1/64

Contents

1.Preface 2

2.ProgressofRelevantStandardOrganizations 3

2.1.3GPP 4

2.2.IMT-2030PromotionGroup 5

2.3.NextGAlliance 5

2.4.Hexa-X 6

2.5.6GANA 7

3.6GDataServiceScenariosandRequirementAnalysis 7

3.1.SensingData 7

3.2.AIModelTrainingData 10

3.3.AIModelData 11

3.4.DataServicesBasedonDistributedArchitecture 14

3.5.User-CenteredDataManagementandControl 15

3.6.DatainIntegratedSatellite-TerrestrialCommunicationScenarios 17

3.7.SOEDataAcquisition 18

3.8.Self-GeneratedDataServices 20

3.9.DataBenefits 21

3.10.Summaryof6GDataRequirements 23

4.DefinitionandFrameworkofthe6GDataPlane 24

4.1.Definition 24

4.2.NecessityoftheDataPlane 26

4.3.6GDataPlaneFramework 27

5.KeyTechnologiesof6GDataPlane 34

5.1.DataBearerandTransmissionProtocol 34

5.2.AirInterfaceDataPlane 36

5.3.CoreNetworkDataPlaneFunctionsandArchitecture 39

5.4.CoreNetworkDataPlaneTransmissionProtocol 40

5.5.AIModelDataCompressionTechnology 42

5.6.DistributedDataTechnology 44

5.7.ServiceInterfaceTechnologyBasedonSemanticGraph 46

5.8.CoordinationbetweenDataServicesandOtherServices 52

6.DataPlanePrototype 54

6.1.DataPlanePrototype1 54

6.2.DataPlanePrototype2 55

7.SummaryandOutlook 57

8.References 59

9.Abbreviations 61

10.ContributorstotheWhitePaper 64

2/64

1.Introduction

5Gsupportsthreescenarios:enhancedMobileBroadband(eMBB),Ultra-ReliableLow-LatencyCommunications(URLLC),andmassiveMachineTypeCommunication(mMTC).Exposureofcapabilitiesandeventsissupportedthroughthenetworkexposurefunction(NEF)orcommonAPIframework(CAPIF).Applicationfunctions(AFs)canobtainthedataofthe5GsystemviaNEF/CAPIF,whichpromotescross-layerinnovationbetweenthenetworkandAF.Asthevolumeofdatageneratedbymobilenetworkscontinuestogrow,theneedfordatacollectionanddataprocessinghasbecomeincreasinglyprominent.Correspondinglynetworkserviceshaveevolvedfromcommunicationservicetomultidimensionalservicesofcommunication,sensing,andartificialintelligence(AI).Traditionalmethodsofdatacollectionanddataprocessingarenotadaptedtonewchanges,anditisdifficulttomeettheadditionaldatarequirements.Thisisbecause5Gnetworksdonothaveaunifieddatamanagementframework.Asaresult,multipletypesofdatacannotbeintegratedorcoordinatelymanaged,whichmayincreasethecomplexityandcostsofdatagovernance.Focusingondatapipelines,5Gnetworksdonotfullyexploreorutilizethevalueofdata.Basedontheexistingmethods,datacollectionanddataprocessingrequirementsofvarioustypesofdatacannotbemet.Thedatavalueofcommunication,sensingandAIcannotbeexplored.Moreover,rightsandinterestsofdatacannotbeguaranteed.Thereisnotaunifieddatamanagementandcontrolframeworkfordataqualitymanagementtocannotguaranteethelegality,authenticity,andintegrityofdata.Therefore,5Gnetworkscannotmeettherequirementsofdataregulatorytheexpectationofprivacyandsecurityofusers.Intheaspectsofnewservices,itisdifficultfor5Gnetworkstoefficientlycollect,transmit,processandanalyze

largeamountsofdatainmobilenetworks,suchassensingdata,AImodel,etc.

