How to Efficiently Utilize Multi-hop Wireless Ad hoc Networks - Throughput Improvement and Mobility Support

MISSING IMAGE

Material Information

Title:
How to Efficiently Utilize Multi-hop Wireless Ad hoc Networks - Throughput Improvement and Mobility Support
Physical Description:
1 online resource (175 p.)
Language:
english
Creator:
Huang,Rongsheng
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Electrical and Computer Engineering
Committee Chair:
Fang, Yuguang
Committee Members:
Wu, Dapeng
McNair, Janise Y
Yang, Liuqing
Chen, Shigang

Subjects

Subjects / Keywords:
cross -- mac -- mcn -- mesh -- mobility -- throughput -- wanet
Electrical and Computer Engineering -- Dissertations, Academic -- UF
Genre:
Electrical and Computer Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract:
Due to its low-cost, flexible in construction and self-organizable features, wireless ad hoc networks (WANETs) have provided many applications in a lot of areas. However, the multi-hop transmissions and lack of infrastructure bring the low throughput performance and poor mobility support. Aiming at employing WANETs, I strive to tackle these two problems in the recent years. Throughput performance in WANETs mainly relates to the spectrum wastage caused by improper MAC design. For multi-channel MAC schemes, due to the limited number radios, the spectrum wastage is mostly created by the channel assignment signallings. Observing that channel assignment is not necessary before each transmission, we can let nodes cache the previous channel assignment and lift the signalling burden of the control channels. We proposed this self-adjustable multi-channel MAC scheme (SAM_MAC) to increase the MAC throughput in multi-channel systems. Only high MAC throughput cannot guarantee high end-to-end throughput in multi-hop wireless networks because if the area with high traffic load cannot acquire more spectrum congestions will appear and lead to poor overall throughput performance. We proposed a semi-centralized semi-distributed resource allocation scheme for wireless mesh networks, with which central areas can easily acquire more spectrum and the outer areas can easily reuse the spectrum. Within one neighborhood, nodes with high traffic load should acquire more spectrum. Previous distributed scheduling schemes which aim at solving this problem are too complicated in algorithm and cannot suit multi-hop, contention-based networks. Moreover, they ignore the fact that the traffic demands of neighboring nodes are correlated with each other. We proposed "2-hop MAC" scheme for WANETs to allocate spectrum resource among neighboring nodes in a more intelligent way. Spectrum allocated previously will not be wasted by packet droppings in the following hops. Using the collected traffic dependency information, each node can allocate resource accordingly so that a locally optimized end-to-end throughput can be achieved. In this scheme, although the detailed flow information is not necessary for MAC layer since only the traffic in the neighborhood is considered, a cross-layer information exchange is required. Mobility issues are not well addressed in WANETs due to its lack of infrastructures, compared to cellular systems. By adding some functional entities or modules to the system, we can support mobility in WANETs similarly. Mobile IP can be used as the solution for macro-mobility (inter-domain roaming). For micro-mobility, we proposed the mesh mobility management (M^3) scheme. This scheme stores the locations (serving mesh routers) of mesh nodes in the gateway as the anchors and use temporary routing entries between the anchors and the exact locations to reach the destinations. This approach combines the advantages of tunnel-based approach and routing-based approach and provides a suitable micro-mobility solution for wireless mesh networks. When integrating WANETs into cellular systems, the advantages of WANETs and cellular systems can potentially be both achieved. This type of networks, known as multi-hop cellular networks (MCNs), can potentially have better mobility support. For handoff performance, we propose an ad hoc networks-embedded handoff assisting (ANHOA) scheme, which utilizes the embedded ad hoc networks in MCNs to exchange information of different handoff options and assists multi-hop handoffs. The handoff reservation in each BS can be lowered by considering the handoff attempts to different cells. We proposed the information exchange among neighboring BSs and the algorithm of finding the minimum handoff reservation. For location management, we proposed to utilize aggregative devices to aggregate location management signallings. For the high capacity transit (HCT) systems, we specially proposed a grouping method to reduce the location management signallings. Through these methods, the congestions caused by heavy location request arrivals can be greatly relieved. The performance results have verified the contributions of these works.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility:
by Rongsheng Huang.
Thesis:
Thesis (Ph.D.)--University of Florida, 2011.
Local:
Adviser: Fang, Yuguang.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2011
System ID:
UFE0042751:00001


This item is only available as the following downloads:


Full Text

PAGE 2

2

PAGE 3

andtomywifeandkids 3

PAGE 4

Iowethankstoalotofpeopleuponpresentingthisdissertation.Icannotaccomplishthistaskwithoutthem.Amongthem,Ineedtorstthankmyadvisor,Prof.YuguangMichaelFang.Allmyresearchachievementsaretheoutcomeofhislab,ourlab.Withoutthechancesandadviceshegavetome,Iwouldnotaccomplishanyofthem.Heshowedmetotheworldofresearch.HeguidesmewheneverIamlost.Iamsoluckyofbeingonememberofhislab.Ialsowanttothankmycommitteemembers,Prof.LiuqingYang,Prof.DapengWu,Prof.JaniseMcNairandProf.ShigangChen,fortheiradvicesandsupports.GreatthankstomyfellowmembersinWINETlab.Myresearchdeeplyrelatestoeveryone'seffort.IwouldthankDr.HongqiangZhai,Dr.JianfengWang,Dr.YanchaoZhang,Dr.ShushanWen,Dr.YunZhou,Dr.XiaoxiaHuangforshowingmehowtodoresearch.ThankstoDr.PanLi,Dr.FengChen,Dr.YangSong,Dr.JinyuanSunforthegreatresearchinteractionwehad.ThesamethankstoallthecurrentfellowWINETmembers.IwanttoexpressmyspecialthankstoChiZhangforgivingmeadviceswhenIstruggledwithmyresearch.HehelpedmetobuildmyownresearchmethodologywhenIneededitmost.IalsowanttothankDr.SunmyengKimandallthevisitingmembersfortheirhelp.Personally,Iwanttotakethischancetothankmyfatherandmysisters.Familyisalwaysthemotivationofmydiligence.TheirsupportshavecarriedmethroughmytoughPhDyearsandtheirlovesoothesmypainoflosingmom.Atthe8thyearofmarriage,weareexpectingoursecondbaby.IguessIowemorethanthankstomywife,forkeepingmeawayfromloneliness,forbringingmethisfamilyandforaccompanyingmeinthislonglongjourney. 4

PAGE 5

page ACKNOWLEDGMENTS .................................. 4 LISTOFTABLES ...................................... 9 LISTOFFIGURES ..................................... 10 ABSTRACT ......................................... 12 CHAPTER 1OVERVIEWOFRESEARCHTOPICS ....................... 15 1.1WirelessAdhocNetworks ........................... 15 1.2ObservationsandMotivations ......................... 17 1.3MyResearch .................................. 18 1.3.1ThroughputImprovement ....................... 18 1.3.2MobilitySupport ............................ 20 1.4Outline ...................................... 21 2SAM-MAC:ANEFFICIENTCHANNELASSIGNMENTSCHEMEFORMULTIPLECHANNELADHOCNETWORKS ......................... 23 2.1Background ................................... 23 2.2RelatedWorks ................................. 24 2.3MotivationsandBasicIdea .......................... 27 2.3.1Motivations ............................... 27 2.3.2BasicIdea ................................ 29 2.4ProtocolDescriptionofSAM-MAC ...................... 30 2.4.1BasicMessages ............................ 30 2.4.2BasicOperationalProcedures ..................... 31 2.4.2.1Choosingtrafcchannel(Initialchannelselection) .... 31 2.4.2.2Transmittingtoanodelisteningonanunknowntrafcchannel ............................ 31 2.4.2.3Transmittingtoanodelisteningonaknowntrafcchannel ................................ 32 2.4.2.4Adjustingtoadifferenttrafcchannelwhenbusy/collision(Self-adjustmentprocedure) ................ 33 2.4.2.5Transmittingdataonthecommonchannel ........ 34 2.4.3OtherIssues .............................. 35 2.5PerformanceAnalysis ............................. 36 2.5.1Multi-channelHiddenTerminalProblem ............... 36 2.5.2ControlChannelSaturationProblem ................. 37 2.5.3MissingReceiverProblem ....................... 37 2.5.4OverheadComparisonandThroughputImprovement ........ 38 5

PAGE 6

............................... 41 2.6.1Single-hopTopology .......................... 43 2.6.2Multi-hopTopology ........................... 44 2.7Summary .................................... 45 3EXPLOITINGTHECAPACITYOFMULTI-CHANNELMULTI-RADIOWIRELESSMESHNETWORKS ................................. 47 3.1Background ................................... 47 3.2RelatedWorks ................................. 50 3.3ProblemFormulation .............................. 52 3.3.1PortalCapacity ............................. 52 3.3.2ProblemFormulation .......................... 53 3.3.3ProblemSimplication ......................... 55 3.4ProposedSolution ............................... 57 3.4.1PartI:CentralizedPart ......................... 58 3.4.1.1Overview ........................... 58 3.4.1.2Centralizedresourceallocationandchannelassignment 60 3.4.1.3Actualtotalresourceforallocation ............. 63 3.4.1.4Revisionofweightedresourceallocation ......... 66 3.4.2PartII:DistributedPart ......................... 67 3.4.2.1Overview ........................... 67 3.4.2.2Distributedchannelassignment .............. 68 3.4.2.3Dynamicadjustmentmechanism .............. 71 3.4.2.4Infrastructureformation ................... 72 3.4.3PartIII:SupportofMulti-gateway ................... 73 3.5PerformanceEvaluation ............................ 73 3.6Summary .................................... 76 4UTILIZINGMULTI-HOPNEIGHBORINFORMATIONINSPECTRUMALLOCATIONFORWIRELESSNETWORKS ........................... 77 4.1Background ................................... 77 4.2RelatedWorks ................................. 79 4.3SpectrumUsageIssuesinMulti-hopWirelessNetworks .......... 81 4.3.1IdealSpectrumUsage ......................... 81 4.3.2TrafcDependency ........................... 83 4.3.3AllocationInefciencyRatio ...................... 84 4.3.4AsymmetricNeighborhood ....................... 85 4.4ProposedScheme ............................... 85 4.4.1OverviewofTheProposedScheme ................. 85 4.4.2SchemeDescription .......................... 88 4.4.2.1Datastructureandmessageformat ............ 89 4.4.2.2Procedure .......................... 91 4.5PerformanceEvaluation ............................ 94 4.6Summary .................................... 98 6

PAGE 7

99 5.1Background ................................... 99 5.2RelatedWorks ................................. 100 5.3M3Description ................................. 102 5.3.1ModelDescription ........................... 102 5.3.2ProposedSolution ........................... 104 5.3.2.1Power-up ........................... 104 5.3.2.2Handlingdownstreampackets ............... 105 5.3.2.3Handlingupstreampackets ................. 105 5.3.2.4Handlinghandoff ....................... 105 5.3.2.5Periodiclocationupdate ................... 106 5.3.3ExtendedDiscussionofTheProtocol ................. 107 5.4PerformanceAnalysis ............................. 108 5.5Summary .................................... 112 6IMPROVINGHANDOFFPERFORMANCEBYUTILIZINGADHOCLINKSINMULTI-HOPCELLULARSYSTEMS ........................ 113 6.1Background ................................... 113 6.2RelatedWorks ................................. 116 6.3SystemModelandBasicIdea ......................... 117 6.3.1SystemModel .............................. 117 6.3.2BasicIdea ................................ 121 6.4ANHOA:Ad-hoc-Network-EmbeddedHandoffAssistingScheme ..... 122 6.4.1Overview ................................ 122 6.4.2ArchiectureandRoles ......................... 123 6.4.3InformationMaintenanceofEAN ................... 124 6.4.4HandoffsviaANHOA .......................... 127 6.4.5HandoffDecisionAlgorithm ...................... 130 6.5ResourceManagementforMulti-hopHandoff ................ 132 6.5.1MinimumReservation ......................... 132 6.5.2LoadSharingamongAdjacentCells ................. 135 6.6PerformanceEvaluation ............................ 137 6.7Summary .................................... 140 6.8Appendix .................................... 140 7RELIEVINGTHELOCATIONMANAGEMENTCONGESTIONVIAAGGREGATIONINMULTI-HOPCELLULARNETWORKS ..................... 142 7.1Background ................................... 142 7.2RelatedWorks ................................. 144 7.3SystemModel ................................. 146 7.4GroupingToSaveTheSignalingCost .................... 147 7.4.1Overview ................................ 147 7.4.2ProceduresofGroupLocationManagement ............. 148 7

PAGE 8

....................... 149 7.4.2.2GHLMassociation ..................... 150 7.4.2.3Groupinformationupdate .................. 151 7.4.2.4Grouplocationupdate .................... 152 7.4.2.5Dissociation ......................... 152 7.4.3GainAnalysis .............................. 153 7.4.3.1Signalingcostreduction ................... 153 7.4.3.2FindingTag 155 7.4.3.3Otherprospectivebenets ................. 155 7.5GenericAggregativeLocationManagementinMCNs ............ 155 7.5.1Overview ................................ 155 7.5.2SchemeDescription .......................... 156 7.6LocationManagementCostAnalysisandApplicationinMCNs ...... 160 7.7PerformanceEvaluation ............................ 161 7.8Summary .................................... 165 8CONCLUSIONS ................................... 166 REFERENCES ....................................... 169 BIOGRAPHICALSKETCH ................................ 175 8

PAGE 9

Table page 2-1SimulationParametersForSingle-hopTopology ................. 42 3-1AlgorithmofCentralizedResourceAllocationandChannelAssignment .... 62 4-1LocalParameterSet ................................. 89 4-2NeighborParameterSet ............................... 90 5-1TypicalValueSetofAveragePathLength ..................... 111 6-1PortalNodes'InformationTable ........................... 125 6-2AttachingNodes'InformationTable ......................... 126 6-3SubordinateNodes'InformationTable ....................... 126 6-4SimulationSettingforEachCell ........................... 138 7-1Parameters'SettingsForEvaluatingTheGroupingScheme ........... 162 7-2Parameters'SettingsForEvaluatingTheGenericAggregationScheme .... 163 9

PAGE 10

Figure page 2-1ComparisonofDCCTypeandSPType ...................... 27 2-2IllustrationofSAM-MACProcedures:UnawareCase ............... 32 2-3IllustrationofSAM-MACProcedures:AwareCase ................ 33 2-4IllustrationofSAM-MACProcedures:Self-AdjustmentCase ........... 34 2-5Multi-channelHiddenTerminalProblem ...................... 36 2-6ThroughputGaininSingle-hopTopologywith3+1Channel,withandwithoutcommonchanneltransmittingdata ......................... 43 2-7ThroughputGaininSingle-hopTopologywithMultipleChannels ........ 44 2-8ThroughputGaininMulti-hopTopologywithMultipleChannels ......... 45 3-1TheCommonStructureofaWMN ......................... 47 3-2AnExampleofLinkDistributioninWMNsandtheProposedFrequencyReusePattern ........................................ 54 3-3IdealCaseWhenFrequencyReuseisPerfect ................... 59 3-4IllustrationofChannelBusynessIndicatorPhase(CBIP) ............. 70 3-5ComparisonofPortalCapacityamong3Schemes:CW(oursolution),SC(singlechannel),RM(randomly-choosingmulti-channel). ................ 74 3-6PortalCapacitywithFrequencyReuse ....................... 76 4-1InefciencyofFairRandomAccess ........................ 81 4-2FlowChartforMessageProcessing ........................ 92 4-3FlowChartforSpectrumAllocation ......................... 93 4-4Thetopologiesforevaluation ............................ 95 4-5Comparisonofend-to-endthroughput(a),fairnessindex(b)andAIR(c)forRFA,DSSand2hopMACschemes ......................... 96 5-1MeshModelandIllustrationofM3Scheme .................... 104 5-2GainofM3overpreviousschemes ......................... 111 6-1ModelofMulti-hopHandoffs ............................ 118 6-2EmbeddedAdhocNetworkinCellularSystem .................. 123 10

PAGE 11

............................... 128 6-4HODrate ....................................... 138 6-5ChannelReservation ................................. 139 6-6MarkovChainModelforHandoffReservationSystem .............. 140 7-1LocationManagementPreliminary ......................... 146 7-2GroupAssociationProcedure ............................ 151 7-3GenericAggregativeLocationManagement .................... 157 7-4GroupingGainsofSignalingCostWithDifferentGroupSizes(K) ........ 162 7-5GroupingGainsofSignalingCostWithDifferentTripLength(L) ......... 163 7-6TheAggregationInterval'sRangeswithDifferentLUTrafc ........... 164 7-7TheAggregationSignalingGainswithDifferentLUTrafc ............ 165 11

PAGE 12

Duetoitslow-cost,exibleinconstructionandself-organizablefeatures,wirelessadhocnetworks(WANETs)haveprovidedmanyapplicationsinalotofareas.However,themulti-hoptransmissionsandlackofinfrastructurebringthelowthroughputperformanceandpoormobilitysupport.AimingatefcientlyutilizingWANETs,Istrivetotacklethesetwoproblemsintherecentyears. ThroughputperformanceinWANETsmainlyrelatestothespectrumwastagecausedbyimproperMACdesign.Formulti-channelMACschemes,duetothelimitednumberradios,thespectrumwastageismostlycreatedbythechannelassignmentsignallings.Observingthatchannelassignmentisnotnecessarybeforeeachtransmission,wecanletnodescachethepreviouschannelassignmentandliftthesignallingburdenofthecontrolchannels.Weproposedthisself-adjustablemulti-channelMACscheme(SAM MAC)toincreasetheMACthroughputinmulti-channelsystems.OnlyhighMACthroughputcannotguaranteehighend-to-endthroughputinmulti-hopwirelessnetworksbecauseiftheareawithhightrafcloadcannotacquiremorespectrumcongestionswillappearandleadtopooroverallthroughputperformance.Weproposedasemi-centralizedsemi-distributedresourceallocationschemeforwirelessmeshnetworks,withwhichcentralareascaneasilyacquiremorespectrumandtheouterareascaneasilyreusethespectrum.Withinoneneighborhood,nodeswithhightrafcloadshouldacquiremorespectrum.Previous 12

PAGE 13

MobilityissuesarenotwelladdressedinWANETsduetoitslackofinfrastructures,comparedtocellularsystems.Byaddingsomefunctionalentitiesormodulestothesystem,wecansupportmobilityinWANETssimilarly.MobileIPcanbeusedasthesolutionformacro-mobility(inter-domainroaming).Formicro-mobility,weproposedthemeshmobilitymanagement(M3)scheme.Thisschemestoresthelocations(servingmeshrouters)ofmeshnodesinthegatewayastheanchorsandusetemporaryroutingentriesbetweentheanchorsandtheexactlocationstoreachthedestinations.Thisapproachcombinestheadvantagesoftunnel-basedapproachandrouting-basedapproachandprovidesasuitablemicro-mobilitysolutionforwirelessmeshnetworks.WhenintegratingWANETsintocellularsystems,theadvantagesofWANETsandcellularsystemscanpotentiallybebothachieved.Thistypeofnetworks,knownasmulti-hopcellularnetworks(MCNs),canpotentiallyhavebettermobilitysupport.Forhandoffperformance,weproposeanadhocnetworks-embeddedhandoffassisting(ANHOA)scheme,whichutilizestheembeddedadhocnetworksinMCNstoexchangeinformationofdifferenthandoffoptionsandassistsmulti-hophandoffs.ThehandoffreservationineachBScanbeloweredbyconsideringthehandoffattemptstodifferentcells.WeproposedtheinformationexchangeamongneighboringBSsandthealgorithm 13

PAGE 14

Theperformanceresultshaveveriedthecontributionsoftheseworks. 14

PAGE 15

Wirelessadhocnetworks(WANETs)areonespecialtypeofwirelessnetworks,whichcanbedenedasdecentralizedwirelessnetworkswherethereisnopreexistinginfrastructuresuchasBSsincellularsystems,orAPsinWLANs.Unliketheinfrastructure-basedwirelessnetworks,thenodesinWANETsarecapabletorelaythepacketsfromtheupstreamnodestothedownstreamnodes.Therefore,multi-hoptransmissionshouldbeseenasoneofthemostimportantfeaturesofWANETs.Thisfeaturecancreatealotofusefulapplicationswhenthetelecommunicationinfrastructuresarenotavailable.Forexample,inadisasterrescue,whenalltheinfrastructuresaredestroyed,withthewell-functioninghand-helddevices,communicationscanstilltakeplaceifeachterminalcanreceivetransmissionsfromothersandrelaythem.Thesenetworksthereforeseemveryattractiveforvariousmilitaryandcivilpurposes. WANETshavebeenexistingforovertwodecades.Mostlyweseethemin3majorcategories: Inallthreetypes,onlyMANETsareapureadhoctypeofnetworks.Bothwirelesssensornetworksandwirelessmeshnetworksassumethetrafcaggregatetoacenterpoint,i.e.,sinkinsensornetworksandgatewayinmeshnetworks.Amongthesethree 15

PAGE 16

WANETscanbeappliedinmanyscenariosduetotheirlow-cost,self-organizing,andexibleproperties.However,therearealotofissuesinWANETsthatneedtobesolved,suchasthelowthroughput,routing,andmobilitysupport.Afterall,thisisatotallydifferenttypeofnetworks.Inordertoutilizethem,itisnecessarytounderstandthemrst.Moreover,besidestheaforementioned3majorforms,WANETshaveapplicationsinotherforms,suchasvehicularadhocnetworks,cognitiveadhocnetwork,anddelaytolerantnetworks.Infact,WANETsarepotentiallybecapabletointegratewithothertypesofnetworksinordertoachievecertainapplications.Thisvarietycreatesmoreinterestingresearchtopics.CurrentresearchtopicsinWANETsincludethefollowingareas. SolvingissuesandunderstandingproblemsintheexistingformsofWANETshelpustoutilizethemeffectivelyandefciently.ThinkingoftheprospectivebenetsthatofferedbynewformsofWANETscanalsoexpandnewwaysofutilizingthem.AlltheexistingformsofWANETswereoncenewandnowtheyourish.However,beforewego 16

PAGE 17

22 ],withthescaleofthenetworksincreasing,duetothegrowinginterferenceandthenumberofintermediatehopsofows,thepernodeend-to-endthroughputperformancewillbedecreasingto0.Thisresultdepictsthetrendofthethroughputperformancewithincreasingnetworkscale.However,forpracticalsystems,thisshouldnotbetheexcuseofthelowthroughputperformance.Thelowspectralefciencyofcurrentprotocolsisinsteadthemainreasonofthecurrentlylowend-to-endperformance.Whenthethroughputperformanceissatisfactory,awirelessnetworkshouldprovidemobilitysupporttoitsclients,whichisoneofthemostimportantreasonsthatattractpeopletoutilizewirelessnetworks.However,duetolackofinfrastructure,itisdifcultforWANETstosupportmobility.ToemployWANETs,thesearethetwoprimaryproblemsrequiringtobesolvedurgently. Asofmyresearches,Ifocusontheprotocoldesignaswellasthesystemarchitectureinordertogivesolutionstothesetwoproblems.Thecurrentlayeredapproachofprotocoldesignhasitsadvantagesinsoftwareengineeringandstandardization.However,sinceeverynodeinwirelessmediuminterfereswitheachotherinthemeantimeoftransmitting/receivingandthewirelesschannelconditionsuctuateallthetime,thisapproachcannotachieveagoodperformanceinmanyscenarios.Incorporatingmoreinformationintotheprotocolentitiescanpotentiallydecreasethespectrumwastageandimprovethethroughputperformance.Althoughend-to-endthroughputisdirectlycontrolledbytransportlayer,thespectralefciencyisdecidedbythedesignofMAClayer,assumingthatPHYtechniquesaregiven. 17

PAGE 18

TherearealotofpendingissuesinWANETs,suchasrouting,transportlayer,andsecurity.However,IcannotstretchmyresearchtoalltheareasalthoughIwishIhaveenoughenergytodoso.Iselectthesetwothemesduetotworeasons.First,thesetwoproblemsarethemostcrucialproblemsforthepracticalityofWANETs.AlthoughPHYtechniquesdirectlydrivethedatarateimprovement,sinceitislessrelatedtonetworkingareas,IchoseMAClayerasmymajorfocus.Second,therearetwomethodstounderstandasystem,microviewandmacroview.Holdingonlyonewillleadtoincompleteandevendistortedunderstanding.IselectMAClayerasthemicroviewandselectmobilitymanagementasthemacroviewtoWANETs.Indeed,IwillnotlimitmyviewofWANETswithinthesetwoproblems.EmployingWANETsisalongtermworkanditincludesmorethanjustthroughputandmobility. 1.3.1ThroughputImprovement 18

PAGE 19

RedesigningMAClayerwithonlyone-hopneighbors'informationcannoteffectivelyboostthesystems'end-to-endthroughputbecauseallocatingthespectralresourceonlyaccordingtoMAClayerinformationcannotletheavy-trafcareashavemoreresourcethanlight-trafcarea,especiallyinthecasewhentherearemultiplechannelstobeassigned.Moreover,spatialreuseiscriticaltothethroughputperformanceinmulti-hopWANETsanditisdifculttodesignbecausethecentralizedcontroliseitherunavailableorhardtoachieve.Weproposedaspatialreuseparadigmformulti-channelmulti-radiowirelessmeshnetworks.Tobestefcientlyutilizetheavailablespectrum,themeshnetworksaredividedintothecentralpartandtheoutsidepart.Thecentralpartusesacentralizedalgorithmtondtheoptimumspectrumallocation.Themeshroutersoutsidethecentercandistributedlyreusetheresourceinthecentralpartbysensingthespectrumavailability.Inthispaper,wealsoproposedaconceptnamedasportalcapacitywhichindicatesthecapabilityameshnetworkcanprovidewiththeassumptionofagiventrafcdistribution.Thiscapacityisofgreathelpinthenetworkconstructions. InWANETs,differentareasmayhavedifferentrafcload,asmentionedinmypreviouswork.Inneighborhoodareas,differentnodesmayhavedifferenttrafcdemand.Moreover,thetrafcofdifferentnodesmayhavecorrelationsduetomulti-hopowstraversingacross.Withouttheseconsiderations,spectrumallocatedcanbeeasilywastedeitherincurrenthoporinfollowinghops.Inanotherworkofmine,Iinvestigatethespectralinefciencyofcontention-basedMACforWANETs.Althoughscheduling-basedstandardisavailabletoprovidehigherMACefciency,e.g.IEEE802.16,itisnotsuitableforaunplannedadhocnetwork.Foraself-conguredadhocnetworks,dowehavetolivewiththelowspectrumefciency?Weproposedtoleteachnodeinadhocnetworksallocatespectrumintelligently,insteadofcontendingthechannelblindlyasthecurrentIEEE802.11.Furthermore,welettheneighboringnodes 19

