utilizattldy.com

HALId:inria-00596224https://hal.
inria.
fr/inria-00596224v3Submittedon31Jul2011HALisamulti-disciplinaryopenaccessarchiveforthedepositanddisseminationofsci-entificresearchdocuments,whethertheyarepub-lishedornot.
ThedocumentsmaycomefromteachingandresearchinstitutionsinFranceorabroad,orfrompublicorprivateresearchcenters.
L'archiveouvertepluridisciplinaireHAL,estdestinéeaudéptetàladiffusiondedocumentsscientifiquesdeniveaurecherche,publiésounon,émanantdesétablissementsd'enseignementetderecherchefranaisouétrangers,deslaboratoirespublicsouprivés.
SURF:ADistributedChannelSelectionStrategyforDataDisseminationinMulti-HopCognitiveRadioNetworksMubashirHusainRehmani,AlineCarneiroViana,HichamKhalife,SergeFdidaTocitethisversion:MubashirHusainRehmani,AlineCarneiroViana,HichamKhalife,SergeFdida.
SURF:ADistributedChannelSelectionStrategyforDataDisseminationinMulti-HopCognitiveRadioNetworks.
[ResearchReport]RR-7628,INRIA.
2011.
inria-00596224v3ISSN0249-6399ISRNINRIA/RR--7628--FR+ENGThèmeCOMINSTITUTNATIONALDERECHERCHEENINFORMATIQUEETENAUTOMATIQUESURF:ADistributedChannelSelectionStrategyforDataDisseminationinMulti-HopCognitiveRadioNetworksMubashirHusainRehmani—AlineCarneiroViana—HichamKhalife—SergeFdidaN°7628July2011CentrederechercheINRIASaclay–le-de-FranceParcOrsayUniversité4,rueJacquesMonod,91893ORSAYCedexTéléphone:+33172925900SURF:ADistributedChannelSelectionStrategyforDataDisseminationinMulti-HopCognitiveRadioNetworksMubashirHusainRehmani,AlineCarneiroViana,HichamKhalife,SergeFdida§ThèmeCOM—Systèmescommunicantsquipes-ProjetsAsapRapportderecherchen°7628—July2011—48pagesAbstract:Inthispaper,weproposeanintelligentanddistributedchannelselectionstrategyforefcientdatadisseminationinmulti-hopcognitiveradionetwork.
Ourstrategy,SURF,classiestheavailablechannelsandusesthemefcientlytoincreasedatadisseminationreliabilityinmulti-hopcognitiveradionetworks.
Theclassicationisdoneonthebasisofprimaryradiounoccupancyandofthenumberofcognitiveradioneighborsusingthechannels.
ThroughextensiveNS-2simulations,westudytheper-formanceofSURFcomparedtothreerelatedapproaches.
Simulationresultsconrmthatourapproachiseffectiveinselectingthebestchannelsforefcientcommunication(intermsoflessprimaryradiointerference)andforhighestdisseminationreachabilityinmulti-hopcognitiveradionetworks.
Key-words:multi-hopcognitiveradionetworks,dynamicchannelselection,datadissemination.
LIP6/UniversitePierreetMarieCurie(UPMC)-SorbonneUniversités,Paris,FranceINRIA,FranceandTU-Berlin,GermanyLaBRI/ENSEIRB,UniversitédeBordeaux,France§LIP6/UniversitePierreetMarieCurie(UPMC)-SorbonneUniversités,Paris,FranceSURF:Unestratégiedistribuéedesélectiondescanauxpourladiffusiondedonnéesdanslesréseauxradiocognitifsmulti-sautsRésumé:Nousprésentonsdanscetarticleunenouvellestratégiedesélectiondefréquencespourladisséminationabledesdonnéesdansleréseauradiocognitifsmulti-sauts.
EnexplorantdynamiquementlesressourcesrésiduellessurlesfréquencesdesprimairesetencontrlantlenombredeCRsurunefréquenceparticuliére,notrestratégieSURFpermetd'augmenterlaabilitédediffusiondesdonnéesdanslesréseauxradiocog-nitifmulti-sauts.
GrceàunelargeétudebaséessurdessimulationssurNS-2,nousétudionslesperformancesdeSURFparrapportàtroisautresapprochesliées.
Lesré-sultatsdesimulationconrmentquenotrestratégiepermetlasélectiondesmeilleuresfréquencesadaptéesàunedisséminationableetefcacedesdonnéesdansunréseauradiocognitifsmulti-sauts.
Mots-clés:Réseauxradiocognitifsmulti-sauts,sélectiondynamiquedefréquence,disséminationdedonnées.
SURF:ADistributedChannelSelectionStrategy31IntroductionDatadisseminationiscommonlydenedasthespreadingofinformationtomultipledestinationsthroughbroadcasting.
Themainobjectiveistoreachthemaximumnum-berofneighborswitheverysentpacket.
Inthiscommunicationscheme,noroutingisrequiredthusneitherroutingtablesnorend-to-endpathsaremaintained.
Amongdifferentapplicationswheredatadisseminationcanbeuseful,wefocusinthisworkonnetworkingscenarioswhereprovidersdisseminatenon-urgentmessagesinordertolimitcostandcomplexitythroughthenetwork,suchas:services,updates(e.
g.
,newcodetore-taskaprovidedservice),oranykindofpublicitymessage.
However,guar-anteeingreliabilityofdatadisseminationinwirelessnetworksisachallengingtask.
Indeed,thecharacteristicsandproblemsintrinsictothewirelesslinksaddseveralis-suesintheshapeofmessagelosses,collisions,andbroadcaststormproblem,justtonameafew.
Inthispaper,wefocusondatadisseminationinadhoccognitiveradionetworks.
Cognitiveradionetworksarecomposedofcognitiveradiodevices.
Theconceptofcog-nitiveradiowasintroducedintheseminalpaperbyJ.
Mitola[1].
Themotivationbe-hindcognitiveradiowasthreefold:(1)availabilityoflimitedspectrum,(2)xedspec-trumassignmentpolicy,and(3)inefciencyinspectrumusage.
Therefore,cognitiveradionetworksaredesignedtoopportunisticallyexploittheunderutilizedspectrum.
Moreover,theregulatorybodies,suchas,theFederalCommunicationCommission(FCC)[2]alsopromotedtheideaofusingthecognitiveradiodevicestoaddressthespectrumshortageproblem.
Inthisregard,theFCChasdesignedaninterference-freeopportunisticspectrumaccesspolicy[2].
AccordingtotheFCC'spolicy[2],channelsareonlyallowedtobeusedbyCognitiveRadio(CR)nodesiftheyareidlei.
e.
,notutilizedbythePrimaryRadio(PR)nodesandCRnodesshouldavoidcausingharmfulinterferencetoPRnodes.
Infact,PRnodesarethelegacyusersandtheyhavehigherprioritytousethelicensedband.
CRnodescantakeadvantageofidlechannelstodisseminatenon-urgentandpublicitymessageswithlowcostandcomplexity.
ParticularlyinthecontextofCognitiveRadioWirelessNetworks(CRN)[3],wherechannelsfortransmissionareopportunisticallyselected,reliabilityisdifculttoachieve.
Thisisduetotheinherentfeaturesofsuchnetworks.
First,inadditiontothealreadyknownissuesofwirelessenvironments,thediversityinthenumberofchannelsthateachcognitivenodecanuseaddsanotherchallengebylimitingnode'saccessibilitytoitsneighbors.
Second,CognitiveRadio(CR)nodeshavetocompetewiththePrimaryRadio(PR)nodesfortheresidualresourcesonchannelsandusethemopportunistically.
Besides,CRnodesshouldcommunicateinawaythatdoesnotdisturbthereceptionqualityofPRnodesbylimitingCR-to-PRinterference[4].
Inmulti-hopcognitiveradioad-hocnetworks,wherecoordinationbetweenCRsishardtoachieveandnocentralentityforregulatingtheaccessoverchannelsispresent,reliabledatadisseminationisevenmorecomplex.
Inthisperspective,theimportantstepinhavingefcientdatadisseminationistoknowhowtoselectbestchannels.
Infact,channelselectionplaysavitalroleinreliabledatadissemination.
IfCRnodesselectthechannelsrandomly,thereareverylesschancesthattheneighborreceiversalsoselectthesamechannel.
Consequently,therandomselectionofchannelsseverelydegradesthedatadisseminationreachability.
Furthermore,whenCRnodesrandomlyselectthechannelfortransmission,itmaybepossiblethataPRtransmissionisgoingonandsubsequently,theCRtransmissioncausesharmfulinterferencetothePRnodes.
Alotofworkshavebeencarriedoutfordynamicchannelmanagementincognitiveradionetworks[5,6,7,8,9,10,11,12,13,14,15,16].
Theseapproachesfocusonsingle-RRn°76284Mubashir&Aline&Hicham&Sergehopcognitiveradionetworks[5,6,10,14]andeitherrequiresthepresenceofanycentralentity[11,16]orthecoordinationwithprimaryradionodesintheirchannelselectiondecision[14,16].
Moreover,theseaforementionedchannelselectionstrategiesarenotspecicallydesignedfordatadissemination,e.
g.
,[5]considerthetrafcdemandsofAccessPoints,[16]discussthroughputmaximization,[11]discussloadbalancing,justtonameafew.
AmorerelatedtoourapproachisSelectiveBroadcasting(SB)[17].
Selectivebroadcastingisproposedformulti-hopcognitiveradionetworks,inwhichnodesse-lectaminimumsetofchannelsi.
e.
EssentialChannelSet(ECS)tocoverallitsge-ographicneighbors.
Therearehowever,severalchallengesinthepracticalityofSB.
Fromthecommunicationperspective,simultaneoustransmissionoveranECSrequiresmorethanonetransceiver,resultinginbiggerandmorecomplexdevices,asformilitaryapplications[18].
Furthermore,transmissionsoverasetofchannelswithoutconsid-eringthePRactivitymayincreasetheprobabilityofinterferencewithprimaryradionodes.
Since,nocentralizedentityispresenttosynchronizenodesintheirchannelselectiondecision,selectinganinappropriatechannelforoverhearingfromtheECSchannelsetbytheneighboringnodesmayleadtopacketlosses.
Therefore,anewchannelselectionstrategyisrequiredwhichworkswellwithsingletransceiver,causelessharmfulinterferencetoPRnodesandtrytomaximizethechancesthatthemes-sageisdeliveredtotheneighboringcognitiveradioreceivers,thusincreasingthedatadisseminationreachability.
Thus,differentlyfromworksintheliterature,wegoastepfurtherhereandbuildupachannelselectionstrategy,SURF,fordatadisseminationinmulti-hopcognitiveradionetworks.
InSURF,theobjectiveofeverycognitiveradionodeistoselectthebestchannelensuringamaximumconnectivityandconsequently,allowingthelargestdatadisseminationreachabilityinthenetwork.
Thiscorrespondstotheuseofchannelshavinglowprimaryradionodes(PRs)activities,aswellashavinghighernumberofCRneighbors.
InSURF,theclassicationofchannelsisdoneonthebasisofprimaryradiounoc-cupancyandthenumberofcognitiveradioneighborsusingthechannels.
Anothermainchallengewedealwithinthispaperresideinmakingefcientandreliablechannelse-lectiondecisionson-the-yandinrecoveringfrombadchannelselectiondecisions.
Todealwiththischallenge,weintroducethemechanismofrecoveryfrombadchannelselectiondecision.
Inthismechanism,SURFkeepstrackofpreviouswrongchan-nelstateestimationandaccordinglyadaptsfuturechannelselectiondecision.
Usuallychannelselectionstrategiesprovideawaytonodestoselectchannelsfortransmission.
Besides,SURFenduesCRnodestoselectbestchannelsalsoforoverhearing.
Thiswillhelptotunebothsenderandreceiverwithhighprobabilitytothesamechannel.
Asaconsequence,SURFmayhavehighnumberofneighborsontheselectedchannel.
Inadditiontothat,SURFprotectsthePRnodesbyconsideringthePRunoccupancyinchannelselectiondecision,foreffectiveandreliabledatadissemination.
WeanalyzetheperformanceofSURFthroughextensiveNS-2simulations.
WeusetheCognitiveRadioCognitiveNetwork(CRCN)patch[19]ofNetworkSimulatorNS-2[20].
TheCRCNpatchofNS-2doesnotsupporttheactivityofthePRnodes.
Thus,weenhancetheCRCNpatchofNS-2toincludethePRactivitymodel.
Wecom-pareSURFwithRandom(RD),HighestDegree(HD)andSelectiveBroadcasting(SB)approach[17].
InordertoevaluateSURF,weusevemetrics:(1)harmfulinterfer-enceratio,whichwechoosetocharacterizetheprobableinterferencecausedbyCRtransmissionstoPRnodes;(2)averagedeliveryratioand(3)ratioofaccumulativeCRreceivers,bothforevaluatingthereliabilityofdatadissemination;(4)ratioofeffectiveINRIASURF:ADistributedChannelSelectionStrategy5neighborsand(5)ratioofaccumulativeeffectiveneighbors,bothofthemarechosentocharacterizethetuningofsender/receivernodes.
WecomprehensivelyanalyzeSURFbyvaryingnodedensity,numberofretries,packetdropreasons,etc.
Simulationre-sultsconrmthatSURFprotectsthePRnodesduringtransmissioncomparedtoRD,HDandSBapproaches.
SURFisalsoabletoachievetheaveragedeliveryratioof40%50%,comparedto0%inRD,1%forSBand2%forHDapproaches.
TheresultsofeffectiveneighborsandaccumulativeeffectiveneighborsshowthatSURFisabletowelltunethesending/receivingCRnodesandthus,abletocreatewithhighprobability,aconnectedtopology.
Besidestheseadvantages,thesimplicityanddecen-tralizednatureofSURFmakesitusableinad-hocCRNsdeployedtoconveyservices,updates,oranykindofpublicitymessage.
Themajorcontributionsofthispaperaresummarizedinthefollowing:WedesignSURF,anintelligentanddistributedchannelselectionstrategyfordatadisseminationinmulti-hopcognitiveradionetworks.
SURFisalsoequippedwiththemechanismofrecoveryfrombadchannelselectiondecision.
WeenhancestheNetworkSimulatorNS-2toincludethePRactivitymodel.
WevalidateSURFthoughdifferentmetricsandcomparewithRD,HD,andSB.
Weprovideadetailedliteraturereviewonchannelselectionstrategiesincogni-tiveradionetworks.
Theremainderofthispaperisorganizedasfollows:wediscusschallengesofdatadisseminationinSection2.
ThenwediscusssystemmodelandassumptionsinSec-tion3.
WegivegeneraloverviewofSURFinSection4.
Section5and6dealwithdetaileddescriptionofSURF.
Performanceanalysisisdoneinsection7.
Section8dis-cussestheactivitypatternimpactofprimaryradionodesonchannelselectionstrate-gies.
Section9discussesrelatedwork,andnally,section10concludesthepaper.
2ChallengesofDataDisseminationinCognitiveRadioNetworksDatadisseminationisaclassicalandafundamentalfunctioninanykindofnetwork.
Inwirelessnetworks,thecharacteristicsandproblemsintrinsictothewirelesslinksbringseveralchallengesindatadisseminationintheshapeofmessagelosses,collisions,andbroadcaststormproblem,justtonameafew.
However,datadisseminationisextremelychallengingissueincognitiveradionetworksduetoitsintrinsicproperties,suchas:theavailabilityofmultiple-channelsi.
e.
,CRnodeshavemorethanonechannelintheavailablechannelset.
Morespecically,theid'softhechannelsintheavailablechannelsetofsenderandreceiveraresame.
thediversityinthenumberofavailablechannelsi.
e.
,CRhavemorethanonechannelintheavailablechannelset.
But,theid'softhechannelsintheavailablechannelsetofsenderandreceiversaredifferent.
theprimaryradioactivityi.
e.
,channelsareoccupiedbythePRnodesandareonlyavailabletoCRnodesfortransmissionwhentheyareidle.
Infact,theRRn°76286Mubashir&Aline&Hicham&SergespatiotemporalutilizationofspectrumbyPRnodes(i.
e.
primaryradionodes'activity)addsanotherdimensionofcomplexitytodatadissemination.
Asacon-sequence,thenumberofavailablechannelstoCRnodeschangeswithtimeandlocationandthisleadstothediversityinthenumberofavailablechannelset.
BecauseofPR'sactivity,theusabilityofthechannelsbyCRnodesbecomesuncertain.
Moreover,withoutanycentralizedentity,asinthecaseofmulti-hopadhoccog-nitiveradionetwork,datadisseminationisevenmorechallengingbecauseCRnodeshavetorelyonlocallyinferredinformationfortheirchannelselectiondecision.
Ifachannelselectionisdoneinanintelligentway,higherdatadisseminationreachabilitycanbeachieved.
Furthermore,theconsiderationofPRactivityduringchannelselec-tioncanenhancetheeffectivenessofdatadisseminationreachabilityandcanreducetheharmfulinterferencetoPRnodesbyCRtransmissions.
Wementionsomekeyrequiredcharacteristicsofanychannelselectionstrategyfordatadisseminationincognitiveradionetworks:1.
Efcientmessagedelivery:Agoodchannelselectionstrategyistheonethatincreasestheprobabilityofhighermessagedeliveryinmulti-hopcontext.
2.
Primaryradioconstraints:Thechannelselectionstrategyshouldensurethatthetransmissionontheselectedchanneldoesnotcreateharmfulinterferencetopri-maryradionodes.
3.
AutonomousdecisionbyCRnodes:Itmeansthatthechannelselectionstrat-egyshouldworkwellwithoutanycentralizedauthorityandchannelselectiondecisionshouldbebasedonlocallyinferredinformation.
4.
Sender/Receivertuning:ThechannelselectionstrategyshouldguaranteethattheCRtransmitterandreceiverselectthesamechannelwithhighprobability.
3SystemModelandAssumptionsInthissection,wepresentthesystemmodelconsideredandthebasicassumptionsrelatedtoourproposal.
3.
1NetworkModelWeconsideraCognitiveRadioAd-HocNetwork[21].
Inthistypeofnetworkset-ting,weassumethatnocentralizednetworkentityisavailable.
Instead,weconsideranetworkingenvironmentwherenetworkoperations(e.
g.
,spectrumsensing,channelselectiondecisionetc)areperformedbytheCRnodesthemselves.
ThenetworkiscomposedofasetofPrimaryRadio(PR)nodesandasetofCognitiveRadio(CR)nodes.
Primaryradionodesarethelicensedusersandtheycanaccesstheirrespectivelicensedbandswithoutanyrestriction.
Indeed,PRnodeshavethehighestprioritytoaccessthechannelsandshouldnotbeinterruptedbytheCRnodes[4].
InordertobeabletocommunicateinaCRN,CRnodesmustcreateamulti-hopnetworkbyusingthelicensedbands.
Theuseoflicensedbandsbycognitiveradionodesarehowever,onlypossiblewhenthebandsareidle,i.
e.
unoccupiedbythePRINRIASURF:ADistributedChannelSelectionStrategy7nodes.
Notethatanidlestatedescribesthetemporalavailabilityofachannel.
Insomecases,itcanhappenthataCRnodestartsatransmissionatthesametimewhenPRbecomesactive.
Since,weconsiderthatCRtransmissionsshouldnotgenerateharmfulinterferenceatPRreceivers[22],CRswillcanceltheirtransmissions.
WefurtherassumethatCRnodesareequippedwithasingletransceiver.
Thistransceivercaneitherreceiveortransmitonasinglechannelatatime.
Theutiliza-tionofsingletransceiverreducestheoperationalcostoftheCRdevice[23],aswellasavoidspotentialinterferencebetweenco-locatedtransceiversduetotheircloseprox-imity[24].
WeconsiderthesetoftotalfrequencychannelsC.
3.
2SpectrumSensingbyCognitiveRadioNodesIncognitiveradioad-hocnetworks,cognitiveradionodesareassumedtoworkinstandalonefashionandmakedecisionsbasedonlocallyinferredinformation.
Asaconse-quence,eachCRnodehastoperformspectrumsensingtodetectthepresenceofthePRsignal.
WeassumethatthespectrumsensingisperiodicallyperformedbyeveryCRnode.
WefurtherassumethatthedetectionofthePRsignalistheresponsibilityofthespectrumsensingblock[25].
Inthiscase,SURFwillworkonthelistofavailablechannelsresultedfromthespectrumsensing.
3.
3PrimaryRadioActivityorWirelessChannelModelTheperformanceofcognitiveradionetworkiscloselyrelatedtotheprimaryradioactivityoverthechannels.
Therefore,theestimationofprimaryradioactivityplaysavitalroleinchannelselectiondecision.
Weassumethattheprimaryradioactivityorwirelesschannelcanbemodelledascontinuous-time,alternatingON/OFFMarkovRenewalProcess(MRP)[26,27,28](cf.
Section5formoredetails).
NotethatsuchanON/OFFPRactivitymodelcapturesthetimeperiodinwhichthechannelcanbeutilizedbyCRswithoutcausinganyharmfulinterferencetoPRnodes[29].
3.
4ExchangeofHelloPackets[30]couldbeusedtohelptheneighbordiscoveryprocessanditusesaCommonControlChannel(CCC)mechanism.
InthisCCCmechanism,CRnodeslocallymakeclustersandthecontrolchannelfromtheISMbandisdynamicallyallocatedwithineachcluster.
ThereasonbehindlocallymakingclustersbyCRnodesisduetotheglobalunavailabilityofcontrolchannel.
In[30],rsttheneighbordiscoveryisper-formed.
Theneighbordiscoveryconsistsofthreephases:(1)eachCRdeterminesthesetofidlechannels,(2)auniversaltimescheduleforchannelaccessisfollowed,and(3)usingthisuniversaltimeschedule,CRscandiscovertheirneighbors.
Inthesecondstep,clusteringisperformedbasedonthesetofidlechannelsthatarecommontoallclustermembers.
Thecontrolchannelfromagivenclusterisselectedfromthisset.
Inthismanner,thegoalofincreasingtheavailabilityofcommonidlechannelineachclusterisachievedbygroupingCRswithsimilarspectrumopportunities.
Weassumetheavailabilityofaout-of-bandCommonControlChannel(CCC)[21]forneighbordiscovery.
DuetothetimevariabilityofPRactivityonthelicensedband,controlchannelisselectedfromtheunlicensedISMband.
ThereasonbehindselectingadedicatedspectrumbandforCCCistominimizetheCCCdisruptionscausedbyPRRRn°76288Mubashir&Aline&Hicham&Sergeactivity.
Itisworthnotingthatthecostofswitchingbetweendatachannelsandcontrolchannelisnon-negligiblebecauseoftheavailabilityofasingletransceiver.
4ChannelSelectionStrategySURFTheSURFchannelselectionstrategyisspecicallydesignedforad-hoccognitiveradionetworks.
ThegeneralgoalofSURFistoincreasereliabilityindatadisseminationoveramulti-hopadhocCRN.
NotethatSURFisapacket-basedchannelselectionschemefordatadisseminationandnotaroutingalgorithm.
Therefore,neithertheroutingtablesnortheend-to-endpathsaremaintainedbytheCRnodes.
CRnodes,uponeachpacketreception,selectthebestchannel,andbroadcastthepacket.
WithSURF,everyCRnodeautonomouslyclassiesavailablechannelsbasedontheobservedPR-unoccupancyoverthesechannels.
ThisclassicationisthenrenedbyidentifyingthenumberofCRsovereachband.
ThebestchannelfortransmissionisthechannelthathasthehigherPRunoccupancyandahighernumberofCRneigh-bors.
Indeed,choosingachannelwithfewCRsmayyieldstoadisconnectednetwork.
EveryCRafterclassifyingavailablechannels,switchesdynamicallytothebestoneandbroadcaststhestoredmessage.
Moreover,SURFalsotriestolearnwithpreviouswrongchannelstateestimation.
ThislearningprocessallowsbettertuningthefutureestimationsandhelpsCRnodestorecoverfromtheirbadchannelselectiondecisions.
Additionally,CRswithnomessagestotransmitimplementtheSURFstrategyinordertotunetothebestchannelfordatareception.
Usingthesamestrategyimple-mentedbythesenderallowsreceiversinclosegeographicareastoselectwithhighprobabilitythesameused-to-sendchannelforoverhearing.
ThiswillalsoincreasethenumberofCRneighborsontheselectedchannel.
Thisisduetothefactthat,intuitively,itislikelythatCRsinthesender'svicinityhavethesamePRunoccupancy,hencechan-nelsavailabletoaCRsenderisalsoavailabletoitsneighborswithhighprobability[8].
Therefore,SURFincreasestheprobabilityofcreatingaconnectedtopology.
Onceapacketisreceived,everyCRreceiverundergoesagainthesameproceduretochoosetheappropriatechannelforconveyingthemessagetoitsneighbor.
Channel'sWeightCalculationFormulaSURFstrategyclassieschannelsbyas-signingaweightP(i)wtoeachobservedchanneliinthechannelsetC.
Thus,everycognitiveradionoderunningSURF,locallycomputestheP(i)wusingthefollowingequation:i∈C:P(i)w=PR(i)u*CR(i)o(1)P(i)wdescribestheweightofachannel(i)andiscalculatedbasedonthePRunoc-cupancy(i.
e.
PR(i)u)andCRoccupancy(i.
e.
CR(i)o)overchanneli(c.
f.
section5andsection6).
Then,thechannelsarerankedaccordingtotheirweightsandthebestchan-nel(i.
e.
,theoneprovidinghighestP(i)w)willbeused.
Notethatwhenthechannelhashighweightbutattimetitisoccupied,SURFreacts(i)bynottransmittingthepacketonthebestweightedchanneland(ii)byselectingthenextbestweightedchannelforpackettransmission/overhearing.
Alsonotethatwhenallthechannelsareoccupied,nomessageissent.
TheincreaseofweightisrelatedtothetwoobjectivestheSURFstrategyneedstosatisfy.
ThemajorobjectiveofprotectingtheongoingPRactivityismappedasaINRIASURF:ADistributedChannelSelectionStrategy9functionofPRunoccupancy.
ThehighertheprobabilityofPRsbeinginOFFstate,i.
e.
PR(i)u,thehighertheweightwillbe.
Thus,SURFgiveshighimportancetonotdegradingtheserviceofongoingprimarycommunications.
ThesecondobjectiveofincreasingconnectivityisimplementedinthesecondtermofEq.
11.
Moreprecisely,theweightincreaseswiththenumberofCRneighborsi.
e.
CR(i)o.
Inthefollowing,wediscussindetailhowtheprimaryradiounoccupancyandcognitiveradiooccupancycouldbeestimated.
5PrimaryRadioUnoccupancyTheprimaryradioactivity,i.
e.
presenceorabsenceofthePRsignal,canbemodelledascontinuous-time,alternatingON/OFFMarkovRenewalProcess(MRP)[26,27,28].
ThisPRactivitymodelhasbeenusedverywidelyintheliterature[26,27,28,31,32,33,34,35].
TheON/OFFPRactivitymodelapproximatesthespectrumusagepatternofpublicsafetybands[34,36].
Thepublicsafetybandisdesignatedforcommercialandpublicsafetyuses[37].
Theauthorsin[38]approximateandvalidatethePRON/OFFactivitymodelforthepresenceofthePRsignalinIEEE802.
11b.
TheON/OFFPRactivitymodelisalsothemostfamousmodelforvoice[39].
AnimportantfeatureofthisON/OFFPRactivitymodelisthatitcapturesthetimeperiodinwhichthechannelcanbeutilizedbyCRswithoutcausinganyharmfulinterferencetoPRnodes[29].
Fig.
1illustratesthewirelesschannelmodel.
TheONi.
e.
busystateindicatesthatthechanneliscurrentlyoccupiedbythePRnode,whiletheOFFi.
e.
idlestateindicatesthatthechanneliscurrentlyunoccupiedbyPRnode.
Figure1:Wirelesschannelmodel:AlternatingMarkovRenewalProcessforPRactiv-ity.
ThedurationofONandOFFstatesofchanneliaredenotedasTiONandTiOFF,respectively.
TherenewalperiodofachanneloccurswhenoneconsecutiveONandOFFperiodiscompleted.
LetZi(t)denotetherenewalperiodofchanneliattimet,suchthatZi(t)=TiOFF+TiON[26,28,29,40].
RRn°762810Mubashir&Aline&Hicham&SergeBothONandOFFperiodsareassumedtobeindependentandidenticallydis-tributed(i.
i.
d.
).
SinceeachPRuserarrivalisindependent,eachtransitionfollowsthePoissonarrivalprocess.
In[26,40],theauthorsprovedthatwheneachPRarrivalfol-lowsthePoissonarrivalprocess,thelengthofONandOFFperiodsareexponentiallydistributed.
Inthispaper,weusetheformulationof[26,28,29,40]thatthechannelsONandOFFperiodsarebothexponentiallydistributedwithp.
d.
f.
fX(t)=λX*eλXtforONstateandfY(t)=λY*eλYtforOFFstate.
ThedurationoftimeinwhichchanneliisinONstatei.
e.
channelutilizationuiisgivenas[29]:ui=E[TiON]E[TiON]+E[TiOFF]=λYλX+λY(2)whereE[TiON]=1λXandE[TiOFF]=1λY.
λXandλYaretherateparameterforexponentialdistribution.
E[TiON]andE[TiON]isthemeanofexponentialdistribution.
LetPON(t)betheprobabilityofchanneliinONstateattimetandPOFF(t)betheprobabilityofchanneliinOFFstateattimet.
TheprobabilitiesPON(t)andPOFF(t)canbecalculatedas:PON(t)=λYλX+λYλYλX+λYe(λX+λY)t(3)POFF(t)=λXλX+λY+λYλX+λYe(λX+λY)t(4)Thus,byaddingEq.
3andEq.
13,wegetPON(t)+POFF(t)=1(5)Sinceourgoalistoselectthechannelthatwillbeunoccupiedattimet,fromhereafterwewillonlyconsiderPOFF(t).
EachCRnodelocallycomputestheseprob-abilities.
ThevaluesofλXandλYcanbeeasilymeasuredbyCRnodesbycollectingthehistoricalsamplesofchannelstatetransitions,asin[29].
Inthispaper,weareusingthevaluesmeasuredbyauthorsin[29](cf.
Table2).
ThebestchannelattimetistheonethathasveryhighprobabilityofbeinginOFFstate.
Itmaybepossiblethattheprobabilisticallyestimatednextchannelstatemis-matchwiththecurrentstateofthechannel,referredhereafteraswrongchannelstateestimation.
Thisfurtherleadstobadchannelselectiondecisionandcauseharmfulin-terferencetoPRnodes.
NotethatCRnodeskeepthehistoryofestimatedandmeasuredstatesofthechannel.
Next,wedetailhowthelearningofpreviouswrongestimationcanhelptotunefutureestimations.
5.
1RecoveryfromBadChannelSelectionDecisionsAnotherchallengewedealwithinthispaperresideinmakingefcientandreliablechannelselectiondecisionson-the-yandinrecoveringfrombadchannelselectionde-cisions.
Clearly,keepingtrackofwrongchannelstateestimationscanhelpCRnodestorecoverfromtheirbadchannelselectiondecisions,whichultimatelyenhancetherelia-bilityandtheperformance.
Duetothememorylesspropertyofthemarkovexponentialmodel,thereisalargedegreeofrandomnessandthisresultinimperfectpredictionofchannelstate[34].
TodealwiththismemorylesspropertyofthemarkovexponentialINRIASURF:ADistributedChannelSelectionStrategy11model,CRnodesalwayskeepcalculatingthenextstateofthechannel,POFF(t),withEquation13.
Inparallel,CRnodescalculatesPOFF(t)whichconsidersthecurrentstateofthechannelandwrongchannelstateestimations.
Toachievethisgoal,nodesmaintainthehistoryofestimatedchannelstatesandtheobservedcurrentstateofthechannels.
CRnodesthencomputewhichestimationswerewrongandkeeptheminhistory.
ThishistoryisthenusedtocalculatetheprobabilitiesPUMandPSM.
PUMisdenedastheprobabilitythattheestimatedchannelstatemis-matcheswiththeactualchannelstate.
EachCRnodeusesPUM,whilecalculatingthenextchannelstate(cf.
Fig.
2).
Conversely,theprobabilityofsuccessfullymatchedstatePSMisdenedastheprobabilitythattheestimatedchannelstatematcheswiththecurrentchannelstate.
Moreprecisely,theaccuracyoftherecoverymechanismofSURFdependsupontheestimatedstateofthechannel(cf.
probabilityvaluegivenbyEq.
(13))andthemeasuredcurrentstateofthechannel.
Table.
5providesthepossiblecombinationsbetweenthevaluesofestimatedstateandcurrentstateofthechannel.
TheprobabilityPSMisexpressedas:P(i)SM=xtN,(6)wherextisthenumberoftimestheestimatedchannelstatematcheswiththeactualchannelstate,andNisthetotalnumberoftimestheestimationoccurs,andTheprobabilityPUMisexpressedas:P(i)UM=xntN,(7)wherexntisthenumberoftimestheestimatedchannelstatedoesnotmatchwiththeactualchannelstatei.
e.
howoftenthechannelstatesestimationwaserroneous,andNisthetotalnumberoftimestheestimationoccurs.
Infact,thePUMmeasurestwodifferenttypesofchannelstatescases(cf.
Table5).
TherstoneisthecasewhenestimatedchannelstateisOFFandthemeasuredchannelstateisONandthesecondoneisthecasewhentheestimatedchannelstateisONthemeasuredchannelstateisOFF.
Thus,wefurtherdecomposedPUMintoPMDandPFAas:P(i)UM=xntN=PMD(i)+PFA(i),(8)wherePMDistheProbabilityofMiss-Detectionandoccurswhenestimatedchan-nelstateisOFFandthemeasuredchannelstateisON.
InPMD,CRnodedeclaresthebusychannelasunoccupied.
ThiswillleadtoharmfulinterferencewithPRnodes.
While,PFAistheProbabilityofFalse-AlarmandoccurswhentheestimatedchannelstateisONandthemeasuredchannelstateisOFF.
InPFA,CRnodedeclaresthattheunoccupiedchannelisbusy.
ThiswillleadtorefrainCRnodefromtransmittingandthus,loosespectrumopportunity.
PFAandPMDaremeasuredbyeveryCRnodeonperchannelbasis.
Infact,CRnodeestimatesthestateofthechannelandthisestimatedstateiscomparedwiththeactualstateofthechannel.
Iftheestimatedstateofthechan-nelisONandthemeasuredchannelstateisOFF,CRnodeincreasethePFAcounter,elseiftheestimatedstateofthechannelisOFFandthemeasuredstateofthechannelisON,CRnodeincreasethePMDcounter.
BoththePFAandPMDcountersarethendividedbythetotalnumberoftimestheestimationoccurs.
Inthismanner,eachCRnodemaintainsthehistoryofPFAandPMD.
RRn°762812Mubashir&Aline&Hicham&SergeFigure2:FlowchartshowingthecorrectivemeasuretakenbytheCRnodesinthecaseofdetectionofunsuccessfullymatchedchannelstatesi.
e.
PUM.
Table1:EstimatedandCurrentStatesoftheChannel.
EventEstimatedStateCurrentStatePSMONONOFFOFFPUMPMDOFFONPFAONOFFConsequently,thelowerthePUM(t),themoreaccuratewillbethechannelstateestimation.
Puttingthingstogether,weestimatePOFF(t),whichconsiderstheproba-bilityofunsuccessfullymatchedstateduringthechannelstateestimation,asfollows:PR(i)u=POFF(t)(i)=P(i)OFF(1P(i)FA)+P(i)MD(1P(i)OFF)(9)Inthecaseofaperfectchannelestimation(i.
e.
,PFA=0andPMD=0),POFF(t)=POFF(t).
Inthepresenceofchannelestimationerrors,theprobabilityofchannel(i)beinginOFFstateisgivenbyEq.
(12).
6CognitiveRadioOccupancyCRoccupancyreectsthenumberofCRneighborsusingthechannel.
Infact,agoodchannelselectionstrategyistheonethattuneCRnodestothechannelthathavehighernumberofCRneighbors.
HighernumberofCRneighborsprovidesgoodlevelofnet-workconnectivityandconsequentlyincreasethetransmissioncoverageofCRnodes.
TheCRoccupancyCR(i)oofchannel(i)iscalculatedas:CR(i)o=CR(i)n(10)INRIASURF:ADistributedChannelSelectionStrategy13where,CR(i)nisthenumberofCRneighborsusingthechannel(i).
InordertocalculatetheCRoccupancy,eachCRnodediscoverstheirneighbors.
NeighborscanbediscoveredinanefcientwaybydenominatingtheCommonControlChannel(CCC),whichwillensuretheavailabilityofcommonidlechannelbetweenCRnodes,andtheneighbordiscoverymechanism,asin[30].
Theauthorsin[30]assumedthatduetotheglobalunavailabilityofcontrolchannel,CRnodeshavetolocallymakeclustersthatdecreasetheoverheadinneighbordiscoveryandmakethecoordinationbetweenCRnodeseasier.
AfterlocallymakingtheclustersbyCRnodes,thecontrolchannelfromtheISMbandisdynamicallyallocatedwithineachcluster.
Inthisman-ner,thegoalofincreasingtheavailabilityofcommonidlechannelineachclusterisachievedbygroupingCRswithsimilarspectrumopportunities.
Consequently,eachCRnodeisabletocalculatetheCRoccupancybyknowingthenumberofCRneigh-borsusingthechannel.
Inadditiontoneighbordiscoverymechanismproposedin[30],SURFcanjointlyworkwithanyotherneighbordiscoverymechanism,suchas[41,42].
7PerformanceAnalysisInthissection,weanalyzetheperformanceofSURFthroughextensivesimulations.
7.
1ImplementationSetupWeusetheCognitiveRadioCognitiveNetwork(CRCN)patch[19]ofNS-2[20].
TheCRCNpatchhasthreebuildingblocksthatsupportscognitiveradiofunctionalitiesinNS-2(cf.
Fig.
3).
Thesebuildingblocksarethecognitiveradionetworklayer,thecognitiveradiomaclayerandthecognitiveradiophysicallayer.
Thecognitiveradionetworklayerisresponsibleformaintainingtheneighborlist.
ItalsomakesthechannelselectiondecisiononthebasisoftheinformationprovidedbythecognitiveradioMAClayer.
ThecognitiveradioMAClayersupportsmultiplechannelsandkeepstrackofPRtrafc,collision,interferenceinformationanditalsomaintainsthechannellist.
Thecognitiveradiophysicallayerhasinformationliketransmissionpower,SINR/SNRphysicalmodel,propagationmodeletc.
Theinformationcollectedatdifferentlayersissharedthroughtheinformationsharinglayer.
ThisCRCNpatchofNS-2doesnotsupporttheactivityofthePRnodes.
Thus,weenhancetheCRCNpatchofNS-2toincludethePRactivitymodel.
Fig.
3showsthehighleveldesignofPRactivitymodel(dottedbox)addedinNS-2.
ThePRactivityblockisresponsibleforkeepingtrackofPRactivitiesineachspectrumband(spectrumutilization)i.
e.
,sequenceofONandOFFperiodsbyPRnodesoverthesimulationtime.
TheseONandOFFperiodscanbemodelledascontinuous-time,alternatingON/OFFMarkovRenewalProcess(MRP)[26],[28].
TheON(busy)statemeansthechannelisoccupiedbythePRnode.
While,theOFF(idle)statemeansthechannelisunoccupiedbythePRnode.
WeconsiderthechannelsONandOFFperiodsarebothexponentiallydistributed,asin[28],[29].
TherateparameterλXandλY(cf.
Table2)oftheexponentialdistributionisprovidedasaninputinthesimulation,whichweremeasuredbyauthorsin[29].
Then,accordingtothisrateparameter,channelsfollowtheONandOFFperiods.
Weconsiderasimplemacprotocol(Maccon.
cc),availablewiththeCRCNpatchofNS-2.
Thismacprotocolisamultiple-channel,collisionandcontention-basedmacprotocol.
Notethatintheoriginalstate,theMaccon.
ccmacprotocolselectschannelRRn°762814Mubashir&Aline&Hicham&Sergerandomlyfromthepredenedsetofchannelsandthechannelselectiondecisionoccursatthemaclayer.
Wenowperformchannelselectionatthenetworklayer.
Thus,wemodifythismacprotocolandprovidethecapabilitytothenetworklayertomakethechannelselectiondecision.
WefurtheraddchannelselectionstrategiesRD,HD,SBandSURFtothenetworklayer,whichwedescribehereafter.
Baseduponanyparticularchannelselectionstrategy,thenetworklayertakesthechannelselectiondecision.
Thischannelselectiondecisionisencapsulatedinthenetworklayerpacketheaderanditispassedtothemaclayer,whichthenswitchtothechannelbasedonthechannelselectiondecisionprovidedbythenetworklayer.
IntheMaccon.
ccmacprotocol,therearetwochannelstates:IDLEandBUSY.
ThesestatesaredependentonthechannelconditionsandtheyhaveusedbythemacprotocoltohandlethetransmissionandreceptionactivitiesofCRnodes.
IDLEmeansthatthechannelisfreetousefortransmissionbytheCRnodeandBUSYmeansthatthechannelisoccupiedbyanyundergoingCRtransmission.
InordertodealwiththeactivitiesofthePRnodes,weincludetwomorestatesatmaclayerforeachchanneli.
e.
,PR_OCCUPIEDandPR_UNOCCUPIED.
ThestatePR_OCCUPIEDmeansthatthechannelisoccupiedbythePRnodeandPR_UNOCCUPIEDmeansthatthechannelisunoccupiedbythePRnode.
Thesetwostatesofthechannelwillbecheckedeachtimebythemacprotocolwhileperformingtransmissionoroverhearing.
Figure3:HighleveldesignofprimaryradioactivitymodelinNS-2.
7.
2PerformanceMetricsWecompareSURFwithrandomstrategy(RD),highestdegreestrategy(HD)andse-lectivebroadcasting,proposedin[17]withmultipletransmissions(SB).
WesuggestedRDstrategy,whichisthesimplestoneandnoinformationisrequired.
InRD,channelsarerandomlyselectedtobeusedbyCRnodesfortransmissionand/oroverhearing,withoutanyconsiderationtotheongoingPRandCRactivityoverthesechannels.
HDapproachonlyconsidersCRactivitiesandisinspiredbySBapproach.
InHD,CRnodesselectthehighestCRdegreechannelfortransmissionandoverhearing,withoutINRIASURF:ADistributedChannelSelectionStrategy15anyconsiderationofPRactivity.
Thehighestdegreechannelcovers,consequently,thehighestnumberofneighborsintheavailablelistofchannels.
InSB,eachCRnodecalculatesaminimumsetofchannels,EssentialChannelSet(ECS),fortransmissionthatcoversallitsgeographicneighbors,withoutconsideringthePRunoccupancy.
InSB,aCRnodetransmitsonmultiplechannelsinround-robinfashionpresentintheECSlist,untilallneighborsarecovered.
Notethatin[17]nothingismentionedabouthownodesoverhearoverthechannels.
Therefore,weconsidernodesselectforover-hearingthehighestdegreechannelfromtheirECSlistonly.
Ifmorethanoneoptionisavailable,arandomchoicefortransmission/overhearingisperformedamongthosechannelswiththesamedegree.
Since,ourgoalistoefcientlydisseminatethedata,tuningofsenderandreceivernodestothesamechannelwithhighprobability,andtoprotectthePRnodesfromharmfulinterference,wedeneveperformancemetrics:1.
HarmfulInterferenceRatio(HIR):ThismetricisdenedinordertocapturethenotionofcollisionwithPRnodes.
HIRisdenedastheratioofthetotalnumberoftimesthechannelisoccupiedbyPRnodeafterthechannelselectiondecisionovertotalnumberoftimesthechannelselectiondecisionoccurs.
2.
AverageDeliveryRatio:Thismetricisdenedtoeffectivelymeasurethedatadisseminationprocess.
ItistheratioofpacketsreceivedbyaparticularCRnodeovertotalpacketssentinthenetwork.
3.
RatioofAccumulativeCRReceivers:Thismetricalsoevaluatesthedatadissem-inationprocess.
ItisdenedastheaverageratioofaccumulativeCRreceiversperhopovertheaccumulativeeffectiveneighborsperhop.
AccumulativeCRreceiversperhoparethenumberofCRreceiversperhopthatsuccessfullyre-ceivedthemessage,whileaccumulativeeffectiveneighborsperhoparetheCRneighborsthatselectsthesamechannelforoverhearingasthesendernodeusedfortransmission.
Notethatbyaccumulativeratiowemean:ateachnewhopn,thereceiversandeffectiveneighborsofallprevioushopsl1LongONShortOFFLowActivityλX>1λY≤1ShortONLongOFFIntermittentActivityλX>1λY>1ShortONShortOFFINRIASURF:ADistributedChannelSelectionStrategy29(a)(b)(c)Figure16:ZeroPrimaryRadioActivity.
(a)CRNodes'IDandaveragedeliveryratioforRD,HD,SBandSURF.
(b)HopcountandaveragenumberofeffectiveneighborsforRD,HD,SBandSURF.
(c)HopcountandaveragenumberofreceiversforRD,HD,SBandSURF.
RRn°762830Mubashir&Aline&Hicham&Serge(a)(b)(c)Figure17:LongTermPrimaryRadioActivity.
(a)PRharmfulinterferenceratioforRD,HD,SBandSURF.
(b)CRNodes'IDandaveragedeliveryratioforRD,HD,SBandSURF.
(c)HopcountandRatioofaccumulativereceiversforRD,HD,SBandSURF.
INRIASURF:ADistributedChannelSelectionStrategy31(a)(b)(c)Figure18:HighPrimaryRadioActivity.
(a)PRharmfulinterferenceratioforRD,HD,SBandSURF.
(b)CRNodes'IDandaveragedeliveryratioforRD,HD,SBandSURF.
(c)HopcountandRatioofaccumulativereceiversforRD,HD,SBandSURF.
RRn°762832Mubashir&Aline&Hicham&Serge(a)(b)(c)Figure19:LowPrimaryRadioActivity.
(a)PRharmfulinterferenceratioforRD,HD,SBandSURF.
(b)CRNodes'IDandaveragedeliveryratioforRD,HD,SBandSURF.
(c)HopcountandRatioofaccumulativereceiversforRD,HD,SBandSURF.
INRIASURF:ADistributedChannelSelectionStrategy33(a)(b)(c)Figure20:IntermittentPrimaryRadioActivity.
(a)PRharmfulinterferenceratioforRD,HD,SBandSURF.
(b)CRNodes'IDandaveragedeliveryratioforRD,HD,SBandSURF.
(c)HopcountandRatioofaccumulativereceiversforRD,HD,SBandSURF.
RRn°762834Mubashir&Aline&Hicham&SergeTable6:HarmfulInterferenceRatio(HIR)(in%)undervariousPrimaryRadioNodesActivity.
RDHDSBSURFCh=5Ch=10Ch=5Ch=10Ch=5Ch=10Ch=5Ch=10LongTerm6353514950502327High9087868389896065Low17161312181355Intermittent61494746585622228.
4PerformanceAnalysisThissectionpresentstheperformanceanalysisofthefourschannelselectionstrate-giesundervaryingPRnodesactivity.
Toachievethis,weperformedextensiveNS-2simulations.
Forthisend,threeperformancemetricsareconsidered:1.
HarmfulInterferenceRatio(HIR):ThismetricisdenedinordertocapturethenotionofcollisionwithPRnodes.
HIRisdenedastheratioofthetotalnumberoftimesthechannelisoccupiedbyPRnodeafterthechannelselectiondecisionovertotalnumberoftimesthechannelselectiondecisionoccurs.
2.
AverageDeliveryRatio:Thismetricisdenedtoeffectivelymeasurethedatadisseminationprocess.
ItistheratioofpacketsreceivedbyaparticularCRnodeovertotalpacketssentinthenetwork.
3.
RatioofAccumulativeCRReceivers:Thismetricalsoevaluatesthedatadissem-inationprocess.
ItisdenedastheaverageratioofaccumulativeCRreceiversperhopovertheaccumulativeeffectiveneighborsperhop.
AccumulativeCRreceiversperhoparethenumberofCRreceiversperhopthatsuccessfullyre-ceivedthemessage,whileaccumulativeeffectiveneighborsperhoparetheCRneighborsthatselectsthesamechannelforoverhearingasthesendernodeusedfortransmission.
Notethatbyaccumulativeratiowemean:ateachnewhoph,thereceiversandeffectiveneighborsofallprevioushopslThenumberofCRnodesisxedtoN=100.
CRsarerandomlydeployedwithinasquareareaofa2=700x700m2andtheirtransmissionrangeissettoR=250m.
Simulationsrunfor1000secondsandatotalof1000packetsaresent,whereeachpacketissentbyarandomlyselectednodeatanintervalof1second.
Allresultsareobtainedwithacondenceintervalof95%.
Weconsider5(Ch=5)and10(Ch=10)totalnumberofchannels,whichallowsvaryingtheneighborhooddensitydavgbetween11.
3(whenCh=5)and20.
1(whenCh=10).
NotethisdensityiscomputedafterthespectrumsensingprovidesthelistofavailablechannelsandbeforetheCRsselectthechanneltotransmit/overhear.
Inthiscase,itisworthmentioningthat,atthefollowingsimulationstudies,theneighborhoodINRIASURF:ADistributedChannelSelectionStrategy35densityvariesinfunctionoftheCRs'channelselectionandislowerthantheaboveones.
Theresultsattesttheobtainedlowdeliveryratiosaremainlyduetothecreationofdifferenttopologiesresultedfromthemulti-channelavailabilityanddistributedchannelselectionbyCRs.
ThiscanbeveriedintheFig.
16,whichshowsresultsfordeliveryratio,numberofreceiversandofeffectiveneighbors,forCh=5andCh=10whennoPRnodesactivityispresentinthechannels.
Ascanbeobserved,evenwhenCRnodesdonothavetocompetewithPRnodestohaveaccesstothechannels,theaveragedeliveryratiorangesfrom35%50%,theaveragenumberofeffectiveneighborsrangesfrom1020andtheaveragenumberofreceiversrangesfrom122(from1stto6thhop)inSURF.
Fig.
17–Fig.
20showthegraphsforvaryingPRnodesactivitypatterns.
Similarly,Table6summarizestheharmfulinterferenceratioofFig.
17–Fig.
20.
InLongTermPRactivity,besidesofguaranteeinglowerHIRcomparedtoRD,HD,andSB,SURFalsoensuresahigherdeliveryratiothansuchapproaches.
InHighPRactivity,allthechannelarehighlyoccupied,andconsequently,verylesschancesforcommunicationislettoalltheapproaches.
Nevertheless,SURFisabletomanageverylowHIRandstillhavesomedeliveryratio(2%around),comparedtotheotherapproaches.
ItisclearthatwhenPRactivityisverylow(cfFig.
19)everystrategybehaveswellintermofHIR(cf.
19(a)).
Inthiscase,SURFhelpsselectthebestchannelintermofCRconnectivity,i.
e.
,deliveryratiotoCR(cf.
Fig.
19(b)),whilegeneratesverylessoralmostzeroHIR,whencomparedtoRD,SB,andHD.
ThereceiversratioisalsothehighestforSURF.
Unsurprisingly,thebestperformancegainisobservedintheintermittentcasewhenusingSURF:LowerHIRandhigherdeliveryratioisprovidedthanRD,HD,andSB.
Itisworthnotingthat,inthecaseswhereshortONforPRnodesisconsidered(i.
e.
,inintermittentorlowactivityscenarios),alltheapproachesperformthebetter.
How-ever,thechannelselectionmechanismprovidedbySURFcouldndthebestspectrumopportunitiesinallconsideredcases,whilerespectingthePRnodesactivities.
MainConclusions.
Conclusionsarequitetypicalandareforeverywirelesssystemingeneral:Whenthesystemisfree(LowPRactivity),everysolutionoffersacceptablegain.
Sometimesacleversolutiondoesnotworthitduetothecomplexityitintroduces.
Whenthesystemisclosetomaximumcapacity(HighPRactivity),allsolutionshavebadperformance.
WhenchannelsarefullyoccupiedbyPRsthereisnorealopportunityfortransmission,herealsothegainisverylowcomparedtothecomplexityofthesolutions.
Intermittentcaseisthecasewherecleversolutionsneedtooperate.
ThisiswhereSURFgivesthebestresultsandthetargetregiontoavailcommunicationoppor-tunities.
8.
5ImprovementsregardingSURFThechannelselectionstrategyprovidedbySURFcanbefurtherenhancedbyconsid-eringtheprimaryradionodesactivitypattern.
Intheprevioussection,wehavepointedoutthattheintermittentcaseisthecasewherecleversolutionsneedtooperate.
Inthisregard,wecanevaluatethe"power"ofusingotherhistory-basedmetrics(thattrytobetterinferthequalityofchannels)combinedwiththecurrentPwofSURF.
SURFisRRn°762836Mubashir&Aline&Hicham&SergethenrequiredtokeeptrackofhistoryofpastPRnodesactivity.
ThishistorycouldbeusedtogivemoreweighttothechannelswithshortONinaverage.
Someexamplesofmetricsaregivenbelow:1)HowoftenthechannelisfreeHere,SURFmaykeephistoryofchannelstates.
SURFthenconsidersan"observationtimewindow".
Inthismanner,SURFwillcom-putetheratioofbeingfreeoverthewindow(thesizeofthetimewindowcouldbevariedtoevaluatetheimpact).
2)HowlongchannelsstayinOFFstateHere,SURFmaycomputethedurationofOFFstateoverthetotaltimeintheconsideredtimewindow.
ThismetricdependsonhowSURFkeepsthehistoryofchannelstates.
Thiscouldbedoneonsingletimeslotbasisorvaryingslotsoftime.
Thismetricalsodependsuponwhenthevericationforafreechannelisperformed(periodicallyoronlywhenapacketevent(receptionortransmission)happens.
3)Whatwastheratioofsuccess(receptionortransmission)overthetimesthechannelwasinOFFstateThismetricwillgivethequalityofthechannelintermsofcontention,variabilityofPRactivities,etc.
Asplanofourfuturework,weintendtoimproveSURFbykeepingthehistoryofPRnodesactivity.
Moreover,wealsoplantobroadenourinvestigationontheimpactofPRnodesactivityonchannelselectionstrategies.
Wewanttoachievethisbyconsid-eringotherPRactivitymodels,suchasBernoulliProcess,BetaDistributionetc.
RealtimePRnodesactivitytracescanalsobeincludedinourstudiestobroadenthescopeofourinvestigation.
9RelatedWorkRecently,alotofchannelselectionstrategieshavebeenproposedforcognitiveradionetworks[5,6,7,8,9,10,12,13,14,15,16].
Thesechannelselectionstrategiesaredesignedtoachievedifferentperformancegoals,forinstance,optimizationofthrough-put,delay,etc.
Besidesachievingthesegoals,eachchannelselectionstrategyhasanature,accordingtoitsreactionwiththeappearanceofPRnodesontheCRcommu-nicatingchannel.
Therefore,channelselectionstrategiescanbeclassiedintothreecategoriesbynature:(1)proactive(predictive),(2)thresholdbased,and(3)reactive.
Fromthecommunicationperspective,channelselectionstrategiescanbeclassiedintocentralizedanddistributed.
Theclassicationofchannelselectionstrategiesincogni-tiveradionetworksisshowninFig.
21.
Table7comparesdifferentchannelselectionstrategiesforcognitiveradionetworksandtheirfeatures.
Inthefollowing,wediscusseachclassicationindetail.
9.
1GoalsofChannelSelectionStrategiesFromtheperformanceperspective,authorsin[5]proposedachannelselectionstrategytosatisfythetrafcdemandsofAccessPoints.
Severalchannelselectionstrategieshaveproposedintheliteratureforthroughputmaximization[16,47,14,8,48].
In[14],theauthorsdeterminedthetransmissionscheduleoftheCRnodesinordertoimprovethenetworkthroughput.
In[16],theauthorsimprovedthethroughputoftheCRusersintheTVbroadcastnetwork.
Infact,theauthorsproposedapredictivechannelselec-tionschemetomaximizespectrumutilizationandminimizedisruptionstoPRnodes.
Theyconsideredasingle-hopnetworkinwhichCRnodescoordinateswiththeTVINRIASURF:ADistributedChannelSelectionStrategy37receivertocollectinformationregardingPRactivity.
Twoopportunisticchannelse-lectionschemes,CSS-MCRAandCSS-MHRA,areproposedin[47].
InCSS-MCRA,thegoalwastomaximizethethroughputwhileminimizethecollisionrate.
InCSS-MHRA,thegoalwastomaximizethethroughputwhileminimizethehandoffrate.
CSS-MCRAandCSS-MHRAbothconsideredsingleuserandarepredictiveinnature.
Loadbalancingisanotherimportantgoalofchannelselectionstrategies[11,49].
In[11],theauthorsproposedachannelandpowerallocationschemeforCRnetworks.
TheobjectivewastomaximizethesumdatarateofallCRs.
Theyconsideredtheavailabilityofacentralizedauthority,whichmonitorsthePRactivityandassignchan-nelstoCRnodes.
Sensing-basedandprobability-basedspectrumdecisionschemesareproposedin[49]todistributetheloadofCRnodestomultiplechannels.
Theauthorsderivedtheoptimalnumberofcandidatechannelsforsensing-basedschemeandtheoptimalchannelselectionprobabilityforprobability-basedspectrumdecisionscheme.
TheobjectiveofbothschemeswastominimizetheoverallsystemtimeoftheCRusers.
Theauthorsin[35]proposedapredictivechannelselectionschemetominimizethechannelswitchingdelayofasingleCRnode.
Otherchannelselectionstrategiesfocusonoptimizingtheexpectedwaitingtime[50,51],remainingidletime[52,53],reducesystemoverheadandimproveCRQoS[54].
Apredictivechannelselectionstrategy,VoluntarySpectrumHandoff(VSH)[52],isproposedtoreducethecommunicationdis-ruptiondurationduetohandoffsandtoselectthechannelthathasmaximumremainingidletime.
However,VSHrequiresthepresenceofSpectrumServer(SS),acentralizedentity,tomonitortheactivitiesofPRandCRnodes.
In[12],theauthorsproposedachannelselectionschemetomaximizethetotalchannelutilization.
Intheirpaper,theauthorsconsidersource-destinationpairsinsingle-hopcontext.
Channelselectionstrategiescanalsobeusedinconjunctionwithroutingprotocolsforreliablepathselec-tion[4]andgoodrouteselectionfordelaysensitiveapplications[55].
Boththechannelselectionschemes[4,55]aredesignedtoworkwithroutingprotocols,whileinSURF,weconsiderchannelselectionschemeforbroadcasting.
9.
2NatureofChannelSelectionStrategiesInproactivechannelselectionstrategies[7,56,34,57,58,16,35],theactivityofPRnodesispredictedandtheCRnodesmovetothechannelaccordingtotheprediction.
In[7,57],theauthorsclassiedthePRtrafcandapplieddifferentpredictionrules.
ThesepredictionruleswerethenusedinthepredictivechannelselectionschemetondthechannelswiththelongestidletimesforCRuse.
In[58],theauthorsexploredtwoapproachesofpredictivedynamicspectrumaccess(PDSA).
Theirbasicgoalwastopredictwhenthechannelswillbeidle,basedonobservationsoftheprimaryradionodeschannelusage.
TheydeterminedtheexpectedchannelidletimesforCRusage.
Twoproactivechannelselectionstrategies,PRO-IandPRO-IIareproposedin[34].
ThegoaloftheseschemesweretominimizedisruptionstoPRsandthroughputmaxi-mizationofCRnodes.
TheauthorsusesasinglepairofCRnodesandtheyignoredtheimpactofotherCRnodescontendingforthechannel.
NotethatinSURF,CRnodesselectthechannelwhichhashighestprobabilityofbeinginidlestate.
Theauthorsin[56]proposedachannelselectionschemethatoptimizesthedelayinndingthechannelsusingthehistory.
Theirschemeisbasedontwosteps:thedatabasestepandthesignaldetectionstep.
Inthedatabasestep,thedatabasecollectsinformationaboutthechannels.
TheCRnode,whenrequiredachannelfortransmis-sion,sendsaquerytothedatabase.
ThedatabasethenprovidesthemostprobableRRn°762838Mubashir&Aline&Hicham&Sergeunoccupiedchannels,whicharethebestcandidatesforsearchingthechannels.
ThesechannelsarethensubmittedtotheCRnode.
TheCRnodethenperformsthepowerleveldetection,andwhenrequired,thefullsignaldetection.
CRnodethenselectsthechannelsbasedonthepriority.
ThresholdbasedschemesarethosechannelselectionschemesinwhichthePRnodesactiveallthetimeandnoidlechannelisavailabletoCRnodes.
Intheseschemes,CRnodesareallowedtosharethechannelaslongastheinterferencecausedbytheCRnodestothePRnodesisbelowacertainthreshold.
Forinstance,[14]isathresholdbasedchannelselectionscheme.
Inthisscheme,theauthorsconsideredtheTVbroad-castnetworkasaprimarynetwork.
EachTVreceiverisequippedwithasensor,whichisresponsibleformonitoringtheactivityofTVreceiver.
Thissensorthencommuni-catesthePRactivityinformationtotheCRnodes.
CRnodesusethishistoricalPRactivityinformationtoselectthechannel.
Inreactivechannelselectionstrategies[59,60,61,62,63],channelswitchingoc-cursafterthePRnodeappears.
Infact,inreactivechannelselectionschemes,CRnodesmonitorlocalspectrumthroughindividualorcollaborativesensing[61,62,63,64,65,66].
Afterdetectingachangeinthespectrum,e.
g.
,channelisoccupiedbyPRnode,CRnodestopthetransmission,returnbackthechanneltothePRnodeandsearchforotherchanneltoresumethetransmission.
In[67],theauthorscomparedtwotypesofspectrumhandoffschemes:proactiveandreactivespectrumhandoffschemes.
Inreactive-sensinghandoffscheme,thetargetchannelisselectedafterthespectrumhandoffrequestismade.
Whileinproactivespectrumhandoffscheme,thetargetchan-nelispredetermined.
Theauthorsmentionedthattheadvantageofreactivespectrumhandoffschemeresidesintheaccuracyoftheselectedtargetchannel,butincursthecostofsensingtime.
Onthecontrary,theproactivespectrumhandoffschemeavoidthesensingtime,butthepre-determinedchannelmaynotbeavailable.
In[68],theco-authorsof[67]providedthemodelingandanalysisofreactivespectrumhandoffschemeinmoredetail.
In[59],theauthorsproposedasensing-basedopportunisticchannelaccessscheme.
TheyconsideredaPrimaryTVbroadcastnetwork.
TheyalsoconsideredasinglePRnodeandasingleCRnodeandabasestationisrequiredforkeepingtheprimarychan-nel'sstatistics.
Areactivemulti-channelmacprotocol,RMC-MAC,foropportunisticspectrumaccessisproposedin[60].
Theirobjectivewastoincreasethebandwidthutilizationandtoreducetheforcedterminationprobability.
However,theyconsideredasingle-hopCRnetwork.
Dynamicfrequencyhoppingcommunities(DFHC)[69]isalsoareactiveapproach,whichisdesignedforIEEE802.
22networks.
DFHCisasinglehopapproachandrequiresthepresenceofbasestation.
9.
3ChannelSelectionStrategiesfromtheCommunicationPerspec-tiveFromthecommunicationperspective,channelselectionstrategiescanbeclassiedintocentralizedanddistributed.
In[71],acomparisonbetweencentralizedanddistributedapproachesforspectrummanagementisdone.
Incentralizedchannelselectionstrate-gies,acentralizedentityispresent,whichhelpsCRnodesintheirchannelselectiondecision,e.
g.
,[72,73,74].
Theauthorsin[75]investigateddifferentstepsforthede-velopmentofcentralizedalgorithmsfordifferentradionetworks.
In[5],theauthorsproposedanefcientspectrumallocationarchitecturethatadaptstodynamictrafcINRIASURF:ADistributedChannelSelectionStrategy39Figure21:ClassicationofchannelselectionstrategiesforCognitiveRadioNetworks.
RRn°762840Mubashir&Aline&Hicham&SergeTable7:Channelselectionstrategiesandtheirfeatures.
StrategyGoalNatureHop/UserVSH[52]RemainingidletimePredictiveCentralized[16]Maximizechannelutilization,throughputmaximizationandminimizedisruptionstoPRsPredictiveSingle-hopSWIFT[10]Combinemultiplenon-contiguousunoccupiedbandstocreateahigh-throughputwidebandlinkworkonunlicensedbandN/ACBH,LH[13]Maximizechannelutilization&decreasemessageoverheadReactiveMulti-hopWAIT[15]MaximizethroughputReactiveSingle-hopCSS-MCRA[47]MinimizecollisionrateandThroughputmaximizationPredictiveSingleuserCSS-MHRA[47]MinimizehandoffrateandThroughputmaximizationPredictiveSingleuser[14]ThroughputmaximizationThresholdbasedCentralizedPS-OSA[48]ThroughputmaximizationN/ACRpairs[11]LoadbalancingReactiveCentralized[49]LoadbalancingPredictive/ReactiveSingle-hop[35]ReducechannelswitchingdelayPredictiveSingleuserSCA-MAC[50]ExpectedwaitingtimePredictiveN/APOSH[51]ExpectedwaitingtimePredictiveN/AFLEX[5]TrafcdemandsofAccessPointsN/ASingle-hopIEEE802.
22[6]InternationalwirelessstandardbasedonCRtechnologytouseTVspectrumwithoutcausingharmfulinterferencetoTVdevicesN/ACentralized[53]RemainingidletimeProactiveCRpairs[54]ReducesystemoverheadandimproveCRQoSN/AN/A[12]MaximizetotalchannelutilizationReactiveCRpairsMPP[4]ReliablepathselectionN/AMulti-hop[55]RouteselectionfordelaysensitiveapplicationsReactiveDistributed[70]RouteselectionfordelaysensitiveapplicationsReactiveDistributed[7]FindlongestidletimechannelPredictiveN/A[56]OptimizedelayinndingthechannelProactiveN/APRO-I,PRO-II[34]MinimizedisruptionstoPRs,throughputmaximizationProactiveSinglepair[57]Reducedelay&channelswitching,maximizethroughputPredictiveN/APDSA[58]TodetermineexpectedchannelidletimePredictiveN/A[59]OutagerequirementofPRuserCRReactiveCentralizedRMC-MAC[60]ReduceforcedterminationprobabilityandincreasebandwidthutilizationReactiveSingle-hopDFHC[69]BetterQoSandmaximizethroughputReactiveCentralizedSB[17]DataReachabilityN/ADistributedSURFDataReachability,minimizedisruptionstoPRsPredictiveDistributedINRIASURF:ADistributedChannelSelectionStrategy41demandsbuttheyconsideredasingle-hopscenarioofAccessPoints(APs)inWi-Finetworks.
Anapproachthatusenon-continuousunoccupiedbandtocreateahighthroughputlinkisdiscussedin[10].
In[14],theauthorsproposedacentralizedalgo-rithmofchannelsharingbetweenCRnodes.
Theiralgorithmisdesignedforsource-destinationpairsandisspeciallydesignedforsingle-hopcommunication.
Intheirpa-per,theauthorsassumedthatallthePRsareactiveallthetimeandnoidlechannelisavailabletoCRnodesfortheircommunication.
AchannelallocationschemeforIEEE802.
22standardisproposedin[76].
However,thisschemeiscentralizedinnature.
Theauthorsin[77]proposedanopportunisticchannelselectionschemeforIEEE802.
11-basedwirelessmeshnetworks.
However,anAccessPoint(AP)isrequiredtoconnectthenodestotheInternetviameshrouter.
Inmulti-hopcognitiveradionetworks,thereisnocentralizedentitythathelpsCRnodesintheirchannelselectiondecision.
Therefore,distributedchannelselec-tionstrategiesarerequired.
Moreover,CRnodeshavetorelyontheirlocallyinferredinformationandnocooperationorfeedbackisexpectedfromtheprimaryradionodes.
DuetoPRactivity,theholdingtimeandthegranularityofwirelessspectrumbandsalsoaffectsonmulti-hopCRcommunications[22].
Thus,anadaptivechannelselec-tionstrategyisrequiredatboththesenderandreceivernode,sothatthereceivernodetunedtotherightchanneltoreceivesentinformation.
Allthesefactorsmakeschan-nelselectioninthesenetworksextremelychallenging,havingveryfewworksbeendonesofar[17,70,55].
In[70,55],theauthorsproposedadynamicresourceman-agementschemeformulti-hopcognitiveradionetworks.
Infact,theirapproachisaroute/channelselectionfordelaysensitiveapplications,suchamultimediastreaming,whileSURFisachannelselectionschemefordatadisseminationandnotforrouting.
Inselectivebroadcasting(SB)[17],eachcognitivenodeselectsaminimumsetofchannels(ECS)coveringallofitsgeographicneighborstodisseminatemessagesinmulti-hopcognitiveradionetworks.
Therearehowever,severalchallengesinthepracticalityofSB.
Indeed,fromthecommunicationperspective,simultaneoustrans-missionoveraECSrequiresmorethanonetransceiver,whichmeanshavingbiggerandmorecomplexdevices,asitisdoneformilitaryapplications[18].
Onthecontrary,usingasingletransceivertotransmitoverminimumsetofchannelsrequiresdetermin-ingthecorrectchanneltooverhearingatransmission,increasesdelay,andbringsfre-quentchannelswitching.
Secondly,fromtheperspectiveofoverhearing,eitherneigh-bornodesneedtosimultaneouslyoverhearovermultiplechannelsorsynchronizationisrequiredamongneighbors,whichincursschedulingoverhead.
Comparedtotheseaforementionedchannelselectionapproaches,SURFispredictiveinnature.
ThegoalofSURFistoachievehigherdatareachabilityandminimizedisruptionstoPRsinthemulti-hopnetwork.
10ConclusionandFutureWorkInthispaper,wehaveintroducedSURF,anintelligentanddistributedchannelselectionstrategyforreliabledatadisseminationinmulti-hopcognitiveradioad-hocnetworks.
ThemaindesignobjectiveofSURFistheprotectionofprimaryradionodesagainstharmfulinterferencebyCRtransmissionsandtheincreaseofdisseminationreliabilityincognitiveradioad-hocnetwork.
Thesetwogoalswereachievedbyclassifyingthechannelsonthebasisofprimaryradiounoccupancyandthenumberofcognitiveradioneighborsusingeachchannel.
SimulationresultsinNS-2conrmedthatSURF,whenRRn°762842Mubashir&Aline&Hicham&Sergecomparedtorandom-based,higherdegree,andselectivebroadcastingstrategies,isef-fectiveinselectingthebestchannels.
Furthermore,weshowthatunlikeothersolutions,SURFperformanceisenhancedwiththeincreaseofthenumberofexistingchannels.
Thisisduetoitsintelligentselectionmechanism.
WeintendinfuturetoconsiderthetrafcanddataratesofCRnodesinthechannel'sweightcalculationformula,aswellastimeneededtodisseminatemessagesinthenetwork.
References[1]J.
Mitola,G.
Q.
Maguire,Cognitiveradio:Makingsoftwareradiosmorepersonal,IEEEPersonalCommunications6,no.
4(1999)13–18.
[2]FCC,Spectrumpolicytaskforce(sptf),reportofthespectrumefciencywg,Nov.
2002.
[3]I.
F.
Akyildiz,W.
-Y.
Lee,M.
C.
Vuran,S.
Mohanty,Nextgeneration/dynamicspectrumaccess/cognitiveradiowirelessnetworks:asurvey,ComputerNet-works:TheInternationalJournalofComputerandTelecommunicationsNet-working50,Issue13(2006)2127–2159.
[4]H.
Khalife,S.
Ahuja,N.
Malouch,M.
Krunz,Probabilisticpathselectioninop-portunisticcognitiveradionetworks,in:proceedingsoftheIEEEglobecomcon-ference,2008,pp.
1–5.
[5]L.
Yang,L.
Cao,H.
Zheng,E.
Belding,Trafc-awaredynamicspectrumac-cess,in:ProceedingsofTheFourthInternationalWirelessInternetConference(WICON2008),Hawaii,USA,2008.
[6]C.
Cordeiro,K.
Challapali,D.
Birru,S.
N.
Shanka,Ieee802.
22:Anintroductiontotherstwirelessstandardbasedoncognitiveradios,JournalofCommunica-tions1(1)(2006)38–47.
[7]M.
Hoyhtya,S.
Pollin,A.
Mammela,Classication-basedpredictivechannelse-lectionforcognitiveradios,in:IEEEICC,2010.
[8]Q.
Zhao,L.
Tong,A.
Swami,Y.
Chen,Decentralizedcognitivemacforoppor-tunisticspectrumaccessinadhocnetworks:Apomdpframewrok,IEEEJournalonSelectedAreasinCommunications25,No.
3(2007)589–600.
[9]D.
Niyato,E.
Hossain,Competitivespectrumsharingincognitiveradionetworks:Adynamicgameapproach,IEEETransactionsonwirelesscommunications7(7)(July2008)2651–2660.
[10]H.
Rahul,N.
Kushman,D.
Katabi,C.
Sodini,F.
Edalat,Learningtoshare:Narrowband-friendlywidebandwirelessnetworks,in:ACMSIGCOMM,Vol.
38,Issue4,SEATTLE,WA,USA,2008,pp.
147–158.
[11]H.
Wang,J.
Ren,T.
Li,Resourceallocationwithloadbalancingforcogntiveradionetworks,in:IEEEGlobeCom,2010.
[12]F.
Hou,J.
Huang,Dynamicchannelselectionincognitiveradionetworkwithchannelheterogeneity,in:IEEEGlobecom,2010.
INRIASURF:ADistributedChannelSelectionStrategy43[13]V.
S.
Rao,R.
V.
Prasad,C.
Yadati,I.
Niemegeers,Distributedheuristicsforallo-catingspectrumincradhocnetworks,in:IEEEGlobecom,2010.
[14]G.
D.
Nguyen,S.
Kompella,Channelsharingincognitiveradionetworks,in:MILCOM,2010.
[15]P.
Anggraeni,N.
Mahmood,J.
Berthod,N.
Chaussonniere,L.
My,H.
Yomo,Dynamicchannelselectionforcognitiveradioswithheterogenousprimarybands,WirelessPersonalCommunications45(2008)369–384,10.
1007/s11277-008-9464-7.
URLhttp://dx.
doi.
org/10.
1007/s11277-008-9464-7[16]P.
A.
K.
Acharya,S.
Singh,H.
Zheng,Reliableopenspectrumcommunicationsthroughproactivespectrumaccess,in:TAPAS,2006.
[17]Y.
R.
Kondareddy,P.
Agrawal,Selectivebroadcastinginmulti-hopcognitivera-dionetworks,in:IEEESarnoffSymposium,Princeton,NewJersy,2008,pp.
1–5.
[18]O.
Younis,L.
Kant,K.
Chang,K.
Young,Cognitivemanetdesignformission-criticalnetworks,IEEECommunicationsMagazine(2009)64–71.
[19][link].
URLhttp://stuweb.
ee.
mtu.
edu/~ljialian/[20][link].
URLhttp://www.
isi.
edu/nsnam/ns/[21]I.
F.
Akyildiz,W.
-Y.
Lee,K.
R.
Chowdhury,Crahns:Cognitiveradioadhocnetworks,AdHocNetworks7,issue5(2009)810–836.
[22]H.
Khalife,N.
Malouch,S.
Fdida,Multihopcognitiveradionetworks:torouteornottoroute,IEEENetworks(2009)20–25.
[23]H.
Harada,Asmall-sizesoftwaredenedcognitiveradioprototype,in:Proceed-ingsoftheIEEEInternationalSymposiumonPersonal,IndoorandMobileRadioCommunications(PIMRC),Cannes,France,2008,pp.
1–5.
[24]K.
G.
Shin,H.
Kim,C.
Cordeiro,K.
Challapali,Anexperimentalapproachtospectrumsensingincognitiveradionetworkswithoff-the-shelfieee802.
11de-vices,in:4thIEEEConsumerCommunicationsandNetworkingConference,CCNC2007.
,2007,pp.
1154–1158.
doi:10.
1109/CCNC.
2007.
232.
[25]T.
Yucek,H.
Arslan,Asurveyofspectrumsensingalgorithmsforcognitiveradioapplications,IEEECommunicationsSurveysandTutorials11,IssueNo.
1(FirstQuarter2009)116–130.
[26]W.
-Y.
Lee,I.
Akyildiz,Optimalspectrumsensingframeworkforcognitiveradionetworks,IEEETransactionsonWirelessCommunications7(10)(2008)3845–3857.
doi:10.
1109/T-WC.
2008.
070391.
[27]G.
Yuan,R.
Grammenos,Y.
Yang,W.
Wang,Performanceanaly-sisofselectiveopportunisticspectrumaccesswithtrafcprediction,IEEETransactionsonVehicularTechnology59(4)(2010)1949–1959.
doi:10.
1109/TVT.
2009.
2039155.
RRn°762844Mubashir&Aline&Hicham&Serge[28]A.
W.
Min,K.
G.
Shin,Exploitingmulti-channeldiversityinspectrum-agilenet-works,in:ProceedingsofINFOCOM,2008,pp.
1921–1929.
[29]H.
Kim,K.
Shin,Efcientdiscoveryofspectrumopportunitieswithmac-layersensingincognitiveradionetworks,IEEETransactionsonMobileComputing7(5)(2008)533–545.
doi:10.
1109/TMC.
2007.
70751.
[30]S.
L.
LoukasLazos,M.
Krunz,Spectrumopportunity-basedcontrolchannelas-signmentincognitiveradionetworks,in:6thAnnualIEEECommunicationsSo-cietyConferenceonSensor,MeshandAdHocCommunicationsandNetworks(SECON2009).
,Rome,Italy,2009.
[31]H.
Kim,K.
Shin,Fastdiscoveryofspectrumopportunitiesincognitiveradionetworks,in:3rdIEEESymposiaonNewFrontiersinDynamicSpectrumAccessNetworks(IEEEDySPAN),October,2008.
,pp.
1–12.
[32]O.
Mehanna,A.
Sultan,H.
E.
Gamal,Cognitivemacprotocolsforgeneralpri-marynetworkmodels,CoRRabs/0907.
4031.
[33]A.
S.
Zahmati,X.
Fernando,A.
Grami,Steady-statemarkovchainanalysisforheterogeneouscognitiveradionetworks,in:Proceedingsofthe33rdIEEEconferenceonSarnoff,Sarnoff'10,IEEEPress,Piscataway,NJ,USA,2010,pp.
107–111.
URLhttp://portal.
acm.
org/citation.
cfmid=1843486.
1843507[34]L.
Yang,L.
Cao,H.
Zheng,Proactivechannelaccessindynamicspectrumnet-works,ElsevierPhysicalCommunicationsJournalvol(1)(2008)103–111.
[35]Q.
D.
XueFeng,S.
Guangxi,L.
Yanchun,Smartchannelswichingincognitiveradionetworks,in:CISP,2009.
[36]B.
Vujicic,N.
Cackov,S.
Vujicic,L.
Trajkovic,Modelingandcharacterizationoftrafcinpublicsafetywirelessnetworks,in:InProc.
ofSPECTS,2005,pp.
214–223.
[37][link].
URLhttp://www.
fcc.
gov/pshs/public-safety-spectrum/700-MHz/[38]S.
Geirhofer,L.
Tong,B.
M.
Sadler,Dynamicspectrumaccessinwlanchannels:Empericalmodelanditsstochasticanalysis,in:ACMTAPAS,2006.
[39]A.
Adas,Trafcmodelsinbroadbandnetworks,IEEECommunicationsMaga-zine35(7)(1997)82–89.
doi:10.
1109/35.
601746.
[40]K.
Sriram,W.
Whitt,Characterizingsuperpositionarrivalprocessesinpacketmultiplexersforvoiceanddata,IEEEJournalonSelectedAreasinCommuni-cations4(6)(1986)833–846.
doi:10.
1109/JSAC.
1986.
1146402.
[41]C.
Arachchige,S.
Venkatesan,N.
Mittal,Anasynchronousneighbordiscoveryalgorithmforcognitiveradionetworks,in:NewFrontiersinDynamicSpectrumAccessNetworks,2008.
DySPAN2008.
3rdIEEESymposiumon,2008,pp.
1–5.
doi:10.
1109/DYSPAN.
2008.
78.
INRIASURF:ADistributedChannelSelectionStrategy45[42]N.
Mittal,S.
Krishnamurthya,R.
Chandrasekarana,S.
Venkatesana,Y.
Zenga,Onneighbordiscoveryincognitiveradionetworks,JournalofParallelandDis-tributedComputing69,issue7(July2009)623–637.
[43]M.
DiFelice,K.
R.
Chowdhury,L.
Bononi,Modelingandperformanceevaluationoftransmissioncontrolprotocolovercognitiveradioad-hocnetworks,in:MSWiM'09,ACM,NewYork,NY,USA,2009,pp.
4–12.
doi:http://doi.
acm.
org/10.
1145/1641804.
1641809.
URLhttp://doi.
acm.
org/10.
1145/1641804.
1641809[44]S.
Bayhan,F.
Alagz,Distributedchannelselectionincrahns:Anon-selshschemeformitigatingspectrumfragmAdHocNetworksInPress,CorrectedProof(2011)–.
doi:DOI:10.
1016/j.
adhoc.
2011.
04.
010.
URLhttp://www.
sciencedirect.
com/science/article/pii/S1570870511000941[45]J.
Riihijarvi,J.
Nasreddine,P.
Mahonen,Impactofprimaryuseractivitypatternsonspatialspectrumreuseopportunities,in:EuropeanWirelessConference(EW2010),2010.
[46]M.
H.
Rehmani,A.
C.
Viana,H.
Khalife,S.
Fdida,Improvingdatadisseminationinmulti-hopcognitiveradioad-hocnetworks,in:3rdInternationalICSTConfer-enceonAdHocNetworks(ADHOCNETS2011),Paris,France–(Toappear).
,21-23Sep2011.
[47]Q.
Xiao,Y.
Li,M.
Zhao,S.
Zhou,J.
Wang,Opportunisticchannelselectionap-proachundercollisionprobabilityconstraintincognitiveradiosystems,Com-puterCommunications32(18)(2009)1914–1922.
[48]Q.
Zhao,S.
Geirhofer,L.
Tong,B.
M.
Sadler,Opportunisticspectrumaccessviaperiodicchannelsensing,IEEETransactionsonSignalProcessing56(2)(2008)785–796.
[49]L.
-C.
Wang,C.
-W.
Wang,F.
Adachi,Load-balancingspectrumdecisionforcog-nitiveradionetworks,IEEEJournalonSelectedAreasinCommunications29(4)(2011)757–769.
[50]A.
C.
-C.
Hsu,D.
S.
-L.
Wei,C.
-C.
J.
Kua,Acognitivemacprotocolusingstatis-ticalchannelallocationforwirelessad-hocnetworks,in:WCNC,2007.
[51]R.
-T.
Ma,Y.
-P.
Hsu,K.
-T.
Feng,Apomdp-basedspectrumhandoffprotocolforpartiallyobservablecognitiveradionetworks,in:WCNC,2009.
[52]S.
-U.
Yoon,E.
Ekici,Voluntaryspectrumhandoff:Anovelapproachtospectrummanagementincrns,in:ICC,2010.
[53]Y.
Song,J.
Xie,Commonhoppingbasedproactivespectrumhandoffincogntiveradioadhocnetworks,in:GlobeCom,2010.
[54]P.
Zhu,J.
Li,X.
Wang,Anewchannelparameterforcognitiveradio,in:Crown-Com,2007.
[55]H.
P.
Shiang,M.
V.
D.
Schaar,Delay-sensitiveresourcemanagementinmulti-hopcognitiveradionetworks,in:IEEEDySpan,2008.
RRn°762846Mubashir&Aline&Hicham&Serge[56]J.
Vartiainen,M.
Hoyhtya,J.
Lehtomaki,T.
Braysy,Prioritychannelselectionbasedondetectionhistorydatabase,in:CROWNCOM,2010.
[57]M.
Hoyhtya,S.
Pollin,A.
Mammela,Performanceimprovementwithpredictivechannelselectionforcognitiveradios,in:FirstInternationalWorkshoponCog-nitiveRadioandAdvancedSpectrumManagement,CogART2008.
,Denmark,2008,pp.
1–5.
[58]T.
C.
Clancy,B.
D.
Walker,Predictivedynamicspectrumaccess,in:Proc.
SDRForumTechnicalConference,Orlando,Florida,USA,2006.
[59]X.
Liu,S.
N.
,Sensing-basedopportunisticchannelaccess,MobileNetworksandApplications11(2006)577–591,10.
1007/s11036-006-7323-x.
URLhttp://dx.
doi.
org/10.
1007/s11036-006-7323-x[60]S.
Fourati,S.
Hamouda,S.
Tabbane,Rmc-mac:Areactivemulti-channelmacprotocolforopportunisticspectrumaccess,in:4thIFIPInternationalConfer-enceonNewTechnologies,MobilityandSecurity(NTMS),2011,pp.
1–5.
doi:10.
1109/NTMS.
2011.
5721056.
[61]S.
M.
Mishra,A.
Sahai,R.
Brodersen,Cooperativesensingamongcognitivera-dios,in:IEEEICC,2006.
[62]N.
H.
A.
Sahai,R.
Tandra,Somefundamentallimitsoncognitiveradios,in:42AllertonconferenceonCommunication,ControlandComputing,2004.
[63]R.
T.
A.
Sahai,N.
Hoven,Opportunisticspectrumuseforsensornetworks:theneedforlocalcooperation,in:IPSN,2006.
[64]G.
Ganesan,Y.
G.
Li,Cooperativespectrumsensingincognitiveradionetworks,in:IEEEDySPAN,November2005.
[65]A.
Ghasemi,E.
S.
Sousa,Collaborativespectrumsensingforopportunisticaccessinfadingenvironments,in:IEEEDySPAN,November2005.
[66]D.
Cabric,S.
M.
Mishra,R.
W.
Brodersen,Implementationissuesinspectrumsensingforcognitiveradios,in:ProceedingsofAsilomarconferenceonsignals,systemsandcomputers,2004.
[67]L.
-C.
Wang,C.
-W.
Wang,Spectrumhandoffforcognitiveradionetworks:Reactive-sensingorproactive-sensins,in:IEEEInternationalPerformance,ComputingandCommunicationsConference(IPCCC2008),2008.
[68]C.
-W.
Wang,L.
-C.
Wang,F.
Adachi,Modelingandanalysisforreactive-decisionspectrumhandoffincognitiveradionetworks,in:IEEEGlobalTelecommunicationsConference(GLOBECOM2010),2010,pp.
1–6.
doi:10.
1109/GLOCOM.
2010.
5683644.
[69]W.
Hu,D.
Willkomm,M.
Abusubaih,J.
Gross,G.
Vlantis,M.
Gerla,A.
Wolisz,Cognitiveradiosfordynamicspectrumaccess-dynamicfrequencyhoppingcom-munitiesforefcientieee802.
22operation,IEEECommunicationsMagazine45(5)(2007)80–87.
doi:10.
1109/MCOM.
2007.
358853.
INRIASURF:ADistributedChannelSelectionStrategy47[70]J.
Zhao,H.
Zheng,G.
H.
Yang,Distributedcoordinationindynamicspectrumallocationnetworks,in:FirstIEEEInternationalSymposiumonNewFrontiersinDynamicSpectrumAccessNetworks(DySPAN),Baltimore,Maryland,USA,2005,pp.
259–268.
[71]G.
Salami,A.
Attar,a.
R.
T.
OliverHolland,H.
Aghvami,Acomparisonbetweenthecentralizedanddistributedapproachesforspectrummanagement,IEEECom-municationsSurveysandTutorials13,No.
2(2011)274–290.
[72]M.
Pereirasamy,J.
Luo,M.
Dillinger,C.
Hartmann,Dynamicinter-operatorspec-trumsharingforumtsfddwithdisplacedcellularnetworks,in:IEEEWirelessCommunicationsandNetworkingConference,Vol.
3,2005,pp.
1720–1725Vol.
3.
doi:10.
1109/WCNC.
2005.
1424772.
[73]M.
Buddhikot,K.
Ryan,Spectrummanagementincoordinateddynamicspectrumaccessbasedcellularnetworks,in:FirstIEEEInternationalSymposiumonNewFrontiersinDynamicSpectrumAccessNetworks,DySPAN'05,2005,pp.
299–307.
doi:10.
1109/DYSPAN.
2005.
1542646.
[74]R.
Kulkarni,S.
A.
Zekavat,Trafcawareintervendordynamicspectrumallocation:performanceinmulti-vendorin:Proceedingsofthe2006internationalconferenceIWCMC'06,ACM,NewYork,NY,USA,2006,pp.
85–90.
doi:http://doi.
acm.
org/10.
1145/1143549.
1143569.
URLhttp://doi.
acm.
org/10.
1145/1143549.
1143569[75]P.
Leaves,K.
Moessner,R.
Tafazolli,D.
Grandblaise,D.
Bourse,R.
Ton-jes,M.
Breveglieri,Dynamicspectrumallocationincompositerecongurablewirelessnetworks,IEEECommunicationsMagazine42(5)(2004)72–81.
doi:10.
1109/MCOM.
2004.
1299346.
[76]J.
Zhu,S.
Li,Channelallocationmechanismsforcognitiveradionetworksviarepeatedmulti-bidauction,in:ICWMMN2006,2006.
[77]D.
Niyato,E.
Hossain,Cognitiveradiofornext-generationwirelessnet-works:anapproachtoopportunisticchannelselectioninieee802.
11-basedwirelessmesh,IEEEWirelessCommunications16(1)(2009)46–54.
doi:10.
1109/MWC.
2009.
4804368.
Contents1Introduction32ChallengesofDataDisseminationinCognitiveRadioNetworks53SystemModelandAssumptions63.
1NetworkModel63.
2SpectrumSensingbyCognitiveRadioNodes73.
3PrimaryRadioActivityorWirelessChannelModel73.
4ExchangeofHelloPackets7RRn°762848Mubashir&Aline&Hicham&Serge4ChannelSelectionStrategySURF85PrimaryRadioUnoccupancy95.
1RecoveryfromBadChannelSelectionDecisions106CognitiveRadioOccupancy127PerformanceAnalysis137.
1ImplementationSetup137.
2PerformanceMetrics147.
3SimulationEnvironment157.
4SURFParametersEvaluation167.
4.
1RetriesinSURF177.
4.
2ImpactofVaryingNeighborhoodDensityonSURF177.
4.
3PRUtilizationoftheSelectedChannel187.
5SURFComparison197.
5.
1ProtectiontoPrimaryRadioNodes197.
5.
2ReliableDataDissemination217.
5.
3TuningofSenderandReceiver227.
5.
4PacketRatio238ActivitypatternimpactofPRNodesonChannelSelectionStrategies268.
1Context268.
2ChannelSelectionStrategies268.
3PrimaryRadioNodesActivityPattern278.
4PerformanceAnalysis348.
5ImprovementsregardingSURF359RelatedWork369.
1GoalsofChannelSelectionStrategies369.
2NatureofChannelSelectionStrategies379.
3ChannelSelectionStrategiesfromtheCommunicationPerspective.
.
3810ConclusionandFutureWork41INRIACentrederechercheINRIASaclay–le-de-FranceParcOrsayUniversité-ZACdesVignes4,rueJacquesMonod-91893OrsayCedex(France)CentrederechercheINRIABordeaux–SudOuest:DomaineUniversitaire-351,coursdelaLibération-33405TalenceCedexCentrederechercheINRIAGrenoble–Rhne-Alpes:655,avenuedel'Europe-38334MontbonnotSaint-IsmierCentrederechercheINRIALille–NordEurope:ParcScientiquedelaHauteBorne-40,avenueHalley-59650Villeneuved'AscqCentrederechercheINRIANancy–GrandEst:LORIA,TechnopledeNancy-Brabois-Campusscientique615,rueduJardinBotanique-BP101-54602Villers-lès-NancyCedexCentrederechercheINRIAParis–Rocquencourt:DomainedeVoluceau-Rocquencourt-BP105-78153LeChesnayCedexCentrederechercheINRIARennes–BretagneAtlantique:IRISA,CampusuniversitairedeBeaulieu-35042RennesCedexCentrederechercheINRIASophiaAntipolis–Méditerranée:2004,routedesLucioles-BP93-06902SophiaAntipolisCedexditeurINRIA-DomainedeVoluceau-Rocquencourt,BP105-78153LeChesnayCedex(France)http://www.
inria.
frISSN0249-6399