Figure1-1LifeCycleManagementofDatainMobileNetwork

3/64

Datahasbecomeoneoftheproductionfactorsinthedigitalsociety.6Gisanimportantinfrastructureofdigitalsociety.Thedataofthe6Gsystemisinevitablyanimportantpartofthisproductionfactor.The"FrameworkandoverallobjectivesofthefuturedevelopmentofIMTfor2030andbeyond"[1],releasedbyInternationalTelecommunicationUnion-RadioCommunicationSector(ITU-R),proposessixmajorscenarios,includingImmersiveCommunication,MassiveCommunication,HyperReliableandLow-LatencyCommunication,UbiquitousConnectivity,AIandCommunication(AIAC),andIntegratedSensingandCommunication(ISAC).Thismeansthat6Gisamobilecommunicationsystemthatgoesbeyondcommunicationservices.Inadditiontotraditionaldatatransmissionpipelines,6Gintroducesinternaldataofmobilenetwork,suchassensingdatainISACandAIdatainAIAC,etc.Dataprovidersordataconsumersofmobilenetworkinternaldataincludeuserequipment(UE),radioaccessnetwork(RAN)nodes,andnetworkfunctions(NFs)ofcorenetwork(CN)andAFs.Incontrasttouserdataintraditionaldatatransmissionpipelines,the6Gsystemistaskedwithmanagingtheentirelifecycleofinternaldata,includinggeneration,securityandprivacy,coordinationofcollection,transmission,processing,

qualitymanagementanddataservice.

Therefore,adataplane[2]isintroducedtoenableaunifiedandefficientlifecyclemanagementofinternaldataofthe6Gnetwork.Forexample,thedataplaneprovidesdatarequiredbythesensingfunctionorNEF,sothatthe6Gsystemcanprovidethesensingserviceornetworkexposureservices.The6GDataPlane(DP),whichoperatesparalleltothecontrolplane(CP)anduserplane(UP),isnotconstrainedbythetransmissionrequirementsofsignalingoruserdata.ThisallowsthefunctionsandconfigurationsofDPprotocolcanbeoptimizedtoprovideabettersolutiontotheaforementionedneeds.Thisavoidsfragmentedsolutionsforindividualusecases.TheDPwillprovideaunifieddatamanagementframeworkforthe6Gsystem,enablingdatacapabilities.Asaresult,the6Gsystemwillbeabletointegrateandmanagemultipletypesofdata,ensuredatasecurityandtrustworthiness,breakdatasilos,improvetheefficiencyofdatagovernance,protectrightsandinterestsofdata.InthisWhitePaper,contributorsillustrateourinitialviewsandlatestachievementsonthescenariosandrequirementsof6Gdataservice,thedefinitionandframeworkofDP,keytechnologies,andprototypes,hopefullycontributingto6G

development.

2.ProgressofRelevantStandardOrganizations

Thischapterdescribesthe6GDPresearchprogressofmultiplestandardorganizationsandresearchinstitutionsaroundtheworld,including3rdGenerationPartnershipProject(3GPP),IMT-2030(6G)PromotionGroup,NextGAlliance,Hexa-X,and6GAllianceofNetworkAI(6GANA).Thesestandardorganizationshavedoneresearchesonthedatacollectionrequirements,

challenges,solutions,andtechnologytrendsof6G.

4/64

2.1.3GPP

Basedondatarequirementsofeachusecase,severalstandardsof5Gnetworkshavebeenstudiedandstandardizedtosupportdatacollectionanddataanalysis.Forexample,[3]isusedtocollectdataforthenetworkdataanalyticsfunction(NWDAF),whichistheAIfunctionofCN.Inaddition,therearestandardsofNEF/CAPI[4][5]fordatacollectionofnetworkexposure.ThereareLTEPositioningProtocol(LPP)[6]andNRPositioningProtocolA(NRPPa),whichareusedtotransmitpositioningdataforthelocationmanagementfunction(LMF).Thereareminimizationofdrivetest(MDT)[7]andqualityofexperience(QoE)usedforwirelessnetworkoptimizationand

management.

