mnovatIveSmartMAClayer protocolshave been proposed to reduce wireless signal collision and energy consumption. InData Dissemination,infor as rep,ositories.tionalways attracts attention in distributed systems. Time synchronization algorithms are used to keep the sensor clocks as tightly synchromzed aspOSSIbleconsidenng the scalabIl ity and energy constraints.Localization Serviceis another very important research area in userOlltlrlg protoc()ls.lOlWl().vLU,rUflJLfI,ro[()colsare to adjust and simplify the network topologytosave energy.ThISdIssertation focuses on LocahzatIOn and Topology Control. A solutIOn to the prob lemofscalable and efficient sink localization in large scale WSNs is introduced first. Then, a topology control algorithm that considers the coexistence and cooperationofheteroge-1.2Sink Localization inWSNsLocation-based routing has recently emerged as an important approach to address the scalablhty and energy effiCIency concerns for data dissemmatIOn m large scale WSNs. For example, m locatIOn-based routmg, nodes do not need to make complex computatIOns to find the next hop as routing decisions are made based on local information. Morelocation-based routing substantially reduces the communication overhead because routing table advertisements, like those found in traditional routing protocols, are unnecessary. For Illstallce,\Vllellge()grapllicalgreeClYr()lltIllgpr()t()c()I,al()catl()ll=paseClr()lltIllg pr()t()c()l,IS


"Figure 1.3 A simple exampleoflocation service. maimtain a one-hop neighbors' location table. The processing node performs a calculation using local mformatIOn and selects from Its one-hop neIghbors a neIghbor geographIcally closest to the packet's destination to forward the data packet to. While these are important features, very well in line with the constraints and charac-rolltUlgpJrot()colsassume onascalable and energy efficient mechanism to distribute the location informationofthe sinks or destmatIOns. Unfortunately, most of the eXIstmg location mechamsms utIlIze some sort of floodmg procedure to spread the smk's locatIOn, whIchISunSUItable for large scale WSNs. Furthermore, this flooding procedure is frequently repeated in cases with multiple Figure1.3illustrates an application scenario, where numerous sensors are deployed in a large network area. Without the supportofany infrastructures, sensors work collaboratively5


to collect elephants' migration information from the sensor network and then provide that mformatIOn to multIple rangers by usmg a locatIOn-based routmg protocol. Once a target elephant moves the sensors surrounding the elephant collect the stimulus. Oneofthese case are moving rangers. However, the common assumption held here by almost all the location-based routing protocolsISthat the destmatIon's locatIOn mfoffilatIOn, mthISmstance, the ranger's locatIOn information, has been disseminated into the entire network in a scalable and energy efficient more oneormore onerYH,,"nararlger, location information. Frequently repeated flooding queries for rangers' locations in a large network lIkethISISclearly neIther a scalable nor an energy effiCIent solutIOn. Although location services have been under investigation for some time, there are few solutions provided in the literature that are suitable for large scale WSNs, especially with in this dissertation provides a complete scalable and energy efficient solution for this type()fappIIcatI()Il'1.3 Topology Control inWSNs Topology controlISone of the most Important mechamsms utIlIzed m WSNs to reduce energy consumption. Topology control is well defined in[3]as the artofcoordinating aneltw()rkthe desired properties while reducing node energy consumption. Although topology control has been studied for some years, current topology control appr()acIles()Illyc()IlsI(ierIl()IIl()geIle()llsseIls()rIlet\V()rk:s,\VIleretIle (iIfIereIlces()fIIlItIal6


energyofbatteries and sensitivityoftransceivers are omitted. However, the wide spectrum of possIble applIcatIons where wIreless ad hoc and sensor networks can be applIed has In creased the possibilityofmixed networks, where devicesofdifferent types and characteris sameaPJ)lic:ati, collaborate, each taking advantageofthe goodnessofthe others. It is this approach to topology control algonthms thatthISresearch takes, where more powerful devIces are set to have a more prominent role in the network connectivity to extend the network's lifetime.1.4 ContributionsThis research introduces the ALS protocol, a grid-based protocol that provides sink a scalabJle routing for large scale WSNs.InALS, each sink builds a global grid madeofspecial locatIon server nodes called anchors that are used by all sources to findItSlocatIOn. Becauseofthis global grid structure, fewer location request messages need to travel through the store the location informationofall sinks and respond to query messages. Considering theSIzeof large scale WSNs, ALS not only reduces thecommUnICatIOncosts but also the locatIonInfOrmatIOnresponse tIme, as sources wIll only have to find the first global anchor grid. selectionofthe anchors are presented.Inaddition, the location dissemination process and the location query process are described. The caseofvery large wireless sensor networks\\lItI1ratl1erstatICallClres()llrcec()llstraIlleClll()ClesallClscellafl()s\\lItl1Il111ltIpleallClIl1(),/Illg7


sinks and targets are considered. This is a very common scenario since many sources from many dIfferent places mIghtbetransmIttmg mformatIOn to one or more smks at the same time. Using a mathematical approach and simulations, not only is the performanceofthe arClutItlgprotl)C01I,overhead, as well as and the communication and state overheadofthe protocols are con SIdered as mam pet10rmance metncs, whIch are presented varymg the number of smks and sources, the network size, the network density and the speedofmobile sink nodes. The re-a"",uwu."with multiple sources. With different network area, ALS reduces location query overhead by at least70%m the worst case and90%m the best case. Although topology control problems have been studied in the contextofheterogeneous wireless ad hoc and sensor networks before, most existing mechanisms have focused on on vices have identical physical characteristics. As a result, topology control problems have been solved as range assIgnment problems, whIch not only neglect the heterogeneIty of the network but alsodon'ttake advantage of the umque capabIlItIes of dItlerent devIces. In this chapter, the READ topology control algorithm and the DREAD topology con alg()ritllms are presen'ted. in heterogeneous wireless scenarios, where sensor nodes, ad hoc nodes, robots with com mumcatIon capabIlItIes and even more powefful mIlItary wIfeless devIces work together m the same applIcatIOn. InthISheterogeneous scenano, the assumptIon of IdentIcal mItIal energy,reslLdu:alenergy,every wlI'elessd,eVI,ces.v,uu.uu...uas a assignment problem. READ considers these aspects in the formulationofthe optimization8


problem to dynamically recruit the links that optimize the workload between different wire less devIces whIle stIll mamtammg network connectIvIty. DREAD proVIdes a dIstnbuted solution with the same considerations in mind. algorilthnls are imJ)lernenteda()UJCJUJ"uvuare cornpared rithms for wireless ad hoc and sensor networks.Itis demonstrated that READ and DREAD extend the network lIfetIme by making more powertul nodesplayamore Important role m the network. The network with READ as topology control algorithm can last much longer1.5 Organization of the DissertationVl!";aUJL'-'UastollmNS:Clllapter existing literature regarding location services and topology control algorithms. Chapter 3 presents the ALS protocol m detail and Its theoretIcalanalySIS.SImulatIon results are also presented in this chapter. Chapter 4 presents the DynaIllic Residual Energy Awareness topology control algorithm. The network model, detailsofthe algorithm, and its performance arepresents direction for future research.9


<:HAPTER2 LITERATURE REVIEW2.1 Location Service ProtocolsLocatIon serVIce mechamsms can be broadly dIvIded mto three categones as showninFig.2.1.They are quorum-based systems, home-based systems, and systems with ap proximate information.Inquorum-based systems, the setofnodes is divided into mutually own",uU'",""rum). As a result, each element in a quorum can respond to queries coming from a different quorum. These subsets are desIgned m such a way that theIr mtersectIOnISnon-empty and the requesting node finds the desired information. In one typeofquorum-based systems, location information is sent in one direction (e.g., messages are sentinthe orthogonal direction (e.g., east and west) from nodes contained in dIfferent sets. Several quorum-based mechamsms have been proposed for wIreless mobIle ad hoc networks. The scheme presented m [7] mamtams the quorum structure as the nodes move.Thenoveltyofthis scheme liesinthe update mechanism, which utilizes link incias location updatestngl;sef1l1gsignalsorapproaches, like the ones utilized in [8,9]. The scheme presented in [10,II]utilizes the same update strategy but organizes the quorum differently. The scheme utilizes a quorum based locatIOn servIce thataVOIdspartIal floodmg overhead, and/or locatIon faIlures m group movement scenarios.Inthis scheme, the destination node distributes its location to 10


LocationServiceProtocolsApproximateinformationHome-basedSystemUtilizerandomapproachQuorum-basedSystemLocationinformationissentinonedirectionFIgure2.1Taxonomy of locatIOnserVIceprotocols. all nodes located to the north and southoftheir current location while sources send messages m the east and west dIrectIOn to search for the locatIon of the destmatIon, whIchISfinally found at the intersection. The authors utilized face routing in the distribution and search mechanisms to guarantee the successofthe location service. They proposed four to determine the success rate and communication overheadofthe strategies in scenarios wIth dIfferent number of nodes and node degree. The schemeISshown to provIde a good locatIon serVIce to statIc and mobIle nodes, mcludmg nodes that move m groups and toward the same direction. Oneofthe drawbacksofthe scheme, however, is that the messages that distribute the location informationofthe destination and the search messages travel through the entire network evenifthe source and destination are relatively close to each other. Other quorum-based schemes utilize a random approach to obtain up-to-date informatI on m several sets [12,13]whIle other schemes buIld and mamtam a vIrtual backbone with server nodes that maintain the location information [14-16]. However, these later not clearifthis overhead is better than using simple flooding.u


Home-based systems are similar to those well known location management mecha msms utIlIzed m cellular networks. Home-based systems dIvIde the network m several zones, and, then, nodes affiliate with a particular zone (home) and share their location in zone. zone so are can be sent to zones insteadofindividual servers.Asit can be inferred, these schemes require a conSIderable amount of overhead and mtroduce routmg mefficIencIes m scenanos wIth high mobility. In this case, nodes havetosend position updates more frequentlytokeep the per zone, which is not suitable for heterogeneous WSNs becauseofenergy and memory constramts. In the case of large scale WSNs,ItISnot clearIfthese schemes proVIde better perfonnance and consume less energy than other approaches, suchasthe ones proposed in Section 4.2 and Section 4.3. Among the most relevant home-based schemes proposed in Several other location services are basedonapproximate infonnation. In [22], for ex ample, the authors present a scheme by whIch nodes update theIrpOSItIOnsm concentncCIrclesof doublmgSIze.Whenever a node moves out ofItSpresentCIrcle,Itbroadcasts its new position to all nodes inside a new circle centered at the current node's position. cover which introduces a considerable amountofoverhead, not particularly suited for large scale WSNs. A SImIlar scheme, whIch utIlIzes a hIerarchy of square regIons mstead ofCIrcles,ISpresented m [23]. In the proposedGndLocatIOn ServIce (GLS) scheme [23],asthe distance increases, location updates are sent to fewer numberoflocation servers. Although moreits efficiency has not been assessed in very large networks. DREAM[8]and LAR[9]are 12


other schemes in which nodes flood the network with their positions in a proactiveorreactIVemanner, respectIvely, and the locatIon of the destmatIOnISestImated withm aregIOn.Ofcourse, the flooding procedure makes these mechanisms unsuitable for large scale netPerhaps the most similar protocol to ALS is the Two-Tier Data Dissemination (TTDD) protocol for large scale WSNs presented m [4=6].TTDDISagndorquorum-based proto col that provides location information and routing in an integrated manner. Upon detectionasourceasensorsensorsthe protocol uses to receive and forward infonnation from source to sink. This virtual back bone for routmg made of dissemmatIon nodes has been cntIcized for not provIdmg optImal paths [25]. Once the grid structure has been built, a query from a sink travels through two tiers to reach the source node.Ifthe positionofthe destination is unknown,TTDDutilizes lower tier, which is within the cellofthe sink's current location. The flooding continues untIl It reaches the closest dissemmatIon node. AtthISpomt, the message reaches the hIgher tIer, whIchISmade up of all dissemmatIOn nodes from the smk's cell to the source's cell. The closest dissemination node to the sink receives the query and forwards it to the next source. through the higher tier until it reaches either the source nodeora dissemination node cur rently receIvmg data on behalf of the source.ThISprocess proVIdes mformatIOn of the path back to the sulk node, enablmg the source mfonnatIOn to traverse the same two tIers but m reverse are sources. trajectory forwarding strategy makes sink mobility transparent to the higher tier whenever13


the sink moves within the current cell.Ifit moves beyond the cell, a new dissemination node dIscovery procedure needs to be trIggered wIth the assocIated overhead. However, this overhead is expectedtobe small,asnew dissemination nodes are likely to be found in areare based or quorum-based systems that assume sensor nodes are stationary and awareoftheir locatIons. Both protocols are scalable m the sense that theyaVOIdglobal floodmgasthe main mechanismtodisseminate data and location information. Global flooding is avoided enschemes are different in several aspects. For example, TTDD is source (not sink) oriented,asItestablIshes one grId per source.ThISISan Important dIfference m terms of the final overhead because the numberofsinks is usually known in advance to the network designer, while the numberoftargets is completely unknown. Also, TTDD utilizes the disseminauses location information. This is alsoanimportant difference. ALS decouples the routing and the locatIOn functIons.Assuch, once the locatIOn of the smkISknown to the source, any locatIOn-based routmg protocol, suchasGFG [26, 27], Greedy PerImeter Stateless Rout ing (GPSR) protocol [28] or Scalable Energy-Efficient Location-Aided Routing (SELAR) areeXI)ecteddrained considerably faster than the other nodes, and therefore a strategy to change them frequently must be mcluded. In ALS, the anchors do not bearthISload,asthey only re spond to locatIon querIes.TwoaddItIonal dIfferences are worth mentIOnmg. Fust, TTDD utilizes greedy forwarding while ALS utilizes greedy forwarding with face routing [26,27]. ond, ALS uses the "in-network processing" strategy called "propagated fusion"tofurther


optimize the performanceofthe protocol. As new anchor systems are being set up, current anchors Include the new locatIons In theIr memones. Query messages do not need to travel further to find the required location then, and the location time is reduced.alitelrature onroUllin!!mech2misms sensor nel:w()rks. references are provided. The interested reader is directed to[30-33]to learn about routing algonthms for WIreless mobIle ad hoc networks. Survey papers on routIng algonthms for WSNs canbefound in[1,34-36].In addition,[37,38]include an extensive and thorough sensor nel:w()"",riP'"aon locatilon-rOUilIng2.2 Topology ControlCurrent topology control algorithms canbecategorized ashomogeneous, nonhomoge neousandheterogeneousas shown In FIgure 2.2. Homogeneous topology control algo rithms assume that all wireless devices use the same transmission range. Correspondingly, the topology control problem becomes a range assignment problem that searches for the or"'U""'C.utl'ansmitting network properties, such as network connectivity. ProblemsoffindingCTRare the sim plest topology control problems to formulate and solve, and were the first to appear In the lIterature. The assumptIOn that all nodes use the same transmISSIOn power, however, holds onlyifall the wireless transceivers have no difference in their technology and finding the the schemes presented in[40]and[41]belong to this category.[40]proposed a distributed topology control algorithm to construct a planar spannerofunit-disk graph. The resulting grapliC()l1tai ll s all I)elallllaytriallg11latI()l1eClgestr()Ill tlie11l1It__ClISJ(grapli111tlie resllltIllg15