PAGE 20

Inordertoadapttodifferentwirelessnetworks,thecustomizedMACishighlyneededandthecross-layerinformationisprobablyneeded.Ibelievethereisstillalongwayforustofullyunderstandthespectrumefciencyofwirelessnetworks. InIPnetworks,themobilityisusuallyaddressedbymobileIPprotocol.However,mobileIPisonlysuitablefortheinter-domainroaming,i.e.,macromobility.ChangingIPaddressinasmallareaisundesiredforacontinuousservice.IfthesameIPaddressisusedinonedomain,locatingthemovingnodewillconsequentlybeanissue.Weproposedtocombinethetunnel-basedapproachandtherouting-basedapproachtodealwiththemicro-mobilityinwirelessmeshnetworks.Themeshgatewaystorestherecentlocationofeachnodeandtunnelthedownlinkpacketstothecorrespondingmeshrouter.Withinoneperiod,thetemporaryroutingentriesineachrouterthatthemovingnodepasseshelptheintermediateroutertoforwardthedownlinkpacketfromthestoredlocationtothecurrentlocation.Periodicalupdatewillremovethesetemporaryroutingentriesandcontroltheroutingdelay. Iamcurrentlydelvingintotheissuesinmulti-hopcellularnetworks(MCNs),suchashandoffperformanceandlocationmanagementscheme.MCNsareanewemergedtypeofnetworks.Introducingadhocmodeintocellularnetworkscanpotentiallyletus 20

PAGE 21

Assuminghandoffnodescanaccessadjacentcellsviamulti-hopconnections,bylettingneighboringBaseStations(BSs)exchangetrafcinformation,wecanreservelessresourceforthehandoffcallstomeettherequiredhandoffdroppingrate.Wecanalsoletthestaionary/semi-stationarynodesformsmall-scaledadhocnetworksandletthemassistthemulti-hophandoff.Inthisway,thespectrumineachcellcanbemoreefcientlyutilizedandthehandoffdroppingratecanbegreatlydecreased.InmyrstMCNpaper,weproposedANHOA(ad-hocnetworksembeddedhandoffassisting)scheme,whichutilizetheembeddedadhocnetworksincellularsystemstoassisthandoffs,andanalgorithmtoreducetheresourcereservationforhandoffcallsbyconsideringthathandoffattemptsthatfailinonecellmaysucceedinanotherthroughmulti-hopconnections.TheextendedframeworkofinformationexchangeamongBSscanfullltheloadbalancingtask,whichisotherwiseverydifcult. MysecondMCNpaperfocusonthebenetsthattheintroductionofadhoctocellularsystemscanbringtothelocationmanagement.Weproposedagroupingmethodforhighcapacitytransitsystemstoreducethelocationmanagementsignalingoverhead.Wealsoproposedagenericaggregativeschemewhichassignssomeaggregationdevicestocollectthelocationupdaterequestsfromthearrivalmobileterminals.Thesetwoschemecanutilizetheadhocmode'srelayfeatureandsignicantlyreducethelocationmanagementsignalingoverhead. Chapter 2 presentsmyworkonthemulti-channelMACschemeforWANETs.Chapter 3 presentsmyworkonthethroughputimprovementeffortonmulti-channelmulti-radiowirelessmeshnetworks.Chapter 4 presentsmyworkonimprovingthe 21

PAGE 22

Chapter 5 presentsamicro-mobilityschemeforwirelessmeshnetworks.Chapter 6 presentstwoschemesforhandoffperformanceimprovinginmulti-hopcellularnetworks(MCNs),amongwhichoneutilizesembeddedadhocnetworkstoassistthehandoffcallsandtheotheraimsatreducingtheBSs'handoffreservationconsideringmulti-hophandoffattempts.Chapter 7 presentsourworkonusingaggregativedevicesinMCNstoreducelocationmanagementsignallings. Chapter 8 concludesthisdissertation. 22

PAGE 23

1 ].ThebenetsofadoptingmultiplechannelsinMAClayerareshowninmanypapers.Themostapparentbenetisthethroughputimprovement. Nasipurietal.[ 48 ]showedthatwiththesametotalbandwidth,dividingasinglechanneltomultipleonesunderCSMAmechanismgainsacertainthroughputimprovement.Thereasonisthattheutilizationofmultiplechannelscanmitigatecollisionsandcontentions.Intheextremecasewhenchannelassignmentisperfectandeachpairofnodeshaveadedicatedchannel,thecontentionandcollisiondisappear.Therefore,inthissituation,thecapacityofthenetworkscanbefullyused. Theconclusionabovecanbeonlypartlytruebecausetheoverheadofchannelassignmentinarealmulti-channelsystemcannotbeignored.Thecarriersensingcoupledwithanefcientchannelassignmentmechanismisalwaysusedtoselectthechannelwiththeleastinterferencefortransmission. Theotherbenetwewillgainfrommulti-channelisthefairness.Weknowthatin802.11protocols,duetohidden/exposedterminalproblem,insometopologyscenarios,somenodesmaybecomemoredisadvantagedandgetlessopportunitytosuccessfullytransmitthanothernodes.Sometimesthissituationwillcausemoresevereproblem.Bymovingthedisadvantagednodestoanotherchannel,thefairnessproblemof802.11systemcanbealleviated.Inotherwords,thenodeshavemorechoicesofchannelsthaninasinglechannelthusbetterfairnesscanbeachievedinmulti-channelsystems. 23

PAGE 24

Inthispaper,weproposeanew802.11-likemulti-channelMACprotocol,calledSelf-AdjustableMulti-channelMAC(SAM-MAC).Thenovelpartofthisschemeisthechannelassignment,whereaself-adjustmentmechanismisusedtobalancethetrafcofmultiplechannels,thusamoreefcientutilizationofchannelsisachievedandthethroughputperformanceisimproved.Italsomitigatesthecontrolchannelsaturationproblemgreatlyandconsequentlyreducestheoverheadfurther. Therestofthispaperisorganizedasfollows.Section 2.2 reviewstherelatedworks.Sections 2.3 and 2.4 presentthebasicideasandprotocoldescriptionofSAM-MAC.Section 2.5 providesadiscussionoftheproblemsSAM-MAChassolvedandtheimprovementsithasobtained.Finallythesimulationresultispresentedandconclusionsaredrawninsection 2.6 and 2.7 ,respectively. 48 ]isoneoftherstmulti-channelCSMAprotocols,whichusessoftchannelreservation.IfthereareNchannels,theprotocolassumesthateachhostcanmonitorallNchannelssimultaneouslywithNtransceivers.Ahostreadytotransmitapacketsearchesforanidlechannelandtransmitsonthatidlechannel.Amongtheidlechannels,theonethatwasusedforthelastsuccessfultransmissionispreferred.Theprotocolisextendedbyotherstoselectthebestchannelbasedonsignalpowerobservedatthesender.Thismulti-channelschemeisasimpleextensionfromthesinglechannelMAC(802.11).ItrequireseachnodehaveNtransceiverswithoneforeach 24

PAGE 25

DBTMA[ 24 ]andDUCHA[ 72 ]alsodivideachannelintomultiplesub-channels,specically,onedatachannelandonecontrolchannel.Busytonesaretransmittedtoavoidhiddenterminalproblems.Throughthiswaythespatialutilizationisincreasedthusabetterthroughputperformancethan802.11canbeachieved.Theseschemesaimatthehidden/exposedproblemsinmulti-hoptopologies. Toexploitthespectrumefciencywithmulti-channelschemes,channelassignmentisthefocusofmanyotherpapers. S.Wuetal.[ 64 ],proposedaprotocolthatassignschannelsdynamically,inanon-demandstyle.Thisprotocol,calledDynamicChannelAssignment(DCA),requiresonededicatedchannelforcontrolmessagesandotherchannelsarefordatatransmission.Eachhosthastwotransceivers,sothatitcanlistenonboththecontrolchannelandthetrafcchannelsimultaneously.RTS/CTSpacketsareexchangedonthecontrolchannel,anddatapacketsaretransmittedonthetrafcchannel.DCAfollowsanon-demandstyletoassignchannelstomobilehosts,anddoesnotrequireclocksynchronization.Thiskindofschemesdoesnotperformwellwhenthenumberofchannelsislargebecauseallthenegotiationsarefullledonthecontrolchannelandtoomuchcontentionwillcausethesaturationproblemoverthecontrolchannel. Similarideasareusedin[ 14 ]and[ 73 ].Additionally,Zhangusedtwocommonchannelstosolvethehidden/exposedterminalproblemsin[ 73 ]. MMAC[ 56 ]usesadifferentwaytoassignthechannels.Thisprotocoldoesnotneedaseparatecontrolchannel.Instead,itutilizesanATIM-likewindowinthedefaultchanneltofullthechannelnegotiation.TheATIM(AdhocTrafcIndicationMessage)windowisthesynchronizationphasewhen802.11PowerSavingMechanism(PSM)isapplied.Eachnodedecidestobeeitherindozemodeorawakemodeaccordingtothe 25

PAGE 26

Shietal.[ 55 ]proposedAMCPschemewhichissimilartoDCAschemeexceptthatitneedsonlyonetransceiver.Thismajorfeaturecomesfromadirecttimeoutmechanismbeforenodesselectthechannels.Thistimeoutmechanismsolvesthemulti-channelhiddenterminalproblem.However,thisschemedoesnotbringgreatimprovementofthroughputperformance.Theothermajorpartofthispaperisthefairnessimprovement.InInformationAsymmetry(IA)scenarioandFlowIntheMiddle(FIM)scenario,somedisadvantagednodeswillstarveduetotheirdisadvantageinthetopologytogetthechannelidleinterval.Usingmultiplechannelscanmitigatesuchastarvationproblemthroughallocatinganotheridlechanneltothedisadvantagedowtimely.However,thispaperonlysolvedthisproblembyadjustingthedisadvantagedowtoanotherorthogonalchannel.Inthecasewherethenumberofchannelsismuchlessthanthenumberofows,suchastarvationbecomesinevitableagain. Aclassicationhasbeengivenrecentlyin[ 46 ].Inthispaper,themulti-channelschemeshavebeendividedintofourcategories: 1. DedicatedControlChannel(DCC); 2. CommonHopping(CH); 3. SplitPhase(SP); 4. MultipleRendezvoususing1radio(MR). CHandMRusetheideaoftimedivisionandfrequencyhopping.RICH-DP[ 58 ]isanexampleofCHandSSCH[ 8 ]isanexampleofMR.Though[ 46 ]showsMRhasabetterperformancethanDCCandSP,weexcludethem(CHandMR)inourpaperinthatthey 26

PAGE 27

ComparisonofDCCTypeandSPType useaverydifferentapproachinwhichtimesynchronizationisneededandachannelhoppingsequenceisfollowed. ItisclearthattheDCAschemebelongstoDCCtypeandMMACbelongstoSPtype,accordingto[ 46 ].Fig. 2-1 illustratesthebasicdifferenceofchannelassignment(CA)betweenDCCtypeandSPtypeofapproaches. 2.3.1Motivations Ifeachnodehasadedicatedtransceiverforeachchannel,thechannelassignmentwillhavezerooverhead.Thisisbecauseeverychannelisvisibletoeverynodeallthetime.However,duetocostconsideration,thetransceiversareusuallyfewerthantheavailablechannels.Therefore,channelassignmentneedstoassigntheavailablechannelresourcetolimitedtransceiverswhentherearedatatransmissionrequests.Itisimportanttoknowthechannelusageinformationbeforeactualchannelassignment. 27

PAGE 28

Botharefeasibleapproachestoachievingthroughputimprovement.Butusingsplitphasesneedstimesynchronization,whichisdifculttorealizeinadhocnetworks.Alsowithsplitphases,howtodividetimeintodifferentphasesisstillatwo-foldproblem.Therstdifcultyisthatdatapacketshavevariablesizewhichprobablydonotutilizethedataphaseefciently.Thesecondishowlongthechannelassignmentphaseshouldbe.Thisisbecausetrafcmaybedifferentandrequestloadofchannelassignmentisalsohardtopredict.Withoutsynchronizationandtimedivisionproblem,thoughadedicatedfrequencybandstillmeansrelativelyhighoverhead,theDCCapproachseemsmoreappealing.(Hereweassumethesynchronizationisnotmandatoryinthesystem). Sincethetransceiversarefewerthanchannels,thetransceiversneedtoswitchamongtheavailablechannels.Therefore,twoimportantissuescannotbeignoredintheschemedesign.Oneisthechannelswitchingdelayproblemandtheotheristheacquirementofthechannelusageinformation.Fortherstproblem,aper-packet-basedchannelassignmentschemeisnotpreferred.Channelassignmentshouldbevalidforalongerperiod.Forthesecondproblem,whenasingletransceiverisused,thechannelassignmentisdifcultafterdatatransmissionbecausetheusageinformationofotherchannelswillbeinvisibleduringthedatatransmissionperiod.AMCP,whichusesonlyonetransceiver,bypassesthisdifcultyintimedomain[ 55 ].Itrequeststhenodestowaitforonedatatransmissionperiodbeforechannelassignmentifthepreferredchannelisnotavailable,whichisalsoanextraoverheadofchannelassignment.Toaddressthesetwoproblems,usingtwotransceiversisabetterchoice.Thisfurthermorehelpstoreducetheoverheadbyallowingdatatransmissiononthecommonchannel. 28

PAGE 29

Thebasicideaisbasedonsuchanobservationthatinallpreviousschemes,handshakesoccuronlyatthecommonrendezvous.Thismechanismmakesthecommonrendezvoussusceptibletobethewholesystem'sbottleneck.SAM-MACdistributesthehandshakestoavailablechannelsandfurthermoremakesthedatatransmissiononthecommonchannelpossible.Inthisway,ourschemeachieveshigher 29

PAGE 30

Whenthedestinationisnotavailable,itcansendaNCTSonchannel0tothesendertonotifytheNAV(NetworkAllocationVector)[11].Itisusedtosolvethereceiverblockingproblem. Ifthedestinationnode'slisteningtrafcchannelnumberisunknown,thesendersendsRTFonChannel0tonditoutandthedestinationnodeanswersitalsoonChannel0; Ifasenderfeelsthetrafcchannelsituationunsatisfying,itwillsendRCTtothedestinationtochangethesender'strafcchanneltoanotherone,onChannel0.ThenthedestinationnodewillanswerbackACTwhenitdecideswhichtrafcchanneltochoose,alsoonChannel0.Theoldtrafcchannelinformationshouldbeincludedinthesemessagestoletothernodesknowthebusystatusofthistrafcchannel. 30

PAGE 31

Nodeskeeplisteningonthesametrafcchanneluntiltheyneedtotransmittosomenodesonotherchannelsorthischannelisgreatlysaturated.Thechangeofonenode'slisteningtrafcchannelshouldbepublishedtoallotherneighborsviathecommonchannel. Howthesendingnodesndthereceivingnodesandhowtheychangechannelsaredescribedinfollowingprocedures. 2-2 shows,thetransmittingnodesendsaRTFonthecommonchannelrst.Sinceeachnodelistensonthecommonchannel,thedestinationnodeanswersanATFonthecommonchannel,whichincludesthetrafcchannelinformation.Hereafter,thetransmitting 31

PAGE 32

2-3 .IfthesendernodeislisteningonadifferentchannelbeforeRTS,itshouldsendaNBCmessagetonotifyallitsneighborsofthechannelchangeafterRTSissent.ThepurposeistobroadcasttheNAVtoallitsneighborsandtoavoidtheMissingReceiverProblem[ 55 ]causedbytheunnotiedchannelchange.ThereceiversarealsorequiredtobroadcastNAVviaNBCframes. Theneighbors'channeltablealsoincludestheavailablechannellistsofitsneighbors.Thisinformationisstoredforthepurposeofchanneladjustment. Figure2-2. IllustrationofSAM-MACProcedures:UnawareCase 32

PAGE 33

Iftheuserskeeponchangingchannels,thesystemthroughputaswellastheQoSoftheuserswillbedegradedduetothecommunicationoverheadandswitchingdelay.Therefore,abetterchannelre-assignmentalgorithmisrequiredtoreducethefrequencyofchannelchanging.Weassumethetrafcoftheusersisnotextremelyunbalancedandburstythusthetrafcloadcanbewellbalancedamongthechannels.Wecaneasilyobservethatonlywhenachannelcontainsagreatlyvaryingtrafccanthefrequencyofchannelreassignmentbebig.Theassumptionaboveisreasonablebecausetheaggregatetrafcinachannelisrelativelyconstantandtheuctuationofsmallnumberofnodes'trafcwillnotcausegreattrafcloadimbalanceamongdifferentchannels. Figure2-3. IllustrationofSAM-MACProcedures:AwareCase 33

PAGE 34

IllustrationofSAM-MACProcedures:Self-AdjustmentCase Weusetwometricsforthechannelre-assignmentalgorithm:thenumberofneighborsandthechannelbusynessratio[ 68 ],whichcanbothbecountedorcalculatedfromtheinformationheardonthecommonchannel. Channelassignmentalgorithmallowsthecommonchanneltobechosenfordatatransmissionwhenthecommonchannelislight-loadedandthechanneladjustmentisneeded. 34

PAGE 35

Sincethechannelreassignmentisdecidedbythereceivers,theasymmetricinformationofchannelsbetweenthesendersandthereceiversmaycauseunreasonablechannelreassignment.Foranexample,areceivermaychooseachannelthatthesendercannottransmitatall.Thisisausualcaseinmulti-hoptopologies.Toavoidthisinconsistency,thechannelbusynessratioofthesendingnodes'channelsshouldalsobeconsideredduringthechannelreassignment.Intheadjustmentprocedures,thesendersshouldalwaysincludeachannellistwithadecreasingorderofchannelbusynessratio.Thereceiversstorethisinformationtotheneighbors'channeltable.Beforethedecisionofchannelreassignment,thereceiverschooseachannelwiththelowestchannelbusynessratiofromitsowntrafcchannelstatustablewhichisalsointheavailablechannellistoftheneighbors'channeltable.Therefore,thereceiverswouldnotchooseachannelthatcannotbeacceptedbythesenders. Allthechannelsareassumedheretohavethesamewirelessstatusandthechannelbusynessratiocanembodytherealchannelstatusofeachchannel.Withouttheconsiderationoffrequency-selectiveinterference,allthechannelusagecanbeacknowledgedbytheNAVbroadcasting,whichisthevirtualcarriersensing.Ifthephysicalwirelesschannelstatusneedstobeconsidered,thesecondtransceivershouldscanaroundallthetrafcchannelstodothephysicalcarriersensing.Inthiscasethisoverheadalsoneedstobeconsidered. 35

PAGE 36

Multi-channelHiddenTerminalProblem InFig. 2-5 ,nodeAiscommunicatingtonodeBinacommonchannel.NodeCswitchestothischannelaroundtimeT1when,unfortunately,itmissestheCTSsentoutbyB.IfnodeC,ahiddenterminaltoA,proceedstotransmittosomeothernodeaftersensinganidlechannel,thesignalscouldcollidewithsignalssentbyAatthereceiverofB.WecanseethereasonofthisproblemisthatnodeCmissesthechannelstatus,theNAVinCTS.Ifeachnodecanhavethesamenumberoftransceiversaschannels,thisproblemisavoided. ForDCA[ 64 ]andMMAC[ 56 ],sincehandshakesoccuronlyatthecommonrendezvousandeachnodecanobtainthechannelusageinformationbymonitoringhandshakes,multi-channelhiddenterminalproblemdoesnotoccur.ForAMCP[ 55 ],withonededicatedchannelandonlyonetransceiver,multi-channelhiddenterminalproblemcouldnotbeeasilyavoided.WithadirecttimeoutmechanisminAMCP,eachnodeavoidsthisproblembyacquiringthechannelusageinformationduringthisperiod. 36

PAGE 37

ForSAM-MAC,sincehandshakesaredistributed,nodesondifferentchannelsareblindabouteachother'schannelusageinformationifthecontrolchannelisabsent.However,withonededicatedtransceiveronthecontrolchannel,SAM-MAChasthecapabilityofknowingallchannels'usageinformation.ThemessageNBCisusedforthispurpose.Withthismessage,everynodeknowstheNAVofallthechannelsandavoidthemulti-channelhiddenterminalproblem. 64 ],thissaturationproblemofDCAschemehasbeenshownbythesimulationresults.NeithercanMMACovercomethisproblemduetothelimitofATIMwindowlength.In[ 55 ],AMCPusesthesameserialcontentionprocedureonthecontrolchannel,soitcannotavoidsuchaproblem,either. ForSAM-MAC,weobservethatmostofthehandshakesbeforeeachtransmissionoccurondifferenttrafcchannels.Themajortrafconcontrolchannel,NBCmessages,isnotrelatedtothehandshake.Consequentlythesaturationproblemcanbeovercome.Onlywhentherearealotofchanneladjustmentswillthehandshaketrafcofcontrolchannelbecomeheavier.Thiscanbecausedbyextremeuctuationoftrafcaggregationoneachchannel,whichisveryrare. 37

PAGE 38

ForSAM-MAC,allthechannelusageinformationcanbeobtainedfromcontrolchannelwithadedicatedtransceiver.Therefore,thisproblemissolvedinSAM-MAC. 2-1 .Theaboveequationshowsthatthemaximumthroughputwithmultiplechannelsandmulti-channelschemesistheproductofthenumberofchanneltimesthemaximumthroughputof802.11protocols,minusthechannelassignmentoverhead.Obviously,thechannelassignmentoverheadisthemajorindicatoroftheperformanceofamulti-channelscheme. ReferringtoFig. 2-1 ,theratiobetweenchannelassignmentphaseandtotalbeaconintervalinSPtypeschemesisdenotedas.Therefore,thechannelassignmentoverheadofSPtypeisandtheoneforDCCtypecanbeeasilytakenas1 38

PAGE 39

56 ]asanexample.=20=100=0.2.Whenthenumberofchannelsislessthan5,theSPtypeofschemeshavelesschannelassignmentoverhead.Ifthiscanbesmallerthan0.2,thisadvantageofSPtypecanholdinthescenarioswithmorechannels. However,DCCtypeandSPtypedonothavetheidenticalchannelassignmentcapability.Althoughthewholecontrolchannelcanbeusedforchannelassignment,only1=nofCAphaseisusefulinSPtypebecauseoftherequirementofacommonrendezvous.ToachievethesamecapabilityofchannelassignmentasDCCtype,SPtypeshouldrequire 55 ],[ 64 ],thattherewillbeasaturationproblemincontrolchannelwhen68channelsarefullyutilizedinDCCsystems.Accordingtothepreviousreasoning,weconcludethatSPsystemscannotsupportsomuchchannelassignmentload.Therefore,theSPsystemsareobviouslymorevulnerabletothesaturationproblemthanDCCsystems. Thecomparisonaboveshowsthatinsingle-hoptopologyitisimpossibleformulti-channelschemestogetntimesthroughputasinasingle802.11channelbecauseoftheexistenceofchannelassignmentoverhead.Whenthechannelassignmentloadisheavier,evenn-1timesthroughputisimpossible. Thislimitisnotthesameinmulti-hoptopology.Themulti-channelschemesallowtransmissiontobehappenedconcurrentlyondifferentchannelswhichmaynotbeallowedinasinglechannelscheme.Inotherwords,tosomeextentmulti-channelschemesincreasethespatialreuse.Evenwiththeseoverheadsaforementioned,thethroughputimprovementismoreapparentwhentrafcisheavyorexposed/hiddenterminalproblemsaregreatlyalleviatedbymultiplechannels. 39

PAGE 40

Furthermore,thedifferencefromotherDCCSchemesisthedistributionofhandshakestotrafcchannels.WecanmodelthecontentionasaM/M/1queue.Assumeeachcontendergeneratesapacketwithexponentialinter-arrivalrateandeachresolutionofcontentionsalsofollowstheexponentialdistributionwiththerate.Thenumberofcontendersinthesystemism.Whenusingpreviousschemes'handshake,theprobabilityofkcontenderspkinthesystemandtheaveragequeuelength )(m )k (1m )1 AfterusingSAM-MAC,withmcontendersdispersedtonchannels,theresultischangedto: (1m Itisshownfromtheresultthatp0increasesand 40

PAGE 41

10 ],thesaturatedthroughputisindependentofmwhenmislargeenough.Thisisbecauseinsaturatedsituationthedifferenceofnumbermcouldnotbringmuchdifferenceofcollisionprobability.Whenmissmallorthesystemisnon-saturated,maffects Forallthepreviousschemesinwhichhandshakesoccuronlyatthecommonrendezvous,thededicatedpart(thecommonchannelortheATIMwindow)cannotbeusedfordatatransmissionevenifitisfarfromsaturationbecauseofthedependencyofdatatransmissiontohandshakes.However,withdistributedhandshakes,ourschemeallowsthecommonchanneltotransmitdata,whichmakesthecommonchannelnolongerdedicatedandachievesgreaterthroughputthananyotherschemes.Therefore,Soverheadisfurtherreducedandthethroughputperformanceisimproved.Additionally,sinceourschemegreatlyalleviatesthesaturationproblemofcontrolchannel,thisschemecansupportmuchmorechannelsthanotherschemes. Shi'swork[ 55 ]isusedasthereferenceofsimulationresultforsingle-hoptopology.Thereasonisthatthispaperisthelatestoneaboutthemulti-channelschemes.Ithasgivenacomprehensivegeneralizationofpreviousworkandperformancecomparison.However,inmulti-hopscenarios,thisworkfocusesmainlyonthefairnessissue 41

PAGE 42

SimulationParametersForSingle-hopTopology SIFS 10s DIFS 50s EIFS 364s TimeSlot 20s PHYheader 192bits MACheader 224bits RTS 160bits+PHYheader CTS,ACK 112bits+PHYheader DATA 8000bits+PHYheader+MACheader RTF,ATF,RCT,ACT 160bits+PHYheader BasicRate 2Mbps DataRate 2Mbps SwitchingDelay 0 TopologyRange 100m*100m FlowNumber 15 Duration 25s withoutaddressingthethroughputperformance.Therefore,weevaluateourschemeinmulti-hopscenariosthroughothermethod. Inthesimulation,differentnodesinthenetworkstarttodeliverpacketsfromdifferentstartingtime.Channelre-assignmentalgorithmwilldistributethetrafcevenlytotheavailabletrafcchannels. Afterthisperiod,thechannelswitchingisnotsofrequentbecausethenodeskeepstayingonthesamechannelunlessthetrafcchangeorconnectivitycannotbesatised.Thustheswitchingdelayplaysalessimportantroleinourscheme,whichisignoredhere. Inthefollowingsimulations,single-hopthroughputperformanceiscomparedwithAMCP[ 55 ]underthescenarioswithdifferentnumberofchannels.Specicallythedetailedthroughputperformanceof3+1(3trafcchannelsand1commonchannel)isshown.Multi-hopthroughputperformanceisprovidedandevaluatedafterthat. 42

PAGE 43

10 ],itisnecessarytoaddressthattheaggregatethroughputinmulti-hopscenariosdependsmuchuponthetopologyandthecoveragearea.Simplysaying,abiggertopologypotentiallyhasbiggeraggregatethroughputthanasmalleronebecauseofmorenodesandmorespatialreuse.Forthisreason,thevalueofaggregatethroughputinmulti-hopscenariosismeaninglesswithoutcomparison.Thebenetofourschemeinmulti-hopscenariosisshownthroughcomparingthethroughputgainwithdifferentnumberofchannelsonacertainrandommulti-hoparea. 5-1 Insingle-hoptopologies,eachnodecanlistentoalltheothernodes.Asmentionedbefore,throughputofmulti-channelschemescannotexceedn-1timesofthesaturatedthroughputofa802.11singlechannel,whichintuitivelyistheupperboundthroughputofDCCtypeschemes.However,SAM-MACcanbreakthisupperboundwhencommonchannel(CCinthegures)isusedfordatatransmission. Figure2-6. ThroughputGaininSingle-hopTopologywith3+1Channel,withandwithoutcommonchanneltransmittingdata Fig. 2-6 showsthethroughputcomparisonof802.11,AMCP,andSAM-MACwithandwithoutCCusedfordatatransmission.Undersaturatedsituation,SAM-MACwithoutCCusedfordatatransmissioncangetaslightlybetterperformancethanAMCP, 43

PAGE 44

Nextsimulationshowsthepotentialmultiplechannels'throughputusingourscheme.Togetthepotentialmultiplechannelsthroughput,moretrafcresourcesareneeded.Inthissimulation,100nodesareused.Areferencelineisusedforthegaincomparison.Wenotethatthecommonchannelisnotusedfordatatransmissioninthissimulationinordertomakethebenetofbeingfreefromsaturationproblemclearer. Figure2-7. ThroughputGaininSingle-hopTopologywithMultipleChannels FromFig. 2-7 wecanseethefollowingresult.Thechannelpointstandsforthestandard802.11protocols'throughput.Withthenumberoftrafcchannelsbeingincreasedupto11,thethroughputofSAM-MACisapproximatelylinearlyincreasing,whichshowsthisschemeisfreefromthecontrolchannelsaturationproblem.When11+1channelsareusedthethroughputcanachieve9.5timesasasingle802.11channel.Thepotentialthroughputcanreach1750pkt/secorevenmore.Inthisscenario,theoverheadofadedicatedcontrolchannelcanbealmostignored. 44

PAGE 45

ThroughputGaininMulti-hopTopologywithMultipleChannels comparisonwiththethroughputinsingle-hoptopology,CCisnotusedfordatatransmissioninthissimulation. FromFig. 2-8 ,itisshownthatinthetopologieswherethehidden/exposedterminalproblemsaffectthethroughputgreatly,usingmultiplechannelscanachievehighergainsthansingle-hoptopologycase.Thesaturatedthroughputofonesinglechannelisonly150pkt/sec,whichismuchlowerthanthe185pkt/secinsingle-hoptopologycase.AfterusingmultiplechannelswithSAM-MACprotocol,thegainismorethaninsingle-hoptopologies.In3+1channelsand6+1channelscases,thegainsaremorethan3or6timesofasingle802.11channel'sthroughput,respectively.Thereasonthat10+1cannotachievemorethan10timesgainisbecausethetrafccannotfullysaturatethechannels. Theresultofourthroughputperformancesimulationsupportstheoverheadanalysisaforementioned. 45

PAGE 46

ThisschemehasobtainedabetterthroughputperformancewithonemoretransceiverthanAMCPandMMAC.Itislessrestrictedandeasilydesignedbecauseofthisextratransceiver.Futureworkincludesthedetailedfairnessanalysisandfairalgorithminchannelassignment.Thispartcanbedoneasonesinglemoduleofchannelassignment.Thereforethesoftwareiseasytobeupgraded. 46

PAGE 47

Akyildizetal.haveprovidedanexcellentsurveyonWMNs[ 5 ].Commonly,aWMNconsistsofalimitednumberofgatewaystoaccessthewirednetworks,manymeshrouterseachofwhichcoversacertainareaandalargernumberofmeshclients.Fig. 3-1 givesanexampleofaWMNstructurewithonegateway.Gatewaysfunctionastheportaldevices,whichprovideaccesstothewiredInternet.Meshroutersconnectwithgatewaysviawirelessmediathroughzeroormoreintermediatemeshrouters.Meshclientsaccessthesystemthroughthelocalmeshrouters.Thelocalmeshroutershelpmeshclientstoforwarddatapacketstothegateways,withoneormultipleintermediatehops. Figure3-1. TheCommonStructureofaWMN 47

PAGE 48

Usually,itisdeemedthatmeshclientsuseWMNsmostlyforaccessingtheInternet.Therefore,theabilitythataWMNprovidesmeshclientstodeliver/receivedataacrossthegatewaysisthefocusofthedesign.Duetolimitednumberofgatewaysandirregulardistributionofmeshclientswithmulti-hopdistancefromthegateways,itisdifculttoconstructaWMNwhichcansatisfybothcoverageandthroughputrequirement.Researchershavebeendelvingintothedesignissues.Jiangetal.,proposedaCDMA-basedresourcemanagementschemeforthewirelessmeshbackbone[ 28 ].InordertoboostthethroughputofWMNs,theydesigntheschemeofresourceallocationbyusinglocationandinterferenceinformationfromthereceivers'pointofview.AttemptingtondoutwhatattributetocurrentlowthroughputperformanceofWMNs,wegeneralizetwomajorfeaturesforWMNs:multi-hopandtrafcaggregation. Duetomulti-hopfeatureofWMNs,alargepartofresourceiswastedinforwardingattheintermediaterouters.Thisfeaturecausesmultiplechannelcontentionsandowdiscontinuityaswell.Theothermajordifcultycomingfrommulti-hopfeatureisthefrequencyreuse,orfromanotherangle,concurrenttransmission.Whenthereisnofrequencyreuse,thesystemthroughputisdenitelylowbecausesomanylinkssharelimitedamountoffrequencyresource,alargepartofwhichisusedforforwarding.Inmulti-hopscenarios,reusingfarawayfrequencyresourcewithproperplanningisabasicandeffectivewaytoimprovethethroughputperformance. Becauseofthetrafcaggregationtowardsagateway,centrallinksalwayshavemoretrafcload.Whentheselinksdonotacquiremoreresourcethanoutsidelinks, 48

PAGE 49

Fortunately,withtheaidofmultiplechannelsandmultipleradios(MC-MR),theWMNs'performancecanbeimprovedwitheffectivechannelallocation[ 35 ].IEEE802.11bandIEEE802.11aspecify3and12non-overlappingchannels,respectively.Tothisend,manypreviousworksproposedarchitecturesoralgorithmswithdifferenttechniques[ 35 ],[ 50 ],[ 6 ],[ 34 ],[ 51 ],[ 53 ],[ 65 ]. Unfortunately,uptonow,nopreviouspapergivesustheknowledgeofthecapabilitythatWMNsprovidethemeshclientstodeliver/receivedataacrosstheportals.ItwillbeveryusefultoknowthemaximumtrafcaWMNcansupporttodeliveracrossthegatewaysattheconstructionphasewithoutthepresenceofmeshclients. Inthispaper,weinvestigatethecapacitythatWMNsprovidemeshclientstodeliver/receivedataacrossthegateways,whichwetermasportalcapacity.Itwillbeshownlaterthatportalcapacityvarieswithdifferentfairnessconstraints.Weassumethattrafcdistributionisknownsothatstatisticallyeachmeshroutercontributesacertainamountoftrafctothegateways.Undertheseassumptions,theobjectivefunctionofoptimalportalcapacityisformulatedwithouttherequirementoftheknowledgeofthereal-timetrafcdemand.Furthermore,weproposeoursolutionofachievingtheoptimalityinportalcapacitythroughthecentralizedalgorithmwhileretainingtheoptimalityviaadistributedmechanism. Thecontributionsofthispaperareasfollows. 49

PAGE 50

Therestofthepaperisorganizedasfollows.Section 6.2 discussestherelatedworks.Section 3.3 formulatesthetargetproblem.Section 3.4 describestheproposedsolution.SimulationandevaluationareprovidedinSection 4-4 .Conclusionsaregiveninthenalsection. 51 ],[ 53 ],[ 65 ],[ 57 ].In[ 53 ],anarchitecturecalledHyacinthisproposedfortheMC-MRWMNs.Eachmeshroutercarriesoutload-balancingroutingandload-awarechannelassignmentinadistributedfashion.In[ 51 ],theinterferencemitigationisintroducedinthedesign.Themulti-radioconictgraphandabreadth-rstsearchingalgorithmareusedfortheinterference-awarechannelassignment.In[ 57 ],thechannelassignmentisjointlydesignedwithmulti-pathroutingandscheduling.In[ 65 ],superimposedcodeisusedinchannelassignmentbecauseofitspropertyofs-disjunct.Byassigningsuperimposedcodetoeachmeshrouterinadvance,thecommunicationchannelcanbedeterminedbythemanipulationoftwosuperimposedcodes.However,thismethodhasarequirementforarelativelylargenumberofchannelstoavoidtheco-channelinterference.Mostoftheseheuristicworksaredistributedschemes,whichcanadaptthetrafcvariationandlinkfailure 50

PAGE 51

Someworksapplyoptimizationtechniquesinsteadtoachievetheoptimalityofthesystem[ 50 ],[ 6 ],[ 34 ],inacentralizedmanner.Both[ 6 ]and[ 34 ]useascalefactortoscaletheowrateundertheconstraintsinWMNs,whiletheyusedifferentinterferencemodelsanddifferentalgorithmstosolvetheoptimizationproblems.In[ 50 ],thetargetproblemistodirectlymaximizethesummationofutilityfunctionofrateforeachow.TheseworksaremostlytoocomplicatedinalgorithmbecauseoftheneedofdecomposingtheNP-hardoptimizationproblemandalsolackofrobustnessforsupportingarealsystem.Meanwhile,theabovealgorithmsdonotgiveustheinformationaboutthecapabilitythataWMNcanprovideitsclientstodeliverpacketsacrossthegateways. Inaddition,toachieveandretaintheoptimalityinthesecentralizedworks,WMNsarealwaysassumedtohavestationarymeshtopologyandstatictrafcdemandwhencalculatingtheoptimalvalue.Althoughgatewaysandmeshroutersaremostlystationary,duetoinevitablewirelesslinkfailures,occasionalnodefailuresandnodemaintenance,thetopologycannotbeconsideredasimmutable.Moreover,althoughmeshroutersusuallyhaveamoreconstanttrafcloadthanmeshclientsbyaggregatingmeshclients'trafc,thetrafcloadsofmeshrouterscanstillvarygreatlyfromtimetotimeduetoavarietyofgroupeventsjustlikecitytrafc.Consequently,withtheabovealgorithms,anyoftheselocallyminorchangesinthenetworkcancausethelossoffairnessandoptimality,whichrequiresfurtherglobalchannelassignmentandresourceallocation.Therefore,thevariationoftopologyandtrafcloadcannotbeignoredinWMNs'design.Thoughtherelativelystableandstationaryinformationofthetopologyandtrafccanbetheinputforcalculatingoptimalperformance,adynamicmechanismisrequiredaswelltoaddresstherobustnessissue. 51

PAGE 52

3.3.1PortalCapacity SimilartoAPsinWLANs,themaximumachievablethroughputofgatewaysinWMNsdenesthecapacityofthewholesystem.ThecalculationofaWLAN'scapacityissimplythecapacityoftheoperatingchannelduetothefactofitsone-hopnature.Inmulti-hopWMNs,thesystemcapacityisnotaconstantasitisinWLANsbecausethespectrumissharedbymulti-hoplinks.Differentspectrumresourcesharingamongmulti-hoplinkswillcausedifferentportalcapacity.Accordingtothetrafcaggregationproperty,thesystemcapacitycanbelearnedviathesummationofthecapacityoftheinterferinglasthoplinksconnectedtogateways.Withouttheconsiderationoffairness,aWMN'sportalcapacityreachesthemaximumvaluewhenthelasthoplinkstakeallthechannelresourcewhileallothermeshclientsandmeshroutersarestarved.Therefore,portalcapacityismeaningfulonlywhenfairnessconstraintisconsidered. Ifthefairnesscanonlybeachievedwheneachlink'strafcdemandissatisedwiththepredenedproportionasin[ 34 ],theretainmentofthisfairnesswillbetoodifcultbecauseeachtimewhenthetrafcofanymeshclientchanges,fairnesswill 52

PAGE 53

Althoughthismodelignoresthediversityoftrafcdemandamongtheclients,itcangiveususefulknowledgeofsystem'sperformanceandguidethedesignofthesystem.Suchinformationcannotbeobtainediftherealtimetrafcdemandisconsideredforfairness.Furthermore,wecanincorporatethediversityoftrafcdemandstatisticallybygivingdifferenttrafcweightstodifferentmeshroutersaccordingtotheirgeographiccharacteristics.Forexample,inbusyareas,suchasshoppingmalls,orimportantareas,suchashospitals,themeshrouterscanhavebiggertrafcweightsthanothers. 53

PAGE 54

3-2 .TheaggregateportalcapacitywithoutconsiderationoffairnesscanbeexpressedasPi2L1Pi.Meshrouterj'sproportionofportalcapacityvianallinkiisdenotedasuij subjectto: channelresourceandradioresourceconstraint interference(concurrenttransmission)constraint fairnessconstraint:Pi2L1uijPi=wj Figure3-2. AnExampleofLinkDistributioninWMNsandtheProposedFrequencyReusePattern 54

PAGE 55

UsuallyinWMNs,thegatewaysaretakenasthecentersandmeshrouterssurroundthegatewayswithdifferentdistances,mostofthetimecountedinthenumberofhops.Duetothetrafcaggregation,theclosertotheportalthelinksare,theheaviertrafctheycarry.ItiseasytoobservethatifavirtualcircleisdrawninaWMNandthecircleislargeenough,linksoutsidethecirclecantotallyreusetheresourceoftheinsidelinks,asshowninFig. 3-2 .Therstreasonofthefeasibilityofthisfrequencyreuseisthatoutsidelinkshavelesstrafcloadintotalbecausealltrafcneedstobeaggregatedinthegateway.ThesecondreasonisthatthedistancebetweenfrequencyreuselinkscanbeguaranteedbytheproperfrequencyreusedistanceD.Finally,becauseoflargerareaandlesstrafc,thisreusecanberepeatedwhenthedistancetothegatewayislarger.Therefore,wecanreducetheoriginalproblemtoaproblemwithsmallerareaofinterest.WetermthiscirclewithradiusDasCircleD.WecallthemeshroutersoutsideCircleDasoutsidemeshrouters. ThecriteriontodiscovertheminimumCircleDisthatallthelinksoutsidethecirclecanreusethefrequencyresourceusedinsidethecircle.Usuallythesensingrangeofanodeisabouttwiceaslongasitstransmissionrange(Simulatorns-2adoptsthetypicalvalue2.2).Therefore,inthispaper,wesetthisdistanceDequalto3-hopcoverage.ItisclearthatlinksoutsidetheCircleDarealwaysabletoreusesomefrequencyresourceusedbysomeinsidelinkswithproperassignment.Thisfrequencyreuseseparationremainsvalidnomatterwhattherealtopologyisbecauseofthereasonsmentionedabove. 55

PAGE 56

subjectto: 56

PAGE 57

Basedonthistreestructure,thelasthoplinksareawareoftheirload,thetotaltrafctheyarecarrying.Therefore,theproportionsofallthePicanbeacquired.Hence,ijcanbeknownifmulti-pathroutingwithinCircleDisignored.Withthegiventreestructureandwithoutconsideringfrequencyreuse,thesimpliedproblembecomesalinearprogramming(LP)problemwithinCircleDsincetheobjectivefunctionandconstraintsarealllinear.Iftheaccumulatedweightsofthenal-hoplinkschange,anewglobalresourceallocationandchannelassignmentisneededtoregainthefairness. Fromtheproblemformulation,wehavefoundanewwaytoachieveWMNs'optimality.AssumingthateachwirelesslinkwithinCircleDhasthesamewirelessconditionandeachwirelessrouterhasatleasttworadioswithoneforup-linkandtheotherforthedown-linkandignoringthefrequencyreuse,theaforementionedLPproblemcanbeeasilysolvedbyallocatingavailablebandwidthofgivenmultiplechannelstolinksamongmeshroutersaccordingtotheirweights.Therefore,thekeytotheoptimalportalcapacityistheresourceallocationandchannelassignmentforthelinkswithinCircleD. Forapracticalsystemscheme,allocatingchannelresourcetolinkswithinCircleDtoachieveanoptimalaggregateportalcapacityisnotenough.Firstly,eachoutside 57

PAGE 58

Finally,theachievedoptimalityrequirestheconditionthateachmeshroutergeneratesthepreplannedtrafcload.Real-timetrafcvariesfromtheexpectedvaluefromtimetotime,thusthecalculatedresourceallocationmaynotresultintheoptimalportalcapacity.Withadistributedanddynamicmechanismtoadjusttheallocatedresource,whentrafcvariestheoptimalportalcapacitycanstillbemaintained.Robustnessisalsogainedthroughthedistributedanddynamicscheme. Theoverallproposedsolutionconsistsofcentralizedpartanddistributedpart,whicharediscussednext,respectively. 3.4.1.1Overview Weproposeacentralizedschemetocarryoutthetaskofresourceallocationandchannelassignment. Asaforementioned,withtheassumptionofuniformtrafcdistributionandidenticalcoverageareaamongmeshrouters,eachmeshroutercontributesthesameamountoftrafctothegateway.Therefore,theweightofeachlinkcanbedenedasthenumberof 58

PAGE 59

IdealCaseWhenFrequencyReuseisPerfect downstreammeshrouters.Withouttheconsiderationoffrequencyreuse,theresourceallocationisstraightforwardwiththeknowledgeoftotalresourceandlinks'weights.Aweightedallocationisenoughforthistask.Afterwardsweneedtofacethechallengesfromfrequencyreuse,exacttotalresourceanddeviationofresourceallocation. Somenotationsareintroducedbeforehand.ThechannelcapacityisrstlyassumedasaconstantBandthereareKavailablechannels.Wedenotethetotalweightofeachtieras!I,!II,!III,respectively.Thetotalweightis!total.ThenumbersofeachtiermeshroutersaredenotedasMI,MIIandMIII,respectively.ThenumberofallmeshroutersisdenotedasM. Conservatively,whenthereisnofrequencyreusewithinCircleD,alowerboundofportalcapacitycanbeachievedbydividingthewholespectrumbythewholeweightsofthelinks.ThelowerboundisKBM M+(MMI)+(MMIMII)KB Anupperboundisachievedwhenthe3rd-hoplinkscanallreusethefrequencyallocatedtotheprecedinglinks,asshowninFig. 3-3 .Toachievethisupperbound,thenetworktopologyandlinks'weightsshouldberoughlysymmetrictothegateway.ThemostidealaggregateportalcapacityisKBM M+(MMI)KB 59

PAGE 60

3 )isknowntovaryfromKB ForanIEEE802.11channel,themaximalthroughputBthatcanbeachievedisnotaconstant.Itdependsonthenumberofcontendingnodes,i.e.,thecollisionprobability.Ifthechannel'smaximalthroughputisnotthesamefordifferentlinks,theweightedresourceallocationisnotaccurate,whichmayleadtounder-utilizationofportalcapacityornon-optimalportalcapacity.Therealchannelcapacitywithdifferentnumberofcontendingnodesshouldbeguredoutandthecapacitydifferenceneedstobeconsideredintheresourceallocation.Arevisedweightedallocationisusedforthispurpose. Werstpresentthealgorithmforthecentralizedresourceallocationandchannelassignmentwiththeconsiderationoffrequencyreuse. Themostcommonwaytoallocateresourceistheweightedallocation.Ri=!iRtotal Basedontherulethatfrequencycannotbereusedbytwo-hopneighbors[ 50 ],wecanderiveanothertworules:frequencyofrsttierlinkscannotbereusedby3rdtierlinks,norcanthe2ndtierbereusedby2ndtier.Therefore,itisclearthatthefrequencyreuseforthelinkswithin3hopsisonlypossiblebetweenfaraway2ndand3rdtierlinks 60

PAGE 61

Afterallocatingtheproportionalresourcetoeachlink,thegatewayisrequiredtoassignthechannelstotheradiosofeachmeshrouteraccordingtotheallocatedresourcetoeachlink.Becausethenumberofchannelisaninteger,thechannelassignmentcanmakeresourceallocationdeviatefromthetargetproportionduetoroundingoperations.Attestallocationstrategyisusedtomitigatethisproblem.Thefrequency-reusinglinksdonotneedtobeassignedthechannelresourcebecauseofpreviousassignmentforthefrequency-reusedlinks. Duetothelimitednumberofchannels,mostlinksinWMNshavetoshareachannelwithotherneighboringlinks.Thisfactimposesextradifcultyonthefrequencyreuse.Underacontention-basedprotocol,ifthefrequencyreusedlinkssharethechannelwithotherlinks,thefrequencyreuseisnotalwaysfeasible.Anexampleisthatchannelassignmentmakesarst-tierlinkshareachannelwithitsdownstream3rd-tierlink.Inthiscase,itsdownstream4th-tierlinkscannotreusethefrequencyresourceofthisrst-tierlinkbecauseitwillbringundesiredinterferencebetweenthe3rd-tierlinkandthe4th-tierlinks. Whenthefrequencyreusecannotbefullledunderacertainchannelassignment,theresourceallocationisrequiredtore-calculateduetothechangeoftotalweight!total.Therefore,theresourceallocationneedstobeconsideredwithchannelassignmentjointlyduetotheuncertaintyoffrequencyreuse.Ourstrategyistocomparethegainfromfrequencyreuseandthecosttoexcludeotherlinksfromthefrequencyreusedchannel. Resourceallocationisrstlydonewiththeknowledgeoftotalweight!totalandtotalresourceRtotal.Duringthechannelassignment,whenafrequencyreusedlinkismet,weattempttoexcludeallthelinksthatcannotbefrequency-reusedfromthecurrent 61

PAGE 62

AlgorithmofCentralizedResourceAllocationandChannelAssignment 1ResourceAllocation(!total) 2ChannelAssignmentstart 3while(END!=LINK) 4ifLINK==REUSE 5Findthefrequencyreuselinkset 6ifgain>price 7!total=!l0price 9ResourceAllocation(!total) 10continue 11else 12AbandonFrequencyReuse 13Clearfrequencyreusemark 14Waitforchannelassignment 15ifRESOURCE>1 16checkthenumberofradio 17andsplitRESOURCEtoradios 18ifRESOURCE>radios 19errorreport 20searchthettestCHANNEL 21if(CHANNEL==UPLINKCHANNEL) 22searchagain 23LINK=LINK!next 24end 25ChannelAssignmentend 62

PAGE 63

3-1 isthepseudocodeofcentralizedresourceallocationandchannelassignmentwiththeconsiderationoffrequencyreuse. AsshowninLine21,animportantrulefortheassignmentisthatthechannelfrequencyfortheup-linkanddown-linkofeachmeshroutershouldbedifferent.TheconstraintoftheradioresourceisalsoconsideredinLine15.Whentherearemorethanoneradiofordown-linkorup-link,thestrategyistospreadtheallocatedresourceevenlytoeachradio.Thisstrategyconsiderstheprospectivedynamicaladjustmentsothatwhenunusedresourcecanbereallocatedanextrachannelassignmentisnotnecessary.Thechannelassignmentshouldbecontinuous,whichmeansthesamechannelispreferredtobeallocatedtoacloseneighborhood.Thepurposeforthisstrategyistoavoidthehiddenterminalproblemaswellastoenhancethefrequencyreuseprobability. 63

PAGE 64

71 ],thisdifferencecanbeupto10%20%.Ifweusethenominalchannelcapacityastheresourceforallocation,thecentrallinks'sallocatedresourcewillgreatlyexceedtheirtargetproportion.Eventuallytheoptimalitywillbelost.Toexploittherealisticoptimalportalcapacity,thersttaskistondouttheactualresourceforallocation. ThesaturationthroughputofIEEE802.11channelshasbeengivenoutin[ 10 ].However,thisthroughputcannotbeusedtoindicatethechildnodes'trafcrate.Thesaturationthroughputmeanstherealdeliveryratein802.11systemswheneachnodehaspacketstosendallthetime.Thetrafcrate(arrivalrate)ofthesendersisusuallymuchlargerthanthedeliveryrateinthiscase.Wedonotwanttousetoobigarrivalratetoachievethetargetdeliveryrateundersaturationbecauseanoverstockedsystemisnotpreferred.Amappingfromarrivalratetodeliveryrateundernon-saturationisrequiredforourpurpose. Foranalysispurpose,weassumethatthearrivalofpacketsforonemeshrouterismodeledasaPoissonprocess.ThepacketdeliveryofastationcanbemodeledasaqueuesystemandtheMACservicetimeistheservicetimeofthisM/G/1/Ksystemwhensystemisnon-saturated.Whenthearrivalrateisgreaterthantheservicerate,thissystemisoverstocked,i.e.,non-ergodic.Wecanndthebiggestnon-overstockedarrivalrateinordertogettheknowledgeofachievabledeliveryrate,whichisthevalueusedto 64

PAGE 65

Wedenotethenumberofcontendersinonechannelasncandeachcontenderisassumedtohavethesamepacketarrivalratec.Accordingtotheresultof[ 71 ],theMACserviceratecanbeacquiredastheinverseoftheaverageservicetime1 Tomakethequeuesystemnon-overstocked,thearrivalrateshouldbelessthantheservicerate. Themaximumarrivalratecanbeachievedwhentheequationholds:c=1 wherepBisthepackets'droppingprobabilityduetonitequeuelength,pcispacketdiscardprobabilityinonetransmissionand+1istheretransmissiontime.Thederivationcanalsobefoundin[ 71 ]. However,wheneachcontenderhasdifferentarrivalrate,thesevaluesmaychange. Werewritetherelationshipbetweenpcandtheconditionaltransmissionprobabilityas wherep0(i)istheprobabilitythatthewirelessstationihasnopackettotransmit,whichcanbederivedfromc(i).Meanwhile,fromtheMarkovchainmodel,theconditionaltransmissionprobabilitycanbederivedas 1pc(i)+1+(1pc(i))WPk=0(2pc(i))k,m2(1pc(i)+1) 1pc(i)+1+pc(i)WPm1k=0(2pc(i))k+W(12mpc(i)+1),>mwheremisthemaximumnumberofthestagesallowedintheback-offprocedure. 65

PAGE 66

Whenc(i)areprovidedpc(i)canbeguredout.AverageMACservicetimeforeachstationcanalsobecalculated.Thetotalthroughputisthesummationofeachstation'sthroughputwhichcanbederivedfromc(i)andpc(i). AnextremecaseofdifferentarrivalrateisthatamongLcontendersonlyonehasintensetrafcandtheotherstransmitssparsely.Themaximalthroughputinthiscaseapproximatestheoneincaseofncequalsto1. Thistime-varyingchannelcapacityaddsmoredifcultyonthecentralizedresourceallocationtask.However,itisshownthattheunequaltrafcpatternleadsthemaximalthroughputtoapproximatingthevaluewithfewercontenders.Therefore,theequaltrafccaseisreasonedtohavethemostconservativechannelcapacityandthisvalueisusedastheinputofourresourceallocationscheme. Thederivationofeachtier'saveragenumberofcontendingnodes(CI,CIIandCIII)canbeobtainedeasilyasfollows. Throughtheweightedallocationprocess,theallocatedresourcefor3tiersisI!I 66

PAGE 67

Thesolutiontocorrectthisdeviationistouse1 Thetotalallocatedresourceisthen Throughthiscorrection,thetargetproportionofallocationcanbeattained. 3.4.2.1Overview Thereareacoupleofreasonsthatthedistributedmechanismispreferredinadditiontothecentralizedpart.Firstreasonisthatadistributedschemecantthedynamicenvironmentbetterthanawholecentralizedone.Foracentralizedscheme,wheneverthisnetworkischangedeitherintopologyortrafcdynamic,anupdatingofchannelassignmentisneeded.Adistributedschemecanupdatethechannelassignmentinresponsetolocalminorchanges.Thesecondreasonisthedifcultyoffrequencyreusedecisioninmulti-hopscenarios.Forcentralizedapproach,todecide 67

PAGE 68

Wehavealreadydemonstratedthatwhenthecentrallinkshavebeenallocatedthechannelresourceappropriately,theoutsidetiersshouldhaveenoughchannelresourcetorealizetheirshareofportalcapacity.Givenameshrouter'sportalcapacity,thetaskofitsdistributedchannelassignmentistoallocateitschildrentheirsharesofportalcapacityandassignthemchannels.Thedistributedalgorithmcanbringthenetworktheadaptivitytopossibleminorlocalchangesintrafcand/ortopology,whichshouldnotcauseanupdateacrossthewholenetwork.Thedifcultiescomefromtheacquirementofstatusinformationofallthechannelsandthecoordinationofchannelassignmentforneighboringmeshrouters. Inaddition,adynamicmechanismisrequiredtoadjustthechannelresourceamongdifferentlinksaftereachofthemeshroutersobtainsitsshareofportalcapacityandtheassignedchannels.Moreover,thisadjustmentcanhelpthemeshroutersrecoverfromthelinkfailuresandlinkdisadvantages. Sincetheportalcapacityofeachmeshrouterisalwayspasseddownfromitsparent,thedistributedchannelassignmentisdeterminedtofollowatop-downsequence.Thismeansthattheoutsidemeshrouterscanexecutethedistributedchannel 68

PAGE 69

Thecentralizedresourceallocationisembeddedintheprocedureoftheglobalchannelassignment.Eachmeshrouterwaitsforitsparenttoallocatetheshareofportalcapacityandassignthechannelresourceforit.Acertainchannelisdenedasthestartingchannel.Beforechannelassignment,allmeshroutersresideonthestartingchannelwaitingforthechannelassignmentfromtheirparentsorthegateway. Themajordifcultyforachannelassignmentschemewithfewerradiosthanchannelsistheacquirementofstatusinformationofallchannels.PreviousworksuseNAV(networkallocationvector)knowledgecollectedfromeachofthechannelstoindicatethechannelusagestatusforMANETs(MobileAd-hocnetworks)[ 64 ].However,per-packet-basedchannelassignmentdoesnottWMNsbecauseoftherelativelystationarytopologyandrelativelyconstantowinthebackbonelinks.Instead,thechannelassignmentinWMNsisalmoststaticexceptthatthecurrentrequirementcannotbesatised.Fortheabovereason,theNAVknowledgeisnotnecessaryandalong-termindicatorofthechannelbusynesslevelismorepreferredinthechannelselectionforWMNrouters.WespecifyacertaindurationatthebeginningofeveryTbeaconsonthestartingchannelasthechannelbusynessindicatorphase(CBIP),asshowninFig. 3-4 .TheCBIPisdividedintoKpartswhereKisthenumberofavailablechannels.Eachmeshroutersetsbusytonesonthecorrespondingpartstosignalthebusynessstatusofitsoperatingchannels,withthedurationlengthasthebusynesslevel.Eachmeshrouterchoosesthemostidleandnon-conictingchannelforthelinkstoitschildren.Thebusytone'slengthisalwaysdeterminedbytheneighborwhichsensesthebusiestchannelstatusofthischannelbecausetheresultoftheadditionofbusytonesistheonewiththelongestduration.Whenameshroutersensesthatsomechannel'sbusytoneisnotset,itcantellthatthereisnoneighborusingthischannel. 69

PAGE 70

IllustrationofChannelBusynessIndicatorPhase(CBIP) Theindicatorofthechannelbusynesslevelisdeterminedbythefollowingformula: whereListhenumberofmini-slotsusedinCBIPforonechannel,TbusydenotesthebusytimedurationofthegivenchannelandTtotaldenotesthewholetimeduration.ThegranularityofthisindicatorisdecidedbyL,whichcannotbetoolarge.However,inaWMN,theoutsidemeshroutersgreatlyunder-utilizethechannelsbecausetheyreusethefrequencyresourceinagreatlylargerareawithsignicantlylesstrafcdemand.Therefore,Ldoesnotneedtobelarge.ForthemeshrouterswithinCircleD,althoughthisindicatorisnotusefulbecauseeverychannelisfullyallocated,thismechanismisalsoincorporatedinthemeshrouterswithinCircleDbecauseoftheneighborhoodrequirementfromoutsidemeshroutersandtherequirementofagloballyuniformprotocol.Consequently,thecentralizedresourceallocationisrequiredtosubtractthispartofresource. Thealgorithmforthedistributedchannelassignmentcanbedescribedasfollows. Afterameshrouter(3hopsorfurther)getsitsportalcapacityandchannelassignmentfromitsparent(orthegateway),itcalculatestheportalcapacitythateachofitschildrencangetafteritsubtractsitsownusage.ItcollectsthebusynessstatusofeachchannelviaCBIPanddecidesthechannelassignmentforitschildrenafterward.Forthecaseofmorethanoneradiofordown-link,themeshrouterspreadstheportal 70

PAGE 71

Whentheneighboringmeshrouterscarryoutthechannelassignmentatthesametime,itispossiblethatacertainchannelisassignedbyalotofmeshroutersandotherchannelsarespared.Inouralgorithm,eachmeshrouterwaitsforallthehighertierrouterstonishtheirchannelassignmentbeforeitstartsitsown.Amongthemeshrouterswiththesametiernumber,themeshroutercarryingthelargestweightstartsrst. TheoverheadofthismechanismistheCBIPandoneextraradioforthebusytoneinthestartingchannel. Ifameshrouter'sshareofportalcapacitycannotbeusedup,itsparentcanletitssiblingstosharethisunusedresource.WithinCircleD,theunusedportalcapacitycanonlybesharedbylinksusingthesamechannelbecausethechannelresourceisalreadyfullyallocated.Outsidemeshroutershavenosuchlimitaslongastheyhavevacantchannelresourcetofullltheunusedportalcapacity. Thisresourceusagechangeistemporarilycausedbytrafcimbalance.Thistypeofchangehasnoimpactontherelationshipamongmeshroutersandistherebydenedasnon-infrastructure-involvedchange.Italsoincludesthecasethatameshrouterchangestoanotherchannelwhilekeepingitsshareofportalcapacity.Thetriggeringconditionofthischangecanbelinkdisadvantagecausedbyhiddenterminalproblem,orpoorlinkqualityduetofrequency-selectivefadingandinterference. Wecanshowthatwithonlynon-infrastructure-involvedchange,theoptimalityofcurrentassignmentisnotchangedassumingthatallthelinkstatusisidentical.The 71

PAGE 72

Ameshroutercanmonitorothermeshroutersofhighertiers.Itcanchangeitsparentwhenitsportalcapacitycannotbefullledviacurrentparentandthereisanalternateparentwhocanprovideacertainbettershareofportalcapacity.Thischangeinvolvesthechangeoftherelationshipofparent-and-child.Itbelongstotheinfrastructure-involvedchange.Thecaseofanewrouter'sjoiningthesystemandthecaseofarouter'sdetachingitsparentarealsoincludedinthistype.Thedynamicadjustmentforinfrastructure-involvedchangemakesthesystemself-recoverableandmightcausesomeunfairnessinreturn.However,unfairnessisthepricefortherobustness.Anotherglobalchannelassignmentcanregainthefairnessofthesystem. Forthesakeofdynamicadjustment,ameshroutershouldbroadcastitsportalcapacityandcurrentusageratio.Ameshrouterisrequiredaswelltobroadcastitstiernumber,weightandthenumberofdownstreammeshrouters. 72

PAGE 73

Unlikeothercentralizedapproaches,oursolutionrequirescentralizedresourceallocationandchannelassignmentwithinCircleD.Duetothesmallercentralizedarea,supportingmultiplegatewayswithinoneWMNbecomespossiblewithoursolution.Whenthereareseveralgatewaysintheeld,aslongastheCircleDofeachrouterisnotoverlapped,linksinsideeachcirclecanachievetheirportalcapacityfromthegatewaysandlinksoutsidecanacquireportalcapacityfromanygatewaydistributedly.Toguaranteethechannelassignmentofeachcirclehavingnointerferencewitheachother,theseparationdistancebetweeneachcircleshouldbeatleasttwohopsaway.Althoughtherearemoreissuestobeconsideredwhenapplyoursolutiontomulti-gatewayWMNs,oursolutionprovidesagoodoptionwhichmanypreviousworksdonotsupportatall. WithmultiplegatewaysconnectedtotheInternet,aWMNcanprovidebiggerportalcapacityandlargercoveragearea. 73

PAGE 74

ComparisonofPortalCapacityamong3Schemes:CW(oursolution),SC(singlechannel),RM(randomly-choosingmulti-channel). channelassignment(CW)tocomparetheportalcapacityperformancewiththesetwoschemes,SCandRM.Forthepurposeofcomparison,allthreeschemesareappliedtotopologieslimitedinCircleD.Wealsoassumethatthereisnofrequencyreuseinallthreeschemesandthereisnogeographicaldisadvantageamongalllinks.Forbothcomparedschemes,SCandRM,portalcapacitycomesfromtheaggregationofthemaximumthroughputoverthelasthoplinks.Inbothdistributedschemes,sinceeachnodecontendsforthechannelresourcewithoutconsideringtheportalcapacityshareofitsdownstreamnodes,thefairnesshasnoguarantee.Thusafairnessmetricisincludedinthesimulationaswell. ThissimulationsetK,thenumberofavailablechannelsto12foroursolutionandthedistributedmulti-channelscheme;I,II,IIIto1.0,0.8,0.6,respectively.ThechannelbandwidthBisignoredinthesimulationsinceitisaconstant.Therefore,theallocatedresourceandportalcapacitycanbequantiedbyvaluesrangingfrom0to12.Wecomparethevalueoftherealizedportalcapacityamongourcentralizedweightedallocationapproach(CW),andthecomparedschemes,SCandRM. Fig. 3-5 showsthatinSCcase,onlyasmallproportionoftotalchannelresourceistransferredtoportalcapacity,orend-to-endthroughput.RMschemecangetabetter 74

PAGE 75

Asmentionedabove,eachmeshrouterisdesignedtobeabletodeliverthesameamounttrafctothegatewaysgeneratedbyitself.Ifeachlinkdoesnotacquiretheallocatedresourceaccordingtothetargetproportionofweight,theportalcapacitycannotbefairlysharedbyallmeshrouters.Therefore,weuseXi,theratiobetweentheallocatedresourceandthecorrespondingweighttocalculatethefairnessindex. Iftheresourceallocatedtoeachlinkmatchesthecorrespondingweight,withtheexistenceofadmissioncontrolandmessageexchangebetweenmeshrouters,thefairshareofportalcapacitycanbeeasilyreached.Withthepropersettingoftheweightforeachlink,oursolutionachievesthebestfairnessasshowninFig. 3-5 Ifthethreeschemesareappliedtolargertopologies,duetoless-plannedresourceallocation,SCandRMwillhavemuchpoorerperformancethanoursolution. Thesecondpartofoursimulationisfocusedonthefrequencyreuseissue.Frequencyreusecanbringmoreactualchannelresourceandtheimprovedportalcapacity.Thispartofthesimulationshowstheportalcapacityimprovementowedtofrequencyreuseinoursolution.Inoursolution,centralizedresourceallocationonlyconsidersfrequencyreusewithinCircleDbecauseoutsidemeshrouterscanreusetheresourceinsidethecircle.Therefore,frequencyreuseinsideCircleDdeterminestheoptimumfrequencyreuseofthewholesystem,whichcanbeeasilyguredoutthroughtheeldmeasurement. Wetesttheportalcapacitywithidealfrequencyreuseusingouralgorithmunderthescenariosofdifferentnumbersofoutsidemeshrouters.Forcomparison,theportalcapacitywithoutfrequencyreuseisalsoprovided. 75

PAGE 76

PortalCapacitywithFrequencyReuse Fig. 3-6 showstheportalcapacityimprovementusingidealfrequencyreuse.FromtherstpartofFig. 3-6 wecanseeasignicantincrease(roughly40%)oftheproportionoftheportalcapacity,whichmeanstheratiooftherealportalcapacityandtherealallocatedchannelresourcenotincludingtheoverheadandtheunallocatedchannelresource.Whenthenumberofoutsidemeshrouterschanges,thereisaslightdecreaseofbothportalcapacityproportionandrealportalcapacity.Thereasonisthatwhenthenumberofmeshroutersincreases,theweightsofsecond-andthird-tierlinksincreases.Therefore,morechannelresourcewillsufferfromsmaller. 76

PAGE 77

22 ].However,forpracticalmulti-hopwirelessnetworks,suchasWMNs,WSNsandsomebattle-eldadhocnetworks,thereisplentyofroomtoimprovethethroughputperformancesincethecurrentpoorthroughputperformanceofthesenetworksismainlyduetotheinefciencyofspectrumusage.Mostofrecentresearchworksarefocusingonthespectrumefciencyforthisreason. Asthedefactostandardofmostofthemulti-hopwirelessnetworks,IEEE802.11wasoriginallydesignedforthesingle-hopWirelessLANs.Itsperformanceinmulti-hopscenariosismuchbelowourexpectationduetoitsblindly-contendingandmechanicallyforwardingproperties[ 69 ][ 72 ].ThisrandomaccesspropertyofIEEE802.11isoneofthemajorreasonsfortheinefcientresourceusage.Scheduling-basedprotocols,likeIEEE802.16,canprovidebetterspectrumefciencybecauseitdoesnotrequirenodestocontendforthechannelbeforeeachtransmissionwiththeassumptionofrelativelyconstanttrafcows.However,sincethistypeofprotocolsneedsaxedframeformatwhichisvulnerabletotheschedulingconict,itisnotsuitableforthemulti-hopwirelessnetworks. RandomaccessMACprovidesaroughlyfairmechanismforwirelessnodestoaccessthemedium.TheeffortofdifferentiatingtheuplinkanddownlinkresourceallocationhasbeenrstappliedtoWLANsin[ 33 ]becauseoftheobservationthatas 77

PAGE 78

70 ],[ 66 ],[ 31 ].Theseschemesheuristicallysearchforbetterspectrumsharingmechanismamongwirelessnodes,bydifferentiatingtheforwardingpriorityaccordingtotheprioritytagsofpacketsorows.However,whenthetrafcpatternismorecomplicated,theseschemescannotguaranteesignicantperformanceimprovement. Ontheotherside,withcentralizedapproaches,scheduling-basedMACcanallocatetheresourceinamoreefcientway.Thisapproachcanndtheoptimalsolutionwithknowledgeofthetopologyandtrafcwhenthenetworkisnotlarge.However,forlarge-scalenetworks,thisapproachbecomesinfeasibleduetotheNP-hardnessandthedifcultyofinformationcollection.Therefore,thedistributedschedulingapproachesareproposedtoaddressthisdilemma,[ 40 ],[ 47 ].Thesetwoapproachesgiveustheinsightofhowgoodperformancethenetworkscanachieve.However,theyalwaysrequireaperfectscheduling,aMACwithnocollisionandnohidden/exposedterminals,whichisalmostimpossibleinmulti-hopwirelessnetworks.Previousdistributedschedulingschemesalsoignorethemulti-hopnatureofowsinmulti-hopnetworks,whichcausesalotofwastageinspectrumallocation.Moreover,withinaneighborhoodrange,differentneighborscansensedifferentconditionofchannels,resultinginpotentialconictsofdistributedscheduling,whichisdifculttosolveindistributedschedulingschemes. Inthispaper,weproposeanewMACwithadifferentspectrumallocationmechanismbasedonIEEE802.11.Similartopreviousdistributedapproaches,theefcientspectrumusageofthisschemecomesfromthecollectionofneighbors'trafcinformation.Differentfrompreviousworks,ourschemecollectstrafcdependencyinformationfromneighborsaswell.Byaddressingasymmetricneighborhoodandtrafcdependencyissues,thispapergivesacomprehensivewaytoimprovethethroughputperformanceinmulti-hopwirelessnetworks.Upontheobservationofefciencyofspectrumallocation,a 78

PAGE 79

Therestofthispaperisorganizedasfollows.Section 6.2 introducestherelatedworks.Section 4.3 discussesthespectrumusageissuesinmulti-hopwirelessnetworks.Section 4.4 describestheproposedschemes.Section 4.5 providestheevaluationstudyforourscheme.Finally,conclusionisdrawninSection 4.6 In[ 70 ],forwardedpacketsaregivenhigherprioritythanupstreampacketsbyashorterIFS,thusthepacketaccumulationattheforwardingnodesisalleviatedandthedeliveryfailureduetoforwardingcongestioncanbereduced.YangandVaidyatriedtoensuremediumaccessforhighprioritysourcestationsintheirpriorityschedulingscheme[ 66 ].Twonarrowbandbusy-tonesignalsareusedtoensuremediumaccessforhighprioritynodes.Kanodiaetal.furtherproposedadistributedpriorityschemetodifferentiatethedifferentpackets'priorityintransmission[ 31 ].Accordingly,theforwardedpacketshavetheirpriorityincreasedforthesamereasonasaddressedin[ 70 ].Alltheaboveworksattempttoimprovetheperformancebyintroducingdifferentiationamongdifferentnodes,differentpackets,ordifferentows.However,suchdifferentiationscannotimprovetheoverallperformanceofthesystemremarkablywhenthetrafcandtopologybecomemorecomplicatedorwhenowscannotbeprioritized.IFAscheme 79

PAGE 80

18 ]basedontherandomaccessMAC.Theneighbors'trafcinformationarecollectedandthespectrumresourceisallocatedaccordingly.Bythrottlinginputtrafctoitssystemwidefairtimeshare,randomaccessMACcangreatlyimprovethefairnessandthroughputperformance.However,thetrafcdemandofeachnodesimplyusesthevalueofarrivedtrafcandonlyparkinglottopologiesareanalyzed. Thedistributedschedulingapproachfocusesontheoverallthroughputperformanceanalyticallysincetheyarederivedfromcentralizedschedulingalgorithms.Theschemein[ 29 ]requireseachnodetocollecteachneighbor'squeue-lengthinformationanduseaprobabilitytotransmititspackets.Theprobabilityisderivedfromtherelationshipofeachnode'squeue-lengthwithitsneighbors'.Thispaperprovesthatwhenthispolicyofdistributedschedulingisapplied,thelargestcapacityregioncanbeachievedwithanefciencyratio.Lin'sworkhastakenmulti-channelandroutingintoconsiderationbesideseachneighbors'queue-length[ 40 ].Accordingtothecalculatedcontentioncost,radiocostandcongestionlevel,whicharederivedfromneighbors'queue-length,channelconditionandnodes'otherinformation,eachnodedecidestheassignmentofpacketstodifferentchannelsanddifferentslots.Thispaperalsoshowsthattheprovideddistributedalgorithmisprovablyefcient,whichmeansthataprovablefractionofthemaximumsystemcapacitycanbeachieved. Althoughtheseworksprovidetheoreticalresultsofoverallthroughputperformance,theassumptionsofaperfectMACandthelackofarealprotocolsupportpreventthemfromtransitingthetheoreticalresultstoapracticalprotocol.Moreover,whenallqueuesarefullandthetrafcisbackloggedduetoover-injectionoftrafc,thesedistributedschedulingalgorithmscannotaddressthecongestionproblem. Furthermore,theanalysisoftheseworksdonotconsiderthemulti-hopnatureofows.Theyconsiderthetrafcload(queuelength)independentlywhiletheinherenttrafccorrelationamongneighboringnodesisignored.Similarto[ 18 ],theytakethe 80

PAGE 81

Nomatterwhichapproachisconcerned,thepurposeistoefcientlyutilizethespectrumresource.Thereasonofthecollectionofneighbors'informationistoleteachnodeefcientlysharethespectrumresource.Formulti-hopwirelessnetworks,onlyneighbors'trafcloadinformationisnotenoughfortheidealspectrumusageallocation. 4.3.1IdealSpectrumUsage Figure4-1. InefciencyofFairRandomAccess 4-1 shows3owsinasimple4-node2-hoptopology.Flow1and2sharethesameforwardingnode,Node3.WithIEEE802.11protocols,Node3canshare1=3ofthechannelcapacitylikeNode1andNode2,withNode4onlyreceiving.Inthiscase,apartofthechannelcapacitythatNode1andNode2holdiswastedbecauseofNode3'sdeliverylimit.Consequently,wecanexpectthatwhenthetrafcconsistsofstableandcontinuousows,adjustingeachnode'mediumaccesscanbringsignicantimprovementofthethroughputperformance. Usuallytheperformanceofmulti-hopwirelessnetworkscanbeevaluatedbytwometrics:fairnessandend-to-endthroughput.SupposetheowdemandvectorisfF1,F2,F3,...,FNg,whereNdenotesthenumberofexistingows.Withacertain 81

PAGE 82

18 ].However,thisdoesnotguaranteeanoptimumallocationsincetheaggregateend-to-endthroughputisnotconsidered.Usingmax-mindeliveryratioasthecriteriaasin[ 15 ]includestheend-to-endthroughputintoconsideration.Inthispaper,thedeliveryratioisdenedasfollows. Theobjectiveis: LetusconsiderFig. 4-1 again.WhenlegacyIEEE802.11isapplied,eachactivenodeobtainsthesameshareofwirelesschannel.Theoverallend-to-endthroughputofthissmallnetworkisroughly1=3,witheachowtaking1=9equally.Whenschedulingisusedandnofairnessisconcerned,themaximumoverallend-to-endthroughputcanreach1whenFlow3takesallthespectrumresource.However,theminimumowrateis0,withow1and2totallystarved.Ifweapplythemax-mincriteriontothissimpletopology,wecanobtaintheidealspectrumusageallocationsimplybyobservation.Themaximumoverallend-to-endthroughputis3=5,witheachowtaking1=5andtheremaining2=5takenbythersthopofFlow1andFlow2.Eachowobtainsthesameowrate1=5,thusthemax-mingoalisachieved. Forcomplicatedtopologies,itisnoteasytondtheidealallocationbysimpleobservationsincemorecomplicatedinterferenceandfrequencyreuseneedtobeconsidered.Whentheknowledgeofthewholetopologyandowinformationisgiven,therearemanyalgorithmstondthisidealallocationwhileNP-hardnessisabigobstacleforthesolutions.Thedistributedschedulingproposestoutilizethequeue-lengthortrafcloadinformationofneighboringnodesastheinputofdistributedschedulingalgorithm[ 40 ],[ 47 ],[ 29 ].Thedistributedschedulingisinnatureanother 82

PAGE 83

However,mostofthedistributedalgorithmshaveahiddenassumptionthateachnode'strafcloadisindependent,whichdoesnotholdwhenmulti-hopowsexist. 40 ],[ 47 ],[ 18 ],[ 29 ],thetrafcdemandofeachnodeisassumedtobethelocaltrafcinput.However,thistrafcinputdependsonnotonlythearrivaltrafcfromlocalupperlayerentities,butalsotheforwardingrequirementfromtheupstreamnodes,whichdependsoncurrentschedulingorspectrumallocation.Therefore,usingthelocaltrafcastheinputofthealgorithmcanmakethespectrumallocationdeviatedfromtheidealone.Previouspapersignorethistrafcrelationshipintheiralgorithms,whichwetermitastrafcdependency. WeuseFig. 4-1 astheexampletoillustratetheproblemofignoringtrafcdependency.SupposeNode1,2and3havethesameoriginaltrafcload,valuedas1,andNode3needstoforwardthetrafcfrombothNode1and2.Supposeeachnodecollectsonlythearrivaltrafcinformationofeachneighborandallocatesthespectrumaccordingly.Theinitialarrivaltrafcrateofeachnodeisassumedtobetheresultofrandomaccess.Theresultturnsouttobefardeviatedfromtheidealresult,witheachofNode1and2occupying3=11ofthechannelcapacityandNode3occupies5=11.WecanseethatNode3'sallocatedresourcecannotevencoveritsforwardingrequirement,andthusabigwastageiscreated.Thisallocationdeviationcomesfromtheignoranceoftrafcdependencyofneighboringnodes. Trafcdependencycomesfromows'multi-hopdeliveryproperty.Lihasraisedsimilarconcernsofthisinherentcorrelationofupstreamanddownstreamsub-owsin[ 39 ].Forcentralizedalgorithms,thecentralpointisassumedtohaveallthe 83

PAGE 84

whereNisthenodeset,Tiistheoutowrateofnodei,Fistheowset,Kjisthehopsetofowj,Te2ejistheend-to-endrateofowj,andPHYj,kisthephysicallayerrateofowjathopk. Apparently,thismetricgivesustheknowledgeofhowmuchallocatedspectrumiswasted.Thiswastagemainlycomesfromallocationdiscrepancyatdifferenthopsforoneow.AgoodspectrumallocationschemeshouldhaveanAIRwiththevalueof0.WecanverifythisclaimaccordingtotheAIRcalculationinaboveexamples.TheAIRvalueintheschemesin[ 47 ],[ 29 ],[ 40 ],[ 18 ],whichignoretrafcdependency,is8=33.Thismeans8=33oftheallocatedspectrumiswastedbecausethecorrespondingtrafcamountoriginallytransmittedisnotnallydelivered. 84

PAGE 85

4.4.1OverviewofTheProposedScheme Thebasicprocedurecanbedescribedasfollows.Eachnodeisrequiredtobroadcastitstrafcdemandtoitsneighbors,whichisthesameaspreviousworks.Meanwhile,eachnodeisrequiredtonotifyitsneighborsaboutthetrafcdependencybetweenthem,whichdifferentiatesourworkfromothers.Afterwards,eachnodeallocatesthespectrumindividuallyaccordingtotheinformationcollectedandapplythecalculatedtrafcratetoitstransmission. Trafcdependencyinformationfromdifferentneighborsaffectstheestimationofaccuratetrafcloadindifferentways.Threedifferentrolesofneighborsaredenedinthispaper.Whenoneneighborhastrafcforthecurrentnodetoforward,wenamethis 85

PAGE 86

Thetrafcdemandfromeachnodeconsistsoftwoparts:thetrafcthatrequirestobeforwardedfromitsupstreamneighborsandthetrafcoriginatedfromitsupperlayerlocally.Itcanbeexpressedbythefollowingformula: whereTDiisthetrafcdemand,TDOiisthetrafcoriginatedfromlocalupperlayer,TDfwdingj,iisthetrafcthatrequirestobeforwardedfromitsupstreamneighborjandNiisthesetofneighborsfornodei.Thelatterpartisdependentwithitsneighbors'trafcdemandandtheformerpartisnot.Therefore,anaccuratetrafcdemandofonenodeshouldbebasedontheknowledgeofallupstreamneighbors'trafcdependencyinformation.Inthisscheme,upstreamnodesshouldnotifytheirdownstreamneighborsabouttheirforwardingrequest.Consequently,thedownstreamnodesupdatetheirtrafcdemandaccordingly.Theknowledgeofaccuratelocaltrafcdemandisnotenoughforidealspectrumallocation.Itisalsoimportanttoacquirethecorrecttrafcdemandoftheneighbors,TDi.Whendownstreamnodesbroadcasttheirnewtrafcdemands,sincethedownstreamneighbors'trafcincludesforwardingrequirementfromtheupstreamnodes,theupstreamnodesshouldbeabletoextractthedependenttrafcfromthemessages,thusthepurechangeoftheoriginaltrafcofthedownstreamnodescanbeknown.Thisknowledgeisimportantinobtainingtheaccuratetrafcdemandofneighbors.Sinceeachnodeconsidersbothone-hopupstreamneighbors'andone-hop 86

PAGE 87

Itisobviousthatduetocomplicatedtrafcpatterns,eachneighborcanplaydifferentrolessimultaneously.However,wekeepthesetermsinthissectiontoillustratetheschemeclearly. Inthisscheme,onenode'strafcchangewillaffectthetrafcdemandofcorrelatedneighbors.Thechangeofonenode'strafcdemandispassedtoothernodesasifthereisnochannellimit.Inthisway,thetrafcdemandinthisschemecanreectthetruetrafcdemandoftheneighborhood.Therefore,thespectrumallocatedtoonepacketinthecurrenthopwillalsobeallocatedtothispacketinotherhops,andthusthebandwidthwasteduetoallocationdiscrepancyisreducedfromthebeginningoftheallocationphase.Inthespectrumallocationphase,thechannellimitcomesintoplaytogiveeachnodetheidenticalallocationratio,andthusthefairnessiswell-addressed.ThereexistsanothertypeoftrafcdependencywhenCSMA/CAisapplied,becausethereceivingnodesneedtosendCTSorACKmessagesuponreceivingpacketsfromthesendingnodes.Inourspectrumallocation,wecountthispartintosendingnodes'spectrumusage. Ifeachnodeintheneighborhoodhasthesameinformationaseachother,thedistributedspectrumallocationsshouldbeidentical.Inmulti-hopscenarios,itiscommonthatneighborshavedifferentneighborhood,thusthedistributedspectrumallocationsareprobablydifferent.Ifthetotalspectrumallocationdoesnotexceedthechannellimit,thedistributedalgorithmcanbeacceptableeventhoughsomefairnessissacriced.However,spectrumallocationconictsusuallyexistwiththeasymmetricneighborhood.Theproposedschemeusesafeedbackmechanismtoregulatetheneighboringnodes'trafcratefromover-injection. Whenasinglerateisused,fairnessrequirestherealtrafcrateofeachnodeisproportionaltotheratiobetweenitstrafcdemandandthetotaltrafcdemand.For 87

PAGE 88

30 ].Eachnode'strafcloadisexpressedbythetrafcrate.Basedonthelinks'achievablePHYrate,thistrafcloadcanbeconvertedtoairtime,whichisusedforfairspectrumallocationinlocalarea.Weignorehowtherateadaptationmechanismisimplementedinthispaper.WealsosimplyassumethatduringoneperiodeachnodeusesoneconstantPHYrateinonelinktoitsneighbors.Theproposedschemerequirestheair-timefairnessamongneighbors,whichmeanstheair-timefractionofeachnode'sallocatedtrafcrateisproportionaltotheratiobetweenitsair-timeoftrafcdemandandthesummationoftheair-timeofeachnode'strafcdemand.ThiscanbeexpressedasFormula 4 Inthisscheme,eachnodemaintainsthreetableswhichrecorditsowntrafcinformation,itsneighbors'trafcinformationandthetrafcdependencyinformation.Eachnodeperiodicallygetsknowledgeofitsoriginaltrafcloadandforwardingtrafcloadfromitsupperlayersandupdatesthesetables.Italsoupdatesthesetableswhenitreceives/overhearsmessagesfromitsneighbors.Eachnodecalculatestheachievabletrafcrateforitselfanditsneighborsdistributedly,withthealgorithmwhichwillbe 88

PAGE 89

WedenotethemaximumnormalizedIEEE802.11one-hopMACthroughputasC,whichalsomeansthemaximumbusyair-timefraction(C=RC 71 ],canprovidewaysonhowtodeterminethevalueofC. Inthefollowingsubsections,werstpresentthesupportingdatastructureandmessages'format.Wethendescribethedetailedprocedureandalgorithm. 4-1 whichincludesitsowntrafcinformation: Table4-1. LocalParameterSet Eachnodeshouldalsomaintainatablewhichrecordsitsneighbors'correspondinginformationasthepotentialinputofdistributedspectrumallocation.ThedetailinformationislistedinTable 4-2 Thetrafcdependencyinformationisstoredinan(n+1)(n+1)matrixD,withthelocaltrafcdemandincluded.Di,jmeanstheforwardingrequirementfromnodeitonodej.Wheniequalstoj,Di,jstorestheoriginaltrafcexcludingitsforwardingdemandfromitsneighborsinthisneighborset.Notethatthisoriginaltrafcdemandmaynotpurelybeoriginal.Innodei'sstorage,ifneighborjhassomeforwardingrequestfromits 89

PAGE 90

NeighborParameterSet neighbork,whichisnotaneighborofnodei,neighborj'soriginaltrafcdemandinthematrixincludesthispartofforwardingrequest.Obviously,wehavetherelationshipasfollows: 4 meansthebroadcastedversionofTDi.Throughoutthispaper,superscript0standsforthebroadcastedversionandsuperscriptstandsfortheregulatedversion.Apparently,matrixD'sstoragehassomeoverlappedinformationwiththeformertwotables.ThedifferenceisthatmatrixDonlystoresthebroadcastedversionandtheothertwotablesgatherthelatestinformation. Thebroadcastedmessagesshouldcontainthefollowinginformation:thetrafcdemandofthenode(TD),thetrafcload/rate(R)andtheadjustedtrafcrate(R)ofthenode,alistoftrafcthatthehostneedstheneighbortoforward(TDtfwi),anindicatorthatatrafcregulationisnecessaryorthisregulationistobedeactivated.Thereisonemoreparameter,unitdemandallocation",isimpliedinthebroadcastmessages. TD(4) Thisvalueisusedtoeliminatetheconictsduetoasymmetricneighborhood,whoseusagewillbeillustratedinalatersection. 90

PAGE 91

Forupstreamneighbors,trafcdemandisasbroadcastedsincenoothernodesintheneighborhoodcanaffectthisvalue.However,localnode'strafcdemandisaffectedbyitsupstreamneighbors.Thelocaltrafcdemandcalculationcanbefoundinequation( 4 ).Fordownstreamneighbors,thetrafcdemanddependsonlocaltrafcchange.Itcanbecalculatedasfollows: whereTDtfw0icanbefoundinmatrixDaccordingtothecurrentnodeindex(IfitiscalculatedbyNodek,TDtfw0i=Dk,i).Sinceeachnodecanplaydifferentrolessimultaneously,theoveralltrafcdemandcanbecalculatedinthisway: Iflocalnodeindexisnoti,TDOiusesthevalueofDi,i.WecanseethatTDtfwichangeswhenupstreamneighbors'trafcchanges,whichembodiesthetrafcdependency.Throughthisway,theaccuratetrafcdemandcanbecalculatedbyeachnodelocallyandtheidealspectrumallocationcanbefullled. Theasymmetricneighborhoodinformationcancausedifferentspectrumallocationatdifferentneighbors,leadingtospectrumusageconicts.Whenonenodesensesthetotalspectrumallocationbyindividualneighborsexceedingthechannellimit,aregulationindicatorissentout.Theregulationmechanisminthisschemeusesthe 91

PAGE 92

FlowChartforMessageProcessing parameterof".Thisparameterstandsfortheuniformallocatabletrafcair-timeperunittrafcdemandinair-timeform.Eachindividualnodeallocatesthetrafcratebasedontheair-timefairness,whichmeansRi 4-2 showstheowchart. Beforeeverydatatransmission,ifparametersintherecordhavebeenupdated,adistributedspectrumallocationalgorithmiscarriedout.Whenthenodestartstotransmit,itpiggybacksitsbroadcastinformationinRTSorDATApackets.Afterbroadcasting,thenewlycalculatedvaluesupdatetheoldrecord.IfTD,RandTDtfwiarenotchangedsincelastbroadcasting,noinformationisattachedtothedata.ThealgorithmexecutedbeforebroadcastingisshowninFig. 4-3 Thegrossspectrumallocationcanbebrieydescribedasfollows.IfthetotaltrafcdoesnotexceedthechanneltrafclimitC,thenodeusesitsrequiredtrafcrate. 92

PAGE 93

FlowChartforSpectrumAllocation Otherwise,thenodeusesthetrafcrateinproportiontothechanneltrafclimit.Thebasicallocationalgorithmcanbeexpressedinthefollowingequation. TDi PHYi+Xk2N,k6=iXj2NDj,kTDtfw0k+TDtfwk 93

PAGE 94

Aswementionedabove,theend-to-endthroughputperformancedependsonthetopologyandthetrafcpatternheavily.Theend-to-endthroughputvarieswhenthetopologyorthetrafcpatternisslightlychanged.Inthissimulation,weconstruct4differentandtypical3-hop(4-hopinthelasttopology)topologiesforevaluation,asshowninFig. 4-4 .Sincethebenetsoftheproposedschemeismoreprominentwhenthetrafcpatternislesshomogeneous,weabandonthecommonly-usedgridtopology.Thecirclenodeshaveoriginaltrafcwhiletherectangularnodesonlyforwardtrafcfortheneighbors.Theellipticalnodesonlyreceivetrafc.Consequently,therearebothlong-hop(3-hopor4-hop)owsandshort-hop(2-hop)owsintheconstructednetworks.Inthelasttopology(topology(d)),thereisacaseforbi-directionalows.Theoriginaltrafcdemandismarkedundereachnode.Toillustratethespectrumallocationresultclearly,thetrafcdemandissettoexceedthechannellimit,whichisassumedtobe11Mbps.Thetrafcdemand'sunitisinkbps. 94

PAGE 95

Thetopologiesforevaluation Forcomparisonpurpose,weapplytwootherschemestothesetopologies.OneisrandomfairaccessMAC(RFA)andtheotherisadistributedschedulingscheme(DSS).RFAletseachnodeaccessthemediumrandomly,thusfairly(assumed).DSSrequireseachnodecollecttheneighbors'trafcdemand,however,withnoconsiderationoftrafcdependencyandallocatespectrumaccordingly.Thetrafcdemandinthisschemeissimplytakenasthetrafcarrivalrate.Sincethetrafcarrivalratedependsonapredenedschedulingoutput,wesettheinitialschedulingasfairmediumaccess.Althoughthersttwoschemeslackofmechanismstoovercometheasymmetricneighborhoodproblem,inthissimulation,weassumethereexistsaregulationmechanismwhichcanregulatethetrafcwithineachneighborhoodsothatitwillnotexceedthechannellimit.Thersttwoschemesrequireeachnodetodeliverthetrafcfromdifferentsourcesaccordingtothearrivalproportions.Theoverheadofall3schemesisignored.Thefairnessindexinthesimulationiscalculatedinthefollowing 95

PAGE 96

whereXiisthetheratiobetweentherealizedowrateandthecorrespondingowraterequestedforowi. Figure4-5. Comparisonofend-to-endthroughput(a),fairnessindex(b)andAIR(c)forRFA,DSSand2hopMACschemes FromFig. 4-5 ,wecanseetheperformancecomparisonforRFA,DSSand2-hopMAC.Inthisgure,thebenetsofend-to-endthroughputandfairnessof2-hopMACareshownbythecomparisonwithRFAandDSS.Part(a)comparestheend-to-endthroughputperformance.Theproposedschemehasachievedsignicantgainineverytopologysetting.Itisinterestingtonotethattheend-to-endthroughputofRFAisbetterthanDSSinTopology(a),(b)and(c).ThereasonisthatDSSdoesnotconsiderthemulti-hoppropertyofthetrafc,thusthespectrumallocationcandeviatefarfromtheidealcase.ForTopology(d),becausethereisacenter-likepoint,DSScanallocatethespectrumbetterthanRFA.NotethattheregulationmechanismforasymmetricneighborhoodproblemisassumedforRFAandDSS,andthustherealisticperformanceshouldbeworse.Theothercommentforthissimulationisthatthetrafcdemandisintentionallysetsothatthetotaltrafcdemandbarelyexceedsthechannellimitandeachowhasidenticaldemand.Whenthetrafcdemandofeachowandthe 96

PAGE 97

Part(b)showsthefairnessperformancecomparison.Thefairnessamongdifferentowsisguaranteedonlywith2-hopMAC.Sinceeachnodeisassumedtodeliverthetrafcaccordingtotheproportionofthetrafcfromdifferentsources,nodeswhichdonotknowtherealtrafcdemandfromdifferentsourcesbutthearrivaltrafcamount,cannotallocatethespectrumusagefairlyaccordingtotheirknowledge. End-to-endthroughputisnotagoodindicatorofsystem'sperformanceinthesensethatitcannottellhowgoodaschemeis.Itdependsgreatlyonthetrafcdemandofeachnodeaswellastheowdistributioninthenetworks.AIRisintroducedastheindicatorofhowgoodthespectrumisallocatedusingthegivenscheme.IfAIRreaches0,theschemeallocatesthesystem'sspectrumtotheservicednodeswithoutanywastage,whichmeansthepacketsobtainingthespectrumtodeliverattheirrsthophaveobtainedthevirtuallyreservedspectrumfortheirfollowinghops.Thereisnodoubtthatwhentheavailablespectrumisallocatedwithoutwastage,theefciencyofthespectrumallocationachievesitsbestperformance.However,duetothecomplexityoftopologyandtrafcdistribution,non-zeroAIRisusuallyinevitable.Part(c)showstheAIRcomparisonamong3schemes,throughwhichwecanobservethespectrumallocationefciencyofeachscheme.ItcanbeseenthatbothRFAandDSSwastealotofspectrumtheyallocate.SinceDSSusesinaccuratetrafcinformationtoallocatespectrum,itsAIRishigherthanRFA'sinsomescenarios.For2-hopMAC,AIRisalmost0inallfourtopologies.Thisresultisobviousinthateachneighborincorporatestheexacttrafcinformationfromitsneighborsandpassthisinformationtothe2-hopneighborsviatheseneighbors.Throughthismethod,theknowledgeofaccumulatedtrafcdemandforeachnodeintheneighborhoodiscorrectlyacquired.NotethatinTopology(d),theAIRhasanon-zerovaluebecauseoftheregulationforasymmetric 97

PAGE 98

98

PAGE 99

Akyildizetal.[ 5 ]haveproposedafewmodelsofWMNs.Usually,WMNconsistsofvarioustypesofentities:gateways,meshrouters,accesspoints(AP)andmeshclients.Gatewaysaretheconnectionpointstothewire-linenetworks.Meshclientsaretheterminaluserswhichhavenoorlimitedroutingfunction.WirelessAPsaretheentitiesinchargeofthewirelessaccessforthemeshclients.Stationarymeshroutersformawirelessmultihopbackbonewithlong-rangehigh-speedwirelesstechniquessuchasWiMAX.Indifferentmodels,ameshnodecancontainoneormorefunctionalentities,e.g.,meshroutersusuallyimplementAPfunctionalities. Whenthemobileclientsarestationary,withthesupportofbackbonerouting,thewirelessaccessforthemcanbeaccomplishedwithinafewhops.However,difcultyariseswhenthereareneedsforthemeshclientstomoveacrossthecoverageareaofdifferentAPs.Howtomaintaintheongoingconnectionandhowtoforwardthedownstreamandupstreampacketsarenotsolvedbythecurrentstandards.IEEE802.16eaddsamendmentstotheoriginalstandardtosupportmobility,butonlyspeciesMACandPHYlayer[ 2 ].IEEE802.11sattemptstoextendtheWiFitosupportthemeshmodeandprovidemobilitysupport,whichisstillunderdevelopment. Mobilitymanagementisnotanewtopicinotherexistingnetworks.Akyildizetal.[ 4 ]presentedasurveyonthistopic.Incellularsystems,thisparthasalreadybeenacriticalparttothecontinuousserviceofthemobileclients.Handoffqualityisoneofthemostindispensabletestingitemsineacheldtrialtest.However,wirelessmeshnetworks, 99

PAGE 100

49 ].ThemainideaisverysimilartotheHLR/VLRmechanismincellularsystems.HomeAgent(HA)andForeignAgent(FA)playtherolesofhomedatabaseandvisitingdatabaseintheIPnetworks,respectively.HomeaddressisusedastheIDofamobileclientandtheCare-of-Address(CoA)isusedtolocatethecurrentpositionofthemovingmobileclients. MobileIPcanprovideasolutiontotheinter-domainmovementinWMNs.However,itisnotsuitablefortheintra-domainmovement,whichismuchmorefrequentthantheinter-domainmovement.ThereasonisthatifFAisimplementedineveryAP,signalingcostandhandofflatencybecomethemajorproblemstothemobilitysupport.Therefore,thesolutiontocopewiththelocalmovementisrequired.ProtocolsforIPmicro-mobilityhavebeenproposedtosolvethemobilitydilemmainsmall-scalenetworks[ 23 ],[ 27 ],[ 52 ],[ 59 ].ThoughtheseprotocolscanbeappliedtoWMNs,heaviersignalingcostandlongerhandofflatencyduetomorefrequentlocalmovementinWMNsstillimpedethepracticalmobilitysupport. Inthispaper,weproposeamobilitymanagementschemeinWMN,termedMeshMobilityManagement(M3).SomefeaturesofWMNs,suchasmulti-hop,meshtopologyandcontinuouscoverage,havebeentakenintoconsiderationtobettersupporttheIPmicro-mobilityinWMNs. Therestofthispaperisorganizedasfollows.SectionIIdiscussessomerelatedworks.SectionIIIdescribestheproposedscheme.PerformanceanalysisiscarriedoutinSectionIV.Conclusionisgivenattheend. 100

PAGE 101

54 ]mentionedthemobilitymanagementissueintheircomprehensivework.Theexperimentresultsconrmthathandofflatencyusingatunnelingschemeismuchlongerthanthatusingatrouting.However,sincemobilitymanagementisnotthefocusofthispaper,theauthorsdiscussonlythefeasibilityofmobilitysupportanddonotincludedetailedanalysis.InSMesh[ 7 ],multipleAPsmonitorthemovingmobileclientstoachieveseamlesshandoff.Thisschemeeliminatesthehandofflatencyatthepriceofhighsignalingcost. However,previousworksonIPmicro-mobilityarepossibletobeappliedtoWMNs,sinceWMNcanbetreatedasonetypeofmobileIPnetworks.WenowreviewsomeIPmicro-mobilityprotocols. InCellularIP[ 59 ],mobileclientsusethegateway'sIPaddressastheirCoAandeachrouterinthisdomainusethehomeaddressesofthemobileclientstoroutethedownstreampackets.Thedefaultroutesforeachroutertothegatewayareusedtodirecttheupstreampackets. HAWAIIisanotherimportantframeworkofIPmicro-mobility[ 52 ].TheCoAofeachmobileclientinHAWAIIisauniqueIPaddressallocatedbythegatewayofthedomain.DifferentfromtheCellularIP,HAWAIIusestheCoAofeachmobileclienttoroutethedownstreampackets.ThisdifferencemakesHAWAIIlesscoupledwithMobileIPprotocolandalsoenablestheper-owQoSsupportinthebackbonenetwork. Inbothschemes,eachdomainisidentiedbyasinglegatewayandtheentiredomainisconstructedtoatree-likestructure.BothschemesrequireeachroutertomaintainaroutingentryforeachmobileclientinthedownstreamAPs'coverage.Whenhandoffoccurs,thecorrespondingroutingentrieswillbeupdatedinalltheroutersinvolvedfromthenewAPtothecrossoverrouterwhichissharedbythenewAPandoldAP.Theinvalidroutingentriesintheroutersoftheoldpathneedtoberemoved.Due 101

PAGE 102

11 ]. AnotherimportanttypeofIPmicro-mobilityprotocolsisthehierarchicaltunnelingapproach[ 11 ],anexampleofwhichisMobileIPRegionalRegistration(MIP-RR)[ 23 ].Hsiehetal.[ 27 ]proposedanotherscheme,namely,HierarchicalMobileIPv6.Thistypeofschemesreplacesthemobile-specicroutingbyintroducingthetunnelingtechnique.Throughthehierarchicalregistrationprocedure,thehigher-levelFAknowsthelocationinformation(IDofthelower-levelFA)ofthemobileclientsandencapsulatesthedatapacketswiththedestinationaddressofthislower-levelFA.Per-hostroutingentryisnotrequiredfortheroutersintheseschemeswhileper-hostlocationinformationisstillstoredinFAs.DuetotheextraprocessingofencapsulationanddecapsulationasinMobileIP[ 49 ],largerdelayisintroducedtoeachow.AdditionalcostofthistypeofschemesisthattwoormoreCoAshavetobeused.Whenhandofftakesplace,theregistrationwithadifferentCoAalsoaddsextradelay.TheintuitiveideaofthisapproachistoextendtheMobileIPmechanismtolocalmovements. 5.3.1ModelDescription ThegatewayisrequiredtoassignauniqueIPaddressinitsdomaintoamobileclientwhenitispoweredup.ThisuniqueIPaddressofamobileclientcanbetheCoAwhenmobileIPisprovidedfortheinter-domainroaming.Theforeignagent(FA)andhomeagent(HA)canresideinthegateway.Inthescenariowheremorethanone 102

PAGE 103

Weusea3-levelhierarchicalstructuretoillustrateourscheme,asshowninFig. 5-1 .ThethreeAPsconnectingtothegatewayhavesuperiorstatusthantheirdownstreamnodes.TheyarerequiredtocollectthelocationinformationofthemobileclientsinthecellsofthesubordinateAPs.WenametheseAPssuperiorrouters(SR)hereafter.TherestoftheAPshaveequivalentstatus.SRsactasthedelegatesofthegatewayandsharethesignalingtrafc.Inasmallermeshnetwork,ifthegatewayisnotthebottleneck,thesesuperiorrouterscanberemovedwhichyieldsonly2-levelhierarchicalstructure. Asdiscussedin[ 53 ],aWMNcanbeconstructedinatree-likestructure.Eachrouterhasitsonlyparentnodeandmayhaveanumberofchildren.Thiskindofmodelinghasitsbenetfortheroutingwhereonlythetrafcowsbetweenthegatewayandeachmobileclientareconsidered.Thismodelshowsitslimitationwhenthemobilitymanagementistakenintoaccount.ThetreestructureisextractedfromtherealgeographicaltopologybasedonthecriterionoftheshortestpathfromeachAPtothegateway,whichcannotbeusedtoobtaintheoptimalpathbetweenanytwogeographicalneighboringAPs.Theroutingofpreviousschemesstrictlyfollowsthetreestructureevenwhenthereexistshorterpaths.UnlikeotherWMNmodels,ourschemeallowsthecommunicationalongthepathswhicharenotinthetree.Weassumethatmostofthetime,geographicallyadjacentAPshaveshortercommunicationpathsotherthantheonlypathalongthetree.Therefore,thisstructureembodiesameshtopology. Weassumethattheroutinginthebackbone(APs,superiorroutersandthegateway)hasbeensetup.SincethebackbonenodesinWMNsaremostlystationary,thisassumptionisreasonable.Theremainingproblemisonensuringamobileclienttomovearoundinthisareawithoutincurringhighpacketloss,longhandofflatencyandhighsignallingcosttothesystem. 103

PAGE 104

MeshModelandIllustrationofM3Scheme 5.3.2.1Power-up DatabaseofeachAPcontainsonlythesubscriberinformationofthemobileclientcurrentlyinthecell.DatabaseofeachsuperiorrouteradditionallycontainsthelocationinformationofallthemobileclientsresidinginthesubordinateAPs'cells. 104

PAGE 105

InFig. 5-1 ,theboldlinesillustratethedownstreamprocess,withthedashedlinesandsolidlinesindicatingtheroutingpartandthetunnelingpart,respectively.Fromthegateway(GW)totheSRs,thepacketsareroutedaccordingtothelocationinformation.Theotherroutingpartindownstreamforwardingwillbediscussedinsection4). 105

PAGE 106

43 ]andPOFLA[ 44 ]fordifferentapplications.Incellularsystem,thismethodiscalledpointerforwardingandweborrowthisname[ 44 ]. SupposemobileclientAmovesfromposition3(A(3))toposition4(A(4))inFig. 5-1 ,thedownstreampacketsarersttunneledtoAP3.AP3forwardsthedecapsulatedpacketstoAP5accordingtothepointerofmobileclientA.UpstreampacketsfromAareroutedtogatewaybythedefaultroutesofAP5,AP7andSR3,sequentially. TopreventtheencapsulatedpacketsfrompassingthenaldestinationAPtoaformerAP,theattachedIPheadercanincludethemobileclient'sIPaddressintheoptioneld.Therefore,whentheencapsulatedpacketsreachthenaldestinationbeforetheendofthetunnel,theAPofnaldestinationcandecapsulatetheminsteadofsimplypassingthemon. 49 ].HMIPusestheideathatafteracertainnumberofthehopsthemobileclienttriggersalocationupdatetotheHA[ 43 ].Ourschemeusesthetimeintervaltobethetriggeringcondition.WeassumethatthemobileclientsintheWMNarenotsofastthatduringtheperiodofTlutheycrosslessthanNhndfAPs.Thereafter,onceeveryTlu,themobileclientscantriggerthelocationupdatetocontrolthetriangularroutingproblem.Tomakethisupdatemoreefcient,weleteachAP,insteadofeachmobileclient,betheinitiatortotriggerthisupdate.EachAPreportsthecurrentsetofmobileclientstothesuperiorrouters.ThesuperiorroutersselectanotherintervalThutoperiodicallyupdatethesettothegateway.ThuisobviouslyrequiredtobenolessthanTlu. Afterthisperiodiclocationupdate,downstreampacketscanbetunneledtotheAPwherethemobileclientexactlylocateswithouttraversingalltheAPsthemobileclienthasvisited. 106

PAGE 107

11 ]. Havingcomparedthesetwotypesofapproaches,wenowdiscusstheirprosandcons.Byusingthemobile-specicrouting,thenecessityofencapsulation/decapsulationiseliminated,andviceversa.Thereasonthatmobile-specicroutingcannotbeappliedtomacro-mobilityisthedifcultytondacrossoverrouterwhichcanmaintainthemobile-specicroutingentry.Inotherwords,thescalabilityproblemmakesitinfeasible.Moreover,thisapproachhighlydependsontheroutingprotocols.Anotherproblemofthisapproachispointedoutin[ 11 ],whichis,whenupdatemessagesarelostduetophysicalreasonssuchasradioblack-out,theroutingentriesindifferentroutersmightbeinconsistent.Maintenancesignalingmightbeanadditiontoguaranteetheconsistency.Forthehierarchicaltunnelingapproach,ifthenumberofhierarchicallevelsarenotsmallenough,theencapsulation/decapsulationwillcausethedelayperformanceintolerable.However,ifthenumberofhierarchicallevelsaresmallenough,thesignalingcostofhandoffandhandofflatencymaybeinsteadintolerable. Ourschemeachievestheadvantagesofbothpreviousapproaches.TunnelingthedownstreampacketsinthebackbonelowertheroutingrequirementforeachintermediateAPs.Withoutthemultiple-levelregistrationprocedureinthehierarchicaltunnelingapproach,ourschemeachievesshorterhandofflatency.Consequently,thepacketlossproblemisgreatlyalleviated.Asimplebufferingtechniquecaneliminate 107

PAGE 108

ThefeaturesofM3canberelatedtosomefeaturesofWMNs.Continuouscoverageisthereasonforapplyingthemobile-specicroutinginthelastfewhops.ThestationarycharacteristicofWMNs'backboneyieldssimplerbackbonerouting.Thehierarchicalstructureandthesimplebackboneroutingrendertunnelingmoreappealing.Lowspeedofmobileclientslimitsthedelayofdownstreamforwardinginthisscheme.Moreover,duetothecontrollabledelay,theperiodiclocationupdatecanbeappliedwithoutsideeffects.Unlikethestricthierarchicalstructureofcellularsystem,mobilitymanagementunderthephysicalmeshstructureofWMNscanberealizedinamoreexibleway. 11 ].Aswementionedintheprevioussection,theencapsulation/decapsulationintroducedbythetunnelingbringsusextradelaytodownstreampackets.Thisisthepricewhenwewanttouselessper-hostrouting.However,thisdelayisnotsignicantandshouldbetolerablebecauseencapsulationanddecapsulationonlyhappenonceforeachdownstreampacket.Therefore,ourperformanceanalysiswillnotincludethispart. Thesignalingcostisdenedasthetotalamountoftheextrasignalingtrafcduetothemobility.Thehandofflatencyisdenedastheservicedisruptiontimeofthemobileclientsduetohandoff. 108

PAGE 109

Indifferentmobilitymanagementschemes,theupdateprocedurealwaysstartsfromthenewAP/BStosomeanchorpoint,suchasVLR/HLRorHA,thenfromtheanchorpointtotheoldAP/BS.Ifweassumetheupdatesignalingprocedureisthesameindifferentschemes,thecostdependsonlyontheupdatepathlength,whichisthefocusouranalysis. Letn,m,,C,Cu, Forourscheme, Next,weprocure 109

PAGE 110

Excludingthelinksinthetree,thenumberofwhichism1(excludingthegateway),wedenethehandoffpathstobethesequentialsiblingvisiting.Thenumberofneighborhandoffpathswith2j(j2[1,l])hopscanbeexpressedasfollows. Therefore,assumingthatthehandoffofeachcaseisequallikely,theaveragepathlengthcanbeexpressedas: (m1)l(l+1) 2(5) Thegainofourschemeisdenedasg=Cps 5-2 110

PAGE 111

GainofM3overpreviousschemes Wenowdiscusstheaveragepathlength.Intheliterature,themaximumnumberofhopsisnotrecommendedtobelarge,with4or5preferred.Theaveragenumberofchildrenkshouldberelativelysmallinordertoavoidtheperformancebottleneck.TableIshowsthetypicalvaluesetoftheaveragepathlength. Table5-1. TypicalValueSetofAveragePathLength parameters k=2,l=3 k=3,l=3 k=2,l=5 k=3,l=5 k=m,l=1 averagepathlength 4 2.9 4.21 3 2 Iftherearenormalrouters(withoutAPfunctionality)inthenetwork,thehandoffbetweenparentsandchildrenbecomeslesspossible.Therefore,theaveragepathlengthforthepreviousschemeswillbelarger.ThegainofM3willbehigher. ThissignalingreductionismainlyfortheSRsandthegatewayoftheWMN,whicharethebottleneckmostofthetime. Handofflatencycanbeexpressedas: whereDhndfdetcisthedelayofhandoffdetection,Dupathestbistheunit-stepdeliverydelayinpathupdateand 111

PAGE 112

112

PAGE 113

21 ]andthespecicationoftheinterworkingbetweenWCDMAnetworksandWLANswasstandardized[ 20 ].Itisnowwell-knownthatthesewirelessadhocnetworks,usuallyoperatedintheadhocmode,canbeconguredmoreexiblywithmuchlowercostandpotentiallyhigherdatarateduetomuchshortertransmissionrange,thoughthecoverageandmobilitymaynotbeguaranteed.Consequently,itisexpectedthatMCNscantakeadvantagesofbothtraditionalcellularnetworksandwirelessadhocnetworks. AMCNiscommonlyconsideredasacellular-basedintegrationofacellularnetworkandadhocnetworksbecauseadhocmodeistreatedasanadditivecomponenttocellularsystems,whichdoesnotimpairthetraditionalcellularinfrastructure,whichmeansthatmobileuserscanaccessBSsthrougheitherdirect(one-hop)connectionsormulti-hopconnections.Withthecellular-basedMCNsunderconsideration,therstadvantageoverthetraditionalcellularsystemsweobserveisthelowcostofextendingservicecoverage.Multi-hopwirelessrelayingcaneasilyprovidefaraway 113

PAGE 114

61 ],toimplementloadbalancingamongdifferentcells.Finally,withtheadhoclinksbearingpotentiallyhigherdataratewithextraspectrum,thesystemcapacitycanbeexpectedtoimprove.Ofcourse,werealizethatthehiddenassumptionisthatmobileterminals,inadditiontotheoperatoraddedadhocrelayingstations,alsoparticipateinrelayingandhelping,whichmaynotbepracticalifthereisnoincentiveforthehelpingmobiles.Therefore,certainincentiveprotocolsshouldbedesignedtostimulatemobileterminalstohelptheMCNstooperate.Thistopicwillbeinvestigatedinaseparatepaper. Inthispaper,wefocusonthebenetsthattheadhocmodecanbringtothehandoffmanagement.HandoffmanagementisanimportantpartofmobilitymanagementinPersonalCommunicationServices(PCS)systemsandanywirelesscellularsystemingeneralinmaintainingseamlessconnection[ 16 ].Handoffdropping(HOD)rateisthemajormetricforassessinghandoffperformanceanditusuallyrequiresthedesignedsystemtokeeptheHODlowerthancertainthresholdinordertomeetthecustomers'satisfaction.Inwirelesscellularnetworks,HODrateisusuallyguaranteedbybandwidthreservationforhandoffcallsincellsamobileusermostlikelyvisitsduringitscallconnection.MorebandwidthreservationcanindeedmeettheHODraterequirement,butatthepriceofblockingmorenewcallconnectionsbecauselessresourceismadeavailablefornewcallconnections.Thus,agoodhandoffmanagementshouldnotonlykeeptheHODratelow,butalsotieuplessbandwidth.Asasupplementtocellularsystems,adhoclinks(linksviarelayingbyusingharvestedeitherinbandoroutbandresourceopportunistically)createmoreopportunitiesforMNstoconnecttoBSs.Therefore,HODrateisexpectedtodecreasewithadhocmodeintroducedintothe 114

PAGE 115

Aswealludedearlier,theintroducedadhoclinkscreateplentyopportunitiesforaccesstothecellularsystems.Bychoosinggoodlinkconncections,wecandecreasethecalldroppingratewhileinappropriatechoiceoflinksmayleadtopoorperformance.Carelessdesignofhandoffdecisionprocessmayincreasethehandofffrequency,increasethecalldroppingrate,orleadtotheinefciencyofresourceusage.HowtodesignthehandoffstrategyintheMCNsbecomesamorecomplicatedtaskthanpurelycomparingthereceivedsignalstrength(RSS)becauseoftheoptionsofferedbythemulti-hopconnections.ThisproblemforMCNshasnotbeentoucheduponpreviously.Mostpreviousworksassumethattherelayingismadebyspecicplanteddevices(relaystations)accordingtopropernetworkplanning[ 13 63 ].Thisassumptionreallylimitstheutilizationofbenetsofferedbytheself-organizingadhocnetworkinganddiminishtheexibilityoftheadhocmodeintroduced.Inthispaper,weproposeanewscheme,calledANHOA(Ad-hoc-Network-EmbeddedHandoffAssistingScheme),whichutilizestheembedded,self-organizedsmall-scaleadhocnetworkstoassistthehandoffs.Byexchanginginformationinsidetheembeddedadhocnetworks,relaynodescanhelphandoffcallschoosebetterhandoffoptionsandthusreducethecalldroppingprobability.Consequently,tomeetacertainHODrate,BSscanreservelessbandwidthforhandoffcallsthanbefore.WealsodesignanalgorithmtoenableadjacentBSstolowerthebandwidthreservationaccordingtothetrafcinformationoftheircellstoloosenupmorebandwidthfornewcallswhilemeetingtheHODraterequirement.Furthermore,wehavealsoproposeaframeworkforadjacentBSstoexchangeinformationthroughwhichtheloadbalancingamongthesurroundingcellscanbeimplemented.AlthoughtherearesomeworksondifferentissuesinMCNs,theyrarelyaddressthehandoff 115

PAGE 116

Therestofthepaperisorganizedasfollows.Section 6.2 discussestherelatedworks.InSection 6.3 ,wedescribethesystemmodelandthebasicideasofthispaperandinSection 6.4 ,wepresenttheproposedANHOAscheme.ThealgorithmforndingminimumreservationinMCNsandtheframeworkofinformationexchangeamongadjacentBSsaregiveninSection 6.5 .wecarryouttheperformanceevaluationinSection 6.6 andconcludethepaperinthenalsection. 41 ].Inthispaper,theauthorssuggestedtousemulti-hopcommunicationsfrommobilestoabasestationviapossiblymultiplehopswithlowertransmissionpowersothatmoresimultaneouscommunicationscanbeaccommodated.IniCAR(IntegratedCellularandAdHocRelayingSystems),Wuetal.proposedtoproactivelydeployanewsetofrelayingnodesinareaswheretrafccongestionstartformingandusetheserelayingnodes,calledadhocrelayingstations(ARSs),torelaythetrafcfromthecongestedcelltothenon-congestedcellswheretrafccanbeserved[ 63 ].ToinvestigatehowmuchadhocrelayingcouldenhancethesystemcapacityinMCNs,Lawetal.studiedthecapacitybyassumingthatacellisdividedintotwoco-centeredareaswheredirectcommunicationsarecarriedoutwithinthenearrangebetweenmobilesandtheBSwhilemulti-hopcommunicationswillbecarriedoutonlywhenmobilesareoutsidethenearrange[ 36 ]andfoundthatforcertainscenarios,thesystemcapacitycanindeedbeincreased. 116

PAGE 117

42 ].Thisworkwasmotivatedbytheobservationthatahigherdownlinkdataratemaybeneededformanyapplications.Byallowingwirelessclientstorelaythedownlinktrafcinthisscheme,thesystemcanachievebetterthroughputperformance.Theauthorsdevelopedthediscoveryalgorithmofwirelessproxy(relayclients),whichplaysveryimportantroleinimplementingtheirproposedarchitecture.Inviewofthepotentialgaininusingadhocmode,futuregenerationwirelesscellularstandardshavealsoconsideredthisissueandproposedODMA(OpportunityDrivenMultipleAccess)protocol[ 19 ],whichisasimilarschemetoUCAN[ 42 ],butfocusesontheimprovementofdataratebyallowingrelaying. Therearemanyproposalsonad-hoc/cellularintegrationarchitectureintheliteraturewhichaddressdifferentaspectsofvariousintegratednetworks.Cavalcantietal.providedasurveyofalltheseintegratednetworks[ 12 ].Someofthefeaturesoftherelatedworksarecompared.LeandHossainhavealsogivenasurveyontheexistingMCNs[ 37 ],whichfocusesmoreonresourcemanagement.Choetal.dealtwithhandoffissuesofvarioustypesinMCNs[ 13 ].In[ 9 ],Bhargavaetal.tookadifferentapproachandintendedtousecellularnetworkstohelpthemanagementofadhocnetworksbyutilizingthebenetofcoverageofferedbycellularsystems.Aswecanobserve,mostintegrationresearchworksarebasedoncellularsystems,focusingmoreonenhancingtheperformanceofcellularsystemsbecausecellularsystemsaredeeplycommercializedandtheycanprovideseamlesscoverageandmobilityservice. 6.3.1SystemModel 117

PAGE 118

SincethefocusofthispaperisthehandoffissueinMCNs,inourmodelhere,eachMN(mobilenodeormobileterminalsoruserequipment)isalwaysboundtoonecertainBS,nomatterwhetherviaonlycellularlinksorviamulti-hopconnections.Wecallthehandoffsinvolvingmulti-hopconnectionsasthemulti-hophandoffs.WhenanMNmovesfromonecelltoanotherandchoosesthetargetBSasitsservingBS,thehandoffprocessisnodifferentfromthatinthetraditionalcellularsystems.Onlywhenamulti-hopconnectionexistseitherbeforeorafteranMNaltersitsconnection,dothehandoffsfacethenewtypeofnetworkchallenges.Fig. 6-1 illustratesthemulti-hophandoffsinamulti-hopcellularsystem. Figure6-1. ModelofMulti-hopHandoffs InFig. 6-1 ,MNAmovesfromcell1tocell0.Unfortunately,theBSincell0hasnosparespectrum.Intheillustratedscenario,MNAhastwooptions.ItcaneitheraccesstheBS5throughMNBoraccessBS7throughMNCandMND.EitheroptioncanenableMNtoaccesstothecellularsystem. 118

PAGE 119

Intraditionalanalysisofcellularnetworks,itisassumedthatthecalldroppingscausedbylinkfailureareusuallyignoredandonlythecalldroppingiscausedmainlybyhandofffailure.Infact,aconnectioncanbedisconnectedduetowirelesslinkfailure,especiallywhentheconnectionismulti-hop.ThuscalldroppingscausedbythefailureofadhoclinkscannotbeignoredinMCNsforouranalysis.Thecalldroppingratecanbeexpressedinsuchaformulaasfollows. fail))(1P(HOD))(6) 119

PAGE 120

fail)istheprobabilityofcalldroppingcausedbyconnectionfailures.Multi-hopconnectionsisdisconnectedmoreeasilyduetothesmallertransmissionrange,mobilityofrelaynodesandnatureofmultiple-hop.Obviously,thecalldroppingrateisrelatedtothehopcountsofthemulti-hopconnectionandthemobilityofrelaynodesaswellastheroamingnodes. TomaintainacertainP(HOD)level,certainbandwidthshouldbereservedforhandoffcalls.Ifthereisonlyonecell'sresourceunderconsideration,accordingtoMarkovChainmodel,suchas[ 26 ],theHODprobabilitycanbeobtainedwhenthetrafcarrivalanddepartureratearegivenby wheref()isthederivationfunctionofHODratewhichisdescribedindetailinAppendix.WedenotetheHODrateasP0whenonlyonecell'sresourceisconsidered.Thearrivalratesofnewcallsandhandoffcallsaredenotedasnandho,respectively.ThedeparturerateofallcallsisdenotedasandRsvisthenumberofreservationchannels. WeknowthatthetransmissionrangeofanMNismuchsmallerthanthatofaBSandmanyMNsarehighlymobile.Itseemsadhocmodecouldnotsignicantlyhelpthecellularsystemduetomuchshorterspanofcertainlinks.However,therearestillmanyrelativelystationaryMNsexistinginthesystembecauselowmobilitycanachievehigherdatarateorpurelybecauseofusers'behaviors.DevicessuchasARSin[ 63 ]canalsobeintroducedtoincreasethestabilityofmulti-hopconnections.Therefore,duetotheexistenceofplentyofpotentialrelaynodes,wecanrelyonmulti-hopconnectionstoimprovetheperformanceofcellularsystems. 120

PAGE 121

ThetraditionalwaytocontroltheHODrateincellularnetworksistoadjustthebandwidthreservationforthehandoffcalls[ 17 ].HigherreservationmeanslowerHODrate.However,blindlyincreasingthebandwidthreservationalsoleadtothehighercallblockingratefornewcalls,thusdecreasethespectrumefciency.Therefore,searchingfortheoptimalreservationwhichcanalsosatisfytheconstraintoftheHODrateshouldbethegoalforeachBS.ForMCNs,duetotheexistenceofmulti-hophandoffsandmultiplehandoffattempts,eachBScanpracticallyhaveasmallerbandwidthreservation.Theknowledgeofadjacentcells'trafccanhelpeachBSmakingresourcemanagementmoreeffective,furthermore,balancetheloadamongdifferentBSs. 121

PAGE 122

6.4.1Overview 122

PAGE 123

EmbeddedAdhocNetworkinCellularSystem WerstdescribethenetworkarchitectureandrolesoftheassistingnetworksinFig. 6-2 .Inthisnetworkarchitecture,eachoftheseembeddedadhocnetworkscoverstheareawhichmaycoverseveraladjacentcells.AsFig. 6-2 shows,severalMNsformoneembeddedadhocnetwork(EAN)inthecellularsystemswhichspansacrossseveralcells.InFig. 6-2 ,thecirclednodesarethebackbonenodesandthesolidlinesbetweenthemstandfortherelativelystationarylinks.ThetriangularnodesstandfortheroamingnodeswhicharesearchingforBSconnectionswiththehelpoftheembeddedadhocnetworks.Wecallthesenodesattachingnodes(ANs).ThenodesinthebackboneconnectingdirectlytotheBSsaremarkedwithinsidethecircles.Wecallthese 123

PAGE 124

EachnodeinthebackboneofEANsisapotentialDNforroamingMNs.WhenaroamingMNattachestoanEANandsearchesforamulti-hopconnectiontoaBS,itsDNintheEANprovidesconnectioninformationforitandthenthehandoffattemptsfollow.PortalcapacityinthispaperisdenedasthemaximumachievablebandwidthalongthepathintheEANfromDNstotheconnectedBSs. ForeachPN,itsownportalcapacityistheaverageavailablebandwidthobtainedfromtheMarkovchainmodel.Giventhearrivalrateanddeparturerateofacell,accordingtoMarkovchainmodelas[ 17 26 ],theexpectedunusedchannel/bandwidthcanbederived.Foreachrelaynode,theportalcapacitytodifferentPNsalsodepends 124

PAGE 125

70 ],thepathcapacityisatmost1=4ofminimumlinkcapacity.Therefore,weusethefollowingequationtoestimatetheportalcapacity. whereEstCellBWdenotestheestimatedavailablebandwidthincellularspectrum,HCisthehopcountfromDNtoPN,LidenotesthelinkcapacityoflinkiandLdenotesthelinksetofthecorrespondingpath. ThisinformationisusedbyeachANtodeterminewhichPNtoconnect.TheinformationabouttheconnectiontodifferentBSsviadifferentPNsismaintainedineverybackbonenodeoftheEAN.Besidesthelinkcapacitystatustableaforementioned,eachbackbonenodeinEANsshouldmaintaintwotableswhichstoretheinformationofavailableconnectionresourceandtheinformationofitsANs,respectively. Table6-1. PortalNodes'InformationTable TheportalcapacityentriesinthetableareusedbyANstochoosePNswithbetterdroppingprobability.Theseentriesarecorrelatedwitheachother.Inotherwords,ifone 125

PAGE 126

AttachingNodes'InformationTable PNischosenbysomeAN,theportalcapacityofotherpathsaswellasthispathwillprobablyalsochange.Thereasonisthatdifferentpathsmaysharesameportalnode,orsharesamecertainlink,or,linksfromdifferentpathsinterferewitheachotherwithhighprobability. ForeachPN,itisnecessarytocollectandupdatetheinformationofitsSNs.Therefore,thefollowingtableisrequiredtobemaintainedineveryPNbesidestheabovetwotables. Table6-3. SubordinateNodes'InformationTable Whenthelinkcapacityofeachlinkchanges,theportalcapacityofeachpathisrequiredtoadaptaccordinglyinordertoreecttheBSconnectioncapabilityinrealtime.ThechangeofthearrivalrateorthedepartureratecausesthechangeofEstCellBW,andthusthechangeofPC.ThesechangesshouldbebroadcastedintheEANandeachbackbonenodeshouldupdateitsentriesaccordingly.WhenoneANacquiresacertainamountofbandwidthfromonePN,alltheintermediatelinkssubtractthesameamountofbandwidthfromeachlinkcapacity.Moreover,alltheinterferedlinks 126

PAGE 127

6-1 andOPCinTable 6-2 andTable 6-3 arecalculatedbasedonthelinkcapacityinformationviaFormula 6 InadditiontothesePCrelatedupdates,whenanMNentersorleavestheEAN,thecorrespondingrelationshipshouldalsobeupdated.Theseupdateswillbereferredtoastherelationshipupdateinthefollowingsection.Uponthearrival(bandwidthgranting)ordeparture(bandwidthreleasing)ofoneSN,aPNnotiesalltheintermediatebackbonenodesaboutthiseventhopbyhoptilltheDN.ThePNadds/clearsthecorrespondingSNentry.ThePNandallothercorrespondingbackbonenodesupdatethePCinformation,andtheDNadds/clearsthecorrespondingANentry.Besidestheupdateofthese3tables,thechangeofPCshouldtriggerthechangeoflinkstatustableofthecorrespondinglink,andthusthelinkstatustableofinterferedlinks. Moreover,thetimingforahandoffalsobecomescomplicated.Inthetraditionalcellularsystems,thetimingofhandoffsiswhenthecurrentBS'ssignalstrengthisbelowacertainthresholdandthealternativeBS'ssignalstrengthisaboveacertainthreshold(ormorecomplicatedstrategy).Whenmulti-hopconnectionsareallowedincellularsystems,handoffdecisionaregivenmoreoptionsandcanbemoreexiblethanbefore. 127

PAGE 128

Figure6-3. EAN-involvedHandoffs ThebasicprocedureofEAN-supportedhandoffshasseveralelementsjustasthetraditionalhandoffsasshowninFig. 6-3 .UpdaterequestsaresenttothenewBSs.NewBSsneedtoretrieveusers'informationfromtheoldBSs.TheconnectionsarethenestablishedbetweenthenewBSsandMNs.FinallytheoldBSsarerequiredtocleartheoldrecord.Smalldifferencefromthetraditionalcellularhandoffproceduresistheextrainformationthattheupdaterequestshouldcarry,thePortalNodeID.Thisinformationisusedforthereleaseoftheoldmulti-hopconnectionthroughthesameEANoranotherone. AlthoughtheextrasignallingswithinEANsbringextracosttothehandoffprocedures,itcanbeseenasthepricetopayforalternativehandoffoptions,whichisintuitivelybetterthanthetraditionalhandoffsonly.ThesignallingswithinEANssimplydealwith 128

PAGE 129

ANHOAisproposedtoprovidemechanismstosupportmulti-hophandoffsincellularsystemswiththeaidofEAN.Althoughtherearealotofissuesinhandofftimingandhandoffprocedures,wefocusonhowtheMNschoosehandoffoptionssothatthecalldroppingcanbegreatlyreduced. First,letusdescribethehandoffproceduresviaANHOA.WithinthetransmissionrangeofanyEANnode,theANcanacquiretheknowledgeofallPNsintheEANandtheEstCellBWofeachconnectedBS.TheANusesthehandoffdecisionalgorithmtochooseasuitablePNandrequeststheDNtoforwardtheupdatetothechosenPN.IfthenewEANisdifferentfromtheoldEAN,whichcanbelearnedfromtheEANIDinformationintheupdaterequestmessage,therelationshipupdateoftheoldEANisseparatedfromtheoneofthenewEAN,eveniftheyconnecttothesameBS.TheDNforwardstheupdaterequesttothePN,withtherequestedQoSinformationandtheoldBSandoldPN(null-valuedifthereisno).AfterQoSnegotiation,thePNsendstheupdaterequesttothenewBSwiththenegotiatedQoS.Beforegrantingtheupdaterequest,thenewBSneedstorequestthesubscriberinformationfromtheoldBSandallocatecorrespondingresourcetotheAN,careofthePN.Especially,ifBSdoesnotchange,thehandoffonlyrequiresBStoupdatetheMN'sPNinformation.Ontheotherhand,whenEANdoesnotchange,theaddingandremovingoftherelationupdatecanbeprocessedwithinthesameEAN.WhentheBSdoesnotchangeandthePNchanges,theconnectedBSonlyneedstoupdatetheMN'sPNinformation.WhenneitherBSnorPNchanges,BSdoesnotneedtodoanythingandallthehandoffprocedureiscarriedoutbytheEAN.Therefore,ANHOAcanrelievealotofsignalingpressureoffBSsbytakingcareofalotofhandoffswithinEANs. 129

PAGE 130

Noticethatwhenamulti-hophandoffhappenswithinanEAN,theprocedurecanbeverysimpleandconvenient.Theclientscangettheseamlessserviceatverylowprice.Forthecontinuityoftheservice,withinoneEAN,theMNcanchoosethesamePNasbefore.Inthisway,thehandoffprocedureonlyinvolvesthepathinformationupdateinthecorrespondingnodesintheEAN. ANHOAcanalsohelpnewcalls.Thismechanismdoesnotdiscriminatenewcallsfromhandoffcallsalthoughweshouldlethandoffcallshavehigherpriorityasdoneinthetraditionalcellularsystems. 6 ),calldroppingcanbecausedbyhandofffailureorbyphysicallayercommunicationfailure.Foreachconnectioncandidate,i.e.,oneDN'sconnectiontoonePN,thecorrespondingcalldroppingrateiscalculated.TheANcomparesthecalldroppingratesofallcandidatesandndtheonewiththelowestcalldroppingrate.NotethatoneDNcanhavemorethanonecandidatebecauseitcanconnecttodifferentPNs. Thereareseveralfactorsthataffectthehandoffperformance.TherstistheexpectedbandwidthofthedestinationBS.WiththeknowledgeofarrivalrateanddeparturerateofthecorrespondingBS,thesteadyprobabilityofeachstatecanbe 130

PAGE 131

HOD)=f(!i,!i,Rsv) wherePi(BS HOD)istheHODratederivedbasedontheknowledgeofreservationRsv,theinitial/handoffcallarrivalrate!ianddeparturerate!iofBSi,f()andg()denotefunctionsofP(BS HOD)andEstCellBW,respectively. Thesecondfactoristhehandoffcostintermsofsignalingtrafc.IfahandofftakesplacewithinthesameEANwithoutchangingBS,thehandoffcostisinfactverysmall.ItismoreappealingthanthosehandoffsthatchangesBS,PN,orevenEAN.WedenotetheHODprobabilitywhenonlyBSchanges,onlyPNchanges,andonlyDNchangesasP(BS HOD),P(PN HOD),andP(DN HOD),respectively.WithP(PN HOD)andP(DN HOD)ofeachcandidateobtainedfrommeasurements,theone-timemulti-hopHODrateofeachcandidatecanbederivedviathefollowingformula. HOD)=1(1Pk(BS HOD))(1Pk(PN HOD))(1Pk(DN HOD)) IftheBSorthePNdoesnotchange,thecorrespondingHODratetakes0value. Thethirdfactoristhehop-countofthemulti-hopconnection.Ifthemulti-hopconnectionhashigherhop-count,itwillbemorevulnerabletodisconnect.Inthispaper,weassumeallEANnodeshavesimilarmobility,thuswecanderivetheconnectionfailureprobabilityviathefollowingformula. fail)=1(1Pk(one hop fail))hop countsk(6) ThelastfactoristherelativepositionofdestinationBStotheMN'smovement.IftheBSlocatesalongthedirectionoftheMN'smovement,theMNcanavoidsomeunnecessaryhandoffsinthefuture,thusthecalldroppingprobabilitycanbefurther 131

PAGE 132

HOD)=(1P(BSijBSh)) where 6 ). HOD)(1Pk(one HOD)))(6) AfterwecalculatethePk(call dropping)foreachcandidatek,wesimplychoosethehandoffoption(thecombinationofBSandPN)withthesmallestcalldroppingrate.Althoughthereexistotherfactorsthatcanaffectthehandoffperformance,likeQoSsatisfaction,wewillnotpresentthedetailsherebecausewethinkwecaneasilyextendourframeworktoaccommodatemorefactors. 132

PAGE 133

InMCNs,agivenHODrateistheinputofareservationschemedesign.ThisinputspeciestheexpectedHODrateforeachMNwhenitisphysicallywithinacertaincell.WedenotetherequiredHODrateforeachMNincertaincellasPr(HOD).OurgoalistondtheminimumfeasiblereservationRsv.Inotherwords,weneedtondtherelationshipbetweenPr(HOD)andP0inEq.( 6 ).DifferentfromwhatisdescribedintheANHOAscheme,Pr(HOD)cannotbederivedfromcertaincandidates'informationbecauseitisanexpectedvalueforthewholecell.WeassumeeachhandoffMNrstattemptstoaccessitscurrentcellandthentrytheneighboringcellsonebyonewithacertaintimelimit.Therefore,besidesP0,theHODratesoftheadjacentcellsarealsoneededinndingPr(HOD). Intheproposedalgorithmofndingtheminimumreservation,eachBSisrequiredtocollecttheHODrateinformationoftheadjacentcells.Tosimplifytheanalysis,weassumethateachBSonlyconsidersthemulti-hophandoffstoitsone-hopneighbors.TheperiodicmessageexchangeamongtheneighboringBSscanprovideamechanismtoacquirethisinformation.BSsusethepastarrivalrate,departurerate,andthereservationRsvoftheneighboringcellsasthepredictivevalueofnexttimeinterval.Accordingtothepreviouslymeasuredvalue,andthecurrentreservationRsv,eachBScalculatesitsownone-attemptHODrateandbroadcastthisvalue.Notethat 133

PAGE 134

Withthesebroadcasted~Piandmeasuredqi,wecanndtheexpressionofthenalHODrateasinEq.( 6 ).WhilewetaketherequiredHODrateasPr(HOD),wecanderivetheminimumreservationbyincorporateformula( 6 ).WeuseNtodenotethemaximumnumberofattemptsafterthedirecthandoffattemptfailsandMtodenotethenumberofneighborsofthecurrentcell. 134

PAGE 135

6 ),therequiredHODrateissimplythesummationoftheprobabilitiesthatafterahandofffailsinthecurrentcell,itfailsindifferentnumbersofothercells. InactualMCNswithANHOAscheme,roamingMNsattempttoconnecttotheproperBSsinamoreintelligentwayratherthanconnectingtotheadjacentcellsbyusingtrialanderrorapproach.Therefore,theoverallHODratecanbereducedfurther.Moreover,arejecteddirecthandoffcallcanattempttoconnecttonotonlytheadjacentBSsbutalsoBSsfurtheraway,aslongasthemulti-hoppathsexist.Eq.( 6 )considersonlytheadjacentBSsinordertosimplifythederivationoftheminimumreservation. LetusrevisitEq.( 6 ).Wheneachqihasvalue1,whichmeansMNsinthecurrentcellcanaccesstheresourceofalltheadjacentcells,thissystemcanbeseenasalargercellconsistingofalladjacentcellswithaggregatedresourcefromalladjacentcells.Obviously,thissystemhashighertrunkingefciency,whichcanlowerthecalldroppingrateandtherequirementforresourcereservation.Whenmorethanonehopneighboringcellsareconsidered,evenhighertrunkingefciencyandlowercalldroppingratecanbeexpected. 135

PAGE 136

6 )isnotvalidanymore.Whenoneofthecellsintheneighborhoodhasbeenoverloadedwithtrafc,thisproblemmightbeinevitableandevenmoreprotuberant.Therefore,acertainthresholdshouldbesettoensurethatnoneofthecellisoverloaded.Furthermore,loadbalancingmechanismisnecessaryinMCNstoutilizethewholeresourcemoreefciently.Whenonecellisheavilyloaded,loadbalancingmechanismcanutilizetheadjacentcells'reservationtomaintaintherequiredHODrate. Inthispaper,webuildaframeworkforthispurpose.Inthisframework,thetrafcinformationinadjacentcellsisexchangedandsharedandtheminimumreservationiscalculatedinadistributedfashion.Moreimportantly,whensomecellisheavilyloaded,thisframeworkprovidesamechanismforBSsfromothercellstotakeoverpartofthetrafcsothateachcellcanmaintainarelativelylowHODrate,~Pi.ThisframeworkiscalledTrafcInformationExchangeforBSs'ReservationCalculationProcedure.ThereservationcalculationisdoneperiodicallybyeachBS.Betweentwoconsecutiveperiodsofcalculation,therearethreephases:informationcollection,loadbalancing,andreservationcalculation/informationbroadcasting. Intherstphase,eachBScollectsitsneighbors'trafcloadinthepreviousperiod,theneighbors'~Pi,andthecorrespondingaccessprobabilityqi.EachBScanpredictthetrafcloadaccordingtothepreviouslymeasuredtrafc,whichconsistsoflocaltrafcandpermeatedtrafcfromadjacentcells.ThetrafcloadismeasuredinErlang,i.e.,trafcintensity.Ifthetrafcloaddoesnotexceedapredenedthreshold,th,thesecondphasecanbeskipped.Otherwise,theoverloadedBSrequiresitsneighborstotakeoverpartofitstrafcload.TherequeststodifferentneighboringBSswillbebasedontheaccessprobability,qi.Whenrequestshavebeengranted,indicationwillbe 136

PAGE 137

Duringreservationcalculationphase,eachBScalculatesitsownP0beforehand,accordingtothelatesttrafcloadandthereservation.IfanyBSisrequestedfortakingoverpartialtrafcfromcurrentBS,thecalculationofcorrespondingP0shouldincludethispartialtrafc.With~Piandqiknown,theminimumreservationcanbecalculatedviaEq.( 6 )andEq.( 6 ). Firstly,welookintotheHODratewhenthereservationofeachBSremainsunchanged.Westudyacellularsystemwithseven-cellfrequencyreuse.Forasinglecell,wecalculatetheHODrateaccordingtoEq.( 6 ).ThedetailedderivationisbrieyintroducedintheAppendix,basedonthemodelin[ 26 ].ThebasicsettingofresourceparametersislistedinTable 6-4 EachBScalculatestheoverallHODrateaccordingtoEq.( 6 )withtheknowledgeofitsadjacentBSs'single-cellHODrates.Theaccessprobabilitytotheneighboringcellsaresetasthesamevalue.ThiscalculationgivesthestatisticalHODrateinacell.Thechoicesofmulti-hophandoffsareassumedtobechosenrandomly. 137

PAGE 138

SimulationSettingforEachCell NumberofChannels WithANHOAscheme,MNsareexpectedtoachievebetterperformancebecausemorehandoffopportunityandtheintelligenthandoffdecisionalgorithmshouldmitigatetheongoingconnectioncalldrop. Figure6-4. HODrate InFig. 6-4 ,theoverallHODrateofthecentercellhasbeenshown.Part(a)showstherelationshipbetweentrafcloadandHODrate.FromPart(a),wecanseethatwhentrafcloadincreases,theHODrateincreasesaccordingly.Withthehelpofadjacentcells,HODratecanbereducedgreatly.FromPart(a),wecanseethatevenwithasmallaccessprobability,suchasq=0.25,theHODratecanbeimprovedwith20dB.WecanalsoseefromPart(a),whentrafcloadincreasestoacertainlevel,theHODratewilldeterioratebadly.Thereasonisthatthecurrentreservationcannotsupportthe 138

PAGE 139

Figure6-5. ChannelReservation Secondly,weevaluatehowthechannelreservationisrelievedwiththehelpofadjacentcells.Lowerreservationcanalsogreatlyreducethecallblockingrateofnewcalls.Inthispartofsimulation,wesettherequiredHODrateto0.5103.Underthisconstraint,eachBSndstheminimumchannelreservation.InPart(a)ofFig. 6-5 ,thenumbersofreservedchannelsareshownwithdifferenttrafcloads.Wecanseethatwithoutthehelpofadjacentcells(SingleCellCaseinthegraph),morechannelsneedtobereservedtoachievetherequiredHODrate.Whentrafcloadbecomeslarger,reservationdoesnotworkfortherequiredHODrate.InPart(a)ofFig. 6-5 ,30reservedchannelsstandforthiscase.Withtheincreasingaccessprobability,thechannelreservationcanbegreatlydecreased.Inthecaseofq=0.35,undermostofthetrafcload,thereisnoneedforchannelreservationtomeettheHODraterequirement.Part(b)ofFig. 6-5 showsthecorrespondingcallblockingratewhentheminimumreservation 139

PAGE 140

26 ],the0reservationinthissimulationmeansthateventhereisnochannelreservedforthehandoffcalls,thesharedchannelscanstillsatisfytherequiredHODrate. 26 ],HODrateinacutoffprioritizedreservationsystemcanbemodeledasanite-stateMarkovchain. Figure6-6. MarkovChainModelforHandoffReservationSystem InFig. 6-6 ,Sidenotesthestateofichannelsbeingoccupied.NandHOstandforthearrivalrateofnewcallsandhandoffcalls,respectively.CisthetotalnumberofchannelsandRisthenumberofreservedchannels.isthedeparturerateforbothtypesofcalls.Inthismodel,handoffcallsandnewcallsstarttousethesharedchannels 140

PAGE 141

Wecanwritethestateequationsasthefollows. ThecallblockingprobabilityisequaltothesummationofprobabilitiesthatthestatesoccupystateCRtostateC. 141

PAGE 142

Inmoderncities,populationbecomesdenseranddenserandsodothePCSdevices.Peopletendtomovetogetherduetotheirsimilarworkingandlivingpatterns.Additionally,occasionalbigeventsattractlargenumbersofpeopletogathertocertainlocationsatacertaintime.Toreachalocation,peopletendtotakepublictransportation,especiallyinthemetropolitanareas,becauseofenergyandtrafcconcerns.Themovingmasseseasilycreatebiguctuationofthepopulationofcertainlocations.Consequently,peopleliveinachainofcrowdedenvironments,suchasofcebuildings,stores,subwaystations,stadiums,theaters,etc.Wecaneasilyfailtoreachacontactwhentheotherendisinacrowdedenvironment.Forthecellsalongthelocation 142

PAGE 143

Researchershavenoticedthisproblemandsomemethodsareproposedtoaddressit.F.Wangetal.proposeagroupLUschemeforcellularsystemsin[ 60 ].Y.Zhangetal.proposeabufferingschemetorelievethecongestioncausedbybatcharrivalofLUrequests[ 74 ].However,althoughtheseschemesareinspirational,theycannoteffectivelysolvethisproblem.FortheindispensableLUrequests,themosteffectivewaytomitigatethepotentialwirelesscongestionistoaggregatethesemessagesandreducethechannelcontentions.Unfortunately,observingthearchitectureoftheLMsignalingintraditionalcellularnetworks,wecanseethattherigorousone-hopwirelessstructureforbidsthepossibleaggregationofwirelesssignaling. Fortunately,cellularsystemsevolvethemselvesinthemeantime.Theintroductionofadhocmodetocellularsystemsshedslightonthepossiblesolutionofthisproblem. Theideaofintegratingadhocintocellularsystemscomesfromthemotivationthatcellularsystemsshouldbecomeanall-IPplatformsothatalltheserviceofnewwirelessnetworks(WiFi,WiMAX)canbeincorporatedtogetherwithoutdifculties[ 21 ].Theintegrationcanbeeitherbasedonadhocsystems,suchasCAMAin[ 9 ],orbasedoncellularsystems.SinceourfocusofthispaperisontheLMissues,weconsideronlythecellular-basedintegration.Thecellular-basedintegratedsystems,usuallycalledMulti-hopCellularNetworks(MCNs)[ 41 ][ 42 ][ 12 ][ 36 ],areexpectedtobringtheadvantageofadhocnetworksintocellularsystems,suchasexibility,robustness,self-organizing,andlowcost. InMCNs,eachMTcanbeothers'relaynodeandtheyforwardsignalinganddatapacketsforeachother.Sincetheadhoclinksareallowedtouseextraspectrum 143

PAGE 144

OurresearchinLMincludesthecostanalysisaftertheproposalsoftwoschemes.WeformulatethecostfunctionconsistingofLUpartandpagingpart.BychangingtheLUpartofthecostwithourproposedscheme,wecanndthenewoptimumlocationareasizeinMCNswithouranalyticalmodel. Therestofthepaperisorganizedasfollows.Section 7.2 discussestherelatedworks.Section 7.3 givesouttheunderlyingsystemmodel.Section 7.4 presentsthegroupingscheme.Section 7.5 proposesthegenericaggregationscheme.Section 7.6 showsthesimplewaytoformulateLMcostandndtheoptimumlocationareasize.Section 7.7 providestheperformanceevaluation.Conclusionsaregiveninthenalsection. 144

PAGE 145

44 ]andalleviatethesignalingburdenofHLR(homelocationregister).Dynamicschemescanreportthelocationupdatemoreexiblyinmorevariousways.Distance-basedapproachrequiresthesystemtorememberthelocationofoneMT'slastLUandwhenthemovingdistanceexceedsacertainthresholdanewLUistriggered.Thisapproachhasbeenstudiedin[ 62 ]extensively.Movement-basedapproachisanotherdynamicLUapproach[ 3 ],wheretheLUistriggeredwhenthemovementofacertainMTreachesathreshold.Besidesthese,Misraetal.proposedaninformationtheory-basedLMforverticalroamingusersin[ 45 ]. Forthesignalingcostofpaging,Zangetal.proposedtodelaythepagingwhichleadstoareducedpagingcost[ 67 ]. However,forthepotentialLUcongestionproblem,therearenotsomanyworksexistingincellularliterature.F.Wangetal.proposeagrouplocationupdateschemeforcellularsystemsin[ 60 ].Bylettingthegroupheadreportthelocationofitsmembers,wirelesscostcanbegreatlyreduced.However,eventhoughutilizinggroupingmethodcandispersethelocationupdaterequests,theincurredextragroupingmaintenancesignalingmakesthisdispersionlessattractive.Additionally,thebasestation(BS)isnotagoodchoicetotakechargeofthegroupingmaintenancesinceitisdifcultandinefcientforBSstogroupthemobileuserswithsamemovement.Y.Zhangetal.proposeabufferingschemetorelievethecongestioncausedbyheavyarrivalsofLUrequests[ 74 ].BylettingBSsbufferpendingLUrequestsduetotheexhaustionofDCCH(DedicatedControlChannel),LUretryattemptscanbereducedandhencethecongestioncanberelievedtosomeextent.ThismethodattemptedtospreadtheLUrequestsalongthetimelinebystoringthearrivedLUrequestsinBSs'buffers.However,thecongestioninRACHcannotbeeffectivelyrelievedwiththisbufferingapproach.Hanetal.similarlyproposeagroupLMschemeforone-dimensionalnetworks(transportationsystems)in[ 25 ].Byspecifyingvirtualvisitorlocationregister(VVLR)astheregisterofgroupinformationandanadditionaltierbetweenVLRandHLR,the 145

PAGE 146

32 ].Thefocusisonacceleratinghandoffandcellre-selectionprocedurebypredictingthelocationsofMTsmovingalongtherailways. LocationManagementPreliminary Beforewestarttodescribetheschemes,letusrefreshsomepreliminariesabouttheLM.AsshowninFig. 7-1 ,inLTE(longtermevolution)architecture,wheneachMTdetectsthechangeoflocationarea,itreportsthelocationtoHSS(homesubscribersystem,HLRinLTE).ItrstcontendstoaccesstheRACHandthenaskstheeNB(enhancednodeB,BSinLTE)forthechannelassignment.BesidesLUtrafcLU,newcallsNandhandoffcallsHOalsocontendsfortheRACH.ThetotaltrafcinRACHRACHcanbeexpressedasEq.( 7 ). 146

PAGE 147

Asaforementioned,thecostofLMisusuallyregardedasonlythewirelesscost.Therefore,weignorethecostofwirelinesignalinginouranalysis.AmongthewirelesscostofLUandpaging,sincetheyconsumetheresourceofdifferentchannels,theyarenotequivalentintheoverallcost. LUisnothappeningineachcellsincethelocationareaconsistsofseveralcellsandonlywhenthelocationareachanges,isthelocationupdatenecessary.However,pagingprocessconsumesthepagingchannelofallthecellswithinthelocationarea.Intheproposedschemes,thepagingprocessinMCNsinnotchanged.TheeffortwearemakingistoreducetheLU'sconsumptiononRACH. 7.4.1Overview WhenaHCTvehiclearrivesatanewlocation,theMTsinsidestarttoupdatetheirlocationstogether.TheseLUrequestscaneasilyover-injecttheRACHandcausethecongestions.Apparently,themostdirectwaytosolvethisproblemistogrouptheseMTsandrelievethecontentionsofRACH.Previouswork,GLU[ 60 ],proposedtousegroupingmethodin3Gand3GbeyondsystemstoreducetheLMcost.Differentfromit,weproposeanewgroupLMschemeforMCNswhichutilizesspecialMTsforgroup 147

PAGE 148

Ingroupingapproaches,therearetwohardproblemswhichrequirethedesignerstothinkover.Oneishowthegroupsareformedandtheotherishowtoreducethegroupmaintenancesignalingcost.Moreover,duringtheLUprocess,thesystemswillallocatenewtemporaryidentitiestoreplacetheoldtemporaryidentities.Thegroupingschemeshouldhavethecapabilitytofulllthistaskaswell. 148

PAGE 149

Weproposetouseaguardingperiodastheconditiontostarttheassociationwithanynewgroupmember.WedenotethelengthofthisperiodasTag.EachGHusesitsadhocinterfacetocommunicatewithitscurrentmembersandatthemeantimeprobestheexistenceofprospectivemembers.AtimerwiththelengthofTagwillbestartedwhenanewMTisemerging.IfthisMTisstillinrangeafterTagexpires,theGHwillconrmittodeposititsLMtask.Duetotheimperfectofwirelesschannels,thelinksbetweenGHsandthegroupmemberscanbeinterferedorevenbroken.Therefore, 149

PAGE 150

Thesetwoguardingtimersareusedtomakethegroupconstituentstable.ThevalueofTdgdependsonthechannelcondition.Therefore,itisbettertouseapracticalvalueforthistimerorallowthesoftwareconguration.Ontheotherhand,thevalueofTagdecidesthegainofusinggroupingapproach.IfTagistoosmall,manypassing-byMTscaneasilyassociatewiththeGHsassumingthateveryonecongurestheassociationasautomaticmode.Theunnecessaryassociationswastethechannelresourceandthegainofusinggroupsdecreases.IfTagistoobig,itisdifcultforMTstobecomegroupmemberswhichleadstoasmallergroupsize.Withasmallergroupsize,thegainofgroupingisnotfullyexploited.WewillleavetheprocessofndingpropervalueofTaginthefollowingsection. Besidestheguardingtimers,scenariosofco-existingsignalsofmultipleGHsshouldbealsoconsidered.Thestrategyweproposeisthesimplestone.IftheMTispreviouslyassociatedwithoneGH,itwillmaintaintheassociationevenifsomeotherGHhasstrongersignalstrength.IftheMTisnotassociatedwithanyGHwhenmultipleGHsareavailable,itchoosestheonewiththestrongestsignalstrength.IfthelinkoutagedurationreachesTdg,boththeGHandtheMTdeemthemselvesdissociated.AlthoughinsomescenariosthisstrategywillleadtowrongGHselection,suchaswhenapassengerchangestotheconnectinglinenearby,theassociationwillbecorrectedaftertwovehiclesseparateandthedelayedassociationwillnotimpactthesystemperformance. 7-2 .FirstitsendsadepositrequestmessagetotheGHusingitsadhocinterface. 150

PAGE 151

GroupAssociationProcedure AftertheguardingdurationTag,theGHresponseswithitsidentitywhichisusedbytheMTafterwardstoverifytheGHandreporttheassociationrelationshiptotheeNB.Withthisprocesscompleted,theGHtakesovertheLMtaskofthecorrespondingMT,includingtheregularlocationupdateandperiodiclocationupdate,andthesystem(eNBandMME)hastheknowledgeofthegroupandtheassociatingMT.NotethatthecommunicationsbetweenMTsandGHsareviaadhoclinksandthecommunicationsbetweenGHs/MTsandeNBsareviacellularlinks. Periodically,theupdatedgroupinformationinoneMMEwillbebroadcastedtoitsneighboringMMEs.SinceHCTvehiclesalwaysfollowacertainpath,thebroadcastcanbereplacedwithmulti-casttoreducetheunnecessarycommunications.TheseinformationupdatesdonotinvolvetheHLR(orHSS). 151

PAGE 152

38 ].However,withthegroupunderconsideration,thesignallingoverheadinourschemeisfarless. Uponarrivingatthenewlocationarea,theGHlaunchesthelocationupdateprocedurewiththegroupID.Iftheconstituentofthegroupischangedsincelastreport,alistofcurrentmembersissentoutwiththeLUrequest.Ifthereisnochangeofthegroup'sconstituent,agroupIDisenoughforthegrouplocationupdate. ThenewMMEchecksouttheMTsrequiringLUaccordingtothepreviouslyreceivedgroupinformationandthensendstheseLUrequeststothecorrespondingHSS. SinceallthegroupmembersstayinthesameoldLAbeforeenteringthenewLA,theirprocessoftemporaryID(TMSIinGSM)reallocationcanbegroupedaswell.IfsomeMTsarerequiredwithsomespecialprocess,suchasIMSIverication,thefollowingprocedurescanbecarriedoutinanindividualway.Asofthesystemperformance,thegroupLUrequestshavemitigatedtheRACHcongestionproblemnomatterthefollowingproceduresarecarriedoutinagroupedwayornot. WhentheHCTvehiclearrivesatthedestination,theGHfullsthedissociationtaskwithallthegroupmemberstocleartheinformationstorageinMMEs. 152

PAGE 153

7.4.3.1Signalingcostreduction 7-1 ,ifweutilizethegroupingmethodtoreportthelocation,multiplelocationupdaterequestscanbecombinedasone.TheutilizationofgroupingcanrstenhancethesuccessrateofLU.Asweknow,LUrequestscontendtheRACHwithnew/handoffcalls,followingslotted-ALOHAMACmechanism.WedenotetheLUtrafcrateasRLUandcalltrafcasRc,respectively.TheaveragedgroupsizeisdenotedbyK.TheimprovedLUsuccessratio(weassumeLUfailonlybecauseofRACHcongestion)canbeexpressedasfollows. K)(7) WecanseefromEq.( 7 ),thatwhentheLUtrafcoverwhelmstheRACH,groupingcangreatlyenhancetheLUsuccessrateaswellasthesuccessrateofnew/handoffcalls. InEq.( 7 ),thetrafcofLUisapproximatelyreducedtoKtimeswhengroupingisexecuted.However,theexactwirelesscostoflocationupdateshouldbeanalyzedinordertoevaluatetheperformancequantitatively.WerstinvestigatethewirelesscostofregularLUwithoutgroupinginonelocationareawithMMTsinunittimeduration.TheperiodicalLUcostisnotincludedinthisanalysisforthesimplicitypurpose.TheaveragespeedoftheHCTvehiclesisassumedasvHCTinnumberofcellsperunittime.Weassumethelocationareaconsistsofaclusterofcellswhichformaseriesofconcentrichexagonswiththeradiusofdcells.TheexpectedregularLUsignalingcostcanbeexpressedasfollows. 153

PAGE 154

Withourgroupingmethod,locationupdatetrafcisreducedKtimesalongwiththeincurredMT-eNBregistrationtrafcandtheGH-eNBgroupinformationupdatetrafc.Sinceadhocspectrumisnotcriticalformobileuserstoaccessthesystem,weignorethewirelesscostinadhocspectrum.Foranalysis,weassumethatallMMTsaremovinginKgroupstogether,creatingtheKtimes-reducedlocationupdatetrafc,thegroupinformationupdateforeachgrouponceeveryinterval,andtheassociationtrafcforeachMTonceeverytrip.Ldenotestheaveragetriplengthofpassengersintheunitofnumberofcells.TgristheGHs'intervalofreportinggroupinformationupdate.!gmand!assstandfortheweightsofgroupmaintenancemessagesandindividualassociationmessages,respectively.Therefore,wecanhavethewirelesscostofgroupLMasfollows. KvHCT K1 Thewirelessgainthroughgroupingisherebydenedasfollows. TherstiteminEq.( 7 )standsforthegroupLUtrafccost,whichisreducedKtimescomparedtothetraditionallocationupdate.Sincedifferentmessagesinwirelesshavedifferentimpactonsystemperformance,weputdifferentweightsforthem.ThisparthasunitweightinordertobeconsistentwithEq.( 7 ).TheseconditemstandsfortheincurredtrafcofupdatinggroupinformationbyGHs.ThelastitemistheMTs'associationcost,whichneedsonlyonceduringthewholetrip.Obviously,withtheaimofmitigatingcongestion,!gmand!assshouldbelessthan1.ItcanbeobservedfromEq.( 7 )thatwhengroupsizeKandLarebig,thetotalcostisgreatlyreduced.WecaneasilyconcludethatwhenKisrelativelybigger,thewirelessgainismainlydetermined 154

PAGE 155

Besidesthisgain,groupinggreatlydispersesthesignalingalongthetimeline,leadingtothemitigationofRACHcongestion,whichisamoreimportantbenetofgroupingmethod. ThereisnoavailablecurrentmodelwhichcandenethefunctionsofLandKintermsofTag.Thisisataskwhichnecessitatesmanyrealdata.Supposeweacquiretherelationship,i.e.L=f(Tag)andK=g(Tag),wecanderivetheoptimumTagwithEq.( 7 ),( 7 )and( 7 ). 7.5.1Overview 155

PAGE 156

Inthispaper,weproposetoutilizetheaggregationdevices(ADs)alongthelocationareaboundariestocollectthelocationupdaterequestsviaadhoclinksperiodically.AftereachtimeintervalTaggr,theseADscombinethecollectedlocationupdaterequestsintoonemessageandsendtotheeNBssothattheRACHcongestioncanbegreatlymitigated.EacheNBisrequiredtodecidetheTaggrfortheADsconnectingtoitselfaccordingtothetrafcloadinthecell.ToencouragemobileuserstodelegatetheirLUstotheADs,Taggrshouldbesetsmallerthantheexpecteddelaycausedbyprospectivecollisionsandretransmissions.ShorterdelayofLUmeanssmallerprobabilityofmissingincomingcalls.Therefore,ADswhichcanprovidesmallerdelaywithaproperTaggrcanattractMTstodelegatetheirlocationupdatesandthusmitigatetheusageofRACH. SincetheseADsperiodicallyaccessthesystemtodotheLUformultipleMTs,eNBscanfurtherallocatexedchannelresourceforthemandthentherepeatedcontentionforRACHisunnecessary.NextwedescribethegenericschemeofaggregativeLMinMCNs. 156

PAGE 157

Figure7-3. GenericAggregativeLocationManagement TheADsareappointedbytheeNBsinthisscheme,althoughmoreexibleanddynamicmethodscanbeadoptedinthefuturetomanifestthebenetsofadhocmode.TaggrofeachADwithinthesamecellissettothesamevaluebytheeNB.EachcelladjuststhevalueofTaggraccordingtotheinformationofLUtrafcandothertrafcinRACHoflastperiod.EachMTobtainsthetrafcinformationofRACHfromBCCH.ThecollisionprobabilityPcolcanbederivedwiththeknowledgeofaveragedpacketdurationinsecond,denotedas,andthearrivalrateofRACHtrafc,denotedasRACH,asshowinEq.( 7 ). 157

PAGE 158

7 ).TheseconditemofDRACHissimplytheproductoftheexpectedretransmissiontimesandtLU. ThetotalLUdelayDLUisthesummationofDRACHandwirelinesignalingdelayDwire,whichcanbeexpressedasDLU=DRACH+Dwire.Furthermore,withtheknowledgeofcall-to-mobileratio,wecanknowtheexpectednumberofmissedcalls. OndecidingtheaggregationperiodTaggr,werstneedtondtheconstraints.DALUdenotesthewirelessdelayoftheproposedaggregativeschemewhichismainlyduetotheaggregationinterval.DCSMA,theCSMAaccessdelay,alsocontributingtoDALU,canbeignoredwhenthetrafcislight. 7 ),wherePfisthefailurerateofonetransmissioninCSMA,Wiisthebackoffwindowsizeof(i+1)thtransmission,andEIFSisthetimeoutlengthforonefailedtransmission. Obviously,whenDALUhassmallervaluethanDRACH,i.e.DALU
PAGE 159

Thereisanotherconstraintoftheaggregationinterval,whichisthattheaggregatedtrafcshouldbenomorebeyondthecapacity.AlthoughtheperiodicalaggregatedlocationupdateviaADscanbypassthecontentionofRACH,thereisstillanupper-boundofacceptableLUtrafc,denotedas^LU,whichisrelatedtothetrafcofnewcallsandhandoffcalls.RegardingtheaggregativeLUtrafcasthePoissonprocess,wehavethesecondconstraintexpressedasfollows.NADdenotesthenumberofADs,LUistheLUtrafc,andistheaggregationrateoftheLUtrafc. Ineq.( 7 )indicatesthat,whenalltheLUtrafcgoesthroughADs,i.e.=1,largeraggregationintervalcangathermoreLUtrafcandreducethewirelesscostmore. Withthesetwoconstraints,wecanroughlydecidetheaggregationinterval.ItshouldbenobiggerthanacertainvalueasinIneq.( 7 )inordertostimulatemobileuserstouseADs.ItalsoneedstobenosmallerthananothercertainvalueasinIneq.( 7 )sothattheaggregativetrafcwouldnotexceedthehandlingcapabilityoftheeNB.AlthoughsomecertainADsmaynotcollectanyLUdelegationduringsomeperiodsduetoskewedLUtrafcdistribution,weignoreintheanalysisthepossibilitythattheseADswouldreleasetheassignedchannelsandre-contendtheRACHwhennewLUtrafcarrives.InthecasethatthevalueofTaggrcannotsatisfybothinequalitiesatthesametime,wesimplyremovetherstconstraints(Ineq.( 7 )).ThiscasemeansthattheLUtrafchasnotcongestedRACHandthedelayviaRACHisacceptablysmall.Therefore,thereisnoneedtostimulatetheaggregationbyblindlyshorteningthe 159

PAGE 160

7 )required. Accordingtothevaluedecidedbasedontheabovetwoinequalities,wecanhavethewirelesscostgainasfollows. ToformulatethecostfunctionofLM,weneedtohavetheknowledgeoftwoparameters.Oneisthecall-to-mobileratio,,whichshowstheprobabilitythatamobileuseristobecalled.Theotheristheaveragedmovingspeed,v,whichshowshowsoonanMTistochangethelocationareainaverage.Weassumethatthelocationareaconsistsofaclusterofcellswhichformaseriesofconcentrichexagonsasaforementioned.Wedenotetheradiusasdinnumberofcells.Fortheconvenienceofunderstandingwealsodenetheaveragenumberofusersinthelocationarea,M,andthetimeintervalforthecostmeasurement,. TheLMcostconsistsofthewirelesscostoflocationupdateandthewirelesscostofpaging,asshowninEq. 7 Tominimizetheoverallcost,weneedtoputdifferentweightsonthesetwoparts.!LUand!pagingstandfortheweightsforLUandpaging,respectively.Apparently,since 160

PAGE 161

LUshappenwhentheMTsdetectthelocationareachanged.AssumingeachMTmovesstraightlywiththeaveragespeedv,wecanexpectanLUtooccuratmostafteritmovesacrossthediameterlengthofthelocationarea,whichis2d1.Therefore,forMMTsinduration,thewirelesscostofLUcanbeexpressedasfollows. DifferentfromLU,pagingwillhappenateachcellwithinthelocationarea,wheneverthereisanincomingcallorpacket.NotethatMandareinbothequationsandhencecanberemovedbeforefurtheranalysis. ByndingtherstandsecondderivativeofCtotal,wecanobtaintheminimumvalueofCtotalandthecorrespondingd. Sincethesecondderivativeispositive,bylettingtherstderivativeequalto0,wecanknowdwhichismakingCtotalminimum. !LUv+1 2(7) 161

PAGE 162

7 )andEq.( 7 ).ThesignalinggainisdenesinEq.( 7 ).ThebasicsettingsarelistedinTable 7-1 Table7-1. Parameters'SettingsForEvaluatingTheGroupingScheme 7-4 showstheresult.WecanseethatthegainsincreasewiththegroupsizeK,however,notlinearly.BiggervHCTbringsbiggergainsincetheperiodicalgroupreportingoccupieslessandlesssignalingcostcomparedwithothervHCTrelatedsignaling.Becausetheindividualassociation'scostisnotrelatedtoK,thegainsdonothavealinearrelationshipwithK. Figure7-4. GroupingGainsofSignalingCostWithDifferentGroupSizes(K) Fig. 7-5 showsthecomparisonofthegainswithdifferentaveragetriplengthL.WesetKas10,50and200forthecalculations.Apparently,biggerLalongwithbiggerKbringsbiggersignalingcost'sgains.WhenKisasbigas200,thegainsalmostincrease 162

PAGE 163

Figure7-5. GroupingGainsofSignalingCostWithDifferentTripLength(L) AlthoughbigKandbigLtogetherbringgreatgains,theyusuallydonotcometogether.Wehavementionedearlierthatinacertainscenario,whenwesettheassociationtimerwithabigvalue,thebigaveragetriplengthcanbeacquiredwhereaswiththegroupsizesacriced.Shortassociationtimerisonthecontrary.Therefore,ifwewanttooptimizethesignalingcost'sgain,weneedrstgatherrealdataandndtherelationshipbetweenK,Landthetimer. Forthegenericaggregationscheme,weneedtorstlookintotheintervalTaggr'srange.BasedonthelongestallowedTaggr,wethenobservethegainofsignalingcost.Beforewecanevaluatethescheme,welisttheparameters'settingasTable 7-2 Table7-2. Parameters'SettingsForEvaluatingTheGenericAggregationScheme tLU

PAGE 164

7 )andthelowerboundisdenedinIneq.( 7 ).BychangingtheLUtrafcload,wecanseethechangeoftherangeoftheaggregationinterval. Figure7-6. TheAggregationInterval'sRangeswithDifferentLUTrafc Fig. 7-6 showsthecalculationresultswithdifferentLUtrafcload.WecanseethatthelowerboundoftheaggregationintervaldoesnotchangeapparentlywiththeLUtrafc.Thisisbecauseonlynon-aggregatedLUtrafc,whichisonlyasmallproportion,affectsthelowerboundoftheaggregationinterval.TheupperboundoftheaggregationintervalincreasesexponentiallywiththeLUtrafcinthatitdependsonthecollisionprobabilityinaslotted-ALOHAchannel. Fig. 7-7 showsthesignalingcost'sgainofthegenericaggregationscheme.ThisschemedoesnotbringotherextratrafcandaggregatestheLUtrafconly.Bysettingtheaggregationtimerwithitsbiggestallowedvalue,wecalculatethegains.ItshowsthatfewerADscanapparentlybringbiggergains.ThisisbecausefewerADsarerequiredtoaggregatemoreLUtrafc.However,ifADsarenotenough,highaggregationratecannotbeguaranteedandmorecongestionswillhappeninADaccess. ItisworthytonotethatthisperformanceevaluationdoesnotincludeperiodicalLUforthepurposeofsimplicity.However,periodicalLUcanbeeasilyaggregatedwithouraggregativeapproachandthushighersignalingcostgainwillbeachievedactually. 164

PAGE 165

TheAggregationSignalingGainswithDifferentLUTrafc 165

PAGE 166

Wirelessadhocnetworks(WANETs)haveavarietyofapplications.Theycanappearinstand-aloneformssuchasMANETs,wirelesssensornetworks,orwirelessmeshnetworks.Theycanalsoappearasoneembeddedcomponentinhybridsystems,suchasMCNs.Nomatterinwhichforms,WANETshavemulti-hoptransmissionsandself-organizationastheiroutstandingcharacteristics.Thesecharacteristicsbringusthebenetsoflow-cost,convenienceandexibilityandinthemeantimetheyarethereasonsofsevereinterference,badspatialreuseandlowthroughputperformance.Itneedsalotofeffortstoemploythemefciently. ThisdissertationpresentstheeffortsonutilizingWANETsfromtwoaspects,throughputimprovementandmobilitysupport. AlthoughPHYlayertechniquesarethedrivingforceofraisingdatarate,itisMAClayerthatcontrolsspectrumefciencyinmulti-userscenarios.Duetothemulti-hopproperty,schedule-basedMACwillndsitlimitationinWANETsandcontention-basedMACappearstobeabetterchoice.Inmulti-channelsystems,channelassignmentalwayscostasubstantialpartofspectrum.ReducingthispartofsignalingwillgreatlyincreasetheMACthroughput.However,highMACthroughputdoesnotalwaysmeanhighend-to-endthroughput.Aow'send-to-endrateisdecidedbythehopwiththelowestrate.Therefore,itisimportanttoallocatemorespectralresourcetotheareawithheavytrafcload.Moreover,inmulti-hopWANETs,spatialreuseiscrucialtoresourceallocationsinceifmorespectrumcanbereusedthenlargerthroughputcanbeexpected.InthespecialcaseofWANETs,wirelessmeshnetworks,whereallthetrafcgoestothegatewaysnally,resourceallocationcanbedoneinasemi-centralizedsemi-distributedwaysothatthecentralareascaneasilyacquiremorespectrumandtheouterareascaneasilyreusethespectrum.Astepcloser,nodeswithheavytrafcinaneighborhoodshouldacquiremoreresourcethantheothersforthesame 166

PAGE 167

Duetothelackofinfrastructure,itisdifcultforWANETstosupportmobility,whichisusuallymanagedinahierarchicalstructure.Tosolvethis,wecanaddsomefunctionalmodulesintothesystemstobuildthehierarchicalstructure.Forexample,wecanaddforeignagents(FA)andhomeagents(HA)intowirelessmeshnetworkssothatmobileIPcanbesupported.However,mobileIPisnotasuitablesolutionforintra-domainroamingbecausefrequentlychangingIPbringsfrequentserviceinterruption.Wecanstorethelocations(servingmeshrouters)ofmeshnodesinthegatewayastheanchorsandusetemporaryroutingentriesbetweentheanchorsandtheexactlocationstoreachthedestinations.Thisapproachcombinestheadvantagesoftunnel-basedapproachandrouting-basedapproachandprovidesasuitablemicro-mobilitysolutionforwirelessmeshnetworks.WhenweintegrateWANETsintocellularsystems,wecanachieveWANETs'advantagesincellularsystems.Itisinterestingtolookintothemobilityissuesinsuchsystems.Withmulti-hopconnectionsexisting,roamingclientscanaccesstheresourceoftheadjacentcellsandbetterhandoffperformancecanbethusexpected.Ifweformembeddedadhocnetworksinthesystems,theinformationofhandoffoptionscanbecollectedandmaintainedbeforehandofftakesplace.Wecanchoosebetterhandoffoptionstoreducethehandoffdroppingrate.Wecanadjustthehandoff 167

PAGE 168

Surely,manymoreissuesneedtobesolvedinordertomakeuseofWANETswell.IamcurrentlylookingintotheroutingalgorithmsinWANETsandthinkingoftheadhocroutingissueswhenembeddingWANETsintocellularsystems. 168

PAGE 169

[1] WirelessLANMediumAccessControl(MAC)andPhysicalLayer(PHY)specications.IEEE802.11WorkingGroup,1999. [2] IEEE802.16e:AirInterfaceforFixedandMobileBroadbandWirelessAccessSystems.IEEE802.16WorkingGroup,2005. [3] Akyildiz,I.,Ho,J.,andLin,Y.Movement-basedLocationUpdateandSelectivePagingforPCSNetworks.IEEE/ACMTransactionsonNetworking(TON)4(1996).4:629. [4] Akyildiz,I.,McNair,J.,Ho,J.S.M.,Uzunalioglu,H.,andWang,W.Mobilitymanagementinnext-generationwirelesssystems.ProceedingsoftheIEEE87(1999).8:1347. [5] Akyildiz,I.,Wang,X.,andWang,W.Wirelessmeshnetworks:asurvey.ComputerNetworks47(2005).4:445. [6] Alicherry,M.,Bhatia,R.,andLi,L.Jointchannelassignmentandroutingforthroughputoptimizationinmulti-radiowirelessmeshnetworks.Mobicom'05.Cologne,Germany,2005. [7] Amir,Y.,Danilov,C.,Hilsdale,M.,Musaloiu-Elefteri,R.,andRivera,N.Fasthandoffforseamlesswirelessmeshnetworks.Proceedingsofthe4thinternationalconferenceonMobilesystems,applicationsandservices(MobiSys'06).Uppsala,Sweden,2006. [8] Bahl,P.,Chandra,R.,andDunagan,J.SSCH:SlottedSeededChannelHoppingforCapacityImprovementinIEEE802.11Ad-HocWirelessNetworks.Mobicom'04.Philadelphia,PA,2004. [9] Bhargava,B.,Wu,X.,Lu,Y.,andWang,W.IntegratingHeterogeneousWirelessTechnologies:ACellularAidedMobileAdHocNetwork(CAMA).ACMSpecialIssueoftheJournalonSpecialTopicsinMobileNetworkingandApplications(MONET)9(2004).4:393. [10] Bianchi,G.PerformanceanalysisoftheIEEE802.11distributedcoordinationfunction.IEEEJ.Sel.AreasCommun.18(2000).3:535. [11] Campbell,A.T.,Gomez,J.,Kim,S.,Wan,C.-Y.,Turanyi,Z.R.,andValko,A.G.ComparisonofIPMicro-mobilityProtocols.IEEEWirelessCommunications9(2002).1:72. [12] Cavalcanti,D.,Agrawal,D.,Cordeiro,C.,Xie,B.,andKumar,A.IssuesinintegratingcellularnetworksWLANs,andMANETs:afuturisticheterogeneouswirelessnetwork.WirelessCommunications,IEEE16(2005).3:30. 169

PAGE 170

Cho,S.,Jang,E.,andCiof,J.HandoverinMultihopCellularNetworks.IEEECommunicationMagazine47(2009).7:64. [14] Choi,N.,Seok,Y.,andChoi,Y.Multi-channelMACprotocolformobileadhocnetworks.VehicularTechnologyConference2003.Orlando,Florida,USA,2003. [15] Dong,Q.,Banerjee,S.,andLiu,B.ThroughputOptimizationandFairBandwidthAllocationinMulti-hopWirelessLANs.Infocom'06.Barcelona,Spain,2006. [16] Fang,Y.Modelingandperformanceanalysisforwirelessmobilenetworks:anewanalyticalapproach.IEEE/ACMTransactionsonNetworking(TON)13(2005).5:989. [17] Fang,Y.andZhang,Y.CallAdmissionControlSchemesandPerformanceAnalysisinWirelessMobileNetworks.IEEETransactionsonVehicularTechnology(TVT)51(2002).2:371. [18] Gambiroza,V.,Sadeghi,B.,andKnightly,E.End-to-EndPerformanceandFairnessinMultihopWirelessBackhaulNetworks.Mobicom'04.Philadelphia,PA,2004. [19] Group,3GPP.OpportunityDrivenMultipleAccess.3GTR25.924version1.0.0,2008. [20] Group,3GPP.GPPsystemtoWirelessLocalAreaNetwork(WLAN)interworking.3GPPTS23.234V7.7.0,2009. [21] Group,3GPP.All-IPNetwork(AIPN)feasibilitystudy.3GPPTS22.978V7.1.0,2009. [22] Gupta,P.andKumar,P.R.Thecapacityofwirelessnetworks.IEEETransactiononInformationTheory46(2000).2:388. [23] Gustafsson,E.,Jonsson,A.,andPerkins,C.MobileIPRegionalRegistration.InternetDraft,2000. [24] Haas,Z.J.andDeng,J.Dualbusytonemultipleaccess(DBTMA)-amultipleaccesscontrolschemeforadhocnetworks.Communications,IEEETransactionson50(2002).6:975. [25] Han,I.andCho,D.GroupLocationManagementforMobileSubscribersonTransportationSystemsinMobileCommunicationNetworks.IEEETransactionsonVehicularTechnology(TVT)53(2004).1:181. [26] Hong,D.andRappaport,S.TrafcModelandPerformanceAnalysisforCellularMobileRadioTelephoneSystemswithPrioritizedandNonprioritizedHandoffProcedures.IEEETransactionsonVehicularTechnology(TVT)35(1986).3:77. 170

PAGE 171

Hsieh,R.,Seneviratne,A.,Soliman,H.,andEl-Malki,K.PerformanceanalysisonhierarchicalMobileIPv6withfast-handoffoverend-to-endTCP.Globecom'02.Taipei,R.O.C.,2002. [28] Jiang,H.,Wang,P.,Zhuang,W.,andShen,X.AninterferenceawaredistributedresourcemanagementschemeforCDMA-basedwirelessmeshbackbone.IEEETransactionsonWirelessCommunications6(2007).12:4558. [29] Joo,C.andShroff,N.PerformanceofRandomAccessSchedulingSchemesinMulti-hopWirelessNetworks.Infocom'07.Anchorage,AK,2007. [30] Joshi,T.,Mukherjee,A.,Yoo,Y.,andAgrawal,D.AirtimeFairnessforIEEE802.11MultirateNetworks.IEEETransactionOnMobileComputing7(2008).4:513. [31] Kanodia,V.,Li,C.,Sabharwal,A.,Sadeghi,B.,andKnightly,E.DistributedPrioritySchedulingandMediumAccessinAdHocNetworks.WirelessNetworks8(2002).5:455. [32] Kastell,K.,Bug,S.,Nazarov,A.,andJakoby,R.ImprovementsinRailwayCommunicationviaGSM-R.VehicularTechnologyConferenceIEEE63rd.Melbourne,Vic.,2006. [33] Kim,S.,Kim,B.,andFang,Y.DownlinkandUplinkResourceAllocationinIEEE802.11WirelessLANs.IEEETransactionsOnVehicularTechnology54(2005).1:320. [34] Kodialam,M.andNandagopal,T.Characterizingthecapacityregioninmulti-radiomulti-channelwirelessmeshnetworks.Mobicom'05.Cologne,Germany,2005. [35] Kyasanur,P.andVaidya,N.H.Capacityofmulti-channelwirelessnetworks:impactofnumberofchannelsandinterfaces.Mobicom'05.Cologne,Germany,2005. [36] Law,L.,Krishnamurthy,S.,andFaloutsos,M.CapacityofHybridCellular-AdhocDataNetworks.Infocom'08.Phoenix,AZ,2008. [37] Le,L.andHossain,E.MultihopCellularNetworks:PotentialGains,ResearchChallenges,andaResourceAllocationFramework.IEEECommunicationsMagazine5(2007).9:66. [38] Levine,D.,Akyildiz,I.,andNaghshineh,M.Aresourceestimationandcalladmissionalgorithmforwirelessmultimedianetworksusingtheshadowclusterconcept.IEEE/ACMTransactionsonNetworking(TON)5(1997).1:1. [39] Li,B.End-to-EndFairBandwidthAllocationinMulti-hopWirelessAdHocNetworks.ICDCS'05.Columbia,OH,2005. 171

PAGE 172

Lin,X.andRasool,S.ADistributedJointChannel-Assignment,SchedulingandRoutingAlgorithmforMulti-ChannelAdHocWirelessNetworks.Infocom'07.Anchorage,AK,2007. [41] Lin,Y.andHsu,Y.Multi-hopcellular:anewarchitectureforwirelesscommunications.Infocom'00.Tel-Aviv,Israel,2000. [42] Luo,H.,Ramjee,R.,Sinha,P.,Li,L.,andLu,S.UCAN:AUniedCellularandAd-HocNetworkArchitecture.Mobicom'03.SanDiego,CA,2003. [43] Ma,W.andFang,Y.DynamichierarchicalmobilitymanagementstrategyformobileIPnetworks.IEEEJournalonSelectedAreasinCommunications22(2004).4:664. [44] .Apointerforwardingbasedlocalanchoring(POFLA)schemeforwirelessnetworks.IEEETransactionsonVehicularTechnology54(2005).3:1135. [45] Misra,A.,Roy,A.,andDas,S.Information-TheoryBasedOptimalLocationManagementSchemesforIntegratedMulti-SystemWirelessNetworks.IEEE/ACMTransactionsonNetworking(TON)16(2008).3:525. [46] Mo,J.,So,H.,andWalrand,J.Comparisonofmulti-channelMACprotocols.MSWiM'05.Montreal,Quebec,Canada,2005. [47] Modinao,E.,Shah,D.,andZussman,G.Maximizingthroughputinwirelessnetworksviagossiping.SIGMetrics/Performance'06.SaintMalo,France,2006. [48] Nasipuri,A.,Zhuang,J.,andDas,S.R.AmultichannelCSMAMACprotocolformultihopwirelessnetworks.WCNC'99.NewOrleans,USA,1999. [49] Perkins,C.E.MobileIP.IEEECommunicationsMagazine35(1997).5:84. [50] Rad,A.H.M.andWong,V.W.S.JointChannelAllocation,InterfaceAssignmentandMACDesignforMulti-ChannelWirelessMeshNetworks.Infocom'07.Anchorage,AK,2007. [51] Ramachandran,K.N.,Belding,E.M.,Almeroth,K.C.,andBuddhikot,M.M.Interference-AwareChannelAssignmentinMulti-RadioWirelessMeshNetworks.Infocom'06.Barcelona,Spain,2006. [52] Ramjee,R.,Varadhan,K.,Salgarelli,L.,Thuel,S.R.,Wang,S.-Y.,andPorta,T.L.HAWAII:adomain-basedapproachforsupportingmobilityinwide-areawirelessnetworks.IEEE/ACMTransactionsonNetworking10(2002).3:396. [53] Raniwala,A.andChiueh,T.ArchitectureandalgorithmsforanIEEE802.11-basedmulti-channelwirelessmeshnetwork.Infocom'05.Miami,FL,2005. [54] S.Ganguly,Navda,V.,Kim,K.,Kashyap,A.,Niculescu,D.,Izmailov,R.,Hong,S.,andDas,S.R.PerformanceOptimizationsforDeployingVoIPServicesinMesh 172

PAGE 173

[55] Shi,J.,Salonidis,T.,andKnightly,EdwardW.Mediumaccesscontrol:Starvationmitigationthroughmulti-channelcoordinationinCSMAmulti-hopwirelessnetworks.MOBIHOC'06.Florence,Italy,2006. [56] So,J.andVaidya,N.MultiChannelMACforAdHocNetworks:HandlingMultiChannelHiddenTerminalsUsingASingleTransceiver.Mobihoc'04.Tokyo,Japan,2004. [57] Tam,W.andTseng,Y.JointMulti-ChannelLinkLayerandMulti-PathRoutingDesignforWirelessMeshNetworks.Infocom'07.Anchorage,AK,2007. [58] Tzamaloukas,A.andGarcia-Luna-Aceves,J.J.Areceiver-initiatedcollision-avoidanceprotocolformulti-channelnetworks.Infocom'01.Anchorage,Alaska,USA,2001. [59] Valko,A.G.CellularIP:anewapproachtoInternethostmobility.ComputerCommunication29(1999).1:50. [60] Wang,F.,Tu,L.,Zhang,F.,andHuang,Z.Grouplocationupdateschemeandperformanceanalysisforlocationmanagementinmobilenetwork.IEEE61stSemiannualVehicularTechnologyConference.Stockholm,Sweden,2005. [61] Wang,Y.andSheu,J.Adynamicchannel-borrowingapproachwithfuzzylogiccontrolindistributedcellularnetworks.SimulationModellingPracticeandTheory12(2004).3-4:287. [62] Wong,V.andLeung,V.AnAdaptiveDistance-basedLocationUpdateAlgorithmforPCSNetworks.IEEEJ.SelectAreasCommun.19(2001).10:1942. [63] Wu,H.,Qiao,C.,De,S.,andTonguz,O.IntegratedCellularandAdHocRelayingSystems:iCAR.IEEEJournalonSelectedAreasinCommunications(JSAC)19(2001).10:2105. [64] Wu,S.,Lin,C.,Tseng,Y.,andSheul,J.ANewMulti-ChannelMACProtocolwithOn-DemandChannelAssignmentforMulti-HopMobileAdHocNetworks.ISPAN'00.WashingtonD.C.,USA,2000. [65] Xing,K.,Cheng,X.,Ma,L.,andLiang,Q.SuperimposedCodeBasedChannelAssignmentinMulti-RadioMulti-ChannelWirelessMeshNetworks.Mobicom'07.Montreal,Quebec,2007. [66] Yang,X.andVaidya,N.PrioritySchedulinginWirelessAdHocNetworks.WirelessNetworks12(2006).3:273. [67] Zang,H.andBolot,J.MiningCallandMobilityDatatoImprovePagingEfciencyinCellularNetworks.MOBICOM'07.Montreal,Canada,2007. 173

PAGE 174

Zhai,H.,Chen,X.,andFang,Y.ACallAdmissionandRateControlSchemeforMultimediaSupportoverIEEE802.11WirelessLANs.ACMWirelessNetworks12(2006).4:451. [69] .ImprovingtransportlayerperformanceinmultihopadhocnetworksbyexploitingMAClayerinformation.IEEETransactionsonWirelessCommunications6(2007).5:1692. [70] Zhai,H.andFang,Y.DistributedFlowControlandMediumAccessControlinMobileAdHocNetworks.IEEETransactionsonMobileComputing5(2006).11:1503. [71] Zhai,H.,Kwon,Y.,andFang,Y.PerformanceanalysisofIEEE802.11MACprotocolsinwirelessLANs.WirelessCommunicationAndMobileComputing4(2004).8:917. [72] Zhai,H.,Wang,J.,andFang,Y.DUCHA:ADual-ChannelMACProtocolforMobileAdHocNetworks.IEEETransactionsonWirelessCommunications5(2006).11. [73] Zhang,J.,Wang,Y.,andWang,J.DCC-MAC:anewMACprotocolforad-hocnetworksbasedondualcontrolchannel.PIMRC2003.Beijing,China,2003. [74] Zhang,Y.andFujise,M.LocationManagementCongestionProbleminWirelessNetworks.IEEETransactionsonVehicularTechnology56(2007).2:942. 174

PAGE 175

RongshengHuangreceivedhisBSandMSdegreesinelectricalengineeringfromXi'anJiaotongUniversity,Xi'an,China,in1996and1999,respectively.From1999to2001,hewasworkingwithHuaweiTechnologiesCo.Ltd.asanR&DengineeronGPRSand3Gprojects.From2002to2005,hewasworkingwithUTStarcomResearchCenter,Shenzhen,China,asaseniorengineerandteamleaderon3Gproject.Since2005,hehasbeenworkingonthePhDdegreeintheDepartmentofElectricalandComputerEngineeringatUniversityofFlorida.Hisresearchinterestscovertheareaofmediaaccesscontrol,mobilitymanagement,protocolandarchitecturedesignforwirelessnetworks.HeisnowastudentmemberoftheIEEE. 175