TomeetdatacollectionrequirementsoftheNWDAF,Release17introducestheDataCollectionCoordinationFunction(DCCF)tothe5GnetworktocollectdatafromNFsanddistributeresultsrequestedbyNFs[2].TheDCCFpreventsdataproviders,suchastheAccessandMobilityManagementFunction(AMF)andSessionManagementFunction(SMF)fromhandlingmultiplesubscriptionstothesamedataandsendingmultiplenotificationscontainingthesameinformationduetonocoordinationbetweendataconsumers.ExcepttheNWDAF,5GCNNFs,suchastheAMFandSMF,asmainnetworkelements(NEs)ofthecommunicationnetwork,mainlyprovidecommunicationservicesinsteadofdata.However,theNWDAFgenerallyneedstoobtainalargeamountofdataforbigdataanalysis.Repeatedreportingofalargeamountofidenticaldatadecreasestheperformanceof5GCNNFs.TheNWDAFcansubscribeorunsubscribedatafromtheDCCFthroughtheNdccfinterface.IftheDCCFhasnotpreviouslycollectedthedatarequestedbytheNWDAF,itcanusetheservice-basedinterface(SBI)tocollectthedatafromNFsorcollectdatathroughthemessagingframework.Subsequentlythedatacanbe

transmittedtotheNWDAFthrougheitherSBIormessagingframework.

FortheoptimizationandmanagementofRAN,thenetworkmanagementfunctioncansendanMDTorQoErequesttotheRANnode,whichtriggerstheRANnodetoconfigureMDTorQoEdatacollectiontotheUEs.TheMDTorQoEdatareportedbytheUEissenttothetracecollectionentity(TCE)ormeasurementcollectionentity(MCE)foranalysisthroughtheservicingRANnode.ThenetworkmanagementfunctionoptimizesconfigurationsofRANbasedonthe

self-organizingnetwork(SON).

TheLPPisusedtoexchangepositioningcontrolinformationandpositioningdatabetweentheUEandthenetwork.Generally,theamountofpositioningdataisnotlarge.Therefore,theLPPiscarriedontheCPprotocolstack.Inotherwords,thepositioningprotocolstackconsistsoftheLPP,non-accessstratum(NAS),RadioResourceControl(RRC),PacketDataConvergenceProtocol(PDCP),RadioLinkControl(RLC),MediumAccessControl(MAC),andPhysicalLayer(PHY).BasedonpositioningmeasurementdatatransmittedbytheLPP,theLMFestimates

thelocationinformationofUEsandprovidesitto5GNFsorAFs.

5/64

2.2.IMT-2030PromotionGroup

In6GNetworkArchitectureVision[8],IMT-2030PromotionGroupproposedthatthe6Gnetworkarchitecturerequiresakindofdatafunctionswhicharedifferentfromthetraditionaluserplane.Thedatafunctionssystematicallyaddressthechallengesofmanagingandmonetizingnon-traditionaluserplanedatainthe6Gnetwork.FromtheperspectiveofNFlayer,6Gdatafunctionsconsistofdataorchestrationandcontrol,dataprocessingandforwarding,anddatastorage.Basedonthecapabilityreportedbydataprocessingandforwardingnodesandrequirementsofdataservice,thedataorchestrationandcontrolnodeselectsthedatasourcenodes,processingandforwardingnodes.Basedonorchestration,theselectednodesformadatabearertoprovidedataservices.Thedataprocessingandforwardingnodeprovidesdataservicesasrequired,suchasdatacollection,datapreprocessing,datastorage,dataprivacyprotection,securityandtrustworthiness,dataanalysis,datasharing,anddataforwarding.Thestandardizationofdatalifecyclemanagementcanimprovedatacomparabilityandreusability,whichincludesprocessessuchasdataprivacy,securityandtrustworthiness,datacoordination,generation,andcollection,datastorage,datatransmission,dataprocessing,dataservice,anddataqualitymanagement,rightsand

interestsmanagement,etc.

In6GWirelessSystemDesignPrinciplesandTypicalFeatures[9],IMT-2030PromotionGroupproposedthatefficientdatagovernance,aunifieddatacollectionmethod,anddatalifecyclemanagementshouldbeconsideredatthebeginningof6Gsystemdesign.Nativedataisoneoftheprinciplesfor6Gwirelesssystemdesign.Thedesignofnativedatashouldensuredatasecurityandprivacy,improveefficiencyofdatacollection,transmission,andstorage.Thedesignofnativedatashouldalsoenhancedatasharingandreusability.Nativedatabuildsanopenandunifieddatalifecyclestandardtosupportallprocessesofubiquitousheterogeneousdatacirculation.Withcost-effective,efficient,andreliabledataservices,itenables6Gnetworksandassociatedindustriestoprocessandanalyzevarioustypesofdataaccuratelyandsystematically,so

astomakebetterdecisionsandprovidehigherqualityservices.