Topology ControlrWireless NetworRHomogeneous Non-homogeneous HeterogeneousjRAand variantsEriergy=effiCieritCommunicationFigure 2.2 Taxonomyoftopology controL topology, the shortest path between any two nodesuandvis at most a constant factorofthe shortest path connectinguandvin unit-disk graph. Several other examples canbefound m [3].Ithas been proven that the critical transmitting range with preservationofconnectivity lengthofthe longest EMST. Due to the strong assumptionofknowing the exact node's10-cation and the huge amount of control message overhead to exchange locatIOn mfoffilatIOn network-wide, researchers have devoted their attention to find theCTRwith the presenceofuncertainty about node positions. The typical approach is to study the conditions for asymp,toticaJllj almost sure cormeCtl'IIty a cerltam in the area.Indense networks, geometric random graphs theory has been utilized to solve the problem.In1997, Mathew D. Penrose proved m [42] that Ifnpomts are dIstnbutedunitonnlyat ramlom 16


ing the lengthofthe longest MST edge built on thennodes, then Equation2.1will holdEas(2.1) A further corollary is thatifthe area is unit square andnnodes are distributed uniformly at_ ilOgn+f(n)Ie-nrrwheref(n)could be any function that satisfies the conditionlimn->cxJ(n)=+CX).(2.2)neltw()rkarea Some researchers, therefore, have added one more parameter,l,the side length of the can in from 0 to any constantc,wheredis the orderofdimensionofthe network space. The proposition was proven in[3]thatifthe area is the[QJJdwithd=2,3andnnodes areInnon-homogeneous topology control, on the other hand, different wireless devices can choose dIfferenttransmISSIOnranges withm the same maXImum transmIttmg range. Under the assumption that all nodes have the same path model with the same parameters, the are exchanlge2lble conceplts.desired transmission range for each individual wireless device, while maintaining certain network propertIes and achIevmg specIfic obJectives.In[3], the range assIgnment problem 17


was defined as follows: letNbe a setofnodes in the d-dimensional space, withd=1,2,3.Determine a functionRAsuch that the communication graph is strongly connected, andc(RA) LUEN(RA(u)Y"is minimum over all connect()'is an NP-hard problem in two-dimensional networks and in three-dimensional networks, caseofrouting protocols, unidirectional links incur more overhead, which could override lar, where the resulting communication graph contains only bidirectional links.In[46],131()llgIlet al. pr()ye(l tIlat tIleB,al1ge1\SSIgl1111el1tals()NP=Ilar(l.Severalnon"'homogeneoustopologycontrolmechanismshavebeenproposedinthelit...erature. For example, the algorithms presented in[47-52]also minimize energy consumpfind the maximum transmission power while maintaining connectivity and bi-connectivity. Two centralIzed algonthms are proposed for thestatICverSIon, whIle two heunstIcs al-gOllthmsareprclpo:sedauthorspres1enta,-u",un,unA.

is assigned a different transmission range. In the algorithm, called LMST, each node builds Its one-hop local mmimum spannmg tree wIth the dIstance as the weIght cost assocIated with each edge. The topology derived preserves the network connectivity, and the degreetOPIOIClgycan In [47], Volkan Rodoplu and TeresaH.Meng proposed a location-based topology con trol scheme, R&M for ad hoc networks, where all the nodes have very accurate mfoffilatIon about their location. In R&M, each node eliminates any nodes in its relay region and onlyaifless power is consumed.Ifevery node maintains links with the nodes in its enclosure, ItISshown that the resultmg topologyISa mInImUm power, strongly connected topology. Due to these advantages, LMST and R&M have become widely known and benchmark algorithms for performance comparison. However, they still assume that all the devices are sameCOllh!sUI'ation.Homogeneous and non-homogeneous topology control algorithms differ from the one presented m Chapter 4. InthISdIssertatIOn, however, ItISno longer assumed that the net workdeVIcesare SImIlar; mstead, WIrelessdeVIcesWIthdIflerent capabIlItIes and character istics are considered. Therefore, known homogeneous and non-homogeneous algorithms anet1W'ork,Wireless have different receiver sensitivities, antenna gains, maximal transmission powers, and/or dIfferent battenes, and consequently, homogeneous or non-homogeneous algonthms can notbeused dIrectly. Heterogeneous network topology control problems have not been formulated and solved sofar.19


<:HAPTER3 ANCHOR LOCATION SERVICE PROTOCOL3.1The Anchor Location Service ProtocolThe ALS protocol proposed mthISchapter canbecategorIzed as a quorum-based nodes as location servers (anchors). ALS was designed from the ground up for large scale provide superior performance when compared to the other quorum-based and home-based schemes presented m SectIOn 2.1. For example, the anchor system establIshed m ALSISexpected to provide better search times than the schemes presented in Section 2.1, mainly becauseofproximity. On average, it is expected that search messages will go through fewer ALS is also expected to substantially reduce overhead compared to anyofthe approx Imate mformatIOn-based schemes smce floodmgISreduced to local exchanges. Further more, most of the schemes mcluded m the related work presented m SectIon2.Iwere designed with mobile ad hoc networks in mind. As a result, one can argue that in order to more cOluplex energy efficiency nor scalability to a very large numberofnodes are design considerations in mostofthose schemes. As a result, it is unknownifthese protocols can be used directly m large scale WSNs. These are addItIOnal JustIficatIOns for the comparIson of ALS wIth the TTDD protocol only.20


The ALS protocol was first introduced in [53] and then expanded in [54] and??In sensor rep ,res:enlledasaan axis and divided into equal-sized cells. The predefined geographical crossing pointsoftheroAl'1ctlrllrotcaas asCh''11rotlIWI''are reten'edgrid nodeasitssink agentto distribute its location information. Then, the sink agent selects some specialgndnodes asanchorsand bUilds ananchor systemthat contams the locatIOnofthe sink agent. When the sink moves, asink agent chainis formed dynamically to keepasa a sourceofthe sink agent. After that, data is transmitted using any location-based routing protocol. In the above case, the GPSR protocolISutIlIzed. In the followmg sectIOns, the global grid construction process, the anchor selection process, the query and data dissemination processes, and Sink and Target Mobility and Agent Chain Maintenance willbedescribed.3.1.1 ALS Global Grid Construction Processsensors areae!)lOve

G1G2G31 2 5 6-3 47 8A-...G4G5G9101314--11121516"'87 "'88 "'8Figure3.1The grid node selection process.6Once all sensors are in place, every sensor decidesifit is a grid node. First, each celloflengthais dividedbythe two midlines to create four smaller squaresofequal size (see FIgure 3.1). EverygndnodeWIllbe m charge of Its four nearby squares. Every sensor node then obtains its own location using any existing positioning mechanism and maintains its on ownsensor point the square belongs to. For example, in Figure 3.1,G1...Ggare nine grid points. NodeAbelongs to square 4, whIchISattached togndpomtG5.Ifm the small squares 4, 7, 10, and13(which are also attached toG5),there is no other node that is nearer toG5thanA, Awill sages to each other to establish a neighbor grid node table. In the example, after nodeAconfirmsItSrole asgndnode, It sendsALS_GN_GNJJECLAREmessages to thegndnodes ofG2, G4, G6, Gsto inform them thatithas taken the roleofgrid node. Node A also updates its own 22


table after receiving similar messages from its neighboring grid nodes. These messages are sent usmg geographIc routmg usmg the greedy forwardmg strategy, I.e., sendmg the packets to the closest node to the destination.3.1.2 Anchor Selection ProcessIn ALS, the sink agent distributes the sink's location information using an anchor sysaasloc:atllonservers. The anchor selection process is achieved by meansofpropagatinganchor setup mes sages.At the begmmng of the process, the smk agent, usmg locatIOn mformatIOn about its neighbors, sends out fourFirst Stage Anchor Setup Messagesin four straight orthogo nal directions (North, South, East, and West) and recruits all the grid nodes that lie along asmessages areinterrne<:!iatesensors tween two neighboring grid nodes and the recruited anchors store a copyofthe sink agent's locatIOn. The anchor selectIOn process needs to consIder specIal cases, such as relaymg the setup messages around void areas and the borderofthe network. Face routing mechanism is messages can areas with one another. For instance, once a setup message arrives at the border or at a void area, it is divided into twoSecond AnchorThe borderofthis void areaISthen partItIoned mto two parts and each second stage setup messageISrouted around oneofthese partitions using the right-hand rule or left-hand rule [26,27]. (The proofs that samecan be better explained by looking at the example shown in Figure 3.2. The North first stage anchor setup message, which is represented by the arrow coming from the bottom,l1l<}vesal()llg tlie.t=.]lIlle. }\t p()Illtfl,tlie first=stage allcli()f setllP l1lessage cliec](s tliel()=cal neighbor grid node table and cannot locate the next forwarding grid node in the North 23


j9IJ!""V'-....7Voidarea\.....I-...........6Cj'.......'jll57n u4...e--_L.""../j3\Voidareaf'-.-....AjI'"X=l1Figure 3.2 Left and right hand rule in the anchor setup process. direction. The first stage anchor setup message is then split into two second stage anchor setup messages, which use the right-hand and left-hand rule to navigate around the perimeter of theVOIdarea. Once a second stage anchor setup message comes back to thex=llme, at pointB,itis split again. At that point, one second stage anchor setup message continuesIQTll-rlanoor area setup messages arrive at particular grid nodes which have already been visited, e.g. points C andD,they stop propagatmg and the nonnal setup process resumes. process, once a message a not only checksifit has received this setup message before, but also adds other already strategy is known aspropagatedJusion,which will significantly reduce the time delayoflocatIon propagatIOn mfonnatIon and proVIde better pertonnance. Convergence proof of the anchor setup process is included in Claim 1 and Claim 2 in Appendix A.


Figure 3.3 The anchor system and the query and data dissemination processes.3.1.3 Query and Data Dissemination Processessensor networJe somenh'u'''r'<:ll"UUH.UU'"\....".1">""'1At that point, one sensor node will sense the target and will become the source node that will transmit the sensed information to the sink. In order to do that, the source node will regIster ItselfWIththe nearestgndnode, whIchISknownasthesource agent.The source agent will then send four query packetstofind the locationofthe sink agent. Once the source finally sends the data packets to the sink agent using the GPSR protocol [28]. The query process is quite similar to the anchor system setup process. First stage query packets are sent to four orthogonal dIrectIons usmg the same strategy utIlIzed m the anchor messages messages arenel:w{)rkor a area process.25


the process stops when these messages find each other. Upon reaching the anchor system, the anchorsproVIdethe source agent wIth the smk agent'sgndpomt coordmate. Anchors also send these replies utilizing geographic routing with the greedy forwarding strategy. one serveasThis is also beneficial because it shortens the query time,asthe source will utilize the first response. The globalgndconstructIOn process, the anchor selectIon process, and the query and data dissemination processes, the anchors, and the query and data forwardingare3.1.4 Sink and Target Mobility and Agent Chain MaintenanceAs stated before, a sink agent chain is formed dynamically to manage moving sinks andaVOIdcreatmg a new anchor system every tIme the smk moves beyond Its current cell. After a smk selects Its first agent (the pnmary agent),Itkeeps updatmg Its mstant locatIOn until it finds another grid node that is closer to itself. At that point, the sink selects the newasnew new about its own location. Thus, a sink agent chain is built to keep the anchor system intact while tracking the sink node locally. The protocol is designed to allow the chain to have up a newone. ALS does not keep trackofmoving targets, i.e., no chain is built and maintainedascase sensor sources. Then, source nodes find source agents that query the anchor system in searchofthe sink agent's location.Ifthe target moves, a new sensor will become the source node.Ifthe new source nodeISstIlI wIthm the same source agent's area of coverage, the source agent will not trigger a new anchor query. However,ifthe target moves beyond the current26


cell or outofthe rangeofthe current source agent, the new source node will find a new source agent, and the source agent wIll query the anchor system agaIn.Itis important to consider how the source agent knows the new positionofthe new sink movesh",.rnr,r1sourceInorder to avoid this situation, the protocol includes a process whereby the source agent quenes the anchor system penodlcally. The frequency of the queryISset consldenng the moving speedofthe sink node. This information can be dynamically adjusted or statically casescan can speed estimations and convey that information back to the source agent periodically.IntIlls\V()rk,tIlestatICappr()acIlls Illlplelllel1te(l; tIle(lYl1allllcappr()acIl \VIlllJe part()fflltllre3.2 Theoretical AnalysisofALSInthis section, the scenario and notation utilized to analyze the communication and model will also serve to validate the simulation models and results later. Simulations are utIlIzed to evaluate the ALS protocol more thoroughly,InpartIcular those aspects that In volve tIme, suchasthe average locatIOn tIme and anchor system setup tIme, among others.3.2.1 Scenario and Notationsquare area as areaofEach cell withinAhas the same side and the areaofeach cell is then0'2.The total numberofsensors is assumed to beN,uniformly distributed so


