With Contributions of the following partners (in order of appearance in the document) Partn

A state of the art

InformationSocietyTechnologies(IST)-6thFramework

Programme

DeliverableNo:D.WP-JRA-2.4.1

QoSinmulti-servicewirelessnetworks

Astateoftheart

DeliverableVersionNo:

SendingDate:

Editor’sName:

LeadContractorNo:

Disseminationlevel:

URLReferenceofWorkpackage:Version1.02004/08/24AlexandreProutiere11Con dentialReporthttp://www.77cn.com.cn

A state of the art

ProjectAcronym:

ProjectFullTitle:

Typeofcontract:

contractNo:

ProjectURL:Euro-NGIDesignandEngineeringoftheNextGenerationInternet,http://www.77cn.com.cnworkofExcellence507613http://www.77cn.com.cn

A state of the art

WithContributionsofthefollowingpartners(inorderofappearanceinthedocument):

PartnerNum.andNameContributorNameContributoremail

FranceTelecomR&DT.Bonald

A.Proutiere

J.Roberts

N.Hegde

SalahEddineElayoubi

T.Chahed

A.E.Samhat

J.Martínez

V.Pla

S.Borst

L.Georgiadis

V.Tsibonis

T.Korakis

L.Tassiulas

C.A.Courcoubetis

M.Dramitinos

V.A.Siris

G.D.Stamoulis

D.Kouvatsos

I.Awan

S.Lucetti

LucaTavanti

A.Beben

M.Cesana

K.A.Hummel

HenriKoskinen

A.Penttinen

S.VanDenHeuvel

C.Blondia

D.Remondo

M.C.Domingo

R.Guimarães

J.Morillo

L.Cerdà

J.M.Barceló

J.Garcia

M.Welzlthomas.bonald@http://www.77cn.com.cnalexandre.proutiere@http://www.77cn.com.cnjames.roberts@http://www.77cn.com.cnhegde@eurandom.tue.nlsalah_eddine.elayoubi@int-evry.frtijani.chahed@int-evry.frabedellatif.samhat@int-evry.frjmartinez@upvnet.upv.esvpla@dcom.upv.esSem.Borst@cwi.nlleonid@eng.auth.grvtsib@egnatia.ee.auth.grkorakis@inf.uth.grleandros@inf.uth.grcourcou@aueb.grmdramit@aueb.grgstamoul@aueb.grD.D.Kouvatsos@scm.brad.ac.uki.awan@scm.brad.ac.uks.lucetti@iet.unipi.itluca.tavanti@iet.unipi.itabeben@tele.pw.edu.plcesana@http://www.77cn.com.cnkarin.hummel@univie.ac.athenri@netlab.hut. vpenttin@netlab.hut. sylwia.romaszko@ua.ac.bechris.blondia@ua.ac.beDavid.Remondo@upc.escdomingo@mat.upc.esrpaoliel@ac.upc.esjmorillo@ac.upc.esllorenc@ac.upc.esjoseb@ac.upc.esjorge@ac.upc.esmichael.welzl@uibk.ac.atEurandom,TheNetherlandsINT,FranceGIRBA/ITACA/UPV,SpainCWI,TheNetherlandsCERTH,GreeceAUEB,GreeceUniversityofBradford,EnglandUniversityofPisa,ItalyWarsawUniversityofTechnology,PolandPolitecnicodiMilano,ItalyUniversityofVienna,AustriaHUT,FinlandUniversityofAntwerp,BelgiumUniversitatPolitecnicadeCatalunya,SpainTechnicalUniv.ofCatalonia,SpainUniversityofLinz,Austria

VersionHistory

Version-1.0-

InternalVersion

2004/08/24--alexandre.proutiere@http://www.77cn.com.cn

A state of the art

ProjectAcronym:

ProjectFullTitle:

Typeofcontract:

contractNo:

ProjectURL:Euro-NGIDesignandEngineeringoftheNextGenerationInternet,http://www.77cn.com.cnworkofExcellence507613http://www.77cn.com.cn

A state of the art

Contents

1Introduction

1.0.0.1QoSrequirementsandtraf cdemand....................

1.0.0.2Radioresourcemanagement.........................

1.1Cellularnetworks........................................

1.2WirelessLANs.........................................

1.3AdHocnetworks........................................

1.4Transportprotocolsforwirelessnetworks...........................2AdmissionControlinUMTS-StateoftheArt

2.1CACalgorithmsbasedonamaximalnumberofusers.....................

2.2CDMA-orientedCACalgorithms................................

2.2.1Interference-CAC....................................

2.2.1.1Modelingthesystem.............................

2.2.1.2Introducingthewiredcapacity.......................

2.2.2Power-basedCAC...................................

2.2.2.1Power-CACandcoverage..........................

2.2.2.2ICACorPower-CAC?............................

2.2.3SIR-basedCAC.....................................

2.2.3.1Handlingmultipleclassesoftraf c.....................

2.2.3.2Theeffectivebandwidth:apossiblesolutioninWCDMA.........

2.2.3.3CombinedSIR-andpower-CAC......................

3PerformanceofstreamingservicesinwirelessCDMAnetworks

3.1Feasibilityconditions......................................

3.1.1Other-cellinterferenceapproximations........................

3.1.2Effectivebandwidth..................................

3.1.3Suf cientconditions..................................

3.1.4Powercontrolalgorithms-Rateofconvergence....................

3.2Outageprobability.......................................

3.3Impactofadmissioncontrol..................................4SupportingCircuitServicesinCellularNetworks

4.1DesignAdmissionControlMinimizingBlocking/Forced-termination.............

4.1.1EvaluationScenario..................................

4.1.1.1LossandDelayModels,AbandonmentsandRetrials............

4.1.1.2OverlayCell.................................

4.1.1.3ModelAssumptions.............................

4.1.2PoliciesthatonlyDeploySystemStateInformation..................

4.1.2.1ApproachestoCACdesign.........................

4.1.2.2AdmissionControlPolicies.........................

4.1.3Policiesthatdeployhandoff/handoverpredictionschemes..............

5RadioResourceSharing

5.1Introduction...........................................

5.2WirelessPacketFairQueueing:CopingwithBurstyChannelErrors.............

667789910101010111112121313131415171818192020202223232323242426262627292930

A state of the art

5.3.5

5.3.6

5.4

5.4.1Interactionwithmulti-antennaarrays.........................Cross-layerscheduling.................................StableScheduling...................................

5.4.1.1

5.4.1.2Single-UserTransmission..........................Multiple-UserTransmissions........................424343434346

48

48

49

49

51

51

51

52

52

52

54

54

55

55

56

57

59

62

62

62

62

63

64

65

66

71

71Packet-LevelDynamics.....................................5.5Flow-LevelPerformance....................................5.5.15.5.25.5.3Theimpactoffastfadingonperformance.......................Theimpactofslowfadingonperformance......................Multi-cellscenario...................................6ResourceControlforLoss-SensitiveTraf cinCDMANetworks6.16.2BackgroundandMotivation..................................AnewResourceControlApproach...............................7Auction-basedResourceReservationin2.5/3GNetworks7.17.2BackgroundandMotivation..................................ATHENA:AnewResourceReservationMechanism.....................8TransportintheUMTSRadioAccessNetwork8.18.28.38.48.5QoSintheUTRAN.......................................Transportnetworkarchitecture.................................ATM-basedtransportintheUTRAN..............................IP-basedtransportintheUTRAN...............................IP-basedservicedifferentiationintheUTRAN........................9UMTSNetworkswithPrioritisedRABs10MobilityModels10.1MobilityModels........................................10.1.1Introduction,Terms,andDe nitions..........................10.1.2Classi cationofMobilityModels...........................10.1.3SelectedMobilityModels...............................10.1.4ExtractingMobilityPatternsfromTraces.......................10.1.5ConclusionandOpenIssues..............................11IEEE802.11WirelessLANStandardanditsevolutions11.0.5.1FHSSPHYlayer...............................11.0.5.2DSSSPHYlayer...............................

12Theevolutionof802.11-eandh

QualityofServicesupport,TransmissionPowerControlandDynamicFrequencySe-lection7712.1Introduction...........................................12.2802.11e.............................................

12.2.1ThemotivationoftheTaskGroupe..........................777777

A state of the art

12.2.2802.11eframework...................................

12.2.2.1ProvisionofQoS..............................

12.2.2.2Traf cStreams................................

12.2.2.3Modi cationintheaccessfunctions....................

12.2.3802.11MACandchannelcondition..........................

12.2.4Functionalcharacteristicsofan802.11eTraf cScheduler..............

12.3802.11h.............................................

12.3.1ThemotivationofTaskGrouph............................

12.3.2TransmissionPowerControl(TPC)..........................

12.3.3DynamicFrequencySelection(DFS).........................

12.3.3.1Measurements................................

12.3.3.2Selectingachannel.............................

12.3.4APresponsibilitiesunderthenewframework.....................

13Handlingofreal-timeapplications

14PerformanceEvaluationofIEEE802.11

wirelessLANs

14.1Introduction...........................................

14.2Performanceevaluationof802.11anditsa,b,andgvariants.................787879798284868686909292929699999914.3Performanceevaluationof802.11e...............................10515ImprovingtheFairnessandEf ciencyofMACProtocols10915.1Introduction...........................................10915.2IEEE802.11bOverview....................................110

15.2.1RelatedWork......................................11215.3InterferenceAwareMAC....................................11415.4Conclusions...........................................11716WLANLossModelsforMultimediaServices11816.1WLANLossModelsforMultimediaServices.........................118

16.1.1WLANQoSMeasures.................................11816.1.2LossModelsforWLAN................................11816.1.3Conclusion,OpenIssues,andConnectionstoOtherWorkPackages.........12017Connectivityandcoverageproblemsinadhocandsensornetworks(stateoftheart)12117.1Connectivity...........................................12117.2Sensorcoverage.........................................12218MACprotocolsforadhocnetworks12418.1Introduction...........................................12418.2Classi cation..........................................12418.3IEEE802.11featuresandlimitations..............................12518.4Problemsandpossiblesolutions................................127

18.4.1HTPandETPproblem.................................127

A state of the art

18.4.2Blocking,falseblockingandpseudo-deadlock....................12818.4.3Unfairnessintheadhocnetwork............................12818.4.4MobilityandIEEE802.11...............................12918.4.5Multiplefading.....................................12918.4.6Near-farproblemandCDMAenvironment......................12918.4.7Omnidirectionalvs.directionalandsmartantenna..................12918.5Newschemesandtechniques..................................130

18.5.1PowerControlApproaches...............................13018.5.2Randomscheduleapproachandfan-in/outmethodofseedexchange........13118.5.3Multipacketreception(MPR)networks........................13118.5.4TheBlack-Burst(BB)mechanism...........................13218.5.5TheRandomizedInitializationProtocol........................13218.5.6Tree-SplittingApproaches...............................13218.5.7Neighborhoodawarenessprotocols..........................13318.5.8Synchronousatomicbroadcastprotocol........................13318.6Conclusions...........................................13419Energyef cientroutinginadhocnetworks13519.1Introduction...........................................13519.2Researchonenergyef cientrouting..............................13619.3Energyef cientnetworkoperation,combinedapproach....................13719.4Conclusion...........................................13820End-to-EndQoSSupportinAdHocNetworksConnectedtoFixedNetworks

21QualityofServiceforMobileAd-hocNetworks:anOverview139141

21.1Introduction...........................................14121.2Reservation-LessApproaches..................................142

21.2.1Load-BalancingSchemes................................14221.2.2CourtesyPiggybacking.................................14221.2.3SWAN.........................................14221.3Reservation-OrientedApproaches...............................143

21.3.1INSIGNIA.......................................14321.3.2FlexibleQoSModelforMANETs(FQMM)......................14321.3.3CEDAR.........................................14321.3.4QualityofServiceforAdhocOn-DemandDistanceVector..............14421.3.5Canseveretal......................................14421.3.6AdhocQoSon-demandrouting(AQOR).......................14521.4Conclusions...........................................14522PassingCorruptDataAcrossNetworkLayers:AnOverviewofRecentDevelopmentsandIssues14622.1Introduction...........................................14622.2UDPLite............................................14622.3DCCP..............................................147

A state of the art

22.4Issues...................

22.4.1TheLinkLayerPerspective...

22.4.2IPv6................

22.4.3EncryptionandAuthentication..

22.4.4Congestionvs.Corruption....

22.4.5LinkLayerARQConsiderations.

22.5Inter-layerCommunication........

22.5.1TrigTran:CorruptionExperienced

22.5.2Transportprotocoldetection...

22.5.3Otherpossibilities?........

22.6Conclusion................

22.7Acknowledgments............

References154....................................................................................................................................................................................................................................................................................................................................148148148149149150151151151152152153

A state of the art

1.Introduction

Thisdocumentconstitutesthe rstdeliverableofworkpackageJRA2.4onQoSinmultiservicewirelessnet-works,providingastate-of-the-artontherelatedtopics.Thisdeliverablegatherstogetherthecontributionsofthedifferentpartnersinvolvedinthisworkpackageandisthebaseforfutureresearchwithintheworkpackage.Thedocumentisfarfrombeingcompleteandwilleventuallybeextendedlateron.

1Introduction

A.Proutière

FranceTelecomR&D

N.Hegde

Eurandom,TheNetherlands

Therecognizedgoalinnetworkdesignistobeabletohandletraf cdemandensuringpre-de nedqualityofservice(QoS)requirementsandusingaminimumamountofnetworkresources.Insimplistictelephonenetworksforinstance,Erlang’sformula[151]identi es,foragiventraf cdemand,therequirednumberofcircuitstoguaranteeagivencallblockingprobability.Today,telecommunicationnetworksareexpectedtosupportvariouskindsofapplicationsde nedbytheirdifferentQoSrequirements.Thisneedforaconvergednetworkgeneratesasigni cantcomplicationindesigningitsarchitectureandprotocols.Eveninthecaseofwirednetworks,noconsensushasemergedandtheproposedsolutionsareconstantlychallengedbynewnetworkservicesandnewphysicaltransportfacilities.

Inthecontextofwirelessnetworks,thedesignisfurthercomplicatedbythefactthattheircapacitycannotbeexpressedinasimpleway,likelinkbandwidthinwire-linenetworks.Capacityoftheairinterfaceisacomplicatedfunctionofthetime-varyingchannelqualityofeachactiveuserandofthewayradioresources,e.g.powerandbandwidth,areshared.InitiatedbyShannon[386],informationtheoryidenti esthemaximumcapacityofthesystem,givensomefadingcharacteristics.Thisnotionofcapacityassumesastaticscenario,i.e.,a xednumberofactiveusersandisoftennotabletoaccountfortheactualQoSrequirementsofthevariousofferedapplications.

Thedesignofwirelesssystemsgatherstogetherchallengesofdifferentnature,withtheaimtooptimizetheoftenso-called’erlangcapacity’,de nedasthemaximumtraf cthatthesystemcanhandlemaintainingpre-de nedQoSrequirements.Thedesignofsuchsystemsshouldbeinspiredbyschemesapproachingtheinformationtheoreticallimitsinstaticscenarios,andshouldalsoaccountfortherandomnaturetraf caswellasforthevariousQoSrequirementsassociatedwiththesupportedapplications.

1.0.0.1QoSrequirementsandtraf cdemandTraf cdemandinmulti-servicenetworkscanbeclassi edintwobroadcategories,namelystreamingandelastictraf c,accordingtothecorrespondingQoSrequirements.Streamingtraf ccorrespondstoaudioorvideoapplicationsandrequireslowpacketlossanddelay,whereaselastictraf ccorrespondstodocumenttransfersandrequiresalowtransfercompletiontime.Inaddition,forbothtypesoftraf c,users rstperceiveperformancethroughtheblockingprobability,orequivalentlybytheprobabilitythattheserviceisavailableandthatthesystemallowsthecallortransfertostart.AnotherQoSparameterthathastobemaintainedbelowagiventhresholdistheprobabilitythatthecallortransferisinterrupted,whichmayoccurevenwhenaperformantadmissioncontrolschemeis

A state of the art

1.1Cellularnetworks

implemented,dueforexampletousermobility.Finally,somespeci cQoSrequirementshavesometimestobeensured.Forexample,whenauserreceivespacketsfromdifferentradiotransmittersatthesametime,i.e.theuserisinsofthandover,thepacketarrivalsatthetransmittershavetobesynchronized(incellularnetworks,thisrequirementimpliesstringentboundsforpacketdelayintheUTRAN).

Asinwire-linenetworks,traf cisrandomlygeneratedbyusers:atanytime,acalloratransfercanbegeneratedbetweenasourceandadestination.Incellularnetworks,forthedownlinkforexample,thetransferisinitiatedfromabasestationtoamobileattachedtothisbasestation.Inad-hocnetworks,thesourceandthedestinationmaybeanynodesofthenetwork.Whenthecallorthetransferiscompleted,therequirednetworkresourcesarefreed.Inwirelessnetworks,thistraf cdemandisalsode nedbythemobilitypatternsofeachuser/node:eachuserfollowshis/herowntrajectory(mostoftenindependentlyofotherusermovements)resultinginatimevaryingsource-to-destinationdistanceandroutingpath.

1.0.0.2RadioresourcemanagementRadioresourcesconsistofbandwidthandpower.Theachiev-ablerateforasource-destinationpairdependsinacomplicatedmanner,notonlyonthewaytheseresourcesaresharedamongallsource-destinationpairs,butalsoonthehighlyvariablepropagationcharacteristics,i.e.fading.Therearebasicallytwoapproachestohandlefadingdependingontheconsideredapplication: rst,onemayadapttheusedradioresourcestocompensateforthevaryingchannelqualityandtomaintaina xedrate,asincircuit-switchedvoicenetworks;second,onemaydynamicallyadapttheratesoastomatchthe uctuationsinthechannelquality.Thelatterapproachiswell-suitedfordatatransferapplicationsthatdonothaveastringentraterequirement.

Radioresourcesmanagementinanactualsystemmustalsosatisfysomedesignconstraintsdependingontheconsideredtypeofnetwork.Incellularnetworksforexample,radioresourcesmayoftenbemanagedinacentralizedmanner,whereasinWLANorAdHocnetworks,byde nition,theabsenceofsignalingproceduresimposesadecentralized(andthuslessef cient)waytoshareresources.

1.1Cellularnetworks

Cellularnetworksaredividedupintocells,eachcellbeingservedbyoneormoreradiotransceiverscompos-ingabasestation.Theradioresourcemanagementcanthenbeperformedinacentralizedmanneratthebasestationlevelorperhapsevenatahigherlevelinthenetwork(e.g.theRNCinUMTSnetworks)toallowforspeci cfunctionalities,suchasmacrodiversity.

Cellularsystemsare rstde nedbytheirradioaccesstechnology.Fromthe rstanalogsystemsdeployedinthe70’stotheOFDM/MIMObased4Gproposals,theaccesstechnologyhasconstantlyevolved,notonlytooptimizetheuseofthebandwidth,butalsotosupportdifferenttypesofapplications.A rstissuetoaddressisthedesignofanoptimalradiointerfacebasedontheconsideredapplication.Forexample,traditionalvoicetraf cprovestobebettersupportedbyCDMAsystemsthanbyTDMAsystems,whereasrecentadvancesinscheduling,suchasthedevelopmentofschedulersthattakeadvantageoffastfadingvariations,makeTDMAaccess,ratherthanCDMA,moresuitedtodelaytolerantdatatraf c.Nevertheless,thereisnoconsensusonthewayradioresourcesshouldbeshared,andbuildinganef cientmulti-serviceairinterfaceisstillthemajorissueinwirelessnetworkdesign.

Cellularnetworkswereinitiallydesignedtosupportcircuitservices.Theperformanceofsuchservicesde-pendsonmanyparameters,suchasthetypeofradioaccessorthewayadmissioncontrolisperformed.Unlikeusualtelephonenetworks,theQoSinwirelessnetworksisperceivednotonlythroughtheprobabilitythecall

A state of the art

1.2WirelessLANs

isadmittedbutalsothroughtheprobabilitytheQoSisensuredduringtheentirecall.TheprobabilitythattheQoSisdegradedduringthecallisoftentermedoutageprobability.Indeed,anoutagecanoccurevenundersevereadmissioncriterium,due,forexample,tousermobility.Wemayconsidertwokindsofmobility,intraandinter-cellmobility.The rstkindmainlyconsistsinshadowingandRayleighfading,whereasthesecondkindconsistsinwideusermovementsimplyinghardhand-overs.PredictingandcharacterizingmobilityisanimportanttasksincemobilityrepresentsthemaincontributioninoutageandQoSdegradation.AreviewofmobilitycharacterizationsispresentedinSection10.Theadmissionschemehastorealizeagoodtrade-offbetweenblockingandoutageprobabilities.ThesenotionsarefurtherdiscussedinSection3.Astate-of-the-artonadmissioncontrolinCDMAnetworksanditsperformanceevaluationispresentedinSections2,4and

9.Inconjunctionwithadmissioncontrol,powercontrolalgorithmsareusedtoadaptthetransmittedpowertothechannelqualityvariations.Referto[46]foranexhaustivereviewoftheliteraturedealingwithpowercontrol.

Inadditiontocircuitservices,wirelessnetworksareexpectedtosupportawidevarietyofdatatransfers.Thiskindoftraf c,withoutdelayorraterequirements,opensupthepotentialforde ningvariousschedulingdisciplinessoastomaximizecertainperformanceindicators.Areviewofexistingresultsonsuchschedulingdisciplinesandtheirpacketand owlevelperformanceispresentedinSection5.

Recently,economicmodelinghasbeenproposedasanalternativewayofresourcemanagementinwirednetworks.Thesenewideas,suchascongestionpricingorauction-basedresourcereservation,arenowadaptedtothecontextofwirelessnetworks.RefertoSections6and7forareviewonthistopic.

Finally,itmightbeimportanttoensurespeci cQoSrequirementsintheaccessnetwork,e.g.lowpacketdelaytoallowmacrodiversity.ThearchitectureaswellastheenvisagedQoSmechanismoftheUTRAN(theaccessnetworkofUMTSnetworks)arediscussedinSections8and9.

1.2WirelessLANs

WLANsaredesignedwithacompletelydifferentphilosophythancellularnetworks.TheinitialmotivationoftheIEEE802.11WorkingGroupwasthedevelopmentofaradiointerfacethatcouldreplicatethetraditionalEthernet-basedlocaldatanetworksinwirelessnetworks.Themajorconstraintofsuchaninterfacewasthattheradioresourceneededtobesharedinadecentralizedmanner,withoutanysignalingprocedures.TheWorkingGroupdevelopeda rststandard,the802.11standard,thatactuallycanbeseenasamodi edversionofEthernet,atleastfortheMAClayer.Manyotherstandardshavebeenproposedsince,thatadapttheMACprotocolstothephysicallayerevolutionorsupportnewfunctionalitiessuchasQoSprovisioning.ThevariousstandardsandtheirevolutionaredescribedinSections11and12.

ThedecentralizedradioresourcemanagementinWLANscreatesmanynewissues.First,thedeployedMACprotocolsarenaturallyveryinef cientinusingthescarceradioresources.Second,duetothediversityofradioconditionsofthevarioususers,theMACprotocolalsoprovestobequiteunfair.ManyotherproblemsexistinWLANs,forexampletheinitialdesignofMACprotocolsdidnottakeintoconsiderationseparatede nitionsofuplinkanddownlink.Asaresult,oneuplinkconnectionhasthesamechancetoaccessthemediaasallactiveconnectionsinthedownlink,leadingtoinef cienciesinresourcemanagement.TheperformanceevaluationofusualMACprotocolsispresentedinSection14.SomeproposalstoimprovetheirperformancearepresentedinSection15.

Finally,WLANswereoriginallydesignedtosupportInternet-likebestefforttraf candarethereforenotsuitedtodelaysensitivestreamingservices.Thislattertypeofserviceswasthemotivationofthe802.11e

A state of the art

1.3AdHocnetworks

WorkingGroup.AdescriptionofthestandardsintroducedtoprovideQoSguaranteesinWLANsandtheirperformancearepresentedinSections12and16.

1.3AdHocnetworks

Wirelessad-hocnetworksconsistofmobilenodescommunicatingoveraradiochannel,eachnodebeingallowedtocommunicatewithanyothernode.Duetothenatureofthewirelesschannel,eachnodecaneffectivelytransmittosomenodesonly,typicallyinhiscoveragearea.Thenwhentwonodesindifferentcoverageareaswishtocommunicate,itisnecessarythatintermediatenodescooperatetoforwardpacketsforonenodetotheother.Nodesmobilitycanalsobeusedtoincreasethenodecoverage,butwithanonnegligibledelaycost.Inafewwords,ad-hocnetworksarecooperativemulti-hopmobilewirelessnetworks.Thereisanintenseresearchactivitytowardsthedevelopmentofsuchnetworks.Forexample,considerableeffortsaremadetodesignef cientMACandroutingprotocolaswellashigherlevelfunctionalitiessuchassecurityorQoSprovisioning.Section17isdevotedtothestudyoftheconnectivityandthecoverageofad-hocnetworks.Section18presentsareviewofMACprotocolsforsuchnetworks.Section19addressestheproblemofde ningef cientroutingprotocolsandSections20and21discussthepossibilitytosupportQoSinad-hocnetworks.

Anotherresearcheffortismadetowardstheestablishmentofthecapacityofad-hocnetworks,i.e.,theoreticalboundsonhowmuchtraf cthesenetworkscansupportassumingperfectroutingandMACprotocols(thiskindofworkwasinitiatedbyGuptaandKumarintheirseminalpaper[188]).Thisaspecthasnotbeentreatedsofarinthisdocument.Thereaderwill ndin[418]aquiteexhaustivesurveyonthistopic.

1.4Transportprotocolsforwirelessnetworks

Finally,itisworthnotingthattransportprotocolssuchasTCPfordatatraf cwereoriginallydesignedforwirelinenetworksandarenotwell-suitedtotheradiomediumwhichisinherentlyslowandlossy.SomerecentproposalsthataddressthisissuearepresentedinSection22.Butthereaderwill ndanextensivestateoftheartofthistopicindeliverablesrelatedtoworkpackageJRA2.1.

A state of the art

2.AdmissionControlinUMTS-StateoftheArt

2AdmissionControlinUMTS-StateoftheArt

S-E.Elayoubi,T.Chahed

GET/InstitutNationaldesTelecommunications,France

SeveralCACalgorithmshavebeendevelopedforcellularnetworksduringthepastdecade.However,fewofthesealgorithmswereadaptedtothedynamicnatureofthirdgenerationwirelessnetworks.Infact,theradiointerfaceofthesenetworksisbasedonWideband-CDMAandthecapacityisdetermined,notonlybythenumberofusersandtheirbitrates,butalsobytheirpositioninthecellwhichinturnchangeswithmobility.WewillanalyzeinthisdocumentthemostimportantCACalgorithmspresentedintheliterature,withrespecttothevariousconstraintstheyconsider,aswellastheircomplexityandtheirabilityornottohandledifferentclassesofmultimediatraf c.

2.1CACalgorithmsbasedonamaximalnumberofusers

The rstCACalgorithmsweredesignedfor2Gsystems,especiallyGSM,andwerebasedonamaximalpredeterminednumberofusersinthesystem,orequivalentlyona xedcapacity.Infact,forsystemsbasedonTimeDivisionMultipleAccess(TDMA)orFrequencyDivisionMultipleAccess(FDMA),userscanbeacceptedaslongastherearechannels(timeslotsorfrequenies)available.ThisiscalledNCACandcorre-spondstoahardcapacityallocation,whichisnotthecasein3Gsystems,whosecapacityvariesdynamicallywiththequalityoftheairchannelandmobilityofusers,asshallbeseennext.

Nevertheless,thissimplisticassumptionmakesitpossibletointroduceseveralclassesoftraf c(Leongetal.in[277]andBartolinietal.in[41]),basedontheLimitedFractionalGuardChannel(LFGC)policy, rstproposedbyRamjeeetal.[353]for2Gsystems.Theythusobtainimportantperformancemeasuressuchthattheblockingprobability,i.e.theprobabilitythatacallisblockedduetothelackofresources,orthehandoffblockingprobability,i.e.theprobabilitythatamobile,acceptedinagivencellandmovingtowardsanadjacentone,failstocontinueitsconnectioninthenewcell.

2.2CDMA-orientedCACalgorithms

The rstCDMA-orientedalgorithmsweredesignedfor2GsystemsusingCDMAasaccessprotocolontheradiointerface,i.e.IS95system.TheseearlieralgorithmsconsideredinterferenceasthelimitingfactorwhenmakingaCACdecision.Infact,incontrastwiththehardcapacityallocation,CDMAsystemsarecharacterizedbyasoftcapacitywherethereisnohardtheoreticallimitonthenumberofusers.Ausercanthenbeacceptedifitsadmissiondoesnotcauseanexcessivedegradationforalreadyconnectedusersduetoahighlevelofinterference[31].

TherearethreeclassesofCDMA-orientedCACintheliterature,namelytheInterference-CAC,thePower-CACandtheSIR-CAC.

2.2.1Interference-CAC

InInterference-CAC(ICAC)algorithms,thedecisionisbasedonapredeterminedinterferencethresholdatthebasestation[213][82]:acallisacceptediftheresultinginterferenceremainsacceptable.Letusnotethat,

A state of the art

2.2CDMA-orientedCACalgorithms

byde nition,ICACalgorithmsaredesignedforthereverselink(uplink)astheyconsidertheinterferenceatthebasestationandnotatthemobile.Furthermore,ICACalgorithmsdevelopedintheliteraturedonottakeintoconsiderationmultipleclassesoftraf c.

2.2.1.1ModelingthesystemA rstenhancementtoICACwaspresentedbyHoetal.in[204]whereseveralCACschemeshavebeenbuiltbasedonananalyticalmodelofaDirectSequenceCDMA(DS/CDMA)cellularuplinksystem.TheaimoftheproposedCACschemesistomaximizethecell’sErlangcapacity,de nedastheratiooftheaveragenumberofcallsinthecelltothesystembandwidth,whilesatisfyingtwosystemconstraints:the rstisonblockingprobabilityandisgivenby(PBL≤θ)andthesecondisonoutageprobability,i.e.probabilitythatanongoingcallseesitsQoSdegraded(withouttakingintoaccountthehandoffblocking)(Pout≤ ).The rstandsecondmomentsoftheother-cellinterferencehavebeencalculatedusingatwo-slopepathlosspropagationmodelandanequivalentnumberofactiveusersZisderived,inthesensethattheeffectoftheother-cellinterferenceonthecapacityisequivalenttotheeffectofZlocalusers.Theoutageconditionisthenobtainedas

K+Z+noise>max.Interference

whereKisthenumberofusersinthecell.

Theimportanceofthisequation(1)isthatitde nesanICACalgorithmwiththesimplicityofNCAC.

ThesystemismodeledasaMarkovchainwitheachstaterepresentingthenumberofongoingcalls.TheblockinganddroppingprobabilitiesarethencalculatedusingaGaussianapproximationofK,forfourdif-ferentCACschemes:

Complete-SharingCallAdmissionControlScheme:thenumberofacceptedcallsislimitedtoGmax,Th.ThisnumberandthearrivalrateλarecalculatediterativelytomaximizetheErlangcapacitywhilesat-isfyingthesystemconstraints.ThisisequivalenttoanNCAC,withthedifferencethatthethresholdisvariablewiththeother-cellinterference.

Check-Interference-Upon-Call-ArrivalCallAdmissionControlScheme:Gmax,Th=∞,butacallisacceptedifandonlyiftheinterferencedoesnotexceedathresholdTTh.Again,TThandλarecalculatedbyiterationstosatisfythesystemconstraints.

Position-BasedCallAdmissionControlScheme:eachcellisdividedintotwozones:theinnercellandtheoutercell,andtwodifferentthresholdsTinandToutarede nedfortheacceptanceofthenewcallsinthetwozones.

However,thesealgorithmsremainlimitedtothecaseofasingleclassofvoicecalls,andnodistinctionismadebetweennewandhandoffcalls.

2.2.1.2IntroducingthewiredcapacityAnotherimportantenhancementtoICACwasintroducedbyKaseraetal.[236]whereadmissioncontrolisstudiedasacongestioncontrolpolicyintheIP-basedRadioAccessNetwork(RAN).Thisisbasedontheideathatthecurrentpoint-to-pointlinksusedinconnectingbasestationstonetworkcontrollerswillevolvetoanIP-basedRANforreasonsoflowercostduetostatisticalmultiplexinggains,betterscalabilityandreliability,andtheprojectedgrowthindataapplications.(1)

A state of the art

2.2CDMA-orientedCACalgorithms

TheCACalgorithmisindeedanI-CACasitlimitsthenumberofacceptedusersinthecellbyathresholdPmaxontheinterferenceatthebasestation.However,insteadofusingthismaximumallowablereceivedpowertoperformadmissioncontrol,eachbasestationusesavariablereceivedpowerthreshold,Padm,thatisperiodicallycalculatedbytheRNC,basedontheIPRANlossrate.Theideaissimple:themeasuredpacketlossrateisusedtocalculateapowerscalingfactorα,whichisthenusedtoscaletheadmissioncontrolthresholds,Padm,fortheairinterface.

Despiteitssimplicity,theimportanceofthisstudyliesinthefactthatitwas,withtheworksofElayoubietal.[141],oneofthe rstworksevokingthewiredconstraintsintheRANasalimitationwhenmakingaCACdecision.Indeed,airandwiredresourcesarecomplementary;theformerispreponderantinlargecells,usedinruralenvironmentswherewehaveasmallloadfactorandthesystemiscoverage-limited,whilethewiredcapacityismoresigni cantinurbanenvironmentswheresmallcellswithhighloadareusedandthesystemiscapacity-limited[206].

However,thechoiceofpacketlossasadeterminingfactorinCACdecisionsin[236]wasnotverysuccessful,mostlybecauseofthereal-timenatureonthetransportintheRANinWCDMAthatmakesthedelay,andnotthepacketloss,themostconstrainingfactor[375].Ontheotherhand,thestudywaslimitedtovoiceusersonly(amaximalnumberofusersisderivedfromtheCACthresholdPadm),whiletheIPRANisessentiallyproposedtocarrydatatraf cinthirdgenerationsystems.Thesecharacteristics(delayanddatatraf c)weretakenintoaccountin[141].

2.2.2Power-basedCAC

Whenacceptingacall,onecannotneglecttheeffectofthetransmissionpoweronQoS.Infact,thepowerthatcanbeemittedbyamobileorabasestationisnotin nite,butislimitedtoamaximalvaluePmax.

Inthedownlink,thispowerisconsideredasthelimitingfactor(Huangetal.[213]andCaponeetal.[82])becauseeachnewuserwillconsumeapartofthebasestationpower.Thecallisthenblockedifthebasestationcannotincreaseitspowertosupportit.

2.2.2.1Power-CACandcoverageFortheuplink,therelationshipbetweencoverageandemittedpowerisverytightinapower-controlledCDMAsystem:whenamobilestationgetsfartherfromthebasestation,itisrequiredtoincreaseitstransmissionpowertokeepthepowerreceivedatthebasestationasis.Ifitreachesitsmaximaltransmissionpowerwhilecontinuingitsmovement,itwilleitherhandovertoanadjacentcell,orfailtoinsuretherequiredQoSatitsoriginalbasestationandbedropped.Ontheotherhand,whenanewuserisaccepted,thelevelofinterferenceatthebasestationwillincrease,andconsequentlytheSIRofeachexistentuserwilldecrease.

Dimitriouetal.[132]triedtosolvetheSIRexpressionsinordertodeterminethetargetreceivedpowersatthebasestation.Thepresentedapproachisaniterativeonethatcalculatesthedeterminantsofthesystemwithnactiveusersbasedonthedeterminantswhenn 1activeusersareinthecell.Thereceivedpowersarethendeducedandthenewcallisblockedifthecorrespondingtransmittedpowersexceedthemaximaltransmissionone.AtunableguardfactorFisproposedtocontrolthedroppingratesinthesystem.Fisincreasedwhentheobserveddroppingprobabilityduetotheusers’mobilityistoohigh.However,thisanalyticalstudyhasbeenconductedforasingle-cellcase,anditwasfollowedwithsimulations[133]thatshowthattheother-cellinterferencereducesthecapacity.Alimitationofthisalgorithmisthecomplexity

A state of the art

2.2CDMA-orientedCACalgorithms

ofthecalculationsitrequireswhichmakesverydif culttheintroductionofenhancementslikeprioritizationbetweenhandoffandnewcalls,ortheintroductionofmultimediaservices.

2.2.2.2ICACorPower-CAC?Thetwoapproachesdescribedabove(ICACandPower-CAC)arecom-paredbysimulationsin[213],anditisshownthattheappropriatethresholdsforICACmayleadtoaperfor-mancecomparabletothatofPower-CAC.However,itisshownin[82]thatthesystemmaybeinterference-limitedifthepowerthresholdneverpreventsthepowercontroltoreachanequilibriumpoint,anditispower-limitedotherwise.ItisalsoshownthattheoptimalthresholdvalueofICACdependsonthepropagationenvironment,andthatwecannotde neasimplegeneralcriteriatotunetheinterferencethresholdPthr.TheproblemoftuningPthrwasalsoinvestigatedbysimulationsin[314]todeterminethetradeoffbetweentheblockinganddroppingprobabilitiesinthechoiceofthecapacitybound.Power-basedCACremainsthenthemostaccuratemethodtocontrolthesystem.

2.2.3SIR-basedCAC

AnotherapproachtoCDMA-orientedCACistobasethedecisionofacceptingornotanewarrivalonitseffectontheSIRofexistentusers.ThischoiceisbasedonthefactthatSIRisanimportantmeasureoftheQoSinCDMAradiointerface:itcanbemappedintocorrespondingbitrates,biterrorrates,orbitenergytointerferenceratio(Eb/N0)(Chahedetal.[90]).Thisapproach,combinedwithanadequatepowercontrolmechanism,makesitpossibletoconsiderthemaximaltransmissionpowerofmobilestationsthatisanimportantissueintheuplinkasweshowedabove.

The rstSIR-CACalgorithmwaspresentedbyLiuetal.in[290].ItisbasedontheideathatSIRisinverselyproportionaltothecapacity.Theresidualcapacityforeachcellkisthencalculatedas:

Rk=11 ,fortargetcellTH111

( TH ),foradjacentcells

where X denotesthegreatestintegerlessthanorequaltoX,andαisanempiricalconstantrepresentingtheestimateoftheinterferencecouplingbetweenadjacentcells.

Theoverallresidualcapacityisthentheminimalresidualcapacitiesbetweenalltheadjacentcells,andthenewcallisacceptedifthisquantityispositive.

Onecanseethatthisother-cellinterferenceestimationisnotveryaccurateanddoesnottakeintoaccountthevariousintercellinterferenceconditions.Kimetal.[248]proposedanextensionof[290]withmorerealisticassumptionsontheinterference,astheyconsidernotonlytheinterferencebetweenadjacentcells,butalsobetweencellsthatarenotadjacent,butclosetoeachother.Again,thealgorithmisdesignedforasingleclassoftraf c,andnocoverageconstraintsweretakenintoaccount.

2.2.3.1Handlingmultipleclassesoftraf cTheSIR-CACalgorithmpresentedbyJeonetal.[223]aimstoinsureprioritiesbetweenhandoffandnewcalls,andtosupportmultimediatraf cinbothuplinkanddownlink.Infact,inresourcesharingbetweencallrequests,sinceprematureterminationofconnectedcallsisusuallymoreannoyingthanrejectionofanewcallrequest,itiswidelyrecommendedthatasystemgiveshigherprioritytohandoffcallrequestsascomparedtonewones.Inadditiontothisnew/handoffcleavage,realtimecalls(voice,video,etc)maybegivenhigherpriorityoverelasticones(data).

A state of the art

2.2CDMA-orientedCACalgorithms

Whenaclassicallarrives,thealgorithmin[223]measurestheaverage(Eb/N0)foreachclassofusersandusesittopredictthe(Eb/N0)valueifthenewcallisaccepted.Athresholdisthen xedforeachclassofusersandthenewarrivalisacceptedifthepredicted(Eb/N0)doesnotgobelowthisthreshold.ThisalgorithmcanbeconsideredasaSIR-CAConeas(Eb/N0)canbedirectlyrelatedtotheSIR[253][90].

dMoreexplicitly,letΦuk,iandΦk,irespectivelydenotetheuplinkanddownlinkthreshold(Eb/N0)sofaclass

dkcallforcontrollingadmissionofaclassinewcallrequestandlet uk,iand k,irespectivelybetheuplink

anddownlinkthreshold(Eb/N0)sofaclasskcallforcontrollingahandoffcallrequestfromclassi.Toguaranteethepriorityofhandoffcallsovernewones,wemusthave:

u uk,i<Φk,i,d dk,i<Φk,ifor0≤k≤L 1

Fortheprioritiesbetweencallsofdifferentclasses,CACthresholdvaluesshouldbedeterminedsoastoguaranteetheprioritybetweencallclasses,consideringthedataratesandtherequired(Eb/N0)sofallcallclasses.

ThesystemismodeledasamultidimensionalcontinuoustimeMarkovchainwhose owbalanceequationsaredetermined.Asperformancemeasures,theauthorscalculatedtheblockingandoutageprobabilitiesaswellasthecellthroughput.Thethresholdsarethencalculatedtomaximizethethroughputwhileminimizingthefailureprobabilities.

Thismathematicalstudyisquitecompleteandoffersmanyanalyticalinsights.Itconsidersdifferentclassesofmultimediatraf cwithdistinctionbetweenhandoffandnewarrivals,forbothuplinkanddownlink.However,itisverycomplexandtheCACdecisionisonlyobtainedaftercumbersomecalculations.Italsofailstoconsiderpowerandcoverageconstraintsandtheanalyticalmodelforthe(Eb/N0)isnotaccurateasitisbasedonaveragecalculations.

2.2.3.2Theeffectivebandwidth:apossiblesolutioninWCDMATheeffectivebandwidth,anotionthatoriginatedinCACalgorithmsforATM-basedB-ISDNusingthelargedeviationstheory[362],ismakingitswaystothecontextof3Gsystems.Evansetal.[154],appliedATM’seffectivebandwidthtechniquesinmulti-cellCDMA.Intheirwork,thepresentedCACisaSIR-basedoneastheQoSofmobileiisexpressedintermsoftherequiredbitrateRiandthebitenergytointerferencedensityratio(Eb/I0)i,encapsulatedintheminimumSignaltoInterferenceDensityRatio(SIDR)values

Γi=Ri(Eb/I0)i,i=1...N

TheconditiontoobtaintheeffectivebandwidthisthatallusersmustbegrantedtheirSIDRrequirement,withasmallQoSviolationprobabilityα:

NpkM J mP(ΓpklXpkl>W)≤α,

p=1k=1l=1m=1..M(2)

whereMisthenumberofneighboringcells,Jisthenumberofcallclasses,Npkisthenumberofactivecallsofclasskincellp,andWisthesystembandwidth.ΓpklistheSIDRrequirementofuserlofclasskincellm.TheseSIDRrequirementsareindependentrandomvariableswithdistributionsindexedbyk(users

mofthesameclasshavethesameSIDRdistributions).Xpklsareindependentrandomvariablesthatmodelthe

A state of the art

2.2CDMA-orientedCACalgorithms

interferencecausedatcellsitemwhenmobilelofclasskincellpisreceivedatitstargetbasestationwithoneunitsignalpower.

TheMequations(2)aresolvedusingtheGaussianapproximationorthelargedeviationsbound.Thisleadstotheeffectivebandwidthκmpkwhichistheresourcerequiredatlinkmbyacallofclass(p,k).Theadmissibleregionisthenboundedbya nitenumberofhyperplanesobtainedfrom(2).

Theoretically,theseboundscanbeobtainedformultipleclassesofcalls,butinreal,thecomplexityincreaseswiththenumberofclasses.Thenumericalexamplesin[154]showedthatforonlytwoclassesof"voice"users,namelypremiumandstandardservices(thatcanbeinterpretedashandoffornewcalls),thematrixofeffectivebandwidthvaluesforaseven-cellssystemisofsize(7×14),anditisobtainedaftertediouscalculations.Itwillbea(19×95)matrixfora19-cellssystemwithfourclassesoftraf c(newvoice,handoffvoice,newvideo,handoffvideo,data),andtheadmissibleregionateachnewarrivalistheintersectionof19hyperplanes.

AsoftheCACconstraints,thepresentedmodelwasabletoconsiderthemobilityparameter,includedinthemobilepositioninthecell,andthesourcevariabilityintheeffectivebandwidthcharacterizations.However,themaximaltransmissionpowerofmobilestationswasnottakenintoaccount,althoughithasanimportantimpactonoutagerates.Itwasalsorestrictedtotheuplink.

2.2.3.3CombinedSIR-andpower-CACElayoubietal.[147]developedaneffective-basedCACalgorithmthatinsurestherequiredSIRwhiletakingintoaccounttheabovementionedCACconstraints:maximaltransmissionpowerandcoverage.Thisalgorithmalsofolloweda exiblemodelthatcanhandlehandoverrequestsandmultimediatraf c,inadditiontopossibleextensionstodealwithsharedchannelsthatremainapossiblesolutionfordatatraf c,orthecaseoflimitedcapacityintheUMTSTerrestrialRAN(UTRAN).

Todothis,theybasetheirworkontheeffectivebandwidthcharacterizationsderivedin[422][250]forasinglecellDS/CDMAsystemwithmultiuserreceiverstomitigatetheinterferencebetweenusers,andextendthisworktoamultiplecellcase.Thesetechniquestakeintoaccountthestructureoftheinterferencefromotheruserstodecreasetheinterferencewhendemodulatingauser.Theyconsiderthelinearminimummean-squareerror(MMSE)receiverandthedecorrelatorinadditiontotheconventionalmatched lterreceiver,andderiveeffectivebandwidthcharacterizationsoftheusercapacityinamultiplecellcase.TheSIRrequirementsofalltheuserscanthenbemetifandonlyifthesumoftheeffectivebandwidthsoftheusersislessthanthetotalnumberofdegreesoffreedominthesystem,providedbytheprocessinggain.

ThisalgorithmtakesintoaccountmobilityandcoverageaswellasthewiredcapacityoftheUTRAN,beyondtheNodeB.Elayoubietal.thenenhancedtheirCACalgorithmtosupporthandoverrequestsandmultipleclassesofmultimediatraf c.They rstpresentedameasurement-basedapproachfortheuplinkthatintro-ducesprioritiesbetweendifferentclassesoftraf c(handoffversusnewcallsandvoiceversusdata),anddecreasestheoutageprobabilityofongoingusers.Theythenpresentedasqueezingstrategytodealwithdatacallsandshowhowtomakeuseoftheelasticityofthistraf ctomaximizethecellthroughput.

Asofthedownlink,thisCACalgorithmtakesintoaccountthemaximaltransmissioncapacityofthebasestation,inadditiontothemobilityofuserswithinthecellandbetweenadjacentones.Theyderivedeffectivebandwidthexpressionsandfoundoutthattheydepend,notonlyontheSIRrequirements,butalsoonthepositionoftheuserinthecell:theeffectivebandwidthincreasesrapidlywiththedistancefromthebasestation.Theythendividethecellintonconcentricringsandobtainnsubclassesofeachclassoftraf c,with

本文来源：https://www.bwwdw.com/article/nx2i.html