In6GDataServiceArchitectureResearch[10],IMT-2030PromotionGroupproposed6Gdataservicesandthedataplanearchitectureofthe6Gnetwork,anddescribedthefunctionoforchestrationandcontrol,processingfunctionsandsoon.In5G,therearematureprotocolstacksofCPandUPtosupportcommercialdeploymentandapplications.Theexistingprotocolstacksdecouplefunctionsbetweenplanes,andachievemodular,virtualized,andsoftware-basedfunctionmanagement.The6Gdataplanealsorequirescomplete,flexible,andscalableprotocolstackstosupportdataforwardingandcontrol,dataserviceexposure,dataagentmanagementandcontrol,

etc.

2.3.NextGAlliance

6/64

Inthe6GTechnologiesforWide-AreaCloudEvolution[11],theNextGAllianceproposedthatinadditiontocommunicationservices,computinganddataplaneswithdedicatedcomputinganddatamanagementfunctionsmaybeintroducedtocellularnetworks.The6GTechnologiesforWide-AreaCloudEvolution[11]drivesnetworkarchitecturedesignandintroducesnewfunctionsintermsof6Gnetworkcomputingservicestosupportdistributedcomputingandtightinterworkingbetweenthedistributedcloudand3GPPprotocols.Therefore,changesinthecontrol,management,anddataplanesneedtobestudiedtoadapttothedistributedcomputingprocess.Accordingtothereportcontent,thedataplaneissimilartotheuserplaneofexistingprotocols.It

ispossibletomeetcomputingservicerequirementsbyenhancingthedatabearer.

IntheAI-nativeWirelessNetworks[12],theNextGAllianceproposedthatcomparedwiththeapplicationsofAI/machinelearning(ML)in5G,AI/MLin6Gshouldhavethefollowingfeatures:Datacollectionwillbeatvariouslayerswithinthenetwork.Inadditiontonear-real-timemannerofAI/MLapplicationsof5G,AI/MLapplicationsfor6Gwilloperateinthereal-timemanner.AI/MLwillbeblendedintothedesignof6GandthetransceiversmaybedesignedtobeAI-nativeatthebeginningof6G.Therefore,itisnecessarythatAI/MLbeentrenchedinthedesignofradiolayerswithinterfacingtoAIanddata-collectionframeworks.Theseinteractionsneedastrongemphasisonsecurityandprivacy.ThisAInativemethodensuresrapidevolutionofwireless

technologies,whichmaybepartialindependentofstandardscycles.

2.4.Hexa-X

IntheDraftfoundationfor6Gsystemdesign[13],Hexa-XproposedthatthedatacollectionandAIframeworksarepervasivefunctionalitiesofthe6Gsystem.Thedatacollectionframeworksupportseveraldifferenttypesofdataandinformationtobecollectedfrommultipledomainsandlayersofthenetworkandmovedwithinthenetworkforanalysis.Datatransmissionandintegrationwilltakeplaceconsideringprivacyandownershipconcernsofallstakeholders.Couplingdataacrossapplicationsandnetworkswillprovidetheopportunitytoimprovethenetworkperformanceorenablenetworkawareapplications.Thedatacollectionframeworkcollectsthedatarequiredbythenetworkmanagementfunctionandsupportsreal-timecontroland

operationstoprocessrequireddata.

IntheFoundationofoverall6Gsystemdesignandpreliminaryevaluationresults[14],Hexa-XproposedthatAIneedsnovelarchitecturalelementsthatenableforprivacyawaredatacollectionandlearning.Data-drivennetworkcontrolunitsandUEaggregationunitsareintroducedtosupportdatasharingbetweenthenetworkandUEs,soastoimprovecontrolanddecisions.Basedondataprivacyrequirements,UE-orienteddatacollectionandlearningenableUEstotakeadvantageofnetworkinformationalongwithon-devicecontextualinformation(useractivity,intent,andusagepatterns)toassistthenetworkinconnectivitydecisions,i.e.,toimprove

theconnectivityQoE.