sensor areKSsources. sensors and sulks have the sameradIOranger.ItISalso assumed that every smk receIves a total of D data packets.3.2.2 Communication (berheadThe total communication overheadofthe ALS protocol consistsofthe following five processes:2.Thelocal flooding in the sink's cell that selects the sink agentGSink;3.Thelocal flooding in the source's cell that selects the source's agentGSource;source performs order to obtain the locationofthe sink agent, and the reply messages;5.The transmissionofthe packet from the source to the sink;Inthis analysis, only the communication overheadofprocesses1,4,and5will be considas no3.2.2.1 Process #1: Establishmentofthe Global Grid and Anchor Systemare serve as lOc:auon serversonits behalf. The total overheadofthese processes is given by:ogrid-anch01'_systemKX(2R(m)+l)X(+6xI71xRx(VA))4x171X(R(VA))X(R(28(3.2)


aI I I(o!)(ilU'22'21----Ct2q=-J21Figure 3.4 CalculationofCl.whereRISthe Round functIOn. EquatIOn 3.2 conSIders that there areKsulks m the system and also the moving sink situation.Ifit is assumed that sinks move at the average speed chain will break at most m times and a totalofm+1anchor systems will be established, m course, are The first part of EquatIOn 3.2 represents the overhead of bUIldmg one anchor system times the numberofsinks and the numberoftimes the chain is broken due to mobility. Thefactor is the overhead in numberofhops incurred in transmitting the message betweensinkvandGsinkv.This is shown in Figure 3.4 and calculated in Equation 3.3. which is the maximum value that29can take. Because it is assumed that the


sensor nodes areuniifolrrnllydistlibllte<:l, 'seX1Jeclatlonis:dxdy0.3824(3.3)orwill be established using six linesofgrid nodes, as shown in Figure 3.3 three linesofgrid nodes traversIng the network vertically and three lInes ofgndnodes traversIng the network horizontally. Every traversing pathofthose six lines is approximately partitioned intoR(v;:)segments. Setup messages visit every grid node in their path in order to recruit the anchors. to over to traverse oneThe second partofEquation 3.2 is the overhead for setting up the global grid structure.Dunngthat process,ALS_GN_GNJJECLAREmessages are sent among neIghbonnggndnodes. Every selected grid node sends four declare messages to its four potential neighboring gridortho.golt1aldIrectIOn,s.2 x xsegments between all neighboring grid node pairs, i.e., the total numberofcell sides.rlis the numberofhops between each two neighboring grid nodes. The final coefficient is 4 becausetransmISSIOnsareInboth dIrectIOns. AnalyZIng EquatIOn 3.2,Itcan be seen that+ Process #4: Querying the Anchor SystemThis process consistsofthe communication overhead incurred by the source agent when 30


the anchor system. The overheadofthis process consistsoffour messages from GSourceto f()llr aIlcli()rs all(l tliereplIes,asf()II()\Vs:(3.4) The first term, is the overhead when the source sends a query to GSource'This overheadISexactly the same as the one already calculated by Equation 3.2, when thesmktransmIts a message toGSink.ThISexplams why the same vanable,CI,vISutilIzed, whIch has the same expectation calculated in Equation 3.3. The second term is the overhead source which due to symmetry is The third termofthe equation represents the overhead incurred by the query message when it goes from GSourceto the anchor system. On average, the overheadoffour queries from GSourceto anchors is boundedby6 xrXR ,l.e., no matter where the sink is located, GSourcetransmits two messages in the horizontal axis one. messages encounter the borderofnetwork areaorvoid area, they split into two second stage messages and use the right hand rule and left hand rule to route around, forming six lmes of messages over the entIre area.ThISvalueISat the same time multiplIedbythe numberofsegments and the numberofhops per segment. The last term in Equation 3.4 is cre:l.tes a ansinks' location information. The distance reply messages need to travel through depends()Iltlie relatlye l()catl()Iltlie s()llrce agell! all(l tlie aIlcli()rs. Ilieref()re,C2xare mtro(luc:ed aredetlerrrlinc:d31


and the relative position between GSourceand the network border. The four verticesofthe square are(0,0),an artlitrary areatYl .xo:SYl,Y2:S1 as(3.5) (3.6)Inorder to calculate the message overhead in a network withKsinks andSsources, the following steps are taken. First, the message overhead incurred by the query from aSourceto ItsGSourceand the subsequent reply are calculated usmg EquatIOn 3.7 (3.7) Then, the overhead of the query messages fromGSourceto all the anchors and the theIr subsequent replies is calculated, which is given by Equation 3.8:OCSOllrce<*anchors6xXR('?)(3.8) The first part of EquatIon 3.8ISthe overhead of the query messages fromGSourcetoKanchor systems, which traverses the network like the anchor system setup process. This source are the overheadofthe replies from the anchors to GSource,two messages in the horizontal direction and two messages in the vertical direction. 32


Finally, the total query overhead withKsinks andSsources is:OSource(u)?anchors=LFrom Equations 3.5, 3.6, 3.7, 3.8 and 3.9, the expected total overhead is given by: (3.9)+6xrxrFrom Equation 3.10, the complexityofProcess #4 is0(S+K'?)). Process #5: Data Packet TransmissionInALSdata packets are transmitted from the source to the sink agent using theGPSRtraIISIYlitone packet( 0::;C3,(u,v)::;V2)(3.11)whereC3,(u,v)is determined by the straight distance betweenSourceuandGsinkv 'However, an Therefore, for the theoretical estimation, a two dimensional space(n2)is considered sink receivesDdata packets, the total data forwarding overhead is:Odata=KxDx(u=1,2, ...,S-1,S;v=1,2... K-l,K)(3.12) From Equation 3.12, it can be seen that Process #5 has a complexityofO(KD,?).33


3.2.3 State ()verheadThe total state overheadofthe ALS protocol is givenbythe storage space neededbyHVU'-,:>,aJGctlor:s,source total, there are at most(R(+1)2grid nodes in the whole network area, and each spends at most 4 space umts to save the neIghbors' mformatIOn. Thus, the storage complexIty for a grid node is0(1).ForKsinks andSsources, the total storage overhead for the grid system is: (3.13) For every sink, there are approximately6R ('?)anchors in the network. Each one will use one space unit to store the locationofthe sink agent. Therefore, its storage complexityISQ(1).f()r1\SI111(Sa11Cl5Js()llrces, tIle t()tal st()rageIS:6xRxK(3.14)EveryGSinkandGSourceneeds to storeItSsmk'sand source's locatIOn, respectively. Thus, the storage complexity is0(1).ForKsinks andSsources, the total storage is: The total numberofstates mamtamed m anchors,gndnodes, and numberofGSinkand (3.16)34


The complexityofthe state overhead can be calculated from the equation above as3.3 Performance EvaluationAs stated before, the performanceofthe ALS protocol from the location service pointofmore intere:sting,EXlperilme:ntsare perforrned determine the grid and anchor system setup time and overhead, the sink location time and overhead, and the state overhead of the protocol, whIle varymg the cellSIze,the numberofsinks and sources, the network size, the network density, and the mobilityofthe sinks. The ALS protocol is implemented in the Network Simulator 2 (ns-2) [55] to validate the as mathematical results included in[4,6]and the ns-2 simulation models found in [5] are utilIzed to compare ALS wIth TTDD. Two hundred and fifteen stationary sensor nodes and four stationary sink nodes are uniformly deployed in a two dimensionallOOOrn xlOOOrnnetwork area.Inthis scenario,was a those 215 sensors were chosen as source nodes. Each experiment is run nine times toaVOidcases where the source and smk nodes were very closeorvery far apart from each other only. As a result, everypomtdrawn m the graphsISbased on observations of mne random deployments.Inall simulations, control packets were 36 bytes long while data pa(;kets were investigation, each source node generated one data packet per second. Except in those simulations where the effectofthe cell size is under observation, the parameterofcell size0:ISset to 200 meters smce It was found to be the best pefformmg value. Sulk mobIlIty 35


was implemented using the standard random way-point model with zero pause time and a fixed speed from source to destmatIOn pomt taken from a umform dIstnbutIOn between 0 and maximal speed. are moaelmgmClUamga agation model supporting propagation delay, capture effects and carrier sense, and radio network mterfaces.Inorder to mImmIze collIsIOns among anchor setup messages ongI nated from different sinks, each sink node is assigned a random back-off time interval bethat models the contentionofnodes for the wireless media. The wireless interface worked lIke the 914 MHz Lucent WaveLAN DIrect -Sequence Spread-Spectrum (DSSS) radIo m terface [57]. The signal propagation model combined both a free-space propagation model and a two-ray ground reflection model. The free-space model was used only when the translnitter wasUl1th,ncross-over distarlce model was used. The radio transmission range was set to 250m. Since these models are are stated, these parameters are utIlIzed m all expenments.3.3.1 Optimal Cell SizeThe impactofthe cell sizeonthe performanceofthe location service is investigated first four sources is considered, and the cell size varies from100mto1000min100mincrements. The anchor set up time is analyzed first. As stated before, each point in the plot is the that the setup times are fairly similar regardlessofthe valueof0'.Also, it canbenoticed that 36


0.12,U0.1!j::0.08Figure 3.5 Anchor setup time vs. cellsizea.occursa=occursa=This has to do with the nodes' transmission rangeof250m.Witha 200m,the anchor setup packets can reach the grid nodes in one hop; however in the case ofa 100m,there are about threegndnodes wIthm the250mtransmISSIOnrange, and setup packets need to go through eachofthem. An additional observation is the short amountoftime needed to case ms. can average protocol overhead reduces slightly with the cell size. This slight reduction is due to the reduced number ofgndnodes when the cellSIzemcreases. As a result, fewerALS_GN_GN..DECLAREmessages arewereasindependentofa(see Equation 3.2). The results also show the small overhead introduced Figures 3.5 and 3.6, it is concluded that cell size has little impact on the grid and anchor setup processes. In Figures 3.7 and 3.8, the average location time and overheadofALS are compared with thoseofTTDD. The location timeofthe protocols is definedasthe time elapsed source 37


+ -ALS_ExperimencResults8 -ALS TheoreticaLAnalysisFigure 3.6 Grid and anchor setup overhead vs. cell sizeG ....-........ Cell Size (m)FIgure 3.7 LocatIOn tlme vs. cellSIzeG.receIves the locatIOn of all the smk agents aVailable m the network. The figures show that the pertoffilance of ALSISalmost mdependent of the cellSIze,as the values remam faIrly constant. The slight increase in the location time is due to the relay messages taking longer source source can the figures, ALS outperforms TTDD by reducing 80% the location overhead and 50% the location time for almost all valuesofG.InTTDD, sink nodes flood messages within one cell to find the nearest data dissemmatIOn node; therefore, the bIgger the cellSIze,the more overhead. 38


Figure 3.8 Location overhead vs. cell sizeG.'"0;0.05Q'"2i0.03.'!la.>'"l:!0.01 Cell Size (m)FIgure 3.9 Data delay vs. cellSIzeG.The Impact of the cellSIzeon the average data delayISstudIed as well. The average data delayISthe time perIod from the moment when a data packet appears at the source until the packet is received by the sink.Itconsistsoftwo major parts:sink location timeanddataa sourceprocess to find the locationofall the sinks in the network. After that, the source sends all data packets using the GPSR protocol. The time for every data packet to travel from source to smkISthedata propagation time.Therefore, the average data delayIScalculatedasfollows:39


SinkLocatiorLTime(3.17) whereDis the total numberofdata packets. Figure 3.9 shows the resultsofthese experi-D 100ALS _State_Overhead ExperimenCResuItsALS_ State_ Overhead_ Theoretical_AnalysisnDD_State_Overhead ExperimenC ResultsFigure 3.10 State overhead vs. cell sizeQ.Figure 3.10 shows the state overheadofALS and TTDD.Asexplained before, the This can also be explained by looking at Equations 3.13 and Impactofthe Number of Sinks and SourcesIn order to evaluate the impactofmultiple sinks, the numberofsinks is varied from 1 sources are deployed and the sizeofthe cell sizeQis set to200m.Figures3.11and 3.12 show the average anchor setup time and average grid and anchor setup overhead, respectively. As It can be seen, the anchor setup tIme remams faIrly constant regardless of the numberofsinks in the network. This is because in ALS, each sink sets up its own anchor system 40


independently and without interfering with other sinks. As expected, the grid and anchor setup overhead mcreases wIth the number of smks. The hIgher the number of smks, the higher the overhead,asALS establishes one anchor system per sink. This can also be234 5 6 7NumberofSinks8 910Figure 3.11 Anchor setup time vs. the numberofsinks.a.ja;300ol:<1:200'Cl:"'15+ -ALS_Experiment_Results ALS_ TheoreticaLAnalysisFIgure 3.12Gndand anchor setup overhead vs. the number of smks. Simulations regarding the location time and overhead are also conducted. From Figure 3.13, it can be seen that the location time slightly increases with the number of sinks.ThISISbecause when multIple smks are deployed m the network, locatIOn query messages


0.12,+ -ALS_Location_TimeTTDD_Location_Time0.1-------........-.......FIgure 3.13 LocatIOn tIme vs. the number of smks. always have to travel until they find the farthest anchor system. while a longer average locatIon tImeISexpected, the proposedpropagated-fusionstrategy compensates forthISfactor. This strategy spreads sink's location information in other anchors without incurringa the queries insteadofthe point-to-point query scheme used in ALS. Similar trends can be observed in the caseofthe average location overhead in Figure 3.14. More sinks imply more anchor systems and more query response messages.+ -ALS_Location_Overhead_ExperimenCResultsALS_Location_Overhead Theoretical_AnalysisTTDD_Location_ Overhead_ExperimenCResults.e:-g.f400FIgure 3.14 LocatIOn overhead vs. the number of smks.


