distinctpiggycase

PassiveWi-Fi:BringingLowPowertoWi-FiTransmissionsBryceKellogg,VamsiTalla,ShyamnathGollakotaandJoshuaR.
SmithUniversityofWashingtonCo-primaryStudentAuthorsAbstract–Wi-Fihastraditionallybeenconsideredapower-consumingcommunicationsystemandhasnotbeenwidelyadoptinginthesensornetworkandIoTspace.
WeintroducePassiveWi-Fithatdemonstratesforthersttimethatonecangenerate802.
11btransmissionsusingbackscattercommunication,whileconsuming3–4ordersofmagnitudelowerpowerthanexistingWi-Fichipsets.
PassiveWi-FitransmissionscanbedecodedonanyWi-Fideviceincludingrouters,mobilephonesandtablets.
Buildingonthis,wealsopresentanet-workstackdesignthatenablespassiveWi-Fitransmit-terstocoexistwithotherdevicesintheISMband,with-outincurringthepowerconsumptionofcarriersenseandmediumaccesscontroloperations.
Webuildprototypehardwareandimplementallfour802.
11bbitratesonanFPGAplatform.
OurexperimentalevaluationshowsthatpassiveWi-Fitransmissionscanbedecodedonoff-the-shelfsmartphonesandWi-Fichipsetsoverdistancesof30–100feetinvariousline-of-sightandthrough-the-wallscenarios.
Finally,wedesignapassiveWi-FiICthatshowsthat1and11Mbpstransmissionsconsume14.
5and59.
2Wrespectively.
Thistranslatesto10000xlowerpowerthanexistingWi-Fichipsetsand1000xlowerpowerthanBluetoothLEandZigBee.
1IntroductionOverthepastfewyears,researchershaveexploredtheconceptofWi-Fibackscatter[25,38]thatcreatesanad-ditionalnarrowbanddatastreamtorideontopofexistingWi-Fisignals.
Whilepromising,existingdesignseitherachieveverylowdatarates(100sofbps)atclosebydis-tances(2-4feet)[25]orusecustomfull-duplexhardwarethatcannotbeusedwithanyexistingWi-Fidevices[38].
Inthispaper,wetakeadifferentapproach—insteadofbackscatteringexistingWi-Fisignalstosendanad-ditionaldatastream,weusebackscattercommunicationtodirectlygenerateWi-Fitransmissionsthatcanbede-codedonanyofthebillionsofexistingdeviceswithaFigure1:PassiveWi-Fiarchitecture.
ThepassiveWi-Fidevicesperformdigitalbasebandoperationslikecod-ing,whilethepower-consumingRFfunctionsaredele-gatedtoaplugged-indeviceinthenetwork.
Wi-Fichipset.
Tothisend,weintroducePassiveWi-Fithatdemonstratesforthersttimethatonecangener-ate802.
11btransmissionsusingbackscattercommuni-cation,whileconsuming4–5ordersofmagnitudelowerpowerthanexistingWi-Fichipsets.
WeobservethatwhileCMOStechnologyscalinghasconventionallyprovidedexponentialbenetsforthesizeandpowerconsumptionofdigitallogicsystems,analogRFcomponents,thatarenecessaryforWi-Ficommuni-cation,havenotseenasimilarpowerscaling.
Asaresult,Wi-Fitransmissionsonsensorsandmobiledevicesstillconsumehundredsofmilliwattsofpower[31–33].
Togetaroundthisproblem,passiveWi-FiusesbackscattertodecouplethebasebandWi-Fidigitallogicfromthepower-consumingRFcomponents,asshowninFig.
1.
Inourarchitecture,thepassiveWi-Fidevicesperformdigitalbasebandoperationslikecodingandmodulation,whilethepower-consumingRFcomponentssuchasfre-quencysynthesizersandpowerampliersaredelegatedtoasingleplugged-indeviceinthenetwork.
Thisde-viceprovidestheRFfunctionsforallthepassiveWi-Fidevicesinthevicinitybytransmittingasingle-frequencytone.
ThepassiveWi-Fidevicescreate802.
11btrans-missionsbyreectingorabsorbingthistoneusingadig-italswitchrunningatbaseband.
SincethepassiveWi-Fi1deviceshavenoanalogcomponents,theyconsumeslesssiliconareaandwouldbesmallerandcheaperthanexist-ingWi-Fichipsets.
Moreimportantly,theirpowercon-sumptionwouldbeordersofmagnitudelowersincetheyonlyperformdigitalbasebandoperations.
Torealizethis,however,weneedtoaddressthreemainchallenges.
(a)HowcanWi-Fireceiversdecodeinthepresenceofinterferencefromtheplugged-indeviceTheWi-Fireceiverreceivesthebackscatteredsignalinthepres-enceofastronginterferencefromthetonetransmittedbytheplugged-indevice.
Traditionalbackscattersys-tems[34,38]useafull-duplexradiotocancelthisstronginterferingsignal,whichisnotpossibleonexistingWi-Fidevices.
OurkeyobservationisthatWi-Fireceiversarerequiredtoworkeveninthepresenceofinterferenceintheadjacentbandthatis35dBstronger[12].
Further,asWi-FiandBluetoothradiosarebeingintegratedontoasinglechipset[6],Wi-Fihardwareisbeingdesignedtoworkinthepresenceofout-of-bandBluetoothinter-ference.
Thus,wesettheplugged-indevicetotransmititstoneatafrequencythatliesoutsidethedesiredWi-Fichannel;thisensuresthatexistingWi-Fichipsetscansuppresstheresultingout-of-bandinterference.
(b)Howcanwecreate802.
11btransmissionsusingbackscatterAtahighlevel,werstshifttheout-of-bandtonefromtheplugged-indevicetolieatthecenterofthedesiredWi-Fichannel.
Wethenusethisshiftedtonetocreate802.
11btransmissions.
Intuitively,mul-tiplyingtwosinusoidalsignalscancreateafrequencyshift.
1Thus,bybackscatteringatafrequencyf,wecanshiftthetone.
TosynthesizeWi-Fitransmissions,weleveragethat802.
11busesDSSSandCCKencodingontopofDBPSKandDQPSKmodulation.
Theencodingoperationisdigitalinnatureandhenceisachievedusingdigitallogic.
TocreatethephasechangesrequiredforDBPSKandDQPSK,weapproximateadigitalsquarewaveasasinusoidandmodulateitsphasebychangingthetimingofthesquarewave(see§2.
3).
Thus,passiveWi-Fidevicescanfullyoperateinthedigitaldomainatbasebandandyetsynthesize802.
11btransmissions.
(c)HowdopassiveWi-FidevicessharetheWi-Finet-workTraditionalWi-Fisharesthenetworkusingcarriersense.
However,thisrequiresaWi-FireceiverthatisONbeforeeverytransmission.
SinceWi-Fireceiversrequirepower-consumingRFcomponentssuchasADCsandfrequencysynthesizers,thiswouldeliminatethepowersavingsfromourdesign.
Instead,wedelegatethepower-consumingtaskofcarriersensetotheplugged-indevice.
Atahighlevel,theplugged-indeviceperformscarriersenseandsignalsthepassiveWi-Fidevicetotransmit.
§3describeshowsuchasignalingmechanismcanalsobeusedtoarbitratethechannelbetweenmultiplepassive12sinftsinft=cos(ff)tcos(f+f)t.
Wi-Fidevicesandaddressotherlink-layerissuesinclud-ingACKsandretransmissions.
Toshowthefeasibilityofourdesign,webuildpro-totypebackscatterhardwareandimplementallfour802.
11bbitratesonanFPGAplatform.
Ourexperimen-talevaluationshowsthatpassiveWi-Fitransmissionscanbedecodedonoff-the-shelfsmartphonesandWi-Fichipsetsoverdistancesof30–100feetinvariousline-of-sightandthrough-the-wallscenarios.
WealsodesignapassiveWi-FiICthatperforms1Mbpsand11Mbps802.
11btransmissionsandestimatethepowerconsump-tionusingCadenceandSynopsistoolkits[5,19].
Ourresultsshowthe1and11MbpspassiveWi-Fitransmis-sionsconsume14.
5and59.
2Wrespectively.
Contributions.
Wemakethefollowingcontributions:Wedemonstrateforthersttimethatonecangener-ate802.
11btransmissionsusingbackscattercommunica-tion.
Wepresentbackscattertechniquesthatsynthesize22MHzDSSSandCCKspreadspectrumtransmissionsthatcanbedecodedonexistingWi-Fidevices.
WedesignanetworkstackforthepassiveWi-Fitrans-mitterstocoexistwithotherdevicesintheISMband.
Further,wepresentadetailedanalyticalmodeltounder-standtheoperationalrangeofpassiveWi-Fitransmis-sionsindifferentdeploymentscenarios.
WebuildahardwareprototypeonanFPGAplatformandevaluateitinvariousscenarios.
WealsodesignapassiveWi-FiICandpresentitspowernumbers.
2PassiveWi-FiDesignOurdesignhastwomainactors:aplugged-indeviceandpassiveWi-Fidevices.
Theformercontainspowercon-sumingRFcomponentsincludingfrequencysynthesizerandpoweramplierandemitsasingletoneRFcarrier.
ItalsoperformscarriersenseonbehalfofthepassiveWi-FideviceandhelpscoordinatemediumaccesscontrolacrossmultiplepassiveWi-Fidevices.
ThepassiveWi-Fidevicebackscattersthetoneemittedbytheplugged-indevicetosynthesize802.
11btransmissionsthatcanbedecodedonanydevicethathasaWi-Fichipset.
Intherestofthissection,werstprovideaquickprimerfor802.
11bphysicallayerandbackscattercom-munication.
WethenexplainhowthepassiveWi-Fide-vicesgenerate802.
11bpacketsusingbackscattercom-munication.
Wethentheoreticallyanalyzetherangeofourtransmissionsinvariousdeploymentsscenarios.
2.
1Primerfor802.
11bTransmissions802.
11bisasetofWi-Fiphysicallayerspecicationsthatusespreadspectrummodulation.
802.
11busesDBPSK/DQPSKatthephysicallayerandachievesfour2Figure2:GenerationofWi-Fipacketsusingbackscatter.
Theplotontheleftshowsthe22MHzmainlobeandthesidelobesofthebaseband802.
11bpacketinthefrequencydomain.
TheplotontherightillustratesthebackscatteroperationatthepassiveWi-Fidevice.
Thetwomainlobesareshiftedbyfwithrespecttotheconstanttoneemittedbytheplugged-indevicetogeneratetheWi-Fipacket(inred)atfwifiandamirrorimage(inblue)atfwifi2f.
bitratesusingdifferentspreadingcodes.
Thelowertwobitratesof1and2Mbpsusedirect-sequencespreadspectrum(DSSS)while5.
5and11Mbpsusecomple-mentarycodekeying(CCK).
DSSSusesasinglecodetospreadtheinformationover22MHz,whileCCKusesasetofmultiplecodewordstobothencodebitsandalsoachievea22MHzspreadspectrumsignal.
Weoutlinehoweachofthe802.
11bbitratesareencoded.
1and2MbpsDSSStransmissions.
Togeneratethis,802.
11brstcreatescodedbitsfromtheincomingdatausinga11-bitbarkercode[39].
Specically,802.
11busesasinglebarkersequence,10110111000,thatisgen-eratedatabasebandfrequencyof11MHztospreadthespectrumover22MHz.
Tocreatethecodedbits,802.
11bXORseachofthedatabitswiththebarkersequence.
Thus,thecodedbitsfora'1'databitare10110111000andthatforthe'0'databitare01001000111.
EachofthesecodedbitsisencodedusingDBPSKandDQPSKmodulationtoachieve1and2Mbpstransmissionsre-spectively.
Atahighlevel,thisisachievedbysettingthephaseofthecarrier,sinθ.
DBPSKmodulationencodesa0and1bitbysettingθtoeither0orπ,whileDQPSKencodespairsofbitsbymodulatingthephasebetween0,π/2,πand3π/2.
5.
5and11MbpsCCKtransmissions.
Insteadofusingasinglebarkercode,CCKusesasetof8-bitcodewords.
Atahighlevel,togenerate5.
5Mbpstransmissions,theincomingdatabitstreamisdividedintoblocksof4bits.
ThersttwobitsareusedtopicktheDQPSKphaseandthelasttwobitsareusedtopickaspreadingcodeamongstfour8-bitcodewords.
Togenerate11Mbps802.
11btransmissions,theincomingdatabitsareinsteaddividedinto8bitblockswherethersttwobitsdeter-minetheDQPSKphaseshiftandthelast6bitsareusedtopickaspreadingcodeamongst648-bitcodewords.
Tosummarize,802.
11brequiresgeneratingthecodedbitsusingeitherDSSSorCCKandthenmodulatingthesebitswithDBPSKorDQPSK.
TherstoperationistypicallyimplementedindigitalbasebandlogicwhilethesecondrequirechangingthephaseofIandQcom-ponents.
Finally,wenotealsothatsincetheRFenergyisspreadacrossawideband,spreadspectrumtransmis-sionsareresilienttonarrowbandinterferencebothwithinandoutsidetheWi-Fichannel[39].
2.
2BackscatterCommunicationPrimerUnliketraditionalactiveradiocommunicationthatre-quiresgeneratingRFsignals,devicesusingbackscattercommunicationmodulatetheradarcross-sectionoftheirantennatochangethereectedsignal.
Tounderstandhowbackscatterworks,consideradevicethatcanswitchtheimpedanceofitsantennabetweentwostates.
Theef-fectofchangingtheantennaimpedanceisthattheradarcross-section,i.
e.
,thesignalreectedbytheantenna,alsochangesbetweenthetwodifferentstates.
Now,givenanincidentsignalwithpowerPincident,thepowerinthebackscatteredsignalcanbewrittenas,Pbackscatter=Pincident|Γ1Γ2|24(1)HereΓ1andΓ2arethecomplexconjugatesofthere-ectioncoefcientscorrespondingtothetwoimpedancestates.
Thustomaximizethepowerinthebackscat-teredsignalweneedtomaximizethedifferenceinthepowerofthetwoimpedancestateswhichisgivenby|Γ|2=|Γ1Γ2|24.
Ideally,toensurethatthepowerinthebackscatteredsignalisequaltothatoftheincidentsignal,weset|Γ|2to4whichcanbeachievedbymod-ulatingthereectioncoefcientsbetween+1and1.
Inpractice,however,backscatterhardwaredeviatesfromthisidealbehaviorandincurslosses;ourhardwareim-plementationhasalossofaround1.
1dB.
2.
3802.
11busingpassiveWi-FiGeneratingaWi-Fipacketusingbackscatterischalleng-ingfortwomainreasons.
First,thebackscatteredsignal3ismuchweakerthanthetonetransmittedbytheplugged-indevice.
AWi-Fireceiverwouldsuffersignicantin-bandinterferencefromthistonepreventingitfromde-coding.
Second,thepassiveWi-Fidevicehasasingledigitalswitchthattogglesbetweentwoimpedancestates,resultinginabinarysignal.
ItisunclearhowonemaygenerateWi-Fitransmissionsusingsuchabinarysystem.
Weoutlinehowtoaddressthesechallenges.
Werstdescribethesignaltransmissionsfromtheplugged-inde-viceandthentheoperationsatthepassiveWi-Fidevicethatallowustosynthesize802.
11btransmissions.
Transmissionsattheplugged-indevice.
IttransmitsatoneoutsidethedesiredWi-Fichannel.
OurkeyintuitionisthatWi-Fireceiversaredesignedtofunctioninthepresenceofout-of-bandinterference:802.
11breceiversarerequiredtoensurethatthesensitivityisreducedbynomorethan6dBinthepresenceofinterferenceintheadjacentbandthatis35dBgreaterthanthein-bandsig-nal[12].
Further,asWi-FiandBluetoothradiosarebe-ingintegratedontothesamechipset[6],Wi-Fifrontendsarebeingdesignedtofunctioninthepresenceofout-of-bandinterferencefromBluetoothdevices.
Sincethetonefromtheplugged-indeviceisnarrowerinbandwidththanBluetooth,thiswouldfurtherhelpsuppressthetoneifitisoutsidethedesiredWi-Fichannel.
Wenotehoweverthatexcessiveout-of-bandinterfer-ence,whichoccurswhentheWi-Fireceiverisrightnexttotheplugged-indevice,cansaturateand/orcompresstheRFfrontendresultinginsignicantdegradationofWi-Fiperformance.
Thisiscalledtheinput1dBcom-pressionpointwhichisaround0dBmforcommercialWi-Fidevices[13].
PassiveWi-FiinherentlyavoidsthisissuebyensuringthattheWi-Fireceiver(e.
g.
,smart-phoneorrouter)isnotnexttotheplugged-indevice.
BackscatteroperationsatpassiveWi-Fidevices.
Atahighlevel,thepassiveWi-Fioperationscanbede-scribedasrstshiftingtheout-of-bandtonetransmit-tedfromtheplugged-indevicetolieatthecenterofthedesiredWi-Fichannel.
Wethenusethisshiftedtonetocreate802.
11btransmissions.
Todothis,weleveragethreekeyfacts:(1)Frombasictrigonometry,2sinftsinft=cos(ff)tcos(f+f)t.
Thus,mul-tiplyingtwosinusoidalsignalscancreateafrequencyshift.
(2)Modulatingtheradarcrosssectionofanan-tennaeffectivelymultipliestheincomingsignalbythemodulatedsignal.
Thus,modulatingtheantennaatafre-quencyfwouldcreateafrequencyshiftintheincom-ingsignal.
(3)Allbitratesin802.
11baredifferentiallyphasemodulatedusingDBPSKorDQPSK.
Step1.
Shiftingthetonefromtheplugged-indeviceusingbackscatter.
Saytheplugged-indevicesendsthetonesin2π(fwifif)toutsidetheWi-Fichannel.
PassiveWi-FidevicesuseasquarewaveatafrequencyofftoshiftthetonetothecenteroftheWi-Fichannel.
FromFourieranalysis,asquarewavecanbewrittenas,Square(ft)=4π∞∑n=1,3,5,.
.
1nsin(2πnft)Heretherstharmonicisasinusoidalsignalatthede-siredfrequencyf.
Notethatthepowerineachoftheseharmonicscalesas1n2.
Sothethirdandthefthharmonicarearound9.
5dBand14dBlowerthantherstharmonic.
Thus,wecanapproximateasquarewaveasjustthesinusoidalsignal,4πsin(2πft).
Sincemodulatingtheradarcrosssectionofanantennaeffec-tivelymultipliestheincomingsignalbythemodulatedsignal,thebackscattersignalcanbeapproximatedassin2π(fwifif)tsin2πft.
Sowehaveusedbackscat-tertoeffectivelycreatestwotones,onecenteredatfwifiandtheotheratfwifi2f;thersttoneisatthecenterofthedesiredWi-Fichannel.
Step2.
Synthesizing802.
11btransmissionsusingbackscatter.
NowthatwehaveatonecenteredattheWi-Fichannel,thenextstepistocreate802.
11btransmis-sionsusingbackscatter.
802.
11busesDSSSandCCKencodingwhicharebothdigitaloperationsandhencecanbeperformedusingdigitallogicatthepassiveWi-Fide-vice.
Sothequestionthatremainsis:howdowegenerateDBPSKandDQPSKusingjustasquarewavecreatedatafrequencyfbythebackscatterswitchPassiveWi-FidoesthisbynotingthatDBPSKandDQPSKuseasinewavewithfourdistinctphases:0,π/2,π,3π/2.
Sincethesquarewavegeneratedbyourdigitalswitchcanbeapproximatedasasinewave,wecangeneratetherequiredfourphasesbychangingthetimingofoursquarewave.
Specically,shiftingthesquarewavebyhalfofasymboltime,effectivelycreatesaphasechangeofπ.
Phasechangesofπ/2and3π/2canbeachievedbyshiftingthesquarewavebyone-fourthandthree-fourthofasymboltime.
Thus,passiveWi-FidevicescanfullyoperateinthedigitaldomainwhilerunatabasebandfrequencyofafewtensofMHzandsyn-thesize802.
11btransmissionsusingbackscatter.
Wenotethefollowingpropertiesofourdesign.
Inadditiontocreatinga802.
11btransmissioncenteredatfwifi,asshowninFig.
2,ourbackscattermechanismalsocreatesamirrorcopycenteredatfwifi2fontheothersideofthetone.
Thus,weusetwicethebandwidthofatraditional802.
11btransmission.
Thisisthetradeoffwemaketoachieveordersofmagnitudelowerpowerconsumption.
Wenotethatsuchatradeoffiscommonin802.
11nsystemswhichusechannelbondingofadjacentWi-Fichannelstodoublethethroughput.
802.
11btransmissionshavesidelobes(Fig.
2);thesidelobesofthemirrorcopycreatesinterferenceforthedesiredWi-Fisignal.
Weplotthesignaltointerference4(a)SignaltoInterferenceratio(b)SensitivityLossFigure3:SIRandlossinreceiversensitivity.
Theplotshowstheeffectofdifferentf'sonthequalityandthesensitivityofthesynthesizedWi-Fipackets.
ratiofordifferentfrequencyshifts,f,atthepassiveWi-Fidevice.
Fig.
3(a)plotstheresultsforallfour802.
11bbitratesandshowsthattheinterferencefromthesidelobesofthemirrorcopyreducesasfincreases.
Thisisbecause,asfincreases,themirrorcopiesarefurtherseparatedinfrequency,resultinginlowerinterference.
Aneffectofthisinterference,however,isthatitaddsadditionalnoisetotheWi-Fisignal,reducingthenoisesensitivityatwhicheachofthe802.
11bbitratescanbedecoded.
Fig.
3(b)showsthelossinsensitivityforthefour802.
11bbitrates,asafunctionofthefrequencyoff-set,f.
Theplotsshowthatthesensitivitylossisslightlylargerforhigher802.
11bbitrates.
Thisisbecausehigherbitratesrequireacleanersignaltosuccessfullybede-coded.
Oursystemsetsfto12.
375MHz,wherethesensitivitylossislessthan2dBacrossall802.
11bbitrates.
ThisalsoensuresthatthepassiveWi-Fitransmis-sionsonlyoccupytwoadjacentWi-Fichannels.
NotethatWi-Fiapplieslterstoremovetheinterferingsidelobes.
Ourimplementationhoweverdoesnotdothis.
2.
4AnalyzingPassiveWi-Fi'sRangeInpassiveWi-Fi,thecommunicationrangedependsontwoparameters:thedistancebetweentheplugged-inde-viceandthepassiveWi-FitransmitterandthedistancebetweenthepassiveWi-FitransmitterandtheWi-Fire-ceiver.
Specically,thesignalstrengthatthereceiver,Pr,canbemodeledusingFriispathloss[34]asfollows,Pr=PtGt4πd21λ2G2passive4π|Γ|24αwifi14πd22λ2Gr4πThisequationhasthreekeyparts:theterminrstparen-thesismodelssignalpropagationfromtheplugged-inde-vice,withanoutputpowerPtandanantennagainGt,toapassiveWi-Fitransmitteratadistanced1away.
Thethirdterm,similarly,modelsthesignalpropagationfromthepassiveWi-FitransmittertoaWi-FireceiverwithanantennagainGrandatadistanced2away.
Here,λisthewavelengthoftheRFsignalbeentransmitted.
Fi-nally,themiddleparenthesismodelsthefractionofinci-Figure4:PassiveWi-Fi'sanalyticalreceivedsignalstrength.
ThepassiveWi-FidevicemovesalongthelineconnectingtheWi-Firouterandplugged-indevice.
Figure5:Signalstrengthversusdistancebetweenpas-siveWi-FitransmitterandWi-Fireceiver.
dentsignalfromtheplugged-indevicethatisbackscat-teredbyapassiveWi-FitransmitterwithanantennagainGpassive.
|Γ|2isthebackscattercoefcientwhichisameasureoftheefciencywithwhichpassiveWi-Ficangeneratebackscattersignals.
Asdescribedin§2.
2,thisis1.
1dBinourhardware.
Finally,αwifimodelsthelossinenergyduetosynthesisofWi-Fisignalsusingbackscat-ter.
Thisisaround4.
4dBandincludeshalfthepowerlostinthemirrorcopygeneratedbybackscatterandthelossesduetothesidelobesasdescribedin§2.
3.
Togainabetterintuition,considerthescenarioinFig.
4whereweplacetheplugged-indeviceandtheWi-Fireceiverseparatedby45feet.
Wemovethepas-siveWi-Fitransmitterbetweenthesedevices,alongthelineconnectingthem.
WesetPt,Gt,GrandGpassiveto30dBm,6dBi,0dBi,and2dBirespectively.
Fig.
4showsthereceivedsignalstrength,Pr,aswemovethepassiveWi-Fitransmitterbetweentheplugged-indeviceandtheWi-Fireceiver.
Theplotsshowtwokeypoints.
(1)ThereceivedsignalincreasesasthepassiveWi-FitransmittergetsclosetoeithertheWi-Fireceiverortheplugged-indevice.
Thisisbecause,maximizingthesig-nalstrengthrequiresminimizingtheproductd1d2,whichisachievedeitherbyreducingthedistanced1ord2.
(2)Themid-pointbetweentheplugged-indeviceandWi-Fireceiverhastheloweststrength.
Fig.
4showsthismid-pointsignalstrength,aswechangethedistancebetweentheplugged-indeviceandWi-Fireceiver.
Theplotshowsthatthisdecreaseswithdistancebetweentheplugged-indeviceandtheWi-Fireceiver.
Asexpected,itincreaseswithplugged-indevice'stransmitpower(Pt).
5(a)30ftSeparation(b)50ftSeparation(c)55ftSeparation(d)60ftSeparationFigure6:Theoreticalcoveragemapsfordifferentdistancesbetweentheplugged-indeviceandtheWi-Firouter.
Theblackdotsdenotethepositionsforthesedevices.
Theredregionrepresentspointsinthe2DspacewhereapassiveWi-Fitransmittercanbelocated,whileensuringthatthesignalfromittotheWi-Firouterisatleast-85dBm.
2.
4.
1UnderstandingDeploymentScenarios1.
IwanttodeploypassiveWi-Fidevicesinmyhome.
WheredoIplacetheplugged-indevicesoastomax-imizetheirrangeFig.
5showsthetheoreticalsignalstrengthattheWi-FireceiverasafunctionofitsdistancefromthepassiveWi-Fitransmitter.
WeshowtheresultsfordifferentdistancesbetweenthepassiveWi-Fitrans-mitterandtheplugged-indevice.
WesetGt,Gr,Gpassive,Ptto6dBi,0dBi,2dBi,and30dBmrespectively.
Theplotshowsthat,ingeneral,asthedistancebetweenthepassiveWi-FitransmitterandWi-Fireceiverincreases,thereceivedsignalstrengthreduces.
Moreimportantly,asthedistancebetweenthepassiveWi-Fitransmitterandplugged-indevicedecreases,thecoveragerangein-creases.
Thisisbecause,fromouranalysis,thesignalstrengthcanbeincreasedeitherbyreducingthedistancebetweenthepassiveWi-Fitransmitterandtheplugged-indeviceorthatbetweenthepassiveWi-FitransmitterandtheWi-Fireceiver.
Sinceourgoalistomaximizerange,weshouldreducethedistancebetweenthepas-siveWi-Fitransmitterandtheplugged-indevice.
InthepresenceofmultiplepassiveWi-Fidevices,thiswouldtranslatetominimizingtheworst-casedistancebetweentheplugged-indeviceandallpassiveWi-Fitransmitters.
2.
WheredoIplacemyWi-Firouterandtheplugged-indevice,sothatIcanhavepassiveWi-FidevicesworkfromanywhereinmyhomeFig.
6showsthe2Dcover-agemapsfordifferentdistancesbetweentheplugged-indeviceandtheWi-Firouter.
Theredregionrepresentpointsinthe2DspacewhereapassiveWi-Fitransmit-tercanbelocated,whileensuringthatthesignalfromittotheWi-Firouterisatleast-85dBm.
Thesemapsshowthatthecoverageareaisaunionoftwocirclescen-teredeachattheWi-Firouterandtheplugged-indevice.
So,asageneralruleofthumb,itisbettertodeploytheplugged-indeviceandtheWi-Firouterateitherendsofthecoveragearea.
Notehoweverthatatverylargedis-tancesbetweentheplugged-indeviceandWi-Firouter(Figs.
6(c)and(d)),weendupgettingtwoislandsofcoverage.
SuchlargedistancedeploymentsaresuitableonlywhenthepassiveWi-Fitransmittersaregoingtobeclosetoeithertheplugged-indeviceortheWi-Firouter.
Figure7:Structureofthesignalingpacket.
3PassiveWi-FiNetworkStackDesignWerstdescribehowpassiveWi-FidevicessharetheISMband.
WethenaddresstheissueofACKsandre-transmissionsandnally,presentourprotocoltoasso-ciatepassiveWi-Fideviceswiththenetwork.
3.
1SharingtheISMbandWi-FiusescarriersensetosharetheISMband.
ThishoweverrequiresaWi-FireceiverthatisONbeforeev-erytransmission.
SinceWi-Fireceiversrequirepower-consumingRFcomponentslikeLNA,frequencysynthe-sizers,mixersandADCs,thiswouldeliminatethepowersavingsfromourdesign.
Instead,wedelegatethetaskofcarriersensetotheplugged-indevice,whichalsoarbi-tratesaccessbetweenmultiplepassiveWi-Fidevices.
Weillustratethiswithanexample.
SayapassiveWi-Fitransmitterwantstosentapacketonchannel6andtheplugged-indevicetransmitsitstonebetweenWi-Fichannels1and6.
Beforeanyoftheabovetransmissionshappen,theplugged-indevicerstusescarriersensetoensurethattherearenoongoingtransmissionsonanythefrequenciesincludingandinbetweenchannel1and6.
Oncethechannelsarefoundfree,theplugged-inde-vicesendsapacketsignalingaspecicpassiveWi-Fidevicetotransmit.
Thissignalissentanddecodedus-ingtheultra-lowpowerreceiverdescribedin§3.
1.
1.
ThepacketstartswithanIDuniquetoeachpassiveWi-Fide-vice(seeFig.
7).
WhenthepassiveWi-FidevicedetectsitsID,ittransmitswithinaSIFSdurationattheendofthesignalingpacket.
Thesignalingpacketissentatthecenterofchannel1and6aswellasinbetweenthem.
ThispreventsotherdevicesintheISMbandfromcap-turingthechannelbeforethepassiveWi-Fidevicegetstotransmit.
Thepackethas16bitsandaddsaxedover-headof100sforeverypassiveWi-Fitransmission.
Theabovedescriptionassumesthattheplugged-inde-viceknowswhentosendthesignalingpackettoeachof6thepassiveWi-Fidevicesinthenetwork.
ToseehowthiscanbeachievedletusfocusonourtargetIoTap-plications.
Adevicesendingoutbeaconsisconguredtosendthemataxedrate.
Temperaturesensors,mi-crophonesandWi-Ficameras(e.
g.
,Dropcam[8])haveaxedrateatwhichtheygeneratedata.
Similarly,motionsensorshaveanupperboundonthedelaytheycantoler-ate.
ThepassiveWi-Fidevicesconveythisinformationtotheplugged-indeviceduringassociation(andcanup-dateitlaterusingtheprotocolin§3.
3).
Thisinformationisusedbytheplugged-indevicetosignaleachpassiveWi-Fideviceinaccordancetoitsdesiredupdaterate.
3.
1.
1Ultra-lowpowerreceiverdesignWeencodebitsusingON-OFFkeying.
Weuseapassiveenergydetectorwithanalogcomponentsandacompara-tortodistinguishbetweenthepresenceandabsenceofenergy.
Ourdesignisthesameasthatusedinourpriorwork[25,26]andweskipitforbrevity.
Weimplementthereceiverusingoff-the-shelfcomponentsanditcon-sumes18W,whileachievingabitrateof160kbps.
3.
2ACKsandRateAdaptationACKsandretransmissions.
Theplugged-indevicelis-tenstotheACKsandconveysthisinformationbacktothepassiveWi-Fisensor.
Specically,iftheACKissuc-cessfullydecodedattheplugged-indevice,itsetstheACKbitinthesignalingpacketshowninFig.
7to1andsendsittothepassiveWi-Fisensor,bypiggybackingitduringthenextperiodwhenthesensorisscheduledtotransmit.
IftheACKisnotreceivedattheplugged-indevice,itimmediatelyperformscarriersenseandsendsasignalingpacketwiththeACKbitsetto0.
WhenthepassiveWi-Fisensorreceivesthis,itretransmitsitssen-sorvalue.
Inourimplementation,theplugged-indevicedetectsanACKbydetectingenergyforaACKdurationattheendofthepassiveWi-Fitransmission.
Rateadaptation.
Wi-Fibitrateadaptationalgorithmstypicallyusepacketlossasaproxytoadaptthetrans-mitterbitrate.
Inourdesign,wedelegatethisfunctiontotheplugged-indevice.
Specically,theplugged-inde-viceestimatesthepacketlossrateforeachofitsasso-ciatedpassiveWi-Fidevicesbycomputingthefractionofsuccessfullyacknowledgedpackets.
Itthenestimatesthebest802.
11bbitrateandencodesthisinformationinthebitrateeldofthesignalingpacket.
Sincetheplugged-indeviceknowsthebitrateaswellasthepacketlength(fromassociationasdescribedin§3.
3),itknowshowlongthetransmissionsfromeachofitspassiveWi-Fideviceswouldoccupyonthewirelessmedium.
Thus,itstopstransmittingitstoneattheendofthepassiveWi-FitransmissionandlistensforthecorrespondingACKs.
Figure8:PassiveWi-Fiassociationprocedure.
3.
3NetworkAssociationFinally,wedescribehowthepassiveWi-Fitransmittersassociatewiththeplugged-indeviceaswellaswiththeWi-Firouterinthenetwork.
Thekeychallengeisthatsincetheplugged-indevicedoesnothaveafull-duplexradio(thelackofwhichisdesirabletomakeitpracticalandkeepitlowcost),thereisnodirectcommunicationchannelfromthepassiveWi-Fidevicetotheplugged-indevice.
Instead,asshowninFig.
8,theplugged-indeviceassociateswiththeWi-FirouterwithtwoMACaddress(MAC:1andMAC:2).
Theplugged-indevicethenbroadcastsadiscoverypacketusingON-OFFkey-ingmodulationthatcontainsthesetwoMACaddressesandstartswithabroadcastID.
ThenewpassiveWi-FidevicethentransmitsaWi-FipacketwiththesourceanddestinationaddressessettoMAC:2andMAC:1;thispacketgetsroutedthroughtheWi-Firoutertotheplugged-indevice.
Thepacketpayloadincludesthesen-sorupdaterate,packetlength,supportedbitratesanditsMACaddress,MAC:3.
Theplugged-indevicespoofsMAC:3andassociatesitwiththeWi-Firouter.
ItthenpicksauniqueIDandsendsittothepassiveWi-Fide-vicealongwithotherWi-Finetworkcredentials.
Finally,thepassiveWi-FidevicerespondswithaWi-FipacketwiththesourceanddestinationaddressessettoMAC:3andMAC:1;thispacketgetsroutedthroughtheWi-Firouterandconrmsassociationattheplugged-indevice.
Afterassociation,thepassiveWi-FitransmittercansendWi-Fipacketstotheplugged-indevicethroughtherouter,andchangeitsparametersincludingupdaterateandpacketlength.
NotethatthecredentialsforthespoofedMACaddressescouldbesentsecurelyusingamanufacturersetsecretkeysharedbetweenthepassiveWi-Fidevicesandtheplugged-indevices.
4HardwareImplementationWerstdescribeourimplementationofpassiveWi-Fiusingoff-theshelfcomponentsonanFPGAplatform.
WeusethistocharacterizepassiveWi-Fiinvariousdeploymentscenarios.
WethenpresentourICdesign7Figure9:PassiveWi-Fi'sICarchitecture.
Thefre-quencysynthesizergeneratesbasebandclock.
whichweusetoquantifyourpowerconsumption.
Off-the-shelfimplementation.
Weimplementapas-siveWi-Fiprototypeusingoff-the-shelfcomponentsforbackscatterandanFPGAfordigitalprocessing.
ThebackscattermodulatorconsistedofHMC190BMS8SPDTRFswitchnetworkona2-layerRodgers4350substrate[10].
Theswitchwasdesignedtomodulatebe-tweenopenandclosedimpedancestatesandhada1.
1dBloss.
Alltherequiredbasebandprocessingincludingdatascrambling,headergeneration,DSSS/CCKencod-ing,CRCcomputationandDBPSK/DQPSKmodulationwerewritteninVerilog.
TheVerilogcodewassynthe-sizedandprogrammedonaDE1CycloneIIFPGAde-velopmentboardbyAltera[2].
Weimplementfourshiftsof12.
375,16.
5,22and44MHz.
ThedigitaloutputoftheFPGAwasconnectedtothebackscatterswitchtogener-atetheWi-Fipackets.
A2dBiomnidirectionalantennawasusedonthepassiveWi-Fidevice.
Theplugged-indevicewassettotransmitatanEIRPof30dBm.
Integratedcircuitimplementation.
CMOStechnologyscalinghasenabledtheexponentialscalinginpowerandareaforintegratedcircuits.
Wi-FichipsetshavetriedtoleveragescalingbutwithlimitedsuccessduetotheneedforpowerhungryanalogcomponentsthatdonotscaleinpowerandsizewithCMOStechnology.
However,base-bandWi-Fioperationsareimplementedinthedigitaldo-mainandtendtoscaleverywellwithCMOS.
Forcon-text,Atheros'sAR6003[4]andAR9462[17]chipsetsthatwerereleasedin2009and2012use65nmCMOSand55nmCMOSnodeimplementationsrespectively.
ForpassiveWi-Fidevice'sintegratedcircuitimplemen-tation,wechosethe65nmLPCMOSnodebyTSMC,whichgivesuspowersavingsofbasebandprocessingandensuresafaircomparisonwithcurrentindustrystan-dards.
TheICarchitectureofthepassiveWi-FideviceisshowninFig.
9andhasthreemaincomponents:Basebandfrequencysynthesizer.
Itgeneratesthe11MHzclockrequiredforbasebandprocessingaswellasfourphasesat12.
375MHzoffsetsrequiredforDBPSKandDQPSK.
Wephasesynchronizethe11MHzand12.
375MHzclockstoavoidglitchesduringphasemod-ulation.
WeusedanintegerNchargepumpandringTable1:PassiveWi-Fi'sICPowerConsumption1Mbps11MbpsBasebandFrequencySynthesizer5.
6W5.
6WBasebandProcessor5.
0W48WBackscatterModulator3.
9W5.
6WTotalPower14.
5W59.
2Woscillator-basedPLLtogenerate49.
5MHzclockfroma12.
375kHzreference.
The49.
5MHzclockisfedtoaquadratureJohnsoncountertogeneratethefourphaseswiththerequiredtimingoffsets(correspondingto0,π2,πand3π2phases).
Thesame49.
5MHzcarrierisdividedby4.
5togeneratethe11MHzbasebandclock.
Basebandprocessor.
Ittakesthepayloadbitsasinputandgeneratesbaseband802.
11bWi-Fipacket.
WeusedtheVerilogcodethatwasveriedontheFPGAandusetheDesignCompilerbySynopsistogeneratethetransis-torlevelimplementationofthebasebandprocessor[19].
Backscattermodulator.
ItmixesthebasebanddatatogenerateDBPSKandDQPSKanddrivestheswitchtobackscattertheincidenttonesignal.
Thebasebanddataaretheselectinputstoa2-bitmultiplexerwhichswitchesbetweenthefourphasesofthe12.
375MHzclocktogen-eratethephasemodulateddata.
ThemultiplexeroutputisbufferedandusedtodrivetheRFswitch,whichtogglestheantennabetweenopenandshortimpedancestate.
Table1showsthepowerconsumptionofourdesignat1Mbpsand11MbpswhichwascomputedusingtheCadencespectreandSynopsisDesignCompliertoolk-its[5,19].
PassiveWi-Fi'sICimplementationfor1Mbpsand11Mbpsconsumesatotalof14.
5and59.
2Wofpowerrespectively.
Thedigitalfrequencysynthe-sizerisclockedforDQPSKandconsumesaxedpowerforalldatarates.
Thepowerconsumptionofthebase-bandprocessorthatgeneratesthe802.
11bpacketsscaleswiththedatarateandconsumes30%and80%oftotalpowerfor1and11Mbpsrespectively.
5Evaluation5.
1PhysicalLayerPerformanceWerstevaluatetherangeandthentheeffectofthefre-quencyshiftusedinoursystem.
Finally,wepresentre-sultsforallfour802.
11bbitrates.
5.
1.
1RSSIinLine-of-sightscenariosWerunexperimentsintwoline-of-sightscenarios.
Deploymentscenario1.
WexthedistancebetweenthepassiveWi-Fideviceandtheplugged-indevice.
Wethen8Figure10:RSSIindeploymentscenario1.
WemovethephoneawayfromthepassiveWi-Fidevice.
Figure11:RSSIindeploymentscenario2.
d1(d2)isthedistancebetweenthepassiveWi-Fiandplugged-indevice(Wi-Fireceiver).
ThepassiveWi-Fidevicemovesalonethelinejoiningtheothertwodevices.
movetheWi-FireceiverawayfromthepassiveWi-FideviceandmeasuretheRSSIofthepassiveWi-Fitrans-missionsasseenbythereceiver.
WeruntheexperimentsintheCSEatriumwherethemaximumdistancepossiblewhenthepassiveWi-FideviceandWi-Fireceiverwereplacedoneitherendwasaround100feet.
Inourex-periments,wesetthepassiveWi-Fidevicetogenerate802.
11bbeaconpacketsat1Mbps.
Thesepacketshaveapayloadof68byteswheretheSSIDissettoWiLab_0000andaretransmittedevery15ms.
Wesettheplugged-indevicetotransmititstone12.
375MHzfromthecenterofWi-Fichannel1betweenchannel1and6.
WeuseanHTCOne(M7)phoneasourWi-Fireceiver.
SincethepassiveWi-FideviceistransmittingWi-Fibeacons,itappearsasaWi-FiAPatthesmartphone.
TomeasuretheRSSIvaluesofthesepackets,weuseathirdpartyAndroidappcalledWiAnalyzer[3]thatprovidestheRSSIvalueasshowninFig.
12.
Ineachexperiment,weholdthesmartphoneinourhandandmeasurethereportedRSSIvaluesaswewalkawayfromthepassiveWi-Fidevice.
Themeasurementsaretakenatincrementsof4feet.
Fig.
10plotsthere-sultsforthreedifferentvaluesofthedistancebetweenthepassiveWi-Fitransmitterandtheplugged-indevice.
Thex-axisplotsthedistancebetweenthepassiveWi-FitransmitterandtheWi-Fireceiverwhilethey-axisplotsthereportedRSSIvalues.
Theplotsshowthatasex-pected,theRSSIvaluesreduceasthephonemovesawayFigure12:SnapshotoftheWi-Fianalyzerapp.
WiLab_0000correspondstopassiveWi-Fibeacons.
Figure13:RSSIindeploymentscenario1inthepres-enceofwalls.
Thebrownblocksshowthewallpositions.
fromthepassiveWi-Fidevice.
Further,aspredictedbyouranalysisin§2.
4,therangeofourpassiveWi-Fitrans-missionsreducewiththedistancebetweenthepassiveWi-Fitransmitterandtheplugged-indevice.
WhentheseparationbetweenthepassiveWi-Fitransmitterandtheplugged-indeviceis3or6feet,therangeofthepassiveWi-FitransmissionsspanstheentirelengthoftheCSEatrium.
Therangeisaround55feetwhenthisseparationis12feet.
Thisreducedrangeisduetoacombinationofmultipathandaweakbackscattersignal.
Deploymentscenario2.
Nextweplacetheplugged-indeviceandtheWi-Fireceiveratadistanced1+d2.
WemovethepassiveWi-Fitransmitteralongthelinecon-nectingthesetwodevices.
AsabovethepassiveWi-Fitransmitterissettogenerate802.
11bbeaconpacketsat1Mbpsandtheplugged-indevicetransmitsitstoneat12.
375MHzfromthecenterofWi-Fichannel1.
WecollecttheRSSIvaluesfromaHTCOne(M7).
Fig.
11plotstheresultsforthreedifferentvaluesofthedistancebetweentheplugged-indeviceandtheWi-Fireceiver(d1+d2).
Eachpointonthex-axisdenotesthedistancebetweenthepassiveWi-Fideviceandtheplugged-inde-vice(d1).
TheplotsshowthattheRSSIvaluesarethehighestwhenthepassiveWi-FitransmitteriseitherclosetotheWi-Fireceiverortheplugged-indevice.
Further,theRSSIvaluesareloweratthemidpointbetweenthetwodevices,conrmingourtheoreticalanalysis.
5.
1.
2RSSIinThrough-the-WallScenariosWererunexperimentsintheabovedeploymentscenariosbutnowinthepresenceofwalls.
Intherstdeployment,9Figure14:RSSIindeploymentscenario2inthepres-enceofwalls.
Thebrownblocksdenotethewalls.
d1(d2)isthedistancebetweenthepassiveWi-Fideviceandplugged-indevice(Wi-Fireceiver).
weplacethepassiveWi-Fideviceandtheplugged-inde-vicesatdistancesof1and6feetfromeachother.
AstheWi-FireceivermovesawayfromthepassiveWi-Fide-vice,itisseparatedbymultipledoublesheet-rock(plusinsulation)wallswithathicknessofapproximately5.
7inches.
Asbefore,weuseanHTCOne(M7)phoneasourWi-Fireceiverandsettheplugged-indevicetotrans-mitwitha12.
375MHzfrequencyoffsetfromchannel1.
ThepassiveWi-FideviceperiodicallytransmitsWi-Fibeaconsat1MbpsandwemeasuretheRSSIvaluesasreportedbytheWi-Fireceiver.
Fig.
13showsthattherangeisnowaround28feetwhenthedistancebetweenthepassiveWi-Fideviceandtheplugged-indeviceis6feet.
Thisisexpectedbecausethesignalsgetattenu-atedbytwowallsbeforearrivingattheWi-Fireceiver.
Intheseconddeployment,wexthelocationoftheplugged-indeviceintherstroomandplacetheWi-Fireceiverinthethirdroomatadistanceof25feet.
WethenmovethepassiveWi-Fidevicealongthelinecon-nectingtheabovetwodevicesandmeasuretheRSSIre-portedbytheWi-Fireceiver.
Fig.
14plotstheRSSIre-sultsandshowthattheyfollowasimilartrendasbeforeandworkeveninthepresenceofattenuationfromwalls.
5.
1.
3EffectofdifferentfrequencyshiftsWeevaluatehowdifferentfrequencyshiftvalueseffectpassiveWi-Fiperformance.
Todothis,weplacethepas-siveWi-Fitransmitterandplugged-indevice6feetfromeachother.
WemoveaWi-FireceiverawayfromthepassiveWi-Fideviceina50footlongspace.
Thepas-siveWi-Fidevicetransmits1MbpsWi-Fipacketswithapayloadof512bytesonchannel1.
WeusetheIn-tel5350chipsetasaWi-Fireceiverwhichrunstsharktologallthepacketsthataresuccessfullydecodedbyit.
ThepassiveWi-Fitransmitterconsecutivelytransmits200uniquesequencenumbersinaloopusingwhichwecomputethepacketerrorrateattheWi-Fireceiver.
Werepeattheseexperimentsforthreedifferentshifts.
Fig.
15plotsthePERattheWi-Fireceiverasafunc-tionofdistancebetweenthepassiveWi-FitransmitterandtheWi-Fireceiver.
ThegureshowsthatthePERisconsistentlyaround20%whenweusefrequencyshiftsof44and16.
5MHz.
Forcomparison,wemeasuredFigure15:Effectofdifferentfrequencyshifts.
ThePERsareverystablewith16.
5MHzand44MHzoffsets.
Figure16:All802.
11bbitrates.
Ourdesigncangener-ate802.
11btransmissionsacrossallfourbitrates.
thePERforaconventionalWi-Fitransmitterplaced10feetawayandobservedsimilarPERvalues.
Thein-terestingobservationhoweveristhatwhentheshiftis12.
375MHz,weseealargevariationinthePERasthelocationoftheWi-Fireceiverchanges.
Thisisbecauseoftworelatedreasons.
First,whentheshiftissmall,thetonefromtheplugged-indeviceisveryclosetothede-siredWi-Fichannel.
Second,becauseofmultipath,dif-ferentlocationsseedifferentsignalstrengthdifferencesbetweenthepassiveWi-Fideviceandtheout-of-bandin-terferencefromtheplugged-indevice.
Whentheshiftissmall,thisout-of-bandinterferencecanstillbesigni-cantincertainlocationstocreatelosses.
Wenotethatwhilea44MHzshiftistoohightobewithintheISMband,a16.
5MHzshifthasPERsthatarestableacrosslo-cationsandyetissmalltobewithintheISMbandwhilegeneratingpacketsonallWi-Fichannels.
5.
1.
4Higher802.
11bbitratesFinally,weshowthatpassiveWi-Ficangenerateall802.
11bbitrates.
WeseparatethepassiveWi-Fiandplugged-indeviceby6feet.
WechangetheWi-Fire-ceiverlocationtovespotsina15*24ftroom.
Theplugged-indeviceissettousea12.
375MHzoffset.
ForeachWi-Fireceiverlocation,thepassiveWi-Fidevicetransmits802.
11bpacketsat1,2,5.
5and11Mbps.
Foreachbitrate,thepassiveWi-Fidevicesends200packetswitha512bytepayloadwithdifferentsequencenum-bers.
TheWi-Fireceiver(Intel5350)isconguredtocomputetheeffectivePHYgoodputachievedbymulti-plyingthetransmittedWi-Fibitratewiththefractionofpacketsthataredecoded.
Fig.
16plotsaCDFofthePHY-layergoodputacrossthevelocationsdemonstrat-ingthatwecangenerateallfour802.
11bbitrates.
10(a)IDDetection(b)Wi-FiCoexistenceFigure17:PassiveWi-Finetworkperformance.
5.
2PassiveWi-FiNetworkPerformanceAsdescribedin§3.
1tocoexistintheISMband,theplugged-indevicerstperformscarriersenseandthensignalsthepassiveWi-Fidevicetotransmit.
Inthissec-tion,werstevaluatehowwellthesignalingmechanismworks.
WethendescribehowouroverallcarriersensemechanismworksinthepresenceofotherWi-Fidevices.
5.
2.
1EvaluatingthesignalingmechanismTheplugged-indevicetransmitsapacketwitha10-bitIDthatisuniquetoeachpassiveWi-Fidevice.
Weevaluatetwoaspects:(1)theprobabilitywithwhichthesignalfromtheplugged-indevicetriggertransmissionsfromthecorrectpassiveWi-Fideviceand(2)theprobabilitythatitwouldtriggerthewrongpassiveWi-Fidevice.
Toevaluatethisweconsidertheworst-casescenario:twodevicesthathaveIDsthatdifferbyjustonebit.
Wesettheplugged-indevicetotransmitthesignalingpacketwiththeIDoftherstdevice.
Wemovethetwopas-siveWi-Fidevicesawayfromtheplugged-indevice.
Ateachdistancevalue,theplugged-indeviceisconguredtotransmitthesignalingpacketforatotalof1890times.
ThepassiveWi-FidevicesuseanenvelopedetectortocorrelatefortheirspecicID.
Wecomputethefractionofthe1890signalingpacketsthataredecodedandmatchtheIDofthepassiveWi-Fidevice.
Weruntheseexper-imentsintheUWCSEatriumforincreasingdistancesfromtheplugged-indevice.
Fig.
17(a)showthefractionofsignalingpacketsthatmatchtheIDofthetwopas-siveWi-Fidevicesasafunctionofthedistancefromtheplugged-indevice.
TheplotshowsthatneitherdeviceincorrectlydecodestheID.
Thisisbecauseourreceiverbuildsonourpriorwork[25,26,29,42]andhasgonethroughmultipleiterationstoimproveitsreliability.
5.
2.
2EvaluatingpassiveWi-Fi'scarriersenseTheplugged-indeviceperformscarriersenseandsig-nalsaspecicpassiveWi-Fidevicetotransmit.
TocomparehowourmechanismcomparestostandardWi-Fi,wecomparetheperformanceofaconcurrentWi-Fitransmitter-receiverpairinthepresenceofapassiveWi-FitransmitterwiththatofatraditionalWi-Fitransmit-ter.
WeusetwoIntel5350Wi-FichipsetstotransmitandreceiveWi-Fipacketsusingiperf.
Thedevicesusethechipset'sdefaultbitrateadaptation.
Werunexperi-mentsintwoscenarios:1)weuseaRalinkRT2070Wi-Fichipsettotransmitpacketsat1Mbpsevery15msand2)wesetourpassiveWi-Fidevicetotransmititspacketevery15msat1Mbpsusingourcarriersensemecha-nism.
Wemeasurethethroughputachievedbyacon-currentWi-Fitransmitter-receiverpairinthepresenceofthesetwodevices.
Fig.
17(b)plotstheTCPthroughputandshowsthatpassiveWi-FihasasimilarimpactontheongoingowasatraditionalWi-Fitransmitter.
Thisisbecause,passiveWi-Fiaddsonlyasmallxed100soverhead.
Thissmalloverheadishoweverovershadowedbytransientchangesinnetworkconditions.
5.
3ApplicationsWerstconsiderlowlatencysensorslikemicrophonesandcamerasthattransmitcontinuously.
Wethenanalyzeduty-cycledsensors.
1)Lowpowermicrophonesconsume17W[1]andanADCdigitizingthemicrophoneoutputconsumes33W[1],resultingin50Wforthesensingsubsys-tem.
IfweuseanIoTWi-FichipsetbyGainspanorTItocontinuouslytransmitaudio,theactiveWi-Fitransmitterconsumes670mW[9,21].
Thisresultsinatotalpowerbudgetof670.
05mWwhichisdominatedbytheWi-Fichipset.
However,ifweusepassiveWi-Fiat1Mbps,thepowerbudgetdropsto65W,i.
e.
,a1000xreduction.
2)AlowpowercameralikeOV7690operatingatVGAresolutionandcapturingoneimagepersecondconsumesanaverageof10mW[15].
Thecameraoutputsrawdataat2.
45Mbpswhichcanbetransferredwirelesslywithoutpowerhungryon-boardcompression.
UsinganIoTWi-FichipsetfromGainspanorTI,bringsthetotalpowerconsumptionofthesystemto680mW.
Ifwesubsti-tuteanactiveWi-Fichipsetwhichconsumes670mWofpowerwith11MbpspassiveWi-Fi,wecanimprovethebatterylifeofWi-Fivideocamerabyatleast50x[9,21].
3)DutycycledsensorssuchasiBeacon[11]andhomeproximitysensors[16]periodicallytransmitdatausingBluetoothLowEnergyandZigBeeprotocolsrespec-tively.
Theytypicallytransmitbeacons/datapacketsatarateof100msto900msandlastfor3monthsto3yearsrespectivelyonacoincellbattery[11].
IfwereplacetheBLE/ZigBeetransmitterwhichconsumes35mW[20]intransmitmodewithpassiveWi-Ficonsuming15W,thebatterylifecanbeextendedwellbeyond10years.
6RelatedWorkRFIDsystems.
RFIDtagsbackscatterthesignalbacktoadedicated900MHzRFIDreader.
Theuseofbackscat-11terasageneralcommunicationmechanism,however,hasbeenlimitedtoRFIDsystemsfortwokeyreasons.
First,todecodetheweakbackscatteredsignals,thereadereliminatesthestrongsignalfromthereaderusingfull-duplexradios[27,43].
Thisrequiresexpensivecir-culatorsandhighlylinearanalogRFfrontendatthereaderthatcontributestoitshighcost.
Incontrast,Wi-Fichipsetsdonotrequirethespecializedcomponents,canbefullyintegratedinsiliconandhence,areordersofmagnitudelessexpensive.
Second,enablingbackscat-tercommunicationwithexistingdevicesrequiresacom-pletehardwarechangetotheirchipsetsandincorporatingadedicatedfullduplexradio;thisisahighbarthathaslimitedtheadoptionofbackscatterbeyondRFID.
Wi-Fiandambientbackscattersystems.
Since2013,wehaveintroducedtheconceptsofambientandWi-Fibackscatter[25,28,37]wherebattery-freedevicescom-municatewitheachotherbybackscatteringambientsig-nalssuchasTVandWi-Fitransmissions.
Thebasicdif-ferencebetweenthesedesignsandpassiveWi-FiisthatWi-Fibackscattersystemscreateanadditionalnarrow-banddatastreamtorideontopofexistingWi-Fisignals.
Incontrast,passiveWi-Fiaimstousebackscattertogen-erate802.
11btransmissionsthatcanbedecodedbybil-lionsofexistingdeviceswithaWi-Fichipsets.
Inparticular,ourpriorworkonWi-Fibackscatter[25]demonstratedthatexistingWi-Fichipsetscandecodebackscatteredinformationfromatagusingchangestotheper-packetCSI/RSSIvaluesat1kbpsbitratesanda2mrange.
[38]improvedtherateofthiscommunicationusingafull-duplexradiotocancelthehigh-powerWi-Fitransmissionsfromthereaderanddecodetheweakphase-modulatednarrowbandbackscatteredsignalatthereader.
Thishasallowedthemtoachievedataratesofupto5Mbpsatarangeof1mand1Mbpsatarangeof5m.
Arecentnewsrelease[14]claimstoachieve330MbpsWi-Fibackscattercommunicationat2.
5musingacustomICthatimplementsafull-duplexradio.
Thechallengewiththesefull-duplexdesignsisthattheyhavethesameproblemasconventionalRFIDdesigns—theyrequireacustomfull-duplexradiotobeincorporatedatthereceiverandhencethebackscatteredsignalscannotbedecodedonanyoftheexistingWi-Fidevices.
Finally,[23]createsBluetoothsignalsusingsubcarriermodulationtocreate370kHznarrowband2-FSKsig-nals.
Instead,wecreate22MHzDSSS/CCKtransmis-sionsusingbackscatterandenableWi-Fitransmissions.
Wealsopresentanetwork-layerstackdesignthatenablesustooperatewithexistingdevicesintheISMband.
Duty-cycledradios.
Thekeyideainthesesystemsistodesignacustomlowpowerradiotransmitteranduseawakeupreceivertodutycyclethetransmitterandreducetheaveragetransceiverpowerconsumption[41].
Thepowerconsumptionofsuchtransmittersatsub-milliwattoutputpowerisintheorderof100W[36,44]tofewmWs[22,35,40].
Further,suchradiosusecustompro-tocolssupporting10-100kbpsdataratesthatrequirede-ploymentofspecialpurposereceiversandhardware.
Incontrast,passiveWi-FigeneratesWi-Fitransmissionsattensofmicrowattsofpower.
GiventheubiquityofWi-Fi,thissignicantlylowersthebarforadoption.
Further,thedutycycleoperationisorthogonaltopassiveWi-FiandcanbeusedtofurtherreducethepowerconsumptionofasystememployingpassiveWi-Fi.
LowpowerWi-Fitransceivers.
TheWi-FiindustryhasdesignedchipsetsforIoTapplicationsincludingQUAL-COMMQCA4002andQCA4004[18].
Thesedesignsreducethepowerconsumptionbydecreasingthetrans-mitpowerbyuptoahalfwheninproximityofanotherdevice.
Theyalsooptimizethepowerconsumptionoftheirsleepmodetobelessthan1mW.
GainspanandTIWi-Fichipsetsincorporatea20Wstandbymodeandcanswitchtoactivemodewithintensofmillisec-onds[9,21].
However,theiractivetransmissionpowerisaround600mW[9,21]whichisordersofmagni-tudehigherthanpassiveWi-Fi.
Intel'sMoore'sradio[7]designsdigitalversionsforRFcomponentssuchasfre-quencysynthesizers.
ThisreducesthecostandsizeoftheRFchipsetratherthanitspower—adigitalWi-Fifre-quencysynthesizerconsumes10-50mW[7,24]whichissimilarinpowerconsumptiontoitsanalogcounterpart.
Finally,recentlowpowerWi-Fireceiverdesignsusetechniqueslikedynamicvoltageandfrequencyscal-ing[30]andcompressivesensing[31].
Inparticular,SloMo[31]leveragesthesparsityinherentto802.
11bDSSSsignalsusingcompressivesensingtooperatetheradioatalowerclockrate.
Enfold[30]extendsthistoworkwithOFDMmodulation.
Ourworkonenablingultra-lowpowerWi-Fitransmissionsiscomplimentarytothisworkandcaninprinciplebeintegratedtogether.
7ConclusionWedemonstrateforthersttimethatonecangener-ate802.
11btransmissionsusingbackscattercommuni-cation,whileconsuming4-5ordersofmagnitudelowerpowerthanexistingWi-Fichipsets.
Wi-Fihastradition-allybeenconsideredapower-consumingsystem.
Thus,ithasnotbeenwidelyadoptinginthesensornetworkandIoTspacewherelow-powerdevicesprimarilytrans-mitdata.
Webelievethat,withitsordersofmagnitudelowerpowerconsumption,passiveWi-Fihasthepoten-tialtotransformtheWi-Fiindustry.
Acknowledgements.
WethankBenRansfordandourshepherdJonHowell.
ThisworkwasfundedinpartbyNSFunderawardsCNS-1452494andCNS-1407583andaQual-commfellowship.
12References[1]ADMP801.
http://www.
cdiweb.
com/datasheets/invensense/ADMP801_2_Page.
pdf.
[2]Alterade1fpgadevelopmentboard.
http://www.
terasic.
com.
tw/cgi-bin/page/archive.
plNo=83.
[3]Androidwi-analyzer.
https://play.
google.
com/store/apps/detailsid=com.
farproc.
wifi.
analyzer&hl=en.
[4]Atherostargetscellphonewithwi-chip.
http://www.
eetimes.
com/document.
aspdoc_id=1172134.
[5]Cadencerfspectre.
http://www.
cadence.
com/products/rf/spectre_rf_simulation/pages/default.
aspx.
[6]Co-existenceofwi-andbluetoothra-diosbymarvell.
http://www.
marvell.
com/wireless/assets/Marvell-WiFi-Bluetooth-Coexistence.
pdf.
[7]Connectingthefuture:Thelatestresearchfromintellabs.
http://download.
intel.
com/newsroom/kits/idf/2012_fall/pdfs/IDF2012_Justin_Rattner.
pdf.
[8]Dropcam.
https://nest.
com/camera/meet-nest-cam/dropcam=true.
[9]Gainspangs1500m.
http://www.
alphamicro.
net/media/412417/gs1500m_datasheet_rev_1_4.
pdf.
[10]Hms190bms8byhittitemicrowavedevices.
https://www.
hittite.
com/content/documents/data_sheet/hmc190bms8.
pdf.
[11]ibeacons.
http://beekn.
net/2014/04/will-apple-pull-plug-ibeacon-devices/.
[12]Ieee802.
11standard,2012.
http://standards.
ieee.
org/getieee802/download/802.
11-2012.
pdf.
[13]Max2830bymaxim.
https://datasheets.
maximintegrated.
com/en/ds/MAX2830.
pdf.
[14]Nasanewsrelease:Awi-reectorchiptospeedupwearables.
http://www.
jpl.
nasa.
gov/news/news.
phpfeature=4663.
[15]Ovm7690cameramodule.
http://www.
ovt.
com/uploads/parts/OVM7690_PB(1.
0)_web.
pdf.
[16]Proximitysensors.
https://www.
ia.
omron.
com/products/category/sensors/proximity-sensors/.
[17]Qualcommatheros9462.
http://www.
qca.
qualcomm.
com/wp-content/uploads/2013/11/AR9462.
pdf.
[18]Qualcommqca4002andqca4004.
http://www.
eeworld.
com.
cn/zt/wireless/downloads/QCA4002-4004FIN.
pdf.
[19]Synopsisdesigncomplier.
http://www.
synopsys.
com/Tools/Implementation/RTLSynthesis/DesignCompiler/Pages/default.
aspx.
[20]TICC2541.
http://www.
ti.
com/lit/ds/symlink/cc2541.
pdf.
[21]TICC3100MOD.
http://www.
ti.
com/lit/ds/symlink/cc3100mod.
pdf.
[22]J.
Ayers,N.
Panitantum,K.
Mayaram,andT.
S.
Fiez.
A2.
4ghzwirelesstransceiverwith0.
95nj/blinkenergyformulti-hopbattery-freewirelesssen-sornetworks.
InVLSICircuits(VLSIC),2010IEEESymposiumon,pages29–30.
IEEE,2010.
[23]J.
EnsworthandM.
Reynolds.
Everysmartphoneisabackscatterreader:Modulatedbackscattercom-patibilitywithbluetooth4.
0lowenergy(ble)de-vices.
InRFID,2015IEEEInternationalConfer-enceon.
[24]K.
Greene.
Intel'stinywi-chipcouldhaveabigimpact.
MITTechnologyreview,2012.
[25]B.
Kellogg,A.
Parks,S.
Gollakota,J.
R.
Smith,andD.
Wetherall.
Wi-backscatter:Internetconnec-tivityforrf-powereddevices.
InProceedingsofthe2014ACMConferenceonSIGCOMM,2014.
[26]B.
Kellogg,V.
Talla,andS.
Gollakota.
Bringinggesturerecognitiontoalldevices.
InUsenixNSDI,volume14,2014.
[27]P.
B.
Khannur,X.
Chen,D.
L.
Yan,D.
Shen,B.
Zhao,M.
K.
Raja,Y.
Wu,R.
Sindunata,W.
G.
Yeoh,andR.
Singh.
Auniversaluhfrdreadericin0.
18-mcmostechnology.
Solid-StateCircuits,IEEEJournalof,43(5):1146–1155,2008.
13[28]V.
Liu,A.
Parks,V.
Talla,S.
Gollakota,D.
Wether-all,andJ.
R.
Smith.
Ambientbackscatter:Wire-lesscommunicationoutofthinair.
InProceedingsoftheACMSIGCOMM2013ConferenceonSIG-COMM,2013.
[29]V.
Liu,V.
Talla,andS.
Gollakota.
Enablingin-stantaneousfeedbackwithfull-duplexbackscatter.
InProceedingsofthe20thannualinternationalconferenceonMobilecomputingandnetworking,pages67–78.
ACM,2014.
[30]F.
Lu,P.
Ling,G.
M.
Voelker,andA.
C.
Snoeren.
Enfold:downclockingofdminwi.
InProceed-ingsofthe20thannualinternationalconferenceonMobilecomputingandnetworking,pages129–140.
ACM,2014.
[31]F.
Lu,G.
M.
Voelker,andA.
C.
Snoeren.
Slomo:Downclockingwicommunication.
InNSDI,pages255–258,2013.
[32]J.
ManweilerandR.
RoyChoudhury.
Avoidingtherushhours:Wienergymanagementviatrafciso-lation.
InMobiSys,2011.
[33]R.
Mittal,A.
Kansal,andR.
Chandra.
Empoweringdeveloperstoestimateappenergyconsumption.
InMobiCom,2012.
[34]P.
NikitinandK.
Rao.
TheoryandmeasurementofbackscatteringfromRFIDtags.
AntennasandPropagationMagazine,IEEE,48(6):212–218,de-cember2006.
[35]B.
Otis,Y.
Chee,R.
Lu,N.
Pletcher,andJ.
Rabaey.
Anultra-lowpowermems-basedtwo-channeltransceiverforwirelesssensornetworks.
InVLSICircuits,2004.
DigestofTechnicalPapers.
2004Symposiumon,pages20–23.
IEEE,2004.
[36]J.
PandeyandB.
P.
Otis.
Asub-100wmics/ismbandtransmitterbasedoninjection-lockingandfre-quencymultiplication.
Solid-StateCircuits,IEEEJournalof,46(5):1049–1058,2011.
[37]A.
N.
Parks,A.
Liu,S.
Gollakota,andJ.
R.
Smith.
Turbochargingambientbackscattercommu-nication.
InProceedingsofthe2014ACMConfer-enceonSIGCOMM,2014.
[38]D.
Pharadia,K.
R.
Joshi,M.
Kotaru,andS.
Katti.
Back:Highthroughputwibackscatter.
InPro-ceedingsofthe2015ACMConferenceonSpecialInterestGrouponDataCommunication,2015.
[39]J.
G.
ProakisandM.
Salehi.
Digitalcommunica-tions.
2005.
McGraw-Hill,NewYork.
[40]J.
Rabaey,J.
Ammer,B.
Otis,F.
Burghardt,Y.
Chee,N.
Pletcher,M.
Sheets,andH.
Qin.
Ultra-low-powerdesign.
CircuitsandDevicesMagazine,IEEE,22(4):23–29,2006.
[41]J.
M.
Rabaey,M.
J.
Ammer,J.
L.
daSilva,D.
Patel,andS.
Roundy.
Picoradiosupportsadhocultra-lowpowerwirelessnetworking.
Computer,33(7):42–48,2000.
[42]A.
P.
Sample,D.
J.
Yeager,P.
S.
Powledge,A.
V.
Mamishev,andJ.
R.
Smith.
Designofanrd-basedbattery-freeprogrammablesensingplatform.
InstrumentationandMeasurement,IEEETransac-tionson,57(11):2608–2615,2008.
[43]C.
YingandZ.
Fu-Hong.
Asystemdesignforuhfrdreader.
InCommunicationTechnology,2008.
ICCT2008.
11thIEEEInternationalConferenceon,pages301–304.
IEEE,2008.
[44]F.
Zhang,Y.
Zhang,J.
Silver,Y.
Shakhsheer,M.
Nagaraju,A.
Klinefelter,J.
Pandey,J.
Boley,E.
Carlson,A.
Shrivastava,etal.
Abatteryless19wmics/ism-bandenergyharvestingbodyareasensornodesoc.
InSolid-StateCircuitsConferenceDigestofTechnicalPapers(ISSCC),2012IEEEIn-ternational,pages298–300.
IEEE,2012.
14