You are facing the Mona Lisa_ Spot localization using PHY layer information(您正在面对使用PHY层信息的Mona Lisa_ Spot本地化).pdf
- 文档编号:30838791
- 上传时间:2024-01-30
- 格式:PDF
- 页数:14
- 大小:2.21MB
You are facing the Mona Lisa_ Spot localization using PHY layer information(您正在面对使用PHY层信息的Mona Lisa_ Spot本地化).pdf
《You are facing the Mona Lisa_ Spot localization using PHY layer information(您正在面对使用PHY层信息的Mona Lisa_ Spot本地化).pdf》由会员分享,可在线阅读,更多相关《You are facing the Mona Lisa_ Spot localization using PHY layer information(您正在面对使用PHY层信息的Mona Lisa_ Spot本地化).pdf(14页珍藏版)》请在冰豆网上搜索。
YouarefacingtheMonaLisaSpotLocalizationusingPHYLayerInformationSouvikSenDukeUniversitysouvik.senduke.eduBoidarRadunovicMicrosoftRRomitRoyChoudhuryDukeUniversityromit.rcduke.eduTomMinkaMicrosoftRABSTRACTThispaperexplorestheviabilityofpreciseindoorlocalizationusingphysicallayerinformationinWiFisystems.WefindevidencethatchannelresponsesfrommultipleOFDMsubcar-rierscanbeapromisinglocationsignature.Whilethesesig-naturescertainlyvaryovertimeandenvironmentalmobility,wenoticethattheircorestructurepreservescertainproper-tiesthatareamenabletolocalization.WeattempttoharnesstheseopportunitiesthroughafunctionalsystemcalledPinLoc,implementedonoff-the-shelfIntel5300cards.Weevaluatethesysteminabusyengineeringbuilding,acrowdedstudentcenter,acafeteria,andattheDukeUniversitymuseum,anddemonstratelocalizationaccuraciesinthegranularityof1mx1mboxes,called“spots”.Resultsfrom100spotsshowthatPinLocisabletolocalizeuserstothecorrectspotwith89%meanaccuracy,whileincurringlessthan6%falsepositives.Webelievethisisanimportantstepforward,comparedtothebestindoorlocalizationschemesoftoday,suchasHorus.CategoriesandSubjectDescriptorsC.2.1NetworkArchitectureandDesign:
Wirelesscommu-nicationGeneralTermsDesign,Experimentation,PerformanceKeywordsWireless,Localization,Cross-Layer,Application1.INTRODUCTIONPreciseindoorlocalizationhasbeenalongstandingproblem.WhilethefrontieroflocalizationtechnologyhasadvancedPermissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforprofitorcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthefirstpage.Tocopyotherwise,torepublish,topostonserversortoredistributetolists,requirespriorspecificpermissionand/orafee.MobiSys12,June2529,2012,LowWoodBay,LakeDistrict,UK.Copyright2012ACM978-1-4503-1301-8/12/06.$10.00.overtime,newkindsoflocationbasedapplicationsarerais-ingthebar.Forinstance,theadvertisingindustryisbeginningtoexpectlocationaccuraciesatthegranularityofanaisleinagroceryshop1.Museumsareexpectinguserlocationsatthegranularityofpaintings2,sotouristscanautomaticallyreceiveinformationaboutthepaintingstheystopat.Inad-ditiontosuchhighaccuracydemands,theseapplicationsareinherentlyintoleranttosmallerrors.Ifalocalizationschemeincorrectlyplacesauserintheadjacentaisleinthegrocerystore,ordisplaysinformationabouttheadjacentpainting,thepurposeoflocalizationisentirelydefeated.ThisisunliketraditionalapplicationssayGPSbaseddrivingdirectionswheresmallerrorsaretolerable.Asaconsequence,newlo-calizationschemeswillneedtomeetstrictqualitystandards,withoutincurringaheavycostofinstallationandmainte-nance.Werefertothisproblemasspotlocalization,whereadeviceinaspecific1mx1mboxneedstobeaccuratelyidentified.Localizingthedeviceoutsidetheboxwillbeuse-less,irrespectiveofwhethertheestimatedlocationiscloseorfarawayfromthebox.Thestateoftheartinindoorlocalizationisquitesophis-ticated.Differentschemesoptimizedistinctobjectives,in-cludingaccuracy35,computation4,6,easeofcalibra-tion7,8,energy9,etc.Whiletheliteratureisrich,wesamplefewoftherepresentativeschemestooutlinethefron-tieroftodayslocationtechnology.Cricket10achieveshighaccuracyusingspecial(ultrasound-based)infrastructurein-stalledonceilings.Notingthedifficultiesofinstallingspecialhardware,RADAR,PlaceLabsandHorus4,6,8exploredthefeasibilityofusingsignalstrengthsfromexistingWiFiAPs.WhileRADARandHorusbothrelyonsignalcalibration,EZ7recentlydemonstratedtheabilitytoeliminatecalibra-tionattheexpenseofaccuracydegradation.Summarizingalltheseschemes,wefindthatthestateoftheartachievesmedianlocationerrorof4mand7m,withandwithoutcal-ibration,respectively7.Whilethisaccuracycanenableavarietyofapplications,thereareothersthatneedprecisionatthegranularityof“1mx1m”.ThispapertargetssuchhighaccuracieswhileensuringthatthecalibrationcomplexityisnoworsethanRADARorHorus.WecallourproposalPinLoc,asanacronymforPreciseindoorLocalization.PinLocsmainideaissimple.WhilemostWiFibasedlocaliza-tionschemesoperatewithsignalstrengthbasedinformationattheMAClayer,werecognizethepossibilityofleveragingdetailedphysical(PHY)layerinformation.Briefly,theintu-itionisthatthemultipathsignalcomponentsarriveatagivenlocationwithdistinctvaluesofphaseandmagnitude.WhenaggregatedovermultipleOFDMsub-carriersin802.11a/g/n,theserichdataposesasafingerprintofthatlocation.Sincewedefineeachspotasaclusteroflocations,war-drivingeachspotproducesanarrayoflocationfingerprints.Atrainingalgorithmrunsoneacharrayoffingerprintstolearnthesta-tisticalattributesofthatspot.Later,whenamobiledevicearrivesataspot,itcomputesafingerprint(fromasequenceofoverheardbeacons),andclassifiesittooneofthespotsbymatchingagainstthelearntattributes.Wefindthatde-vicesarereliablyclassifiedtothecorrectspot,despitemove-mentsofpeopleandotherobjectsintheenvironment.Ourwar-drivingeffortiscomparabletoRADARorHoruswemountedalaptoponaRoombarobotandprogrammedittomoverandomlywithineachspotforaround4minutes.Finally,whereseveralotherschemesarestronglyreliantonmultipleAPs,PinLocoffersreasonableperformanceeveninWiFi-sparseenvironments.Insomecases,PinLocisabletolocalizeevenwithsignalsfromasingleAP.Atfirstglance,ourfindingsseemedtoogoodtobetrue.Weexpectedthesignalphasestobesensitivetotheorientationofthelaptop,humanmovements,and/orstructuralchangesintheenvironment(suchasrepositioningofchairs,boxes,shelves).Wesuspectedthatfrequentwar-drivingwouldbenecessarytoadapttosuchenvironmentalperturbations.Whiletheseconcernswerenatural,weweresurprisedtofindthatthefingerprintsactuallypreservedstatisticalpropertiesevenunderperturbations.Forinstance,althoughthechannelre-sponseataspecificlocationvariedwithtimeandenviron-mentaldynamism,theycouldbeconsistentlyorganizedaroundasetoffewtightclusters.Whencombinedacross30subcarri-ersanddifferentAPs(i.e.,high-dimensionaldata),wefoundthateventhesetsofclusterscouldbereasonablyunique.Fur-ther,sincespotsarecomposedofmany“distinctlocations”,thefingerprintofaspotisastringofchannelresponsesfrommultipledistinctlocationsinsidethatspot.Thus,evenifthechannelresponsefromonedistinctlocationisnotunique,theprobabilitythatthestringofchannelresponsesappearsinmorethanonespotisfarlower.Theseandotherfactors(discussedlater)togethercontributetoPinLocsrobustness.RSSI,ontheotherhand,isanaverageofthemagnitudesoneachsub-carrier,whichhidesfine-grainedinformationaboutthatlocation,ultimatelylimitingtheaccuracyoflocalization.Harnessingtheaboveopportunitiesintoaworkingsystem(usingoff-the-shelfwirelesscards)formsthecoreofPinLoc.ThedetailedPHYlayerinformationisfirstextractedfromthedriverandsanitizedusingaphasecorrectionoperation.Thesanitizedparametersarethenfedtoamachinelearningalgo-rithmthatmodelsthechannelresponsedistribution.Later,duringsystemtests,thechannelparametersareextractedfromreceivedWiFibeacons,andclassifiedtooneofthewar-drivenspots.Toaddressenergyissues,PinLocdisablesactivescanning,andonlyusesbeaconsfromAPsinthesamechan-nel.Finally,theindividualmodulesarecombinedintoafullsystem,andtestedoveravarietyofscenarios.Theresultsarepromisingwithlessthan4minutesofwardrivingper-spot,weobserve89%meanaccuracyandfalsepositivescon-sistentlybelow6%.Fromtheapplicationsperspective,Pin-LocwastestedinthemodernartwingofDukeUniversitysmuseum.Spotsinfrontofeachof10paintingswerelocalizedwithhighaccuracy.Tothebestofourknowledge,nopriorworkhasdemonstratedPHYlayer-basedWiFilocalizationonoff-the-shelfplatforms.Zhanget.al.11usedsignalamplitudesandphasesonUSRPplatformstodemonstratelocationdistinction.Wenotethatlo-cationdistinctiondetectswhenanodeslocationhaschanged(e.g.,forsecuritypurposes),butdoesnotneedtoestablishuniquenessforeachlocation.Localizationisnaturallyafarstricterproblem,especiallywhenthetargetissub-meterac-curacies.PinLocmakesanearlyefforttowardsthisgoalthemaincontributionsmaybesummarizedasfollows.Wetargettheproblemofspotlocalizationwheresuc-cessisdefinedastheabilitytoplaceadevicewithina1mx1marea,calledspots.WebreakawayfromRSSIbasedschemesandexplorethefeasibilityofusingdetailedPHYlayerinformation.Weutilizetheper-subcarrierfrequencyresponseasfea-turesofalocation,andrelyonmachinelearningalgo-rithmstoclassifyadevicetooneofthetrainedspots.Weuseoff-the-shelfIntel5300cards;theentiresystemre-liesonexistingWiFideployments,andrequiresnospecialinstallation.WeevaluatePinLocatvaryingaccuracystandards,namely,discriminatingbetweenseatsinalab,chairsinacafe-teria,andadjacentpaintingsinamuseum.Weobserveconsistentaccuraciesundermobile/dynamicenvironments,outperformingHorus4,themostaccurateRSSIbasedlo-calization.Thesubsequentsectionsexpandoneachofthesecontribu-tions,beginningwithdefinitionandapplications,followedbymeasurement,design,andevaluation.2.LOCATIONS,SPOTS,ANDAPPLICATIONSTheabovesectionlooselyusedtheterms“locations”and“spots”weclearlydefinethemhere.Locationsarelikepixelsthatdefinetheresolutionofourlocalizationsystem.Each
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- You are facing the Mona Lisa_ Spot localization using PHY layer information您正在面对使用PHY层信息的
链接地址:https://www.bdocx.com/doc/30838791.html