Figure 3.15 shows the caseofthe state overheadofthe protocols. From this figure, it canbeseen that ALS's overhead mcreases lmearly wIth the number of smks whIle TTDD's remams faIrly constant.ThISISbecause ALSISsmk-onented, whIle TTDDISsourceoriented. Equation3.14explains this behavior.3.3.3ImpactoftheNumberofSources0.1234 5 6 7Ni.iiTIbefrifS6ufCes8 9 10vs. sources.43


1602 34 5 6 7NumberofSources8 910Figure 3.17 Grid and anchor setup overhead vs. the numberofsources. The Impact of havmg multIple sourcesISalso evaluated. As before, the default scenano weremc:re:ase:dare presentedofthem remain fairly constant because anchor setup processes are initiated by the sinks and have very lIttle Impact from the number of sources ...+..-....."""...----+---Figure 3.18 Location time vs. the numberofsources.


The average location time and overhead are shown in Figures 3.18 and 3.19. The number of sources have lIttleIfany effect on the performance metncs becauseInALS, query processes are independently conducted by individual sources and do not conflict with each case is constant because when sinks are static, the global grid system is constructed once and the anchor systemISbUiltKtimes. However, these two major parts of the state overheadIn+ -ALS_Location_Overhead_ExperimenCResults ALS _Location_Overhead Theoretical_Analysis TTDD _Location_ Overhead_ExperimenCResults.e:-g1!400A_FIgure 3.19 LocatIOn overhead vs. the number of sources. Equation 3.16 do not depend on the numberofsourcesS.On the other hand, TTDD floods the network S times to establIsh one globalgndpersource.3.3.4 ImpactofSensor DensityMost results presented so far consider the base scenario where 215 sensor nodes aredelolclvedover a square areaxanseems case scenario has been utilized in many other studies. So for comparison reasons, it is also utilIzedInthISstudy.ThISscenarIOISalso beIng used because the node degreeISnot ex-45


::::i!i:::::ti:::::;l;::FIgure 3.20 State overhead vs. the number of sources.0.1...'"So>-...J:r-..-l3o _..........___ ___25 30 33vs. pected toplayamajor role in the performanceofthe ALS protocol. Unless the network is establish the anchor system going around void areas without any problem. In order to test the hypothesIs, several sImulations were conducted usmg the base sce-nanoutilized throughout the chapter, but varying the numberofnodes, so that average node degreesof4, 6,8,10, 12, 15, 20, 25, 30 and 33 are obtained. From Figure 3.21 to overhead simulation results are presentedasa functionofnode degree. The simulation 46


"0Sl.c1400120Co:::l;;l100,g80()l:-;60\tll4020--s--0 4 6 8 101215 202530 33Figure 3.22 Grid and anchor setup overhead. vs. the network density.0.1!i=Cl>g'0.04!0.02___-D--13'Ei....Figure 3.23 Average location time vs. the network density. results confirm the stated hypothesis, i.e., the ALS protocol is unaffected by the densityofthe network.3.3.5 ImpactofNetwork AreaInthISsubsectIOn, the scalabIhty of the protocolISexplored. The same sensor denSIty is used as in the original scenario, and54,215,464and 815 sensors are deployed in500x xxxneltw()rkareas an



0.12r;=====cc=::=c===.===;:c==C---,--------,--------,----,-----,-+-ALS_Location_Time TTDD _Location_Time0.1i=--_..+..'"..'---o0.51.5 2 2.5Network Area (Km2 )3 3.54FIgure 3.26 LocatIon tIme vs. the network area.e1j..o,]t1l20

protocols are fairly similar. Although the protocols' overhead increase linearly with the network area, both schemes present sImIlar values and slopes.3.3.6 ImpactofSink Mobilityvs.ne1tw()fkarea.20Inordertostudy the impactofmoving sinks on the location service, four sinks are set to move at the speedof5m/s, lOm/s, 15m/s and 20m/s. The maximum sink agent chain 50


160 140Q.:::lll120(f),g100(,)c80"0C640C)..Q.;20.7(10 15Maximal Sink Velocity (mls)20Figure 3.30 Grid and anchor setup overhead vs. sink mobility. a onlsm:al one to chain up to two new agents.Atthat point, the sink will adopt a new agent, and this new onigInalonePO:SItIon,socan new as primary agent, build a new anchor system, and send a cancel message through the previous anchor system to erase It. The addItIOnal overhead and tIme Incurred In the cancellatIOn9III8Ei=76c'iii5'.--(J._----c ,-Q)C)3210 15Maximal Sink Velocity (mls)..'..'.-'20Figure 3.31 Agent chain break times vs. sink mobility. processISconSIdered In the averagegndand anchor setup overhead and average anchor51


setup time. Figures 3.29 and 3.30 illustrate the anchor system setup time and overhead when the sulks are movmg. As expected, these metncs mcrease wIth the smks' movmg speed. The faster the sink moves, the higher the chance to break the sink's agent chain, causes more a new one. can seen5m1s,each sink only sets up the anchor system once, however when sinks move at 20m/s, each smk breaks the agent cham once. Figure 3.32 Location time vs. sink mobility.ALS_Location_Overhead_Experiment_Results ALS _Location_Overhead Theoretical_Analysis TTDD _Location_Overhead _ExperimencResults--_..-..-..Maximal Sink VeloCity(m/s)Locationvs. 52


The impactofthe sink's mobility on location time and location overhead is also ana lyzed. The results are presented m FIgure 3.32 and FIgure 3.33, respectively. From these figures, it can be found that mobility had a little effect on the metric. The average locationare movemove dIe the little extra overhead. Similarly, the state overhead in Figure 3.34 presents a fairly constant behavIOr wIthv.ThISISbecause every time the chamISbroken due to smk mobIl ity, a new anchor system is setup but the old one is then eliminated. So, at the end, there is..................................FIgure 3.34 State overhead vs. smk mobIlIty. just one anchor system per sink. Equation 3.14 also explains this.3.3.7 Total Communication OverheadIn this subsection, the total communication overheadofALS and TTDD will be re ported, and wIll mclude, the data transmISSIon overhead. As explamed earlIer, GPSRISutilized to route packets over WSNs (ALS+GPSR), while TTDD has its own routing mechcOInrrlUnliccltiolno'ver.healdversus 53


10000, 90008000Si70006000 5000 4000 3000 2000 100+ -ALS_Communication_Overhead_ExperimenCResults ALS Comm u nication _Overhead Theoretical_Analysis TTDD_Communication_Overhead_ExperimenCResults---------_..Figure 3.35 Total communication overhead vs. numberofsinks, a=200. overhead is very small and stable compared with TTDD's, which utilizes floodingofquery messages.e,700016000 5000 4000COI1nmlllllc;atl()nO'lerbleadvs.nUITlberIn Figure 3.36, the numberofsources is varied. The trends are similar to the case with multiple sulks. In TTDD, every sourcebUIldsIts owngndsystem and the smk queryISforwarded through each individual grid system, which is why the communication overhead increases dramatically with the numberofsources. On the other hand, ALS builds an sources.54



<:HAPTER4 RESIDUAL ENERGY AWARENESS DYNAMIC TOPOLOGY CONTROL4.1Network ModelInthe tradItional homogeneous network,ItISbetter to transmIt packets through many short links rather than directly. However, this approach only considers the distance be tween the nodes, which is only effectiveifthe network is homogeneous.Inheterogeneousa new apt)roilch ent characteristicsofthe devices and the desire to assign more powerful devices a more promment role m the network.InthISchapter, the network model under mvestIgatIOnISformulated first, then the centralized and distributed versionsoftheREADtopology con trol algorithm are presented along with their performance evaluation. a heterogenleOlls wlfeJless netwolrk resented aGwhere the nodes wir'ele:ssarearerarLdoimJya2-JJnnension nellw()rk area. setcostvalue associated with each edge in the graph G. Theweighted costfunctionwillbeexDe1penldiulgon wireless networks can consistofnodesofdifferent typesofdevices, such as sensors, PDAs, robots, and even more powerful military devices. Becauseofcertain characteristicsofhet er()g el1e()llsl1etVV()rks,tileIl1aX.IIl111Il1tral1sIl1IsSI()l1ral1ge, resI(llialel1ergy,al1tel1l1a gaIl1'al1(lreceiver sensitivity may vary from one nodetoanother. Therefore, nodevis associated with 56


attributes(Pmax_v REv,IBv).The notationPmax_visv'smaximum transmission power in v toEtopackets.f3vis the sensitivityofv'santenna in decibels.RE_Evis current residual energy in node v in Joules, and decreases withv'sactivity from initial valueINI_ENERGYtozero:::Considering two random nodes, u and v, with Euclidean distanced(u, v) between them, It holds thatu'stransmISSIOnISsuccessfully receIved at vIfEquatIOn 4.1 holds as follow:swenJ2;that v clidean distance between the nodes andPuis the transmission power atu.Inaddition, two randomuand from setVare connected a linkEin Gifuv arecOlme:cte:dconsidered in this dissertation, which is supported by the widely used IEEE 802.11 wireless network MedIUm Access Control (MAC) protocol. MAC sends Imk-Ievel acknowledg ments for all um-cast packets, so that all lInksbUIlton top of 802.11 network must be bi -directionaL4.1.1 Maxpower Graphderlote:das(V(G), E(G), Pmax,RE,B),whereE(Gmax)is the edge set when all the nodes work using Ac;co,rdilngto can57


bedefined by Equation 4.2 as follows: Note that the graphGmaxrepresents the setofall possible bi-directional communication is strongly connected, the network topology generatedbythe proposedREADtopology control algorithm also preserves the strong connectivity.4.1.2Ene:rgyModelPf()p{)sedpnpt',nlawaretop101C)gyCOIltf()l",,)O,'-'u,,'uuthe network lifetime by ameliorating energy consumption among different kindsofwireless devIces. Dependmg on the type of devIce, the amount of energy consumed by the radIotraJlS0eIVI=rcanoptimizing the energy used for communication is an important issue, and the energy modelUUJU,,",,",uon on cornmunications, consuming factor [2]. The amountofenergy consumed by a wireless interface canbedescribedbythe following simple model:+whereEe1ecis the energy used to run the transceiver circuitry in signal processing, for mstance channel codmg, mterleavmg and modulatIon;Eamprepresents the energy usedbythe amplifier to transfer the signal and satisfy the receiver's sensitivity requirement; andEsensedenotes the energy for sensing the wireless channel before the transmission takes place. Note thatEampis the productofthe transmission power and the transmission 58


time in Joules. Usually,Ee1ecandEsenseare neglected for the simplicityofdiscussion. on affirms that the higher theEamp,the higher the energy consumptionEcon.mmebut also thatainEconsumeis over-proportional to the increase inEamp.Therefore,Eampis hereafter used to denote the amountofenergy consumed by the wireless interface during communication. Also, the followmg equatIOns hold: How to determine the power consumed during the transmitting and receiving processes still remams unanswered.v(}1Jallclat thesaI11etiI11e[Jr1J=ucapturecl])yll()cleLtis a])()'le(}u,.The relationship between transmission power and transmission range is discussed in a tranlsmlls a me:ssa!;e following model from [58] is used to compute the power consumptionPneeded to send message: wherekandcareto the environment and are constants for thesPE:citicwirelesscincreases with the distance between two communication nodes.59


There are several common path loss models evaluated in the literature to describe the propagatIOn m the WIreless medIUm.IntheTwo-Ray GroundpropagatIOn model, the relationship between the powerP tused by the amplifier to transfer packets and the signalas:PtGtGrh;h; d4L(4.6)IntheFree Spacepropagation model, this relationship is described as: (4.7)InEquations 4.6 and 4.7Gtis the transmitter antenna gain,Gris the receiver antenna gain,Lis the system loss factor and is independent from propagation, and/\is the wavelength inaredis the Euclidean distance between the two transceivers.(4.8)delperldson agatIon. Combmmg EquatIons4.Ithrough 4.8, It can be concluded that m order for the receiver to receive the signal correctly, the following relationship must hold:f3.d2P.t> -'--Of(4.9)Given the above equation, the minimum transmission power required for each communication can be computed. The notationPzt,Vis used to represent the minimum power required60


trmlSlIlitt(;rutrarlsmittiIlg a utIlIzed to descnbe the power used by nodevto receIvethISmessage. Note that:Pu,vlBd(u,v)2Of(4.10)or consumed by the electronics at receivervduring the processofreceiving a packet, is proportional tov'smaximum transmission powerPmax_ v ,and is independentofthe sender's reLatic)llship can where the the ratio between the transmission and reception power con-tsumed at the node, which depends on the typeofwireless card. For example, the CISCO atrmlSITnSSIOn/rec;epl]onora4.1.3 Weighted Cost FunctionIn order to mcrease network longevIty as much as possIble, the proposed READ topology control algorithm considers both the energy for sending and receiving data and the as ahOITIoJgenleolLlswork, a new weighted cost value is introduced for each pairofnodes. Since the maxpower graphISbI-dIrectIOnal, ItISonly necessary todISCUSStwo asymmetnc communIcatIon lInks for each pairofnodes in the heterogeneous network. Assumenodeuandvare within the neIghlbor setofeach for onedtr'ec1tlOll,re1=,res:enltsthe welgntea cost for transmitting and receiving datafromutov;for the reverse direction,wv-+u (e(u,v))61


describes the weighted cost for transmitting and receiving data fromv tou.WU---7V(e('11,v))ISdefined as follows: (4.12) whereREuandREvare the resIdual energy at each node;Pu,vISthe mmlmum power for'11to tovtorer:ei\Tearerf'{'f'l'lTln,aatU1J---7U="----'--REv(4.13) pertorm a succ:esstilJ c()mnaunication. areCOlt1S1ICleI'ed,communication costs in both directions are treatedasa whole. Thus, the weighted cost is defined as:+Therefore, given two edges(Ui,Vi)and(Uj,Vj),it holds:

The above relatIOnshIp guarantees a umque outcome m the edge selectIOn step m the READ topology control algorithm, which is further discussed in the next two sections. areranldOmljdetHO'veaaownmSlxirnUlTItr'an:;;missilonnA'";>1"All the possible edges are presented in maxpower graph and associated considering the mimmum requested power as defined by the weIghted cost. Edge weIghted cost can be calculated from distance and sensitivities. Every node can adjust its transmission power between zero and its maximum transmission therefore different subsetsofmaxcontrol algorithm READ is presented. It selects certain edges from maxpower graph by adJustmg the nodes'transmISSIOnpower to each other to meet the goal of extendmg the network longevity. In Section 4.3, the distributed versionofthe topology control algorithm (DREAD) is presented.4.2TheCentralized Residual Energy Aware Dynamic Topology Control Algorithmcontrol algorithm is presented in detail. This work was first introduced in [60].4.2.1 Centralized Residual Energy Awareness Dynamic AlgorithmThe centralized READ algorithm has two phases: the Initialization phase and the Topol()gyC:()llstruCtI()llpIlase. TIle IllltializatI()ll pIlaseC()llSIStS()ftIlef()II()\VIllgsteps: each transceiver. Note thatGmaxis bi-directional. Without lossofgenerality, it is also assumed thatGmaxis strongly connected.63


2.For each edgee(u,v)EE(Gmax),compute the weighted costw(e(u,v))wu---->v3.Afterstep2,eachedgeeCu,v)EisassoCiatedwithaweightedcostw(e(u,v)).Sort the setE(Gmax)in increasing orderofw(e(u,v))based on the weighted cost relationship described previously in subsection 4.1.3.Toresolve thetIle s()rte(i e(ige seqllellce',VIIIrellallle(iasE) netwolrkempty edge setE(GREAD).Consider every node in the original network graph as an iso lated component setGi ,i.e.Gi{Ui}'During the construction process, two component sets at a one to the nodes have been connected and there is only one component set left. The resulting asInitialization:1:ConstructGmaxalong withE(Gmax)Etom=4: Createaselement set for eachEare more one isolate:d scanned) supposeuEG,andvEGT64


10:u13:Endifis theresultJingtoP'OlClgy.which is also strcmgly Thusconnectedandbi"'directional.ItISworth pointing out that the above algorithm can be applied k-times to preserve the mherIted k-connectivityofthe maxpower graph. BasIcally, READ can simply resort the algorithmISto never node and does not havecreate isolated sets Evenifthemore which are common phenomena in wireless networks. When someofthose components merge,d, k,-cOllnecthrity can the algorithm will be terminated when all the remaining edges have been scanned. The4.2.2 Simulation Results and EvaluationofREADThe simulation setup parameters will be described first, and then the simulation results and the respectIveanalySISconSIderIng SImulatIOnsWIthand WIthout datatransmISSIOnWIllbe deSCrIbed. 65

PAGE 77 Simulation SetupIn the simulation experiments, a heterogeneous wireless network with four typesofdea-LV\JVIICX-L\J\J\JfI',netvv'orkarea aredistribtIte(:l, areHIHHpagatIOnbut determined the characteristicsofeachoftransceivers. Without lossofgenerality, only omnidirectional antennae are considered in this simulation andGtandGrwere set to1.The system loss factor is set toL -1,and the operational frequency equal to 2.4 72GHz. Also from Equation 4.11, it is known parameterISset to a constant value0.6for all the nodes.ttxandtrxare both set to0.01second, meaning that packets are consideredofequal and fixed size. The above simulationrpfi"'n'prlas Table 4.1 Simulation parameters for each typeofdevice.Category/3,f3PLPHI_ELLEHPer,H(dBm) (dBm)(W)(W)( x101)( xlOJ)%Military-81 -67 60 753000200005n I-812.0 180 720 10PDA-81 -67 0.1 0.2367220Sensor-101 -650.0000040.1 0.1 366566

PAGE 78 Simulation Results and Analysis Without Packet TransmissionFIgure 4.1 Maxpower graph wIth 100 nodes. parameter settings is kept, and there are no transmission activities taking place. First, a graphIcal companson of the topologIes that each algonthm producesISprovIded. FIg ure 4.1 shows the maxpower graph generated when each node uses its maximum trans can as arefere:ncle.FIQuresthen plot the topologies generated by LMST with link addition (LMST-add), LMST with lnik removal (LMST-rem), R&M wIth hilk addItIon (R&M-add), R&M wIth lmk removal (R&M-rem), READ with bi-connectivity (READ-K2), and READ uni-connected (READKl),respectively. canj.'\.Uo(.,lVJL-n.uuare connected, while LMST -Rem and R&M-Rem do not generate a connected topology even 67


Figure 4.2LMSTwith link addition. Figure 4.3LMSTwith link removal. though the maxpower graphISconnected, whIch contradIcts theongmalgoal of extendmg the network longevIty wIth preservatIOn of network connectivIty. Figure 4.8 shows the average node degree for each typeoftopology control algorithm. can seen68


FIgure 4.6 READ wIth two-degree connectIvFIgure 4.7 READ wIth one-degree connectIv-Ity.Ity.neltw()rkarea or LMST-add and LMST-rem remain fairly constant and with a very low node degreeofaround 2 to3.Average degrees of R&M-add and R&M-rem are always hIgher than READ and LMST, which can also be also confirmed by looking at Figure 4.4 and Figure 4.5,asthe topologies generated by R&M are denser than the others. gorithms. In general, the average link length of the topology control algorithms decreases wIth the number of nodesIIIthe network area. InthIScase, READ's average lnik length lies between R&M's and LMST's. Results and Analysis with Packet TransmissionAs earlier mentioned, the weighted cost is a functionofthe residual energy, which control algorithms in termsofnetwork lifetime, the data packets will be transmitted after the topology is built. In contrast to the previous simulation scenario without packet trans-69


45 40e35eJj2530"25500mission, this partofthe study will consider packet trafficasthe default simulation scenario. FIrst, the settmg wIthout packettransmISSIOnISkept as before and four packet collectors, or sinks, are randomly deployed in the network area.50==e==;---,----,---,---.--,READ-K2 R&M-Addw.lI<:...R&M-Rem LMST-Add LMST-RemMaxPowerGra h2015Niiriibcf6fN6dcsFigure 4.8 Average node degree.600READ-K2 R&M-Add..lI<:....R&M-Rem LMST-Add LMST-Rem ....<}...MaxPowerGra h" 200100Figure 4.9 Average link length. In this scenario, every device generates four data packets per time unit and each packet onethe routing algorithm, the Dijkstra algorithm is utilized in all cases but with different link metrics. The link metric utilized for LMST and R&M was the energy consumption, while for READ, the lInk metnc was the weighted cost metnc defined m SectIOn 4.1.3. Dunng the simulation, the READ algorithm is triggered every time the residual energy in one node 70


is reduced by 40%ofits last recorded value. Afterwards, the Dijkstra algorithm is run again to establIsh the routmg tables accordmg to the new topology.Inthe case of R&M-add and LMST-add, the topology control algorithms are run every time a node dies, and Dijkstra is run100+=,--"I;;.:ZUu400600BOO1000 1200Figure 4.10 Numberofnodes alive in centralized implementation.1-I0.9-u0.8Luuuuummmm,\,,mmu,axower*J0.7et::0.6L..,Iill0.5.2::II0.4I1:::::::10.3I0.2LI==----0.1-,L...........,IUZUU4UUijUU1UUU120014UUTime (X1Q)to reflect the active abilityofthe network. As the figure illustrates, the numberofnodes aliveofR&M-add and LMST-add start dropping linearly right after the simulation begins, whIle READ-K2 remams unchanged untIltime9000.ThISISvery sIgmficant consIder ing that attime9000 the network has sent around 100 x 4 x 9000 3.6X106packets.71


This is even more significantifFigure 4.10 is seen along with Figure 4.11, which shows the packet delIvery rate of the algonthms.AsItcanbeseen, whIle READ-K2remams at 100% delivery rate, both LMST and R&M drop dramatically from the very beginning. At esting observation is the caseofLMST-rem, which maintains the numberofnodes alive at a faIrly slow decreasmg rate compared to the rest of the algonthms. Although It may be considered the best algorithm from this perspective, when looking at Figure 4.11, howcan seen more algorithms, R&M algorithms provide better delivery rates than LMST algorithms. This baSIcally means that R&M keeps the network more connected than LMST, even wIth fewer numberofnodes. It also can be seen that the numberofnodes alive for maxpower graph drops faster than all other algorithms, which is due to that maxpower graph does not conthe packet delivery rate is lower than the perfonnance achieved by using the maxpower graph, meamng that although the algonthms reduce the topology,thISreductIon does not consIder that the energy consumptIon assocIated wIth edges m heterogeneous networkISnot necessarily proportional to the lengthofedges. Therefore, the topology reduction inaIn order to better understand the network topology evolution, Figures 4.12, 4.13, 4.14 are READ-Kl, four types ofdeVIcessImultaneously drop attime=5500. In READ-K2, the first major drop happens aftertime9000, while in LMST and R&M, they drop at and PDAs start to drop attime2000; and sensors start to drop attime200. In R&M,


PDAs and sensors start to die beforetime=1000; robots start to die aftertime=5500; and mIlItarydeVIcesstart to dIe aftertime=8200. At least two Important observatIOns can be made from these four figures:1)In LMST and R&M, four typeofdevices drop at can that the major dropofdifferent typeofnodes in READ happen simultaneously, which further supports the theory even energy consumptIOn among dIfferent types of nodes can maximize network longevity .5I".II:III,I II I;;;:g4II.L,i."z.i':'IzI30II.0:ll00:bIbi.0READ-KII"",--READ-KIE-READ-K2IE-READ-K2,-R&M-AddIz-R&M-Add1,-R&M-RemLMST-Add-LMST-KI,--LMST-Rem-LMST-RemI........MaxPowerI0in centralized implementation. tralized implementation. cen-4.3 The Distributed Residual Energy Aware Dynamic Topology Control AlgorithmInthISsectIOn,the DREAD topology control algonthmISpresented. DREADISthe distributed versionofthe READ algorithm introduced in the last section.4.3.1 Distributed Residual Energy Aware Dynamic AlgorithmMaintenance phase.Itis assumed that a groupofwireless devices start working approxi73


2515'-\10l.",--D-K1..,5!-Add--"--0centralized implementation.70...-----------------,sensors centralized implementation. mately at the same time and enter the Initialization phase. The Initialization phase consistsofthe following steps:uSelf Advertise Period, denoted as T_SAP.Nodeurandomly chooses a time moment withm T_SAPto broadcast Self Advertise Message, denoted as SAM, at ItsmaXI-mal powerPmaxto allofits potential neighbors. The SAM contains the following information:nodeu'slocation information, which is used to calculate the distance fromuto any potential one-hop neIghbor. ThIs calculatIOnISperformed usmg Equa-are minimum transmission power,Pu,v,from any potential neighborvto nodeu;Equation 4.12 and Equation 4.13; and,nodeu'sCUHentresidual


NeighboLID Position Sensitivity( dBm)Re_En(J) Gain2 (18,40) -70 0.3 200I-2.Establish One-hop Neighbor Table: Upon receiving an SAM, the receiving node v adds the receIVed mformatIOn to the One-hop NeIghbor Table, denoted as Table_ON, which is shown in Table 4.2. Node v performs a simple calculation using Equa-vcanuv erasesuby doing this simple check, directional edges will be eliminated from the network.3.Establish One-hop Edge Weight Table: With the above information from all the onehop neIghbors, the receIvmg node can calculate the weIghted cost for all the one-hop bi-directional edges. The calculation result will be recorded in the One-hop EdgeasLao-""_V'L.

two local one-hop tables with received tables to create a Two-hop Neighbor TableallCla'I\v()=ll()pgClge\yeIgllt IalJ le ""llIcllIllcIllCletlle lleIglllJ()rIllf()fl1latI()llallCleClgeweight information within two hops. These two tables are very similar to the one-hop areasareTable 4.4 Two-hop neighbor table. NeighboL1D 23-70 -900.30.51 1ill_IID_2 Weight 1 2 0.02 3 2 0.12Inorder to better explam the process of buIldmg up two-hop tables, an exampleISprovided in Figure 4.16. A maxpower graph before topology control is displayedD.After step 3, eachofthese nine nodes has its Table_ON available. The one-hopU"""5U'UV'tablesofnodes C andDare aIslpla:yea on the left sideoftheU5'-""""Atstep 4, the DREAD algOrIthm runmngoneach node starts to exchange one-hop neighbor tables. NodeAreceives one Table_ON from eachofits one-hop neighbors, in this case, from nodesB,C,andD,as listed on the left sideofthe Figure 4.16. Once nodeAreceives these one-hop neighbor tables, it merges allofthem into its own Table_TN, as shown at the right sideofFigure 4.16. Table_TEW at nodeAis 76


built in the same manner. After step 4, nodeAhas collected all the information about the nodesandedges withm two hops.A'sOne-hoNeihborTable0.01A'sTwo-hoNeihborTableBBCCDDB'sOne-hoNeihborTableEEFFGC'sOne-hoNeihborTableAGD'sOne-hoNeihborTableA CAA DHB E BFCG0.02C H0.01DI0.0015D H0.00116. Create Symmetric Local Minimal SpanningTree:Once the T _ANP expires, every uses geIlerate aspanning tree. The reason it is called symmetric is becauseifanedge between node 77


uand nodevis selected by nodeu,it is also selected by nodev.The proof of this ClaImISmcluded m AppendIxA.Toillustrate the generationofthe local minimal spanning tree using the Prim algonthm, the same example from FIgure 4.16ISusedasshown m FIgure 4.17. After step 4, nodeAhas a maxpower graph within two hopsasin partaofFig. 4.17. Then,asaspart cofFigure 4.17. The same algorithm is conducted in all the other nodes inde-eaon nodeC.0.003 B 0.0012 BEF............0.004GA 0.002 A .005 0.0015 .02 0.0015C.02DC0.0011H0.01 0.0011Ha)A's two max power graphb)A's two-hop Local minimal spanning tree minimal spanning treeB0.004G0.002 A .005D C0.02....../yO.02H0.0011 .02 0.01 0.002D0.0011d) C's two max power graphe)C's two-hop Local f) C's one-hop Local minimal spanning tree minimal spanning treeFIgure 4.17 GeneratIOn local mmimal spanmng tree. The Mamtenance phase starts after the very first DREAD topologyISgenerated.ItISresponsible for adjusting the topology according to the valueofthe remaining energy. In the current DREAD implementation, the algorithm is triggered once the node's residual 78


energy reaches 60%ofits previous record value. The Maintenance phase consistsofthef()II(}\,yIl1gsteps:ofits previous value,ituses its maximum transmission power to broadcast an Up date Control Message, denoted as a UCM, to its neighbors with the updated residual energy. The structure of the UCMISvery sImIlar to an SAM, except It has a spe cial field to denote the message type. AUCMnot only informs all the potential an2.Reply Self AdvertIse Message: Once nodevreceIVes a UCM from any ofItSpotentIal neighbors, it uses Equation 4.8 to checkifit can reach node?tusing its maximumvuse the residual energy information from the UCM to overwrite the corresponding value in Table_ON and update its Table_OEW accordingly.Italso broadcasts SAMs\\lItl1ItSIl1aX.IIl111Il1tral1sl11IsSI()11p()\\ler.ubrOladl:astsa enters a Waiting Reply Period, denoted as T_WRP,to wait for SAMs replied by all the potential neighbors. 4. Build up new one-hop tables: Upon receiving an SAM, nodeurecords them into its Table_OEW and Table_ON.uchooses a random tIme to broadcastItSOne-hop TablesWIthReply Request Message, denoted as an OTRR message, at the maximum transmission power, which includes Table_OEW and Table_ON.79


6.Build up two-hop tables: Upon receiving the OTRR, all the nodes will reply with theIr own Table_DEW and Table_ONdunngtheTANP,and then update theIr two hop tables. Local minimal spanning tree basedonthe updated Table_TEW and Ta7. Construct new symmetric LMST: Thenodeuthat initiated the topology update pro cess\VIIIrecalclliate Its I()calI11IIlII11alspaIlIlIIlg treebaseCl()Ilall tliellPClateClresIClllalenergy inf6iinati6n as well as Table=TEW and Table=TN. The detailed complete algorithm for node1isasfollows:Iiiitialization:1:Enter T_SAPperiod and broadcast SAM message withPmaxusing random times.2:If(receIve SAM message from nodevdunng T_SAP)and(vcan be reachedby'U'sPmax)4:Endif6:Then1)enter T-i\NPperiod8:Endif9:IfreceIve Table_ON and Table_OEW11:Endif12:IfT-i\NPpenod expIres14:IfTC triggering condition is satisfied16:End if80


vcan18:Then1)use received information to update Table_ON andTab1e_OEW19:2)broadcast SAM message20:Else omit received UCM message22:If(receive SAM message from nodevduringLWRP)and(vcan be reached byu'sPmax)23:Then put received inforiiiation intoand26:Then enter T_ANPperiod and choose a random timetobroadcast OTRR message28:Ifreceive OTRR message29:Then1)update Table_TN and Table_TEW32:EndIt33:IfT_ANPperiod expires34:Endif4.3.2 Complexityofthe DREAD Topology Control Algorithmas can with the numberoftotal nodes alive.Thecomputational complexityofsteps 1,2,3and5are cOIlst,mt, worst'UV.Lv"",which has a cOluputaltional cOIuplexltyof81. In{WID-UlODtables are built


by appending all the received one-hop neighbor tables together, sorting them, and then mergmg the common Items. Smce there arenneIghbors for each node,none-hop neIghbor tables will be received, and the lengthofeachofthose tables isn,so after appending them are at most an aPl)ropriate sorting algorithm, the computational complexityofstep 6 is bounded by0(n2logn2 )0(2n2logn)0(n2logn).In step7,the local minimal spanning tree is built based on two-hop neighbor tables using the Prim algorithm. In the worst case, there aren2nodes in a two-hop neighbor table andn2edges in a two-hop edge weight table. By using binary case0(n2logn2 )=0(2n2logn2 )=0(n2logn).Therefore, the total computational complexityofthe Maintenance phase is bounded by0(1n2logn n2logn+n)0(n2logn).A total of three messages are sent by each node dunng the Mamtenance phase: UCM, SAM, and OTRR message. Therefore, the message complexityofthe Maintenance phase is bounded4.3.3 Simulation Results and Evaluation of DREADsimulation setup parameters are described first and then the simulation results and the anal ysis are presented. Simulation SetupInthISsImulatIOn expenment, a heterogeneous WIreless network wIth four types of devices are considered: sensors, PDAs, robots, and military devices. One hundred nodes ina-'-'H/\JIII,XarenetwoJrkarea are randornly disltributeld, are sensors. areUIUUUlsameconfigure the devices in the earlier READ simulation (See Table 4.1), are used here in the 82


distributed version. Each node utilizes a uniform distribution to randomly draw its maximal transmISSIon powerInWatts, receIVer sensItIvItyIndB, and InItIal energy In Joules from theralnge:s,and that characterize eachofpaJranGetersare on generality, omnidirectional antennae are considered and set toGt -Gr1, the system loss factorL1, and the operational frequency equal to 2.472GHz.0"is subject toI"the wIreless Interface devIce Itself andISassIgned a constant value of 0.6 for all the nodes without lossofgenerality. The valueofttxandtrxare also set to 0.01 seconds, meaningare\Figure 4.18 Topology generated by DREAD. Results and AnalysisditlefClnttopol,ogy control alg()flttlms arescenario where packets are transmitted and distributed topology control algorithms are running constantly. Four sinks are randomly deployed in the network area. Eachofthese 100 devIces generates tour data packets per tIme umt and sends one packet to eachSInk.Inorder to minimize the performance difference caused by the routing algorithm, the Di-83


jkstra algorithm is utilized in all cases but with different link metrics. The link metric utIlIzed for LMST and R&M was energy consumptIOn, whIle for READ, the lIrik metnc was theweighted costmetric defined in Section 4.1.3. Every device is running its own one last recorded value. Afterwards the Dijkstra algorithm is run again to establish the routing tables accordmg to the new topology. Both R&M and LMST are implemented in a distributed manner and therefore run on or tributed LMST, topology control algorithm generates a local minimal spanning tree using the EuclIdean dIstanceasmetnc to select edges and only keeps on-tree nodes that are one hop awayasits neighbors in the final topology. The distributed R&M topology control algorithm selects edges based on the conceptofRelay Region, which does not take resid-pn,"'TOVor antenrla"""rl"iti"'it,,areanode dies and Dijkstra is also run to re-compute the least total energy consumption path to the smks. In DREAD,asexplamed earlIer, the weIghted cost functIonISusedasthe metnc to generate the local mInImal spannIng Tree and to compute the best path to smks by Dijkstra. DREAD, LMST and R&M are all distributed, position-based topology control Figure. 4.18 illustrates the topologyofthe network after running DREAD. The figure shows a topology made of 184 edges after 100 tImeUnIts,whIle there are 983 edges m maxpower graph at the same moment. Figure 4.19 depicts the numberofnodes alive in the network with considerationofalgorithms.Asit can be seen, the curve follows the trend that is observed in Figure 4.10. 84


FIgure 4.19 Number of nodes alIve In dIstributed ImplementatIon. The numberofnodes alive in the caseofLMSTand R&M start dropping linearly right after the sImulatIon begIns. Attime6000,the number of nodes alIve In R&M-add andR&M-rem have decreased from 100 to 60 and the numberofnodes alive in LMST-add andtime7000, 100x4x7000 2.86packets have been sent, this is a significant differenceIIIterms of network lIfetIme compared wIth the other algorithm.ThISnetwork longevIty gain is due to the fact that DREAD always keeps the links that have less weighted costtolJ10IC)gyso0,90,8.&0.70.6-\, 0,5iun+nnnunnunnu'JI-unnnl;nunnunnnunnnunnunnunnuunnunnJ'"q0,3iunnulinunnnnnnnunllnunnunnnunnunnunnnnunJO,lnnnl200400 600BOO100012001400TimeiX10lFigure 4.20 Successful delivery rate in distributed implementation. 85


energy in the data packet transmission procedure. Figure 4.20 shows the packet delivery rate can rate attime7000,while both LMST's and R&M's delivery rate drop dramatically from the areItcan also be seen that the distributed LMST and R&M algorithms do not outperform the maxpower graphInterms of number of nodes alIve and delIvery rate. The reasonsISmaxpower curve same oneasalgorithms deduct energy every time a topology control message is exchanged. FIgures 4.21, 4.22, 4.23 and 4.24 show the number of nodes alIve for each type of device. For DREAD, the numberofeach typeofnodes drops simultaneously aroundtime-7000,which occurs because DREAD keeps energy consumption fairness in mind more residualPTlprCY\Ic4:mtributemore mission process. In LMST and R&M, different devices drop at different times.InLMST, sensors start to drop attime=100;PDAs start to drop attime=500;mIlItary devIces start to drop attime-2000;and robots devIcesstm1to drop attime-3000.InR&M, sensors start to drop attime150;PDAs start to die beforetime-600;robots start to R&M, the four typeofdevices die at different stages of the network lifetime. The lower the InItIal energy, the earlIer the devIces start to dIeInLMST and R&M.InDREAD, all devIcesdIeSImultaneously, whIch further confirms that evemng out energy consumptIon among different typesofnodes can maximize network longevity.86


Figure 4.21 Numberofmilitary nodes alIveIIIdIstnbuted ImplementatIon.,;;:15Z\..:5'1'01:10!j-DREADz500..o.--LMST-Re"Figure 4.23 NumberofPDA nodes alive Figure 4.22 Numberofrobots alive in dIstnbuted ImplementatIOn.50---------,\::--------------IFigure 4.24 Numberofsensors nodes4.3.3.3 Comparison Between READ and DREADTofurther analyze the performanceofREAD and DREAD, the numberofnodes alive and the delIvery rate of DREAD-K2 from SectIon 4.2 are used to compare to that of DREAD. READ is a centralized topology control algorithm and the energy consumption during the exchangeofcontrol messages is not taken into account in the simulation. on a distribtltedc(mtrlol algor'ithrn, runs on each individual node independently and considers all the energy consumption incurredbycontrol packets. Both READ and DREAD use the same weIghted cost functIonmetncto 87


11000.9I-READ-K2..0.82'0.7"roIl::0.6C70..6.S0.460""-0.3 0.2150'READ-K20.1I00200 400 600eD,100012001400020040060080010001200 1400livery rate. DREAD drops faster than READ in both figures. DREAD starts dropping attime-7000, whIle READ starts droppIng attime-9000. The reasons that READ can extend network longevity further are twofold. First, the centralized version selects edges based on the informationofthe entire network and the resulting global spanning tree is means tree is only optimal locally, but not necessarily optimal in the entire network. Secondly, the energy consumptIOn dunng the generatIOn of global READ topologyISnot deductedInthe READ SImulatIOn, whIle the energy consumptIOn Incurred by the exchange of topology control messages is deducted in the DREAD simulation. different type devices between READ and DREAD.InThese four figures, DREAD drops earlier than READ for the same reasons stated before. Crossing these four figures, it can be found that the dIfferent types of devIcesIIIboth DREAD and READ drop sImultaneously due to the fact that both Residual Energy-Awareness Dynamic topology control algorithms use88

PAGE 100

,125104,J,24,iui==l'II-DREADI-READ-K210Q,FIgure 4.27 Number of mIlItary nodes alive of CREAD and DREAD.,,+--------,---,151mmmmmmmmmmmm+ummlmmm_mmm_mummm1zIi'o1] 10imm mu mumm mmm m'L..,mummlummmmmmmmmmmuJ15L-iL,z-DREADlumummmmmumumm-READ-K2Figure 4.29 NumberofPDA nodes aliveofCREADand DREAD.4.4 ConclusionsFIgure 4.28 Number of robots alIve of CREAD and DREAD.eo+uuuuuu.uuumuuuuuuuu.mmmm.L=j...uuuuu+uuuumuuuuuuu_mmmmml\eoimmmmummmmmummmmmumI: 40l.=====.:.::.,mmuuu.uuuuuuuummmmuuuuu.uuuuuul-DREADz-READ-K220o200 400'008001000 1200 1400Time{X1OJFigure 4.30 Numberofsensors nodes aliveofCREAD and DREAD. topology control in heterogeneous wireless scenarios, where sensor nodes, ad hoc nodes, robots wIth commumcatIon capabIlItIes, and even more powerful mIlItary wIfeless devIces work together m the same applIcatIOn.ThISISthe first research that takes mto account difference in initial energy, residual energy, receiver sensitivity, and antenna gain for every89

PAGE 101

<:HAPTER5 CONCLUSIONS AND FUTURE WORK5.1ConclusionsThISdIssertatIon studIes smk localIzatIon problem m large scale WSN s as well as topol ogy control in heterogeneous WSNs. Sink localization problems are considered first with the introductionofthe ALS protocol, a quorum-based mechanism for large scale,WSNs ALS, each sink selects a setofnodes to establish a global anchor system that facilitates the propagatIOn ofItSlocatIOn mformatIOn. InthISmanner, global or dIrected floodmg procedures are avoided substantially reducing the communication overhead. Furthermore, multiple sources use the same global anchor system to find the sink's location, reducing theeven more. means a m,ttbemlatlcal simulations, we demonstrate the effectiveness and scalabilityofthe location service with multIple and movmg smks and sources, dIfferent network denSItIes, and mcreased network areas. ALS also mcludes procedures toaVOIdfrequent floodmg, caused by the mobIlIty of the sinks and targets. In addition, we show that ALS with GPSR, a location-based routingaknown grid-based routing protocol for WSNs.The challengesoftopology control in heterogeneous multi-hop WSNs is considered next. DIfferent types of devIces working on the same applIcatIOn are studIed, and the prob lem is solved as a power assignment problem.Wepropose both centralized and distributed90

PAGE 102

versionsofthe READ topology control algorithm, which considers the receiver's sensitiv Ity, the sender's maxImal transmISSIon power, and the node's resIdual energy to determme the final topology.Sin:mLaticmre sults demonstrate that READ can efficiently increase the network lifetime, presents a low average node degree and mcreases the packet delIvery rate by 40% over R&M and 90% over LMST, two well known position-based topology control mechanisms.5.2 FutureWorkpossible future work include the following:are catTied on therefore, the resulting system in ALS is grid system. In the future, different geometrICshapes could be explored to reduce the commumcatlOn overhead even further. In READ, ratios for each different typeofdevices are fixed at 5% for military nodes, sensors. in this dissertation were conducted with these ratios. In the future, the composing can ratioofdifferent wireless devices. Animportant aspect in the topology control problem is the introductionofmobility in the algorithm. In heterogeneous networks, like the one utilized in this dissertation, ItIS"eryIIk:elytl1atlllIIItarYCle"Ices,allClPJ.)1\.s\VIII111()"e,\VI1ICI1IlllP()sesnew challenge to topology control algorithm design.91

PAGE 103

REFEREN<:ES[I]J.N.AI-KarakI andA.Kamal, "Routmg Techmques m WIreless Sensor Networks: A Survey,"IEEE WirelessCommUnICatIOns,vol.Vol.II,No. 6, pp.6-28, 2004.[3]P.SantI,Topology ControlmWIrelessAdHoc and Sensor Networks.John WIley and Sons, 2005.Int'l Workshop on Wireless Sensor Networks and Applications, 2003.[7] G. KarumanchI, S. Murahdharan, andR.Prakash, "InformatIon DIssemmatIOn m Par titionable Mobile Ad Hoc Networks," inProceedingsofIEEE Symposium on Reliable Distributed Systems,1999. [8]S.Basagni,I.Chlamtac,V.Syrotiuk, and B. Woodward,"ADistance Routing Effect Algorithm for Mobility (DREAM)," inProceedingsofthe 4th Annual ACMIIEEE International Conference on Mobile Computing and Networking (MOBICOM),Oc tober1998,pp.76-84.[9]y.Ko and N. Vaidya, "Location-aided Routing (LAR) in MobileAdHoc Networks,"l>.Jl"UCLl......HJJ"lLJL'LJlIlternatlonal Conference on MobIle[10]I.StoJmenovIc,"AScalable Quorum Based LocatIon Update Scheme for Routmg m Ad Hoc WIreless Networks," mTR-99-09, UnIVersItyotOttawa,aVaIlable at:, 1999.and Sensor Systems (MASS), 2006.92

PAGE 104

[12]A.E.Abbadi, D. Skeen, andF.Cristian, "An Efficient Fault-tolerant Algorithm for Replicated Data Management," in ProceedingsofACMSIGCAT-SIGMOD Sympo sium on PrinciplesofDatabase Systems, 1985, pp. 215-229. [13] M. Herlihy, "Dynamic Quorum Adjustment for PartitionedData,"ACMTransactions on Database Systems, vol.Vol.12,No.2,pp. 170-194, 1987. [14] B. Liang andZ.J. Haas, "Virtual Backbone Generation and Maintenance in Ad Hoc Network Mobility Management," in ProceedingsofIEEE Infocom, 2000.terns,"ACMlIEEE Transactions on Networking, vol.Vol.7, No.2, pp. 228-240, 1999. [16]-,"Ad Hoc Moblhty Management wIth RandomIzed Database Groups," m Pro ceedingsofIEEE ICC, 1999.[17]L.Blazevlc,L.BUUyan,S.Capkun,S.GIOrdano,J.Hubaux, andJ.L.Boudec, "Self orgamzatIOn m MobIle Ad Hoc Networks: The Approach of Termmodes," IEEE Communications Magazine, pp. 166-175, 2001. [18]R.Morns,J.JannoUI,F.Kaashoek, J. LI, andD.Decouto, "CarNet: A Scalable Ad Hoc WIreless Neetwork System,"IIIProceedmgsotACMSlOOPS, 2000. [19]G.Pel and M. GerIa, "Moblhty Management for HIerarchIcal WIreless Networks," MobIle Networks and ApplIcatlOns, vol.Vol.6,No. 4, pp. 331-337, 2001. [20]I.StoJmenovlc, "Home Agent Based LocatIOn Update and DestmatIOn Search Schemes m Ad Hoc WIreless Networks," Techmcal Report TR-99-10, SITE, Um verlSltyofOttawa, September 1999. [21]S.Woo andS.Smgh, "Scalable Routmg m Ad Hoc Networks," WIreless Networks, [22]K.Amouns,S.Papavasslhou, and M. LI,"APOSItIonBased MultI-zone Routmg Protocol for WIde Area MobIle Ad Hoc Networks," m ProceedmgsotIEEE VTC, [23]J.LI, J. JannoUI, D.D.Couto, D. Karger, andR.Morns, "A Scalable LocatIOn SerVIcefor GeographIc Ad Hoc Routmg," m ProceedmgsofACMIIEEE InternatlOnal Conference on Mobile Computing and Networking (MOBICOM), August 2000, pp. 120-130. [24]z..I-I.Mlrall(l,s.i\..KaliIl'"i\.z.()Ile__lJase(lI,,()catI()Il,serylcef()r(Je()castIIlgIIIM()__bile Ad Hoc Networks," in ProceedingsofIEEE lOth Asia-Pacific Conference on Communications, 2004, pp. 955-960. 93

PAGE 105

[25] I. Stojmenovic,HandbookofSensor Networks: Algorithms and Architectures.John Wiley&Sons, Inc., 2005. [26]p.13()se,p.M()fll1' I ..stoJI11el1()YlC,al1(JJ. l}rrlltla,"B:911tll1g\V1tl1Gllaral1tee(J I)ellyery in Ad Hoc Wireless Networks," in3rd int. WorkshoponDiscrete Algorithms and methods for mobile computingandcommunications,Seattle, August 199. [27]--,"Routing with guaranteed delivery in ad hoc wireless networks,"Wireless Networks,vol.7,no. 6,pp.609-616,2001.[28]B.Karp andH.T.Kung, "GreedyPenmeterStateless Routmg for WIreless Net works," inACM/IEEE MOBICOM,2000, pp.243-254.[29] GI,.1lk:acl1al1al1(JMAI,.alJra(J()r,".s;E:I,.AR:.scalable;E:l1ergy=;E:tficlel1tI,.()catl()l1Aided Routing Protocol for Wireless Sensor Networks,"inProceedingsofLCN,2004,pp.694-695.[30] J.13r()cl1, I). Maltz, I).J()l1l1s()l1'y.1I1l'al1(JJ. Jetcl1eya, "}\perf()rI11al1ceC:()I11Par1S()11ofMUlti-Hop WirelessAdHoc Network Routing Protocols," inProceedingsofthe 4th Annual ACMIIEEE International ConferenceonMobile Computing and Networking (MOBICOM),October 1998. [31]S.Ramanathan and M. Steenstrup,"ASurveyofRouting Techniques for Mobile Communication Networks,"Mobile Networks and Applications,vol. Vol.1,No.2,[32] X. Hong,K.Xu, and M. Gerla, "Scalable Routing Protocols for Mobile Ad Hoc Networks,"IEEE Network,voLVol.16,No.4,pp.11-21,2002.[33] J.Y.YuandP.H. Chong,"ASurveyofClustering Schemes for Mobile Ad Hoc Networks,"IEEE Communications Surveys and Tutorials,vol. Vol. 7,No.1,pp.32-[34] I.F.Akyildiz,W.Su,Y.Sankarasubramaniam, andE.Cyirci, "Wireless Sensor Net works: A Survey,"Computer Networks,vol. Vol. 38,No.4,pp.393-422,2002.[35]C.E. Jones,K.M. Sivalingam,P.Agrawal, and J.C.Chen,"ASurveyofEnergy Efficient Network Protocols for Wireless Networks," inWireless Networks,Vol. 7,No.42001 343=358. [36]S.Hedetniemi, S. Hedetniemi, andA.Liestman,"ASurveyofGossiping and Broad casting in Communication Networks,"Networks,vol. Vol. 18, 1988. [37]S.Giordano,I.Stojmenovic, andL.Blazevie, "Position based routing algorithms for ad hoc networks: A taxonomy,",2001.94

PAGE 106

[38] M. Mauve,A.Widmer, and H. Hartenstein,"ASurveyonPosition-based Routing in MobileAdHoc Networks,"IEEE Network,vol. Vol. 15,No.6,pp.30-39,2001.[39]M. Mauve, J. WIdmer, and H. Hartenstem,"ASurvey on PosItion-based Routmg m MobileAdHoc Networks,"IEEE Network,vol. Vol. 15,No.6,pp.30-39,2001.[40]X. LI, G. Cahnescu, andP.Wan, "DIstnbuted ConstructIOn of a Planar Spanner and Routing for Ad Hoc Wireless Networks,"inIEEE INFOCOM,New York, June 2002. [41]X.LI,P.Wan,Y.Wang, andC.YI,"Fault Tolerant Deployment and Topology Control m WIreless Networks," mMoblHoc,Maryland, June2003. [42]M. Penrose,"TheLongest Edge of the Random MInImal SpannmgTree,"mThe AnnalsotApplled Probablllty,1997,pp.340-361. [43]A. Clementi,P.Penna, andR.SIlvestn, "Hardness results for thepowerrange as SIgnment problem m packetradIOnetworks," mProc. 2nd InternatlOnal Workshop on ApproxImatlOnAlgonthmstorCombmatonal OptImIzatlOn Problems,Berkeley,CA, 1999, pp. 197-208. [44]L.M.KiroUSIS,E. Kranak.Is, and D. Knzanc, "Power Consumption m PacketRadIO[46]D. Blough, M.LeOnCInI,G.Resta, andP.Santi,"OnTheSymmetncRangeASSIgnment ProblemInWIrelessAdHoc Networks," mProc. IFIP ConterenceonTheoretI cal Computer SCIence,Montreal,2002,pp.7 1-82. [47]V.Rodoplu andT.H.Meng, "MInImUm Energy MobIle WIreless Networks," mICC'98,Atlanta, GA, U.S., June1998,pp. pp.1633-1639. [48]V.Kawadia andP.R.Kumar, "Power Control and Clustenng m Ad Hoc Networks," mIEEE INFOCOM,SanFranCISCO,March2003.[49] N.Liand J.C.Hou, "Topology Control in Heterogeneous Wireless Networks: Prob lems and SolutIOns," mIEEE INFOCOM 2004,Hong Kong, Chma,2004.[50] N. Li, J.C.Hou, andL.Sha, "Design and Analysisofan MST-Based Topology Con tr()l i\)g()f!tIlIl1,"111IJjJjJj INfQc:QM,a11fra11cIsc(), MarCIl[51]R.Ramanathan andR.Rosales-Hain, "Topology ControlofMultihop Wireless Net works using Transmit Power Adjustment," inIEEE INFOCOM,Tel Aviv, Israel,Mar2000. 95

PAGE 107

[52]R.Wattenhofer,L.Li,P.Bahl, andY.Wang, "Distributed Topology Control for Power Efficient Operation in Multihop Wireless Ad Hoc Networks," inProceedingsofIEEE Infocom 2001,2001. [53]R.Zhang,H.Zhao, and M.A.Labrador, "The Anchor Location Service (ALS) Pro tocol for Large-scale Wireless Sensor Networks," inProceedingsofInterSense,May 2006. [54]--,"AGrid-based Sink Location Service for Large-scale Wireless Sensor Net works," inProceedingsofACMIWCMC,July 2006. [55] "NS-2 Network Simulator (," The InstituteofElectrical and Electronics Engineers, New York, IEEE Std. 802.11 1997, July 1997. [60]R.Zhang and M.A.Labrador, "ResIdual Energy Aware DynamIc Topology Control for Heterogeneous Networks," mproceedmgofInternatlOnal SymposlUm on WIreless PervaSIve Computmg,2007.[6I] http://www.cIsco.comlel17USlproductslhwlwIreIesslmdex.htmi. [62] [63] [65]T.Cormen,C.Leiserson,R.Rivest, andC.Stein,Introduction to Algorithm,2nd ed.96

PAGE 108


PAGE 109

Appendix A Property ProofofALS and READA.IConvergence ProofofALS' Anchor Setup Process and Query ProcessAs mentIOned m SectIOn 3.1, both the anchor system setup and the query processes orthogolnal messages. messages may messa,ges moveto go around the void areas. Since the area outside the sensor network can be viewed as the largestvOIdarea, routmg along the boundary of the sensor networkISconsIdered a convergence messages area. rolitilllgCll'''"lnrlan query processescase following definitions are utilized:Definition1(reaf-grid-pointandvoid-grid-point):Define any grid pomtP(xp yp)as areaf=grid=jJo{nt,ifthere is a sensor node that canbeelected as a grid node for the grid point as aVOI:a-:flrl,(1-lWlirLIno sensor node.pefifliti()fl( (lfla.v() ):(:()llllect eyery paIr()fadJacellt grI(l p()IlltsIJYa virtual edge, onlyifthose two adjacent grid points are both real=grid=pointsorvoid=grid=reaLl-2nOI-0()lllrSasreal-el:Uleand the virtual connection between two adjacent void-grid-points asvoid-edge.Note that real-edges and void-edges are all along the grid lines in the four orthogonal directions. Fig-98

PAGE 110

Appendix A (Continued)ure a graphilcal repres1entaticm void-edges, and four void areas.Definition3(neighbor and far-neighbor):Every grid point has other eight other adjacent grid points around itself. Fourofthem, located in four orthogonal directions, are calledneighborsand the others are calledfar-neighbors.For any two void-grid-points,ifthey are neighborsorfar-neighbors, they are considered to be connected and m the sameVOIdarea, as shown m FIgureA.I.Onthe other hand, real grid-points are connected onlyifthey are neighbors. Any real-grid-pointorvoid-grid-point arnl1!'YUPl1tT?laltL011,reflexive, symmetric,andtransitiveconditions.Itis assumed that the whole network is not partitioned and that all real-grid-points are connected. For example, m FIgureA.I,there are tourVOIdareas, whIch are dIsconnectedWItheach other. TheseVOIdareas are bounded by vOld-gnd-pomts connected to each other via void-edges, such as casesa,b,andd.Ifthere is no void-edge between two void-gridare tarnelghboJrs and connected, as in casec.are one areaDefinition4(Void Polygon, World Void Polygon, and Snode(VPi)):Define everyVOIdareaasVoid Polygon,denoted asVP.Define the Void Polygon containing all the void area nel:w()rk areaasa"1-"_'-'''0'''caseVP.For anydefine the void-grid-point set which includes all the void-grid-points inVasSnode(VPi).99

PAGE 111

Appendix A (Continued).......................o o,...In order to simplify the proof, the caseain FigureA.lwill not be considered because onerea.H2:na-DCnmarea area. the real-grid-point will be considered as a void-grid-point and therefore, it will not make any difference in the area means area an contain some void polygons. In Figure A.2, there are fourRPexamples. The Universal Set ofRPISdenoted asU(RP)and the bIggestRPIIIthe network area asWorld Real PolygonorWRP.100

PAGE 112

Appendix A (Continued)a c1u..'IFf---1.b dFigure A.2 A graphical representationoffour Real Polygon 0 0r--1.()-()--()II>
PAGE 113

Appendix A (Continued)G5 G7Figure A.4 Base case with I void-grid-point.Ue/lnltll')n6orCover(VPi)Define the setofRPs that all contain the sameDefinition7(Envelop E(VPi)):Consider anyV Pi, VP/sEnvelop,denoted asE (VPi),as an a not smallest RP that containsVPibut also unique. SinceE(VPi)is the smallest RP in the setofCover(VPi),the inner areaofE(VPi)contains and only contains the void polygon canto asH10"IlrpA.3 shows four void areas and theirConsl,dera oneofVPi,the second stage anchor setuporquery messages can constructE(VPi)by using tIle fIgIlt Ilall(l flllet()r()llte ar()llll(l tIleVg102

PAGE 114

Appendix A (Continued)Proof: LemmaIwill be proved by induction. Denote the cardinalityofasn,i.e.,In.Werepresent the forwarding and incoming directionofthe message by a vector m the complex plane. Whenn=1,thereISonly one vOld-gnd point inV Pi,and therefore, only possible scenario, like the one shown in Figure A.4. In the same figure, when the first stage setup message reachesVPiatG2withforwardingdirection(x,yi),it splits into two second stage setup messages. The second stage setup message using the right hand rule atG2utilizes the priority sequence(x,yi) x(0,i),(x,yi)x(0,i?,(x,yi)x(0,i)3to select the next grid pointasits forwarding des tination. time we cross to a we tum that vector counter-clockwise tum that vector 90 degrees. As a result, G3is chosen,asillustrated in the figure. Then atG3 ,the setup message withincomingdirection(Xl,yli)uses pri ority sequence(Xl,yli)x(0,i),(Xl,yli)x(0,i?,(Xl,yli)x(0,i)3to select the next grid same manneraccordlm,gtoxtorwurrd,tnfJdwectlonasxto construct uses can tion to calculate the next forwarding direction atG2while it uses theincomingdirectionasthe base direction in the restofthe process. The borderofG2G3G4G5G6G7GI',G9consists of real-edges that canbetraveled aroundbysetup messages. Hence the polygonISclosed.Inother is anRPandccan seen no numberofgrid points or real-edges, or smaller area thanG2G:3G4G5G6G7GSG9'The reason is that deleting any real-edge or real-grid-point from real-polygonG2G3G4G5G6G7GSG9cause103

PAGE 115

Appendix A (Continued)Figure A.5Basecase 1with2 void-grid-points.Whenn2, there aretwopossible scenarios.OneofwhichshowninFigure A.5. In tills case, a first stage setllP Il1essageSplItSIIlt()t\V()sec()IlClstage setllP 11lessages\VlleIlIt meets at withdirection.ThesecondtherIghthandruleatC;usestheprIorItysequence(x,iii)x(0,l),(x,ill)x(0,l)2,(x,ill)x(0,i)3toselect the next gridpointas its forwarding destination. As illustrated in the figure,0:3is chosen.Then,at03 ,the setup message withincomingdirection(x',y'i)uses priority seqlleIlce(:r',y'i)><(0,i),(:r',y'i)><(0,i)2,(:r',y'i)><(9,i)3t()selectIle:xt grici p()iIlt,\Vl1icl1is04 Continuing the process, the setupmessagewillhopover05 ,06 ,07 ,Os,Og,010,011acc;orcjmg to x x xgonG2G3G4G5G6G7GSG9G1OGllis anRPbecauseitsborderconsistsofreal-edges and real-grid-points. Also, because itsbordercanbecycledaround, starting fromandending at thesamereal-gnd-pomt,andItSmnerareacontamsVOIdpolygonsV Pi,thereISnoother104

PAGE 116

Appendix A (Continued)ororarea G2G:3G4GSG6G7GSG9GlOGll. This is true, becauseifany real-edgeorreal-grid-point fromRPG2G3G4GSG6G7GSG9GlOGll is deleted, it willbeunclosed, which conflicts with Definition5.As a realISThe other scenario is shown in FigureA.6.Inthis case, the setup message chooses the first real-grid-pointG2 ,followed byG3 ,G4 ,G s ,Go,G7 ,G s Gg ,GlO,Gll,GI2andG13 TheRPG2G3G4GSG6G7GSG9GlOGllGI2GI3 is then created. Applying the same deduction, it canbeshown thatRPG2G:3G4G5G6G7GSG9GlOGllG12G13is the smallest closed real which contains it isorquery messagescan..arecasesn=the same procedure. Therefore,ifwesupposen=k,a second stage anchor setuporquery create annwean ad(lltIon,alVl[)l(]I-!J"]"Hl-Dcnnlk-void-grid-pointV Pi.If'sneighborsorfar-neighbors.If'seIght neIghbors and far-neIghbors all belong toVPiand thereISa real-gnd-pomt at'sposition in then kscenario, the original real-grid-point at'sposition is disconnected Now,weconsider the case where only one grid-point belonging toVis'sneigh bororfar-neIghbor. There are two cases. Case IISshownIIIFIgure A.7.InthIScase,105

PAGE 117

Appendix A (Continued)G1Figure A.6 Base case 2 with 2 void-grid-points.G3 G4 G5V2G6G2 G7Figure A.7 Step with 1 void-grid-node as far neighbor. 106

PAGE 118

Appendix A (Continued) GI ,G2 ,G3 :G4 :GS :G6 ,G7are real-grid-points. The 4-point starVIbelongs to k-void grid-pointVPi.The 5-point starV2was a real-grid-point when n=k, and now becomes anweE. For the when the arrives at it transmits fromV2toG2using the right hand rule. AfterV2becomes a void-grid-point, the setup message will be forwarded fromGItoG3 ,G4 :Gs ,G6 :G7 :G2 The sequence is shown in Figure A.7. Assume that the boundaryofthe original real polygon constructed by the second stage anchor setup or query messageWIthnghthand rule torVPihas a cyclmg sequellceof. Now the newofthe bound aryoftheRPconstructed by the second stage anchor setup or query message with right hand rule forVPi+V2isGI :G3 ,G4 ,Gs ,G6 :G7 ,G2 :HI,H2 :...:Hj So the borderofthe polygon constructed by the right hand rule is composedofreal-edges, and it canbecycled around, startmg from and endmg at the same real-gnd-pomt. The polygonISanC ... ,There is no otherRPcontainingVPi+V2with less numberofgrid points or real-edges, or less area thanGI ,G3 ,G4 ,Gs ,G6 ,G7 ,G2 ,HI.H2 ,...:Hj According to the induction hypOlthesIs,RPis which does not contain redundant area or real-grid-point, therefore the new partGI ,G3 ,G4 ,Gs ,G6 ,G7 ,G2dose not involve any other new redundant area or real-gnd-pomt or real-edge m the above process. Therefore is the smallestRPwhich contains+we...,The same holds for the other case whent2,as shown in Figure A.8.Wecan also prove this lemma whent3,4,5,6,7 in a similar way.(The rigorous proof is again omitn107

PAGE 119

Appendix A (Continued)G3 G4 G5G2'1,v,Figure A.8 Step with 1 void-grid-node as neighbor.createdieE(VPi),whichisanRPandthesmallestRPwhichcontainsHence, we conclude that using the right hand rule, three setup message will create theE(VPi) foralllSnode(VPi)Iby induction. Lemma2:Consider a void polygon V Pi, starting fromanyoneofreal-grid-point neighborofVPi, the second stage anchor setup or query messages can constructby using Proof: The proof is similar to thatofLemma1.Claim1:Iftwo second stage anchor setup or query messages start from a real-grid-point, which is a neighborofa void-grid-point on aVP,using the right hand rule and left hand Proof: From Lemma 1 and Lemma2,the second stage anchor setup or query messages construct an 108

PAGE 120

Appendix A (Continued)that there is only oneE(VPi)forVPiand it is a cycle, so the messages using the right hand rule and left hand rule will construct the sameRP.Ifthe network size is bounded two sec;ond anchor or on opposite directions, they will meet at some real-grid-point in a finite numberofsteps. or the boundaryofthe WVP, they will terminate the process in a finite numberofsteps. Proof: ConsIdenng that the boundary of the WRPISthe only envelop of the wholeVOIdarea WVP,ifa setup message reaches the borderofthe WRP, it will split into two second stage messages. or query messages using the right hand rule and left hand rule will each construct anE(vVVP)forVVVP.Note that there is only oneE(vVVP)for WVP and it is a cycle, so the second stage anchor setup or query messagesUSIngthenghthand rule and left hand ruleWIllconstruct same are at most that the numberofreal-grid-points on E(WVP)'s boundary is finite, these messages willA.2 Connectivity and Symmetric Property Proof of READClaim3:IfGmaxis connected, the resulting topology is also connected. Proof:Wewill prove it by contradiction.Ifthe resulting topology is not connected, then at least two nodes, let us sayUlanduzare not connected andIndIfferent Isolated sets.Ulis in set SinceISthere exists at least one between andInGmax,let us callit?jJ.Ifwe walk along the path?jJfromUltouz,and suppose the first node which is not in setGc;is the nodeUjin isolated setGTand the node prior toUjon path?jJis 109

PAGE 121

Appendix A (Continued)Ui,setUiUjas READ algorithm is terminated when either all the node sets are connected or it finishes scallllIllg all tIle e(lges.,5Illce\VeassllIl1etIlat resllltIllgt()p()l()gyISll()tc()llllecte(ls()Itmeans READ has finished scanning all the edges and could not find any edge toconnednode sets and which contradict with the existenceofClaim4: Local Spanning Tree constructed by DREAD is symmetric Proof: Smce Onehop Edge WeIght Table&Onehop NeIghbor Table are broadcasted once, therefore those information is propagated and converged within two hops.Wewill prove Claim 4 by contradiction. Suppose node A and node B are neighbors to each other in max asshOlwnOnehop Neighbor Table, every node only knows neighbors' and edges' weight information wIthm two hops and those mformatIOnISIdentIcal. For mstance, node A knows mfoffilatIon regardmg nodes B, C, D, E,F,G, H and edges AB, AC, AD, AE, CH, CE, CB, BH, BG,BENode A has no information about node I or edge HI.Weassume that A chooses edge does not choose edge AB in its LMST generated based on two hop information, it means that B could find another path to reach A within two hop with less weight than to reach A py1\13(llrectly.\\TItIl()lltl()ss()fgelleralItY, letllSsay tIlereISaC()Il1Il1()ll11eIgIlP()rll()(lec:within two hopsofnode B and node A so that EquationA.Iholds as following:>110+

PAGE 122

Appendix A (Continued)FE BSlll1lplemax power netwolrke;iCanlplle.However, since edge AB is selected by node A, which means Equation A.2 holds as wellasfollowing:w(e(A,B))
PAGE 123

ABOUTTHEAUTHORMr. Rui Zhang is a Ph.D candidateinthe DepartmentofComputer Science and En-onwireless networking. Mr. Zhang received the BachelorofScience degree in July 2002 from the DepartmentofComputer Science and Technology, UniversityofScience andIecl1l1()l()gy()fC=I1Il1a(lJIc=),Hefel,C=I1Il1a.AfterI1IS11l1ClergraClllateClegree,l1e\V()rk:eClas a researcher in the SchoolofComputer Engineering, Nanyang Technological University


Download Options

Choose Size
Choose file type
Cite this item close


Cras ut cursus ante, a fringilla nunc. Mauris lorem nunc, cursus sit amet enim ac, vehicula vestibulum mi. Mauris viverra nisl vel enim faucibus porta. Praesent sit amet ornare diam, non finibus nulla.


Cras efficitur magna et sapien varius, luctus ullamcorper dolor convallis. Orci varius natoque penatibus et magnis dis parturient montes, nascetur ridiculus mus. Fusce sit amet justo ut erat laoreet congue sed a ante.


Phasellus ornare in augue eu imperdiet. Donec malesuada sapien ante, at vehicula orci tempor molestie. Proin vitae urna elit. Pellentesque vitae nisi et diam euismod malesuada aliquet non erat.


Nunc fringilla dolor ut dictum placerat. Proin ac neque rutrum, consectetur ligula id, laoreet ligula. Nulla lorem massa, consectetur vitae consequat in, lobortis at dolor. Nunc sed leo odio.