TIG 技術(shù)報告分析_第1頁
TIG 技術(shù)報告分析_第2頁
TIG 技術(shù)報告分析_第3頁
TIG 技術(shù)報告分析_第4頁
TIG 技術(shù)報告分析_第5頁
已閱讀5頁,還剩1頁未讀, 繼續(xù)免費(fèi)閱讀

下載本文檔

版權(quán)說明:本文檔由用戶提供并上傳,收益歸屬內(nèi)容提供方,若內(nèi)容存在侵權(quán),請進(jìn)行舉報或認(rèn)領(lǐng)

文檔簡介

July2022

TITLE

doc.:IEEE802.11-22/0987r

21

Submission page

PAGE

3

XiaofeiWang(InterDigitalInc.)

IEEEP802.11

WirelessLANs

IEEE802.11AIMLTIGTechnicalReportDraft

Date:2022-07-06

Author(s):

Name

Affiliation

Address

Phone

email

XiaofeiWang

InterDigitalInc.

111West33rdStreet

NewYork,NY10120

USA

+1-607-592-2727

Xiaofei.wang@

MingGan

Huawei

Ming.gan@

ZinanLin

InterDigital

RuiYang

InterDigital

AiguoYan

Zeku

JunghoonSuh

Huawei

ZiyangGuo

Huawei

MarcoHernadez

NICT

LiangxiaoXin

Zeku

Abstract

ThisdocumentcontainsthetechnicalreportoftheIEEE802.11AIMLTIG.

R0:initialoutline

R1:insertionofUsecase1

R2:insertionofIntroduction

TableofContents

Introduction

Terminologies

AIML ArtificialIntelligence/MachineLearning

CSI ChannelStateInformation

UHR UltraHighReliability

Backgroundinformation

ArtificialIntelligence/MachineLearning(AI/ML)algorithmshavemadesignificantprogressandarebeingappliedinmanydomains,includingmedicaldiagnosis,speechrecognition,computervision,andintegrationofvisionandcontrolforrobotics.Inaddition,AI/MLalgorithmsareemergingasimportantcomponentsinmanyapplicationssuchasautonomousdriving,languagetranslationandhuman-machineinteractions.

TraditionalAI/MLtechniquesarebasedonacentralizedmodelwhichrequiresexchangingalargeamountofdatabetweendatasourcesandacentralizedserver.Morerecently,distributedAI/MLalgorithmssuchasfederatedlearninghavebeendevelopedthatwillallowmoreanalysisatthesourceandreducetheamountofdatathatneedtobeexchanged,thoughtheexpectedamountofexchangeddataremainssignificant.Withtheprevalenceofwirelessnetworksandcommunications,muchoftheexchangeddataisexpectedtobecarriedthroughwirelessnetworks,suchasIEEE802.11WLANnetworks.

StudieshaveshownthatAI/MLalgorithmscanhelpimprovetheperformanceforwirelesscommunicationnetworks,byprovidingbetterresourceusage,lowerenergyconsumption,higherreliabilityandmorerobustnesstoachangingenvironment.Asthesealgorithmsbecomemorematureandcosteffective,WLANmayleverageAI/MLforenhancednetworkperformanceanduserexperience.

InMay2022,theIEEE802.11WorkingGroup(WG)hasapprovedtheformingoftheAIMLTaskInterestGroup(TIG)bythefollowingmotion[1]:

Motion5:TIGRe:AI/MLusein802.11

ApproveformationofaTopicInterestGroup(TIG)to:

(a)describeusecasesforArtificialIntelligence/MachineLearning(AI/ML)applicabilityin802.11systemsand

(b)investigatethetechnicalfeasibilityoffeaturesenablingsupportofAI/ML.

TheTIGistocompleteareportonthistopicatorbeforetheMarch2023session.

ThistechnicalreportisthefinalreportoftheAIMLTIGtotheIEEE802.11WGdetailingvariousAIMLusecasesdiscussedduringtheAIMLTIG.Foreachusecase,anumberofKeyPerformanceIndicators(KPIs)havebeenidentifiedandrequirementsandtechnicalfeasibilityanalysishavebeenprovided.

AIMLUsecasesforIEEE802.11

Note:usecasespotentiallycanbeorganizedintodifferentcategories

Note:usecasespotentiallycanidentifyKPIs

Usecase1:CSIfeedbackcompression

In802.11ax[1]andthedraftof802.11be[2],theAPinitiatesthesoundingsequencebytransmittingtheNDPAframefollowedbyaNDPwhichisusedforthegenerationofVmatrixatthebeamformee.UponthereceiptoftheNDPfromthebeamformer,thebeamforeeappliesacompressionscheme(i.e.,Givensrotations)ontheVmatrixandfeedsbacktheangelesinthebeamformingreportframe.

Itisindicatedin

REF_Ref118889474\r\h

[4][3]

thathighernumberofspatialstreamshasbeenaninevitabletrendinWiFiformorethanadecade.Theprelimilaryresults

REF_Ref118889474\r\h

[4][3]

REF_Ref118889476\r\h

[5][4]

REF_Ref118889495\r\h

[6][5]

showthatMIMOwithalargenumbertransmitterantennasandalargenumberofspatialstreams(e.g.,16spatialstreams)offerremarkablesystemperformancegainsonbothSU-MIMOandMU-MIMOcases.MultiAP(MAP)maybeonepotentialfeatureinthenext802.11generation,e.g.UHR

REF_Ref118797206\r\h

[7][6]

-

REF_Ref118796138\r\h

[10][9]

.LargenumberofspatialstreamscombinedwithMAPfeaturemayfurtherincreasethesoundingfeedbackairtimeoverheadifcoordinationbetweenAPs(e.g.,jointtransmission/reception,coordinatedbeamforming)isapplied.Largeamountofoverheadorprolongedsoundingproceduresmaynegativelyimpactthelatencyandlimitthesystemperformance.Therefore,thereisaneedtoreducetheCSIoverheadespeciallywhenthenumberoftransmitterantennasgoeshigherormultipleAPsperformjointorcoordinatedtransmission.

Somestudies(e.g.,

REF_Ref118797710\r\h

[11][10]

REF_Ref118797712\r\h

[12][11]

REF_Ref118983623\r\h

[13][12]

REF_Ref118988666\r\h

[14][13]

)haveshownthatAI/MLcanefficientlyreducetheCSIfeedbackandimprovethesystemthroughput.Forexample,motivatedbythenaturethattheCSImayfallintodifferentclustersduetothechannelsimilarityofnearbySTAs,iFORalgorithm

REF_Ref118797710\r\h

[11][10]

appliestheunsupervisedlearning,K-mean,totheCSIcompressiontoclassifytheanglevectorswhicharederivedfromVmatrix.Simulationresultsshowthatfora8x2SU-MIMO,iFORusesaround8%ofthenumberofbitsrequiredbytheexistingfeedbackmechanism(802.11ax)andboostthesystemthroughputbyupto52%.In

REF_Ref118797712\r\h

[12][11]

,anotherunsupervisedlearning,DeepNeuralNetworkAutoencoder(DNN-AE)isappliedtoCSIanglevectorsandfurthercompressesthederivedangles(LB-SciFi)byleveraingthecompressioncapabilityofDNNs.ExperimentalresultsshowthatLB-SciFireducesthefeedbackoverheadby73%andincreasesthenetworkthroughputby69%onaverage.

ThisusecaseproposestoapplyAI/MLtechniquetoCSIfeedbackschemestoreducetheCSIoverheadwithminimumlossofPERperformance.

KPIsconsideredinthisusecaseareproposedasfollows:

Numberoffeedbackbitspersubcarriergroup

AchievedPER

BothSU-MIMOandMU-MIMOcasesneedtobeconsidered

AdditionalAIMLoverheadcompredwithcompressionsaving

OneexampleistheratiobetweenthenumberofadditionalbitsrequiredbyAIMLprocess(includingdatausedformodeltraining/inference

REF_Ref119303357\r\h

[15][14]

themodelparameters,theadditionalsignaling)andthenumberofbitssavedbytheCSIfeedbackscheme.Inthisexample,ifthedatausedformodeltrainingthatisperformedbytheAPfullyreliesonthelegacyCSIreport,thentheadditionalAIMLusedformodeltraining/inferencemaybe0.

Computationcomplexity/Latency:

AdditionaldelayorcomputationisintroducedbyAIMLprocessing

Eveluationmethodologyneedstobeestablished.

Usecase2

UsecaseN

RequirementsandPotentialfeaturesanalysis(highlevel)

Requirements

RequirementsUsecase1:CSIfeedbackcompression

Performanceshouldfollowtheguidiancebelow:

CSIairtimereduction:achievearitimereductionofCSIfeedbackover802.11beforagivenNrxNcMIMO,whereNristhenumberofrowsinthecompressedbeamformingfeeedbackmatrix,Ncisthenumberofcolumnsinthecompressedbeamformingfeedbackmatrix.

AdditionaloverheadusedforAIMLprocess:minimizetheadditionaloverheadusedforAIMLprocess.AdditionaloverheadmayincludethedatausedforAIMLmodeltraining/inference[14],themodelparametersandadditionalsignalling.ThedatausedforAIMLmodeltraining/inference[14]canreusethelegecyCSIreportdata.

PacketErrorrate(PER):guaranteeminimumSNRlosscomparedwith802.11betoachievethetargetPER(e.g.,1%and/or10%)atagivenMCSinalltypesofchannels

REF_Ref119303329\r\h

[16][15]

.

Computationcomplexity/Latency:minimizetheadditionalcomputationcomplexityorlatencyrequiredbytheAIMLprocess

Potentialfeaturesanalysis

Technicalfeasibilityanalysis

Standardsimpact

UsecaseofCSIfeedbackcompression

Thestandardimpactmayinclude:

Additionalsignaling(e.g.,betweenAPandnon-APSTAs)requiredbyAIMLprocessPlaceholderforadditionaltechnicalfeasibilityanalysis

Technicalfeasibility

UsecaseofCSIfeedbackcompression

Thefollowingmetricswillbestudied:

Dataavailabilityandaccesibility:TherearesomeSTAsthatareabletousethedatatoperformAIMLmodeltrainingand/orinference

REF_Ref119086275\r\h

[15][14]

.Thedatausedformodeltrainingand/orinferenceshallbeaccessiblefortheseSTAs.

AP/edgecomputingbasedAIML:Datamaybecollectedfromnon-APSTAs.Thelegeacy802.11CSIreportsmaybeusedastrainingdata.

DevicecomputingbasedAIML:DatashouldbeavailableatallSTAsthatsupportAIMLprocess.

Hardware/softwarecapability:TheSTAsthatuseAIMLtogeneratetheAIMLenabledCSIfeedbackcompressionshallhavethehardwareandsoftwarecapabilitytosupportAIMLalgorithm(s).

AP/edgecomputingbasedAIML

REF_Ref119085527\r\h

[17][16]

:Extradataandmodel(e.g.,modelparameters)exchangemayberequiredtosupportAP/edgecomputingbasedAIML.However,computationisnotexpectedtobelocatedatAPoredgecomputingresourcesforwhichhighercomputationcapabilitiesisexpected.

DevicecomputingbasedAIML

REF_Ref119085527\r\h

[17][16]

:STAsthatsupportAIMLmayberequiredtohaveextracomputationcapability.Extradataandmodel(e.g.,modelparameters)exchangebetweenSTAsmayalsoberequiredtosupportdevicecomputingbasedAIML.

Summary

References

11-22/597r3:May2022WorkingGroupMotions,May18,2022

IEEE802.11-REVmeD2.0,October2022

IEEEP802.11beD2.2,October2022

802.11-18/0818r3,16SpatialStreamSupportinNextGenerationWLAN

802.11-20/1877r1,16SpatialStreamSupport

802.11-20/1535r66,CompendiumofstrawpollsandpotentialchangestotheSpecificationFrameworkDocumentPart2

802.11-22/1515,Acandidatefeature:M

溫馨提示

  • 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
  • 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯(lián)系上傳者。文件的所有權(quán)益歸上傳用戶所有。
  • 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁內(nèi)容里面會有圖紙預(yù)覽,若沒有圖紙預(yù)覽就沒有圖紙。
  • 4. 未經(jīng)權(quán)益所有人同意不得將文件中的內(nèi)容挪作商業(yè)或盈利用途。
  • 5. 人人文庫網(wǎng)僅提供信息存儲空間,僅對用戶上傳內(nèi)容的表現(xiàn)方式做保護(hù)處理,對用戶上傳分享的文檔內(nèi)容本身不做任何修改或編輯,并不能對任何下載內(nèi)容負(fù)責(zé)。
  • 6. 下載文件中如有侵權(quán)或不適當(dāng)內(nèi)容,請與我們聯(lián)系,我們立即糾正。
  • 7. 本站不保證下載資源的準(zhǔn)確性、安全性和完整性, 同時也不承擔(dān)用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。

評論

0/150

提交評論