7/64

2.5.6GANA

Inthe6GDataService-ConceptandRequirement[15],6GANApointedoutthatthe6Gnetworkprovidesnewcapabilities,suchasnativeAI,nativesensing,andnativesecurityinthecontextofintegratedcommunication,sensing,andcomputing.Basedontheinformationtransmissioncapabilitiesoftraditionalmobilecommunicationnetworks,thenewcapabilitiesenhancethedataproductionandconsumptioncapabilitiesofthenetworkandmakethe6Gnetworkaplatformforinformationanddatacirculation.Efficientmanagementofdatacarriedonthe6Gnetworkisakeytechnologyofthe6Gnetwork.Therefore,basedonthetrusteddataserviceframeworkproposedintheWhitePaperon6GDataServiceConceptandRequirements[15],the6Gnetworkwillintroduceanindependentdataplane.The6Gdataplanebuildsarchitecture-levelunifiedandtrusteddataservicestoclarifythedatasource,description,

collection,processing,storage,application,andprivacyprotection.

3.6GDataServiceScenariosandRequirementAnalysis

6Gdataservicesindicatethatthe6Gsystemprovidesdataresourcestointernalfunctions,suchastheUE,RAN,andCN.Thecollecteddatacanbeusedbynetworkcapabilityexposurefunctions(suchastheNEForCAPIF)toprovidedatanetworkcapabilitiesoreventtoexternalfunctionsofthe6Gsystem,suchasAFs.Inaddition,thecollecteddatacanbeusedbythesensingfunctiontoprovidesensingservicestoAFsorNFs,orbeusedbytheAIfunctionstoprovide

assistanceinformationforthecontrolandoptimizationofnetworks.

Withdataasthecoreelement,6Gdataservicesaimtofullyutilizethevalueof6Gdata,soastobreakthroughtheboundariesofsingle-dimensionalmobileservicesandpromoteintegratedinnovationofservices.Thedataof6Gnetworksindicatethedatageneratedorobtainedbythe6Gsysteminsteadofthedatainthetraditionaluserplane.ItincludesthedatageneratedbytheUEs,RAN,CN,andnetworkmanagementfunctionsduringtheoperationsofcommunicationservices.Italsoincludessensingdata(e.g.,sensingmeasurements)andAIdata(e.g.,AImodel)generatedbyUEs,RAN,CNandnetworkmanagementfunctionsduringtheoperationsofnewservices(e.g.,sensingservices,AIservices).Theshareabledataobtainedbythe6Gnetworksfromthirdpartiesisalsoincluded,suchasvarioustypesofsensorinformation(includingtemperature,humidity,andenvironment)andgeographicinformationsystem(GIS)information.Thischapterdescribes

scenariosof6Gdataserviceandanalyzespotentialrequirements.

3.1.SensingData

3.1.1.Description

8/64

Sensingdatainthe6Gsystemisthedatadescribingthephysicalworldstatusobtainedthroughradiowavesorothersensorsduringphysicalworldexploration.Sensingdatamainly

includesnativesensingdata,externalsensingdata,andmulti-modalintegratedsensingdata.

>Nativesensingdata

The6Gmobilecommunicationsystemhasasensingfunction,thatis,nativesensing.Nativesensingdataisthedatageneratedduringnativesensingofthe6Gmobilecommunicationsystem.ItincludesnativesensingmeasurementdataobtainedbytheUEorRANnodebasedonairinterfacesignalmeasurement,andnativesensingresults.Generally,thenativesensingmeasurementdataincludestwotypes.Oneisthedataobtainedduringsensingmeasurementtoassistindatacommunication,suchasthedatameasuredforsensingthechannelenvironment.Theotheristhedatameasuredpurelyforsensingthetargetobjectorenvironment,suchasthesignalstrengthandarrivaltimemeasuredduringpositioning,andtheRFsignaldatameasuredduringranging,speedmeasurement,andimaging.Throughnativesensing,the6Gsystemobtainsthe

sensingresults,suchastheposition,speed,andimagingofthesensingtarget.

>Externalsensingdata

Externalsensingdataisthedatathatthe6Gsystemobtainsfromexternalthird-partyIoTsensorsorGIS,whichincludesintermediatemeasurementdataandsensingresults.Thedataisnotinvisibledatathatistransparentlytransmittedonthenetwork.Itneedstobeprocessedbythesensingmoduleofthe6Gsystemtogeneratesensingmeasurementintermediatedataorevenusablesensingresults.Forexample,the6Gsystemconnectstoautomaticrecognitionequipment,suchasbarcoderecognition,imagerecognition,andradiofrequencyidentification(RFID)toobtaintargetinformation,andconnectstovarioustypesofsensorstoobtainphysicalinformation,suchasbiomass,chemical,heat,pressure,temperature,sound,light,electricity,andvibrationinformation.Thesensorsmainlyincludemechanicalsensors(suchasthedisplacementsensorandlevelsensor),geometricsensors,forcesensors(suchasthepressuresensorandspeedsensor),thermalsensors(suchasthetemperaturesensor),opticalsensors(suchastheimagesensorandinfrared/ultravioletsensor),electromagneticsensors(suchastheelectricfieldsensorandvoltagesensor),acousticsensors(suchasthesoundsurfacewavesensorandultrasonicsensor),raysensors,humiditysensors,gassensors(suchasthegascompositionsensorandgasconcentration

sensor),ionsensors(suchasthePHsensors),physiologicalsensors,andbiochemicalsensors.

>Multi-modalintegratedsensingdata

Comparedwithnativeorexternalsensingdata,multi-modalintegratedsensingdatafocusesonintegratedprocessingofthesetwotypesofdata.Inthe6Gsystem,multi-modalsensingdataincludesnativeandexternalsensingdata,whichcanbeprocessedseparatelyorgeneratenew

sensingdataorsensingresultsthroughintegratedsensing.Multi-modalintegratedsensing

9/64

incorporatesnativesensingandexternalsensing.Itintegratesdatafrommultiplesensingchannelsforunderstandingandprocessing.Varioustypesofsensingdatacollaborate,complement,revise,andenhanceeachothertogeneratebettersensingresultsthansingletypesofsensingdata.Forexample,inanindoorenvironmentcoveredbythelandmobilecommunicationsystemandInternetofThings(IoT)terminals,cellularpositioningdatacanbeusedtogetherwithdataofexternalIoTsensingdevices,suchasWi-Fi,Bluetooth,Zigbee,andUWBpositioningdatato

generatemulti-modalintegratedsensingdata.

3.1.2.PotentialRequirements

Inthe6Gsystem,varioustypesofsensingareavailable,andthesensingdataisextensive.Thenewandoptimizedsensingfunctioncausesahugeamountofdatainthelandmobilecommunicationsystem.Inadditiontonativesensing,the6Gsystemneedstoprocessandtransmit

externalsensingdataandmulti-modalintegratedsensingdata.

Specifically,thefunctionalrequirementsonthe6Gsystemareasfollows:

1)Datacollection:Duringnativesensing,the6GsystemusessignalingtomeasureparameterstransmittedthroughradiowavesandRFchannelsandcollectssensingdatafromvarioustypesofIoTsensingdevicesoverdifferentinterfaces.

2)Dataprocessing:The6Gsystemprocessesthecollectedsensingdata.ThroughAImodeltraining,computing,andotherprocessingmethodsforsensingdataobtainedfromvarioussensingchannels,usablesensingresultsareobtained.Theusablesensingresultsassistindecisionmakingandexecutionofcommunicationorotherfunctionsinthe6Gsystem,orareusedasinputparameterstoassistsomeapplications.Duringmulti-modalintegratedsensing,datafrommultiplesensingchannelscanbecollaborativelyprocessedtoensurebettersensingperformance.

3)Datatransmission:Inthe6Gsystem,massivevolumeofsensingdataneedstobetransmitted.Theintermediatemeasurementdataneedstobetransmittedtonetworkorcomputingnodesforintegrationandcalculation,soastoobtainsensingdataorsