StochasticGeometryandWirelessNetworksVolumeIIApplications.pdf
《StochasticGeometryandWirelessNetworksVolumeIIApplications.pdf》由会员分享,可在线阅读,更多相关《StochasticGeometryandWirelessNetworksVolumeIIApplications.pdf(224页珍藏版)》请在冰豆网上搜索。
StochasticGeometryandWirelessNetworksVolumeIIAPPLICATIONSFrancoisBaccelliandBartomiejBaszczyszynINRIA&EcoleNormaleSuperieure,45ruedUlm,Paris.Paris,December,2009.ThismonographisbasedonthelecturesandtutorialsoftheauthorsatUniversiteParis6since2005,Eu-random(Eindhoven,TheNetherlands)in2005,Performance05(JuanlesPins,France),MIRNUGEN(LaPedrera,Uruguay)andEcolePolytechnique(Palaiseau,France)in2007.Thisworkingversionwascom-piledDecember4,2009.ToBeatriceandMiraiPrefaceAwirelesscommunicationnetworkcanbeviewedasacollectionofnodes,locatedinsomedomain,whichcaninturnbetransmittersorreceivers(dependingonthenetworkconsidered,nodesmaybemobileusers,basestationsinacellularnetwork,accesspointsofaWiFimeshetc.).Atagiventime,severalnodestransmitsimultaneously,eachtowarditsownreceiver.Eachtransmitterreceiverpairrequiresitsownwirelesslink.Thesignalreceivedfromthelinktransmittermaybejammedbythesignalsreceivedfromtheothertransmitters.EveninthesimplestmodelwherethesignalpowerradiatedfromapointdecaysinanisotropicwaywithEuclideandistance,thegeometryofthelocationsofthenodesplaysakeyrolesinceitdeterminesthesignaltointerferenceandnoiseratio(SINR)ateachreceiverandhencethepossibilityofestablishingsimultaneouslythiscollectionoflinksatagivenbitrate.Theinterferenceseenbyareceiveristhesumofthesignalpowersreceivedfromalltransmitters,exceptitsowntransmitter.Stochasticgeometryprovidesanaturalwayofdefiningandcomputingmacroscopicpropertiesofsuchnetworks,byaveragingoverallpotentialgeometricalpatternsforthenodes,inthesamewayasqueuingtheoryprovidesresponsetimesorcongestion,averagedoverallpotentialarrivalpatternswithinagivenparametricclass.Modelingwirelesscommunicationnetworksintermsofstochasticgeometryseemsparticularlyrelevantforlargescalenetworks.Inthesimplestcase,itconsistsintreatingsuchanetworkasasnapshotofastationaryrandommodelinthewholeEuclideanplaneorspaceandanalyzingitinaprobabilisticway.Inparticularthelocationsofthenetworkelementsareseenastherealizationsofsomepointprocesses.Whentheunderlyingrandommodelisergodic,theprobabilisticanalysisalsoprovidesawayofestimatingspatialaverageswhichoftencapturethekeydependenciesofthenetworkperformancecharacteristics(connectivity,stability,capacity,etc.)asfunctionsofarelativelysmallnumberofparameters.Typically,thesearethedensitiesoftheunderlyingpointprocessesandtheparametersoftheprotocolsinvolved.Byspatialaverage,wemeananempiricalaveragemadeoveralargecollectionoflocationsinthedomainconsidered;dependingonthecases,theselocationswillsimplybecertainpointsofthedomain,ornodeslocatedinthedomain,orevennodesonacertainroutedefinedonthisdomain.Thesevariouskindsofiiispatialaveragesaredefinedinprecisetermsinthemonograph.Thisisaverynaturalapproache.g.foradhocnetworks,ormoregenerallytodescribeuserpositions,whenthesearebestdescribedbyrandomprocesses.Butitcanalsobeappliedtorepresentbothirregularandregularnetworkarchitecturesasobservedincellularwirelessnetworks.Inallthesecases,suchaspaceaverageisperformedonalargecollectionofnodesofthenetworkexecutingsomecommonprotocolandconsideredatsomecommontimewhenonetakesasnapshotofthenetwork.Simpleexamplesofsuchaveragesarethefractionofnodeswhichtransmit,thefractionofspacewhichiscoveredorconnected,thefractionofnodeswhichtransmittheirpacketsuccessfully,andtheaveragegeographicprogressobtainedbyanodeforwardingapackettowardssomedestination.Thisisrathernewtoclassicalperformanceevaluation,comparedtotimeaverages.Stochasticgeometry,whichweuseasatoolfortheevaluationofsuchspatialaverages,isarichbranchofappliedprobabilityparticularlyadaptedtothestudyofrandomphenomenaontheplaneorinhigherdimension.Itisintrinsicallyrelatedtothetheoryofpointprocesses.Initiallyitsdevelopmentwasstimulatedbyapplicationstobiology,astronomyandmaterialsciences.Nowadays,itisalsousedinimageanalysisandinthecontextofcommunicationnetworks.Inthislattercase,itsroleissimilartothatplayedbythetheoryofpointprocessesonthereallineinclassicalqueuingtheory.Theuseofstochasticgeometryformodelingcommunicationnetworksisrelativelynew.Thefirstpapersappearedintheengineeringliteratureshortlybefore2000.OnecanconsiderGilbertspaperof1961(Gilbert1961)bothasthefirstpaperoncontinuumandBooleanpercolationandasthefirstpaperontheanalysisoftheconnectivityoflargewirelessnetworksbymeansofstochasticgeometry.Similarobservationscanbemadeon(Gilbert1962)concerningPoissonVoronoitessellations.Thenumberofpapersusingsomeformofstochasticgeometryisincreasingfast.Oneofthemostimportantobservedtrendsistotakebetteraccountinthesemodelsofspecificmechanismsofwirelesscommunications.TimeaverageshavebeenclassicalobjectsofperformanceevaluationsincetheworkofErlang(1917).Typicalexamplesincludetherandomdelaytotransmitapacketfromagivennode,thenumberoftimestepsrequiredforapackettobetransportedfromsourcetodestinationonsomemultihoproute,thefrequencywithwhichatransmissionisnotgrantedaccessduetosomecapacitylimitations,etc.Aclassicalreferenceonthematteris(Kleinrock1975).Thesetimeaverageswillbestudiedhereeitherontheirownorinconjunctionwithspaceaverages.Thecombinationofthetwotypesofaveragesunveilsinterestingnewphenomenaandleadstochallengingmathematicalquestions.Asweshallsee,theorderinwhichthetimeandthespaceaveragesareperformedmattersandeachorderhasadifferentphysicalmeaning.Thismonographsurveysrecentresultsofthisapproachandisstructuredintwovolumes.VolumeIfocusesonthetheoryofspatialaveragesandcontainsthreeparts.PartIinVolumeIprovidesacompactsurveyonclassicalstochasticgeometrymodels.PartIIinVolumeIfocusesonSINRstochasticgeometry.PartIIIinVolumeIisanappendixwhichcontainsmathematicaltoolsusedthroughoutthemonograph.VolumeIIbearsonmorepracticalwirelessnetworkmodelingandperformanceanalysis.Itisinthisvolumethattheinterplaybetweenwirelesscommunicationsandstochasticgeometryisdeepestandthatthetimespaceframeworkalludedtoaboveisthemostimportant.Theaimistoshowhowstochasticgeometrycanbeusedinamoreorlesssystematicwaytoanalyzethephenomenathatariseinthiscontext.PartIVinVolumeIIisfocusedonmediumaccesscontrol(MAC).WestudyMACprotocolsusedinadhocnetworksandincellularnetworks.PartVinVolumeIIdiscussestheuseofstochasticgeometryfortheivquantitativeanalysisofroutingalgorithmsinMANETs.PartVIinVolumeIIgivesaconcisesummaryofwirelesscommunicationprinciplesandofthenetworkarchitecturesconsideredinthemonograph.Thispartisself-containedandreadersnotfamiliarwithwirelessnetworkingmighteitherreaditbeforereadingthemonographitself,orrefertoitwhenneeded.Herearesomecommentsonwhatthereaderwillobtainfromstudyingthematerialcontainedinthismonographandonpossiblewaysofreadingit.Forreaderswithabackgroundinappliedprobability,thismonographprovidesdirectaccesstoanemerg-ingandfastgrowingbranchofspatialstochasticmodeling(seee.g.theproceedingsofconferencessuchasIEEEInfocom,ACMSigmetrics,ACMMobicom,etc.orthespecialissue(Haenggi,Andrews,Baccelli,Dousse,andFranceschetti2009).Bymasteringthebasicprinciplesofwirelesslinksandoftheorgani-zationofcommunicationsinawirelessnetwork,assummarizedinVolumeIIandalreadyalludedtoinVolumeI,thesereaderswillbegrantedaccesstoarichfieldofnewquestionswithhighpracticalinterest.SINRstochasticgeometryopensnewandinterestingmathematicalquestions.ThetwocategoriesofobjectsstudiedinVolumeII,namelymediumaccessandroutingprotocols,havealargenumberofvariantsandofimplications.Eachofthesecouldgivebirthtoanewstochasticmodeltobeunderstoodandanalyzed.Evenforclassicalmodelsofstochasticgeometry,thenewquestionsstemmingfromwirelessnetworkingoftenprovideanoriginalviewpoint.AtypicalexampleisthatofrouteaveragesassociatedwithaPoissonpointprocessasdiscussedinPartVinVolumeII.ReaderalreadyknowledgeableinbasicstochasticgeometrymightskipPartIinVolumeIandfollowthepath:
PartIIinVolumeIPartIVinVolumeIIPartVinVolumeII,usingPartVIinVolumeIIforunderstandingthephysicalmeaningoftheexamplespertainingtowirelessnetworks.Forreaderswhosemaininterestinwirelessnetworkdesign,themonographaimstoofferanewandcomprehensivemethodologyfortheperformanceevaluationoflargescalewirelessnetworks.Thismethod-ologyconsistsinthecomputationofbothtimeandspaceaverageswithinaunifiedsetting.Thisinherentlyaddressesthescalabilityissueinthatitposestheproblemsinaninfinitedomain/populationcasefromtheverybeginning.Weshowthatthismethodologyhasthepotentialtoprovidebothqualitativeandquantitativeresultsasbelow:
Someofthemostimportantqualitativeresultspertainingtotheseinfinitepopulationmodelsareintermsofphasetransitions.Atypicalexamplebearsontheconditionsunderwhichthenetworkisspatiallyconnected.Anothertypeofphasetransitionbearsontheconditionsunderwhichthenetworkdeliverspacketsinafinitemeantimeforagivenmediumaccessandagivenroutingprotocol.Asweshallsee,thesephasetransitionsallowonetounderstandhowtotunetheprotocolparameterstoensurethatthenetworkisinthedesi