/
Achieving Single Channel Full Duplex Wireless Communication Jung Il Choi  Mayank Jain Achieving Single Channel Full Duplex Wireless Communication Jung Il Choi  Mayank Jain

Achieving Single Channel Full Duplex Wireless Communication Jung Il Choi Mayank Jain - PDF document

briana-ranney
briana-ranney . @briana-ranney
Follow
478 views
Uploaded On 2014-10-27

Achieving Single Channel Full Duplex Wireless Communication Jung Il Choi Mayank Jain - PPT Presentation

edu palcsstanfordedu skattistanfordedu Coprimary authors Abstract This paper discusses the design of a single channel fullduplex wireless transceiver The design uses a combination of RF and baseband techniques to achieve fullduplexing with minimal ef ID: 7736

edu palcsstanfordedu skattistanfordedu Coprimary authors

Share:

Link:

Embed:

Download Presentation from below link

Download Pdf The PPT/PDF document "Achieving Single Channel Full Duplex Wir..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.


Presentation Transcript

Figure2:Blockdiagramofawirelessfull-duplexnode.Coloredblockscorrespondtodifferenttechniquesforself-interferencecancellation.Thepowersplittersintroducea6dBreductioninsignal,thuspowerfromTX1is6dBlowercom-paredtopowerfromTX2,withouttheneedforanadditionalattenuator.comefeasible.Abriefoverviewoftheantennacancellationschemefollows.2.2AntennaCancellationThisschemeusestheinsightthattransmissionsfromtwoormoreantennasresultinconstructiveanddestructiveinterferencepatternsoverspace.Inthemostbasicimplementation,thetransmissionsig-nalfromanodeissplitamongtwotransmitantennas.Aseparatereceiveantennaisplacedsuchthatitsdistancefromthetwotrans-mitantennasdiffersbyanoddmultipleofhalfthewavelengthofthecenterfrequencyoftransmission.Forexample,ifthewavelengthoftransmissionis,andthedis-tanceofthereceiveantennaisdfromonetransmitantenna,thentheothertransmitantennaisplacedatd+=2awayfromthereceiveantenna.Thiscausesthesignalfromthetwotransmitantennastoadddestructively,thuscausingsignicantattenuationinthesignalreceived,atthereceiveantenna.Destructiveinterferenceismosteffectivewhenthesignalampli-tudesatthereceiverfromthetwotransmitantennasmatch.Theinputsignaltotheclosertransmitantennaisattenuatedtogetthereceivedamplitudetomatchthesignalfromthesecondtransmitantenna,thusachievingbettercancellation.Ageneralimplemen-tationcouldusedifferentlyplacedormorethanthreeantennastoachievebettercancellation.Antennasareoptimallyplacedonlyforline-of-sight(LOS)com-ponents.Ifantennasareplacedinacorner,forexample,there-ectedsignalsfromeachtransmitantennawillnotnecessarilycan-cel.Whilethisputsafundamentallimitationontheperformanceoftheantennacancellation,signalstrengthofthereectedsignalsistypicallymuchweakerthanLOSduetolongersignalpathandat-tenuationwhenreected.ItispossibletobringthissignalintothedynamicrangeoftheADCbyusingRFinterferencecancellationaftertheantennacancellationstage.Figure2showsablockdiagramofasystemincorporatingallthetechniquesforfull-duplexoperation.Whileeachtechniquehasitsownlimitations,thispapershowsconnectingallthreetechniquesinseriescanovercomethelimitations.3.ANTENNACANCELLATIONThissectionanalyzesthepossiblereductioninself-interferencebyusingantennacancellation.Italsoevaluatesitslimitswithre-specttobandwidthofthesignalbeingtransmittedandthesensitiv-ityofantennacancellationtoengineeringerrors.Itshows,usingactualmeasurements,thatantennacancellationachieves20dBre-ductioninself-interference.Thissectionalsoevaluatestheeffectsofusingtwotransmitantennasforantennacancellationonthecom-municationrange.Itshowsthatantennacancellationdegradesthereceivedsignalatothernodesinthenetworkbyatmost6dBcom-paredtothesingleantennasetup.3.1PerformanceofAntennaCancellationInanidealscenario,theamplitudesfromthetwotransmitan-tennaswouldbeperfectlymatchedatthereceiverandthephaseofthetwosignalswoulddifferbyexactly.However,wendthatthebandwidthofthetransmittedsignalplacesafundamentalboundontheperformanceofantennacancellation.Further,realworldsystemsarepronetoengineeringerrorswhichlimitsystemperformance.Thesensitivityoftheantennacancellationtoampli-tudemismatchatthereceiveantennaandtotheerrorinreceiveantennaplacementisimportanttoconsider.Toanalyzethereductionininterferenceusingantennacancel-lation,welookattheself-interferencesignalpoweratthereceiveantennaafterantennacancellation.ItisderivedinAppendixAtobe:2AantAant+Aantjx[t]j21�cos2dant +Aant2jx[t]j2whereAantistheamplitudeofthebasebandsignal,x[t],atthereceiveantennareceivedfromasingletransmitantenna.Aantistheamplitudedifferencebetweenthereceivedsignalsfromthetwotransmitantennasatthereceiveantenna.dantrepresentstheerrorinreceiverantennaplacementcomparedtotheidealcasewherethesignalsfromthetwoantennasarriveoutofphaseofeachother.Thisequationletsusevaluatethesensitivityofantennacancellationtoreceiveantennaplacement,changeoftransmitfrequency,andamplitudematchingatthereceiveantenna.dantalsocapturestheeffectofbandwidthonantennacancella-tion.Considera5MHzsignalcenteredat2.48GHz.Thus,thesig-nalhasfrequencycomponentsbetween2.4775GHzand2.4825GHz.Ifthereceiveantennaisplacedperfectlyforthecenterfrequency,thereisasmallerrorinplacementfortheotherfrequencieswithinthatbandwidth.Wecanmapthedifferenceinwavelengthtotheerrorinreceiverplacement.Forexample,adifferenceinwavelengthissimilartoa=4errorinreceiverplacement.Thus,dantfor2.4775GHzinthiscasewouldbe1 4�c 2:4775106�c 2:48106,wherecisthespeedoflight.Thisgivesdant0:025mm,correspondingto60.7dBan-tennacancellationforthe2.48GHzcenterfrequency.Thus,60.7dBisthebestantennacancellationpossiblefora5MHzsignalinthe2.4GHzbandusingthe3antennaschemedescribedinthispaper.Similarly,using20MHzand85MHzbandwidthsgivebestcasere-ductionof46.9dBand34.3dBrespectively.Ascanbeseenfromtheeffectofbandwidth,antennacancel-lationdoesnotprovideafrequencyatchannelatthereceiverifthereisperfectamplitudematching.ThisdistortioninthereceivedsignalcanbeaproblemfortheRFanddigitalinterferencecancel- (a)Receivedpowerwithdistancemismatch (b)ReceivedpowerwithamplitudemismatchFigure3:Performanceofantennacancellationwithdistanceandamplitudemismatchforsignalswithdifferentbandwidth.A1mmmismatchcanrestrictthereceivepowerreductionto29dB.Anamplitudemismatchof10%,correspondingto1dBvariation,canrestrictthereceivepowerreductionto20dB.lationstages,sincetheyusetheundistortedtransmissionsignalasreferenceforcancellation.Anyerrorinreceiveantennaplacementaddstodant.Toseetheeffectofreceiveantennaplacementerror,supposethereceivean-tennais1mmofffromtheoptimalposition,i.e.dant=1mm.Withperfectamplitudematchingandwithaof12.1cm(foracenterfrequencyof2.48GHz),weseea28.7dBreductioninpowercomparedtonoantennacancellation.Figure3(a)showsthetheo-reticalperformanceofantennacancellationwitherrorinreceiverplacement,fordifferentbandwidths.Figure3(b)showsthetheoreticalperformanceofantennacan-cellationwitherrorinamplitudematching,assumingperfectcenterfrequencyreceiverplacement,fordifferentbandwidths.Forexam-ple,saytheamplitudeofonesignalis10%higherthantheother,i.e.Aant=0:1Aant.Inthiscase,thepowersofthetwosignalsdifferby1dB.WiththisAant,thereductioninreceivedpowerduetoantennacancellationis23dB,ifweignoretheeffectofband-width.Fora5MHzbandwidth,thesameAantgivesa22.994dBreduction.Thus,asmallamplitudemismatchtendstodominatetheperformancerestrictionsonantennacancellation.Sinceampli-tudemismatchaffectsdifferentfrequenciesequally,theresultingfrequencyresponseisfairlyat,thusgivingalessdistortedinputtothelatercancellationstages.Thus,amplitudemismatchmayenduphelpingthelaterstagesofinterferencecancellation. Figure4:ReceivedSNRfordifferentreceiveantennaplace-ments.ThereceivedSNRisfairlymonotonicwithdistancewhenanyonetransmitantennaisactive.Withbothtransmitantennasactive,thereisasharpreductioninreceivepoweratthenullpoint.3.2AntennaCancellationinPracticeFigure4showstheeffectofantennacancellationwithtransmit-terTX1attenuatedby6dBcomparedtoTX2.ExperimentsshowthatthereceivedpowerfromthetwoTXantennasdiffersbyabout5.1dBwhenthereceiverisplacedatthenullpoint.Thus,thissetuphasanamplitudemismatchof1dBcausingthecancellationtoberestrictedto20dBasshowninthepreviousanalysis.Theaboveanalysisdidnotconsiderthemultipatheffect.However,resultsfromthemeasurementsshowthatthemultipatheffectisnotadom-inantcomponentinourexperimentalsetup.3.3EffectofAntennaCancellationonIntendedReceiversWhileantennacancellationcanreduceself-interferencefromanode'sowntransmitter,animportantquestionishowthisaffectsthereceivedsignalatnodesotherthanthetransmitter.Anotherquestionishowdoesourcancellationtechniquecomparetoasim-pletechniquesuchashavingthesignalsbetweenthetwotransmitantennasphaseshiftedby.Unlikeourtechnique,thephaseshiftapproachdoesnotrequireanattenuatorandgivesanullpointex-actlyatthecenter.ThecontourmapinFigure5(a)showsreceivedpowerwithbothtransmitantennastransmittingasinglefrequencytoneatthesamepowerwithaphasedifferenceofusingasimplesimulationwithafreespacepropagationmodel.Eachcontourlinecorrespondstoaspecicreceivedpower.Figure5(b)showsthereceivedsignalstrengthwithdifferenttransmitpowersfromthetransmitantennassuchthatamplitudesmatchatthenullpointwithoutanyphaseshiftinantennasignals.Thenullpointsachievedinthetwocasesareatdifferentlocations,butbothschemesareequallygoodintermsofsignalreductionatthenullpoint.Thedifferencebetweenthesetwocasesbecomesclearerbylook-ingatthereceivedsignalatlargerdistances.Figure5(c)showsthereceivedsignalstrengthprole,overspace,forasingletransmitantennaoveradistanceof30mfromthetransmitter.Thisisthebaselineforcomparisonofthetwoschemeswithantennacancella- (a)Equalpowers (b)Differentpowers (c)Singletransmitter (d)Equalpowers (e)DifferentpowersFigure5:Freespacesignalstrengthprolesforequaltransmitpowersanddifferenttransmitpowersontwotransmitantennas.Thissimulationusesapathlossexponentof2.Figures(a)and(b)correspondtoashort-rangestudy.Whentransmitpowersareequal,theminimumreceivedsignalisinthemiddleandwhenthetransmitpowersaredifferent,theminimumisclosertothelowertransmitpowerantenna.Figures(c),(d)and(e)correspondtoalong-rangestudy.Whentransmitpowersareequal,receiversequidistantfromthetransmitantennapaircanseehugedifferencesinthereceivedsignalstrength.Whentransmitpowersaredifferent,however,suchdifferencesaremuchsmaller.tion.Figure5(d)showsthecontoursoverlargerdistancesforthesamesetupasFigure5(a).Itisapparentthateveninnormalcom-municationrange,therearelocationswithverylowreceivedpowerduetothedestructiveinterference.Figure5(e)showsthecontoursofreceivedpowerwhenonetrans-mitsignalisattenuatedby6dBcomparedtotheotherandthereisnophaseshiftbetweenthetwotransmittedsignals.Theeffectofdestructiveinterferenceismuchlowerinthiscase.Incaseoftwotransmitantennas,thesignalsfromthetwoanten-nasgetaddedconstructivelyordestructivelyatthereceiver.Atdis-tancesmuchlargerthanthespacingbetweenthetransmitantennas,thesignalsfrombothantennasundergoalmostequalattenuation.Withequalreceivepowerfrombothantennas,aperfectlydestruc-tivecombiningofthetwosignalscausesthereceivedsignaltobezeropower.Incaseofunequaltransmitpowers,thereceivedpoweratthesedistancesisdifferentfromthetwotransmitantennas.Evenwhenthesignalscombineperfectlyoutofphase,theresultingsig-nalisnotzeropower.Comparingwiththesingleantennacase,usingourantennacan-cellationschemeleadstoamaximumdegradationof6dBatanyreceiverlocation.Inarealnetworksetting,diversitygainsduetotwotransmitantennaswouldoffsetthisdegradation.Thus,antennacancellationcangivesignicantreductionatthenullpositionwith-outhavingalargeeffectonreceptionatothernodes.Followingantennacancellation,furtherreductionisobtainedbyRFanddigi-talinterferencecancellationtechniques.4.INTERFERENCECANCELLATIONThissectionexplainstwointerferencecancellationmechanismsusedinfull-duplexingnodesaftertheantennacancellationstage.TherstisRFinterferencecancellationusinganoisecanceler.Thesecondisdigitalcancellationthattakesplace,insoftware,afterthereceivedsignalisdiscretized.4.1RFInterferenceCancellationAsRadunovicetal.[18]exploredfor900MHzbandnetworks,theinterferencecancellationcircuitbasedonQHx220,anoisecan-celerchip,allowsremovingaknownanaloginterferencesignalfromareceivedsignal.TheQHx220chiptakestheknownself-interferenceandreceivedsignalsasinputsandoutputsthereceivedsignalwiththeself-interferencesubtractedout.Thechipallowschangingtheamplitudeandphaseoftheinterferencereferencesig-naltomatchtheinterferenceinthereceivedsignal.AnRFsplitterisusedtogivethetransmitsignaltothecancellationcircuitastheinterferencereference.Figure6showstheeffectofusingtheRFcancellationcircuit.Itshowsspectrumpowersnapshotsatthereceiveantennaforthree Figure6:SpectrumsnapshotsshowingtheeffectofantennacancellationandacombinationofantennaandRFinterferencecancellation.Acombinationofthetwotechniquescangivea50dBreductioninself-interference.cases–themaximumreceiveantennapowerwithonlyonetrans-mittingantenna,thereceivepowerwithantennacancellationandthereceivepowerwithacombinationofantennaandRFinterfer-encecancellation.RFinterferencecancellationachieves20dBreductioninthereceivedself-interferenceontopofthereductionachievedbyantennacancellation.4.2DigitalInterferenceCancellationThereisextensiveexistingworkthatdescribesdigitalcancella-tiontechniques[8,9,10].Traditionally,digitalcancellationisusedbyareceivertoextractapacketfromadesiredtransmitterafterthepackethascollidedwithapacketfromanunwantedtransmitter.Todothis,thereceiverrstdecodestheunwantedpacket,remodulatesitandthensubtractsitfromtheoriginallyreceivedcollidedsignal.Incaseofcancelingself-interferenceforfull-duplex,thetransmit-tedsymbolsarealreadyknown,andthusdecodingisnotnecessaryinordertoreconstructacleansignal.Insteadofdecoding,coherentdetectionisusedtodetecttheself-interferingsignal.Thedetectorcorrelatestheincomingsignalwiththecleantransmittedsignal,whichisavailableattheoutputofthetransmitter.Themainchallengeinsubtractingtheknownsignalisinestimatingthedelayandphaseshiftbetweenthetransmittedandthereceivedsignals.Asthedetectorhasthecompleteknowledgeoforiginallytransmittedsignal,itusesthissignaltocorrelatewiththeincomingsignaltodetectwherethecorrelationpeaks.Thecorrela-tionpeaktechniquegivesboththedelayandthephaseshiftneededtosubtracttheknownsignal.Thus,thistechnique,unlikesomeofthedigitalinterferencetechniques,doesnotrequireanyspecialpreambleorpostambleandisbackwardscompatible.Moreover,thistechniqueismodulation-independentaslongasthecleansig-nalcanbeconstructed.Coherentdetectioncandetecttheself-interferencesignalevenwhenitisweakerthanthereceivedsignal.Therefore,digitalin-terferencecancellationcanimprovetheSINRlevelevenwhenthereceivedsignalisstrongerthanself-interference.ThispropertyisusefulwhenoperatingwithvariabledataratestoallowusinghigherdataratesforhighSNRlinks.Typicalinterferencecancellationalsorequirescompensatingforclockdriftbetweenthetransmitterandreceiver.Sincethetrans-mitterandreceiverdaughterboardsinafull-duplexnodesharethe SectionSummary Section5.2AggregateThroughputforfull-duplexlinksshows1.84xmediangain.Section5.3Full-duplexlinksmaintain88%ofthehalf-duplexlinkreliability.Section5.4Withoutdigitalinterferencecancellation,full-duplexmaintainsonly67%ofthehalf-duplexlinkreliability. Table1:Summaryofevaluationresults Figure7:Mapofnodelocationsfortheexperimentalsetup.Node1isalwayskeptataxedlocationinsideanofceroomandNode2'slocationischangedforeachiterationtodifferentlocationswithinabuildingwing.sameclock,thereisnoclockdrift.However,sincethedaughter-boardsuseseparatePLLlogic,therecanbeajitterintroduced.Webelievethisjitteriswhatlimitstheperformanceofthecurrentim-plementationofdigitalinterferencecancellation.Currently,ourdigitalinterferencecancellationachieves10dBreduction,whichismuchsmallerthanreportedbySIC[9],20dB.Webelieveitcanbeimprovedbyincorporatingachannelestima-tor.Sincetheactualself-interferencesignalsaredifferentfromthegeneratedtransmittedsignalsduetohardwarelimitationsandmul-tipath,correlatingandsubtractingtheestimatedsignalratherthanthecleansignalcanimproveperformance.5.EVALUATIONDoingfull-duplextransmissionshasimplicationstothroughputandpacketdeliveryreliability.Astransmissionandreceptioncangosimultaneously,theaggregatethroughputforanodepaircanbemorethanahalf-duplexsystem.Ontheotherhand,impropercan-cellationcanleadtoastrongself-interferenceandhurtpacketre-ceptionwhiletransmissionisinprogress.Anidealfull-duplexsys-temshouldhaveperfectself-interferencecancellationandachievedoublethethroughputofahalf-duplexsystem.Ourevaluationsusingapreliminaryprototypeshowthatfull-duplexinggivesame-dianaggregatethroughputgainof84%withoutsignicantlossinpacketreceptionreliability.Inotherwords,thefull-duplexpro-totype'sperformanceiswithin8%ofanidealfull-duplexsystem.Morepreciseengineeringcanallowforevenbetterimplementa-tions.Thisevaluationshowsthefeasibilityofmakingafull-duplexwirelesssystem.Table1summarizestheseevaluationresults. Figure10:AninfrastructureWi-Fisetup.AhiddenterminaloccursattheAPwhennodeN1andN2cannotheareachother'stransmissionstherightby6-7dBcomparedtothetypical802.15.4system,forwhichthetransitionregionoccursaround0dB.Besidestheeffectoflongerpackets,webelievethatthisdifferenceisalsoduetothelimitationsoftheimplementationofthe802.15.4receiverinUS-RPsasreportedin[20].5.4DigitalInterferenceCancellationSincedigitalinterferencecancellationisnotpossiblewithanoff-the-shelftransceiver,westudyfull-duplexperformancewithoutus-ingdigitalcancellation.Figures8and9showtheresultsoffull-duplexwithoutperform-ingdigitalinterferencecancellationtounderstandthegainsintheabsenceofdigitalinterferencecancellation.Figure9showsthatfull-duplexwithoutdigitalcancellationhas5dBhigherPRRtran-sitionrange.Thegainofdigitalcancellationisonly5dBsincetheself-interferencesignalisonlyabout5dBabovethenoiseoorafterantennaandRFinterferencecancellationforthissystem.Overall,full-duplexwithoutdigitalinterferencecancellationmain-tainsonly67%ofthelinkreliabilityofthehalf-duplexlinks.There-fore,morelinkswithlowSNRdonotsufcientlycancelouttheself-interference,causing40%ofthelinkstohavelowerthrough-putthanhalf-duplex.Theseresultsrevealthatareasonablefull-duplexoperationwithoff-the-shelfradiosispossibleonlyforhighSNRlinkpairs.AmorecarefullytunedRFcancellationsetupcouldallowfull-duplexoperationwithoff-the-shelfradiosacrossawiderrangeofSNRs.6.APPLICATIONSEarliersectionsshowedthatawirelessfull-duplexsystemthatcannearlydoublethethroughputofasinglehoplinkispracticallyimplementable.Ontheotherhand,theimplementationusesaddi-tionalresourcesthatcouldotherwisebeusedtoimplementa2x2MIMOsystem,thatmayprovidesimilarphysicallayergains.Itisunclearifonlythephysicallayergainsoffull-duplexwouldjustifytheengineeringandcostneededtoimplementthesesystems.However,webelievethatthetruebenetofthefull-duplexsys-temliesbeyondthisgaininthephysicallayer.Practicalfull-duplexingcanmitigatemanyoftheproblemswithwirelessnetworkstoday.Full-duplexinghelpsaddressthreedistinctchallengesincurrentwirelesssystems:hiddenterminals,congestionduetoMACschedul-ing,andhighend-to-enddelaysinmultihopwirelessnetworks.Further,fullduplexcanhaveapplicationstofuturewirelessnet-worksthatusecognitiveradios.6.1ReducingHiddenTerminalsFigure10showsatypicalhomeorofceWi-Fisetup.Endnodesconnecttothebackbonenetworkthroughanaccesspoint.Theclas-sichiddenterminalproblemoccurswhenNodeN2isunabletohearN1'stransmissionstotheaccesspointandstartssendingdatatotheaccesspointatthesametime,thuscausingacollisionattheaccesspoint.Thisproblemcanbesolvedusingfull-duplexnodes.Supposeallnodesalwayshavedatatosendtoandreceivefromtheaccess Figure11:Astartopologymultihopnetwork.NodeN0be-comesacongestednode.ThenetworkthroughputinregularMACoperationis1/nfor2n+1nodes.point.Then,assoonasN1startstransmittingdatatotheaccesspoint,theaccesspointstartstransmittingdatabacktoN1simul-taneously.N2hearsthetransmissionfromtheaccesspointanddelaysitstransmission,therebyavoidingacollision.IftheaccesspointdoesnothaveanypacketstosendbacktoN1,itcanrepeatwhateverithears.ThisrepetitionservesasanimplicitACKforN1andpreventsN2fromtransmitting.Thisschemeformitigatinghiddenterminalsalsoappliestomultihopwirelessnetworks.Full-duplexingdoesnotcompletelypreventthehiddenterminalproblem.Inorderforthereceivertorespond,itneedstoreceivethedestinationaddressofthelinklayerheader.However,typicallythedestinationaddressisprecededbythepreamble,PHYheader,andpartoftheMAClayerheader,wherecollisionscanstilloccur.Forexample,for802.11g,thereceiverneedstoreceive15bytesbeforeitcandecodethereceiveraddress,whichleavestheinitial2.5%ofthepackettimetobevulnerablefor6Mbpsand10%for54Mbps,fora1500bytepacket.Thisvulnerabilityisinevitable,butcanbereducedbychangingthepacketformatsuchthatthedestinationaddressisplacedearlierinthepacket.6.2ReducingCongestionduetoMACSchedulingFigure11showsanetworkinstartopology.NodesN1,N2,andN3havedatatosendtonodesN4,N5,andN6respectively.AlldatahastoberoutedthroughnodeN0,andN0-N3areintheinterfer-encerangeofeachother.Ifallthreesourcenodeshavesaturatedowstobesenttotheirrespectivedestinations,nodesN0-N3con-stantlycontendwitheachotherforchannelaccess.Assumingtypi-calMACscheduling,N0gets1/4ththetotaltransmissionopportu-nities.Thisrestrictstheaggregatenetworkthroughputto1/4ththecapacityofonelink.Inageneralstartopologywith2n+1nodesandnodesN1toNntryingtoroutedatatonodesNn+1toN2nrespectivelyvianodeN0,theaggregatenetworkthroughputis1/n.Withfull-duplexing,N0cantransmitandreceiveatthesametime.Thus,foreachtransmissionfromeithernodeN1,N2,orN3,N0canforwardapackettoadestination.Thus,theaggregatenet-workthroughputisequaltothesinglelinkcapacity.Full-duplexhelpssolvethelossofnetworkthroughputduetocongestionandMACschedulingbyallowingcongestednodestoforwardoutpack-etsandreceivepacketsatthesametime.Existingwork[5,16]hasalsostudiedtheproblemoffairnessbe-tweenupstreamanddownstreamowsinaccesspoint(AP)basednetworks.Since802.11CSMAprovidesthesametransmitoppor-tunitiestoallclientsandtheAP,theAPonlygets1=Nofthechan-nelwhenthereexistNclients.Ifthedownstreamowisequally Figure12:Wormholeswitchinginamultihopnetwork.In-terferencefromforwardinghopscanbecanceledusingdigitalcancellationandcanalsoserveasimplicitACKs.dividedintoNclients,eachowgetsonly1=N2ofthechannelcapacity,whileeachupstreamowgets1=N.SomesuggestedsolutionstothisproblemincludecontrollingthechannelaccesspriorityorincorporatingratecontrolmechanismsabovetheMAClayer[5,16].However,thecongestionreductionofferedbyfull-duplexingextendsseamlesslytoprovideinherentfairnessinAPbasedwirelessLANs.SinceAPscantransmitwhilereceiving,alldownstreamandupstreamowscanget1=Nofthechannelcapacityeach.6.3WormholeRoutinginMultihopNetworksMultihopnetworkssufferfromlongend-to-enddelayscausinglossinperformancefordelaysensitiveprotocolslikeTCP.Further,multihopnetworkshavea1/3rdthroughputscalingcomparedtosinglehopnetworksduetointerferencebetweenforwardinghops.Theideaofreceivingandforwardingatthesametimecanbeex-tendedtosolvetheseproblems.Theinsightisthatasafull-duplexnodeisstartingtoreceiveapacketitcansimultaneouslystarttoforwardit.Thus,insteadofthedefaultstore-and-forwardarchitec-ture,full-duplexnodescouldforwardapacketwhilereceivingit.Thisideaissimilartowormholeswitching[7]usedformultihopwiredcommunicationnetworks.Thistechniquecantheoreticallyreducetheend-to-enddelayforpacketdeliverythroughamultihopnetworkfromapackettimemultipliedbynumberofhopstoalittlemorethanapackettime.Figure12showsthewaywormholeswitchingcanworkforfull-duplexwirelesslinks.N2startsreceivingapacketfromN1.AssoonasN2hasprocessedthepacketheader,itknowswheretofor-wardthepacketandstartstransmittingthepackettoN3.Similarly,N3startsforwardingthepackettoN4.Atthistime,N3'stransmis-sionalsointerfereswiththereceptionatN2.SinceN2knowsthepartofthepacketN3wouldbetransmittingatthistime,itcanusedigitalcancellationtechniquestocancelN3'stransmission.Further,onceN2hasnishedreceivingthepacketfromN1,itcanagainap-plydigitalcancellationtopreviouslyreceivedsamplesfromN1andN3tocancelthesamplesreceivedfromN1.ThisallowsN2tocheckthepackettransmissionfromN3.ThiscanactasanimplicitACKmechanism,thusremovingtheneedofanexplicitARQscheme.ThelastnodeintheroutesendsanexplicitACKtothelastbutonenodeintheroute.Existingworkhassuggestedasimilarim- Figure13:SpectrumatreceivedantennaforaWi-Finodetransmittingatfullpower(18dBm).Antennacancellationgainsareasexpected.RFinterferencecancellationresultsinhighsidelobes.plicitARQschemeforamulti-channelwirelessnetworkusedasaninterconnectbackboneforchipmulti-processors[15].6.4CognitiveRadiosIncognitiveradiotechnologiessuchasWhiteFi[2],theunli-censed(secondary)usersareallowedtouseaspectrumonlyifthelicensed(primary)usersarenotusingit.Oneoftheprimarychal-lengesinsuchsystemsistoidentifywhenitisokayforsecondaryuserstousethespectrum.Specically,whilethesecondaryuserisusingthespectrum,iftheprimaryuserdecidestousethespectrumthenitisusuallyhardforthesecondaryuserstodetectandstopimmediately.Thefull-duplexsystemproposedinthispaperwillenablethesecondaryusertoscanforanyprimaryuserswhileitisusingthespectrum.7.CHALLENGESPrevioussectionshaveshownthefeasibilityoffullduplexfor802.15.4systems.Aswirelesssystemslike802.11have(100x)highertransmitpowerand(4x)widerbandwidththan802.15.4,itisnotcleariffullduplexispossibleinsuchsystems.Preliminaryexplorationshowsthathighertransmissionpowercallsforbetterantennacancellationanddigitalcancellationtechniques,andwiderbandwidthcallsforbetternoisecancellationcircuitry.7.1FullDuplexin802.11Figure13showsthespectrumanalyzeroutputswithandwithoutantennaandnoisecancellationtechniques.Itshowsthatthereduc-tionis48dBwhenthetwoRFcancellationtechniquesareused.TheRFinterferencecancellationstep(usinganoisecancellationcircuit)resultsinseveralhighpowersidelobes,althoughitgivesa15dBreductioninsignalatthecenterfrequency.Thisresultisdif-ferentfromthespectrumobservedfor802.15.4inSection3.Therearetwodifferencesbetween802.11and802.15.4systems;higherpowerandwiderbandwidth.Below,weexplorehowthesetwopropertiesaffectfullduplexin802.11systems.7.2HighTransmitPowerThethreecancellationtechniquespresentedinthispaper,to-gether,give60dBreductionofselfinterferenceforthecurrentimplementation.Fora802.15.4system,atthereceiverlocation, Thenoisecancellationcircuitcurrentlyrequiresmanuallysettingtheamplitudeandphaseforinterferencecancellation.Designingadaptivealgorithmstotrackchannelvariationsandsettingtheam-plitudeandphaselevelforthenoisecancellingcircuitispartoffuturework.Aninterestingfutureresearchdirectionisthedesignofamediaaccesscontrol(MAC)layerthatcantakeadvantageoffull-duplexwireless.Suchprotocolscanaddresssomeoftheperennialprob-lemsinwirelessnetworkssuchasend-to-enddelayandnetworkcongestion.Webelievethisworkprovidesanewresearchdirec-tionforthedesignandanalysisofhigherlayerprotocolsforwire-lessnetworks.10.ACKNOWLEDGMENTSThisworkwassupportedbygenerousgiftsfromDoCoMoCap-ital,theNationalScienceFoundationundergrants#0831163and#0846014,theKingAbdullahUniversityofScienceandTechnol-ogy(KAUST),MicrosoftResearch,ascholarshipfromtheSam-sungScholarshipFoundationandaStanfordTermanFellowship.Finally,wewouldliketothankourshepherd,AshutoshSabharwal,andtheanonymousreviewersfortheircomments.11.REFERENCES[1]P.Bahl,A.Adya,J.Padhye,andA.Walman.Reconsideringwirelesssystemswithmultipleradios.SIGCOMMComput.Commun.Rev.,34(5):39–46,2004.[2]P.Bahl,R.Chandra,T.Moscibroda,R.Murty,andM.Welsh.Whitespacenetworkingwithwi-likeconnectivity.SIGCOMMComput.Commun.Rev.,2009.[3]J.Bicket.Bit-rateselectioninwirelessnetworks.Master'sthesis,MIT,2005.[4]S.BiswasandR.Morris.ExOR:opportunisticmulti-hoproutingforwirelessnetworks.InSIGCOMM'05:Proceedingsofthe2005conferenceonApplications,technologies,architectures,andprotocolsforcomputercommunications,2005.[5]N.Blefari-Melazzi,A.Detti,I.Habib,A.Ordine,andS.Salsano.TCPFairnessIssuesinIEEE802.11Networks:ProblemAnalysisandSolutionsBasedonRateControl.IEEETransactionsonWirelessCommunications,6(4):1346–1355,2007.[6]S.Chachulski,M.Jennings,S.Katti,andD.Katabi.Tradingstructureforrandomnessinwirelessopportunisticrouting.InSIGCOMM'07:Proceedingsofthe2007conferenceonApplications,technologies,architectures,andprotocolsforcomputercommunications,2007.[7]W.J.DallyandC.L.Seitz.Thetorusroutingchip.DistributedComputing,1(4):187–196,1986.[8]S.GollakotaandD.Katabi.ZigZagdecoding:combatinghiddenterminalsinwirelessnetworks.InSIGCOMM'08:ProceedingsoftheACMSIGCOMM2008conferenceonDatacommunication,pages159–170,NewYork,NY,USA,2008.ACM.[9]D.Halperin,T.Anderson,andD.Wetherall.Takingthestingoutofcarriersense:interferencecancellationforwirelesslans.InMobiCom'08:Proceedingsofthe14thACMinternationalconferenceonMobilecomputingandnetworking,pages339–350,NewYork,NY,USA,2008.ACM.[10]S.Katti,S.Gollakota,andD.Katabi.Embracingwirelessinterference:analognetworkcoding.InSIGCOMM'07:Proceedingsofthe2007conferenceonApplications,technologies,architectures,andprotocolsforcomputercommunications,pages397–408,NewYork,NY,USA,2007.ACM.[11]S.KattiandD.Katabi.Mixit:Thenetworkmeetsthewirelesschannel.InHotnets-VI:ProceedingsofACMHotTopicsinNetworksWorkshop,2007.[12]S.Katti,H.Rahul,W.Hu,D.Katabi,M.Medard,andJ.Crowcroft.Xorsintheair:practicalwirelessnetworkcoding.InSIGCOMM'06:Proceedingsofthe2006conferenceonApplications,technologies,architectures,andprotocolsforcomputercommunications,pages243–254,NewYork,NY,USA,2006.ACM.[13]P.KyasanurandN.H.Vaidya.Routingandlink-layerprotocolsformulti-channelmulti-interfaceadhocwirelessnetworks.SIGMOBILEMob.Comput.Commun.Rev.,10(1):31–43,2006.[14]M.Lacage,H.Manshaei,andT.Turletti.IEEE802.11rateadaptation:Apracticalapproach.InstitutNationalDeRechercheenInformatiqueetenAuomatique,2004.[15]S.-B.Lee,S.-W.Tam,I.Pefkianakis,S.Lu,M.F.Chang,C.Guo,G.Reinman,C.Peng,M.Naik,L.Zhang,andJ.Cong.Ascalablemicrowirelessinterconnectstructureforcmps.InMobiCom'09:Proceedingsofthe15thannualinternationalconferenceonMobilecomputingandnetworking,pages217–228,NewYork,NY,USA,2009.ACM.[16]D.J.Leith,P.Clifford,D.Malone,andA.Ng.TCPFairnessin802.11eWLANs.IEEECommunicationsLetters,9(12),2005.[17]QuellanInc.Qhx220narrowbandnoisecancelleric.http://www.quellan.com/products/qhx220_ic.php.[18]B.Radunovic,D.Gunawardena,P.Key,A.Proutiere,N.Singh,H.V.Balan,andG.Dejean.Rethinkingindoorwireless:Lowpower,lowfrequency,full-duplex.TechnicalReportMSR-TR-2009-148,MicrosoftResearch,2009.[19]H.Rahul,F.Edalat,D.Katabi,andC.G.Sodini.Frequency-awarerateadaptationandmacprotocols.InMobiCom'09:Proceedingsofthe15thannualinternationalconferenceonMobilecomputingandnetworking,pages193–204,NewYork,NY,USA,2009.ACM.[20]T.Schmid.Gnuradio802.15.4en-anddecoding.http://nesl.ee.ucla.edu/fw/thomas/thomas_project_report.pdf.[21]M.Vutukuru,H.Balakrishnan,andK.Jamieson.Cross-layerwirelessbitrateadaptation.SIGCOMMComput.Commun.Rev.,39(4):3–14,2009.[22]L.WengandR.Murch.Multi-userMIMOrelaysystemwithself-interferencecancellation.pages958–962,March2007.APPENDIXA.ANALYSISONTHERECEIVEDPOWERAFTERANTENNACANCELLATIONLettheunitpowerbasebandsignalbex[t].ThesignalisscaledbydifferenttransmissionamplitudesA1andA2atthetwotransmitantennas.ThetransmittedsignalsundergoattenuationsAtt1andAtt2andphaseshifts1and2inthewirelesschannelbeforereachingthereceiveantenna.Thereceivedsignalisthengivenby:A1 Att1x[t]ej(2fct+1)+A2 Att2x[t]ej(2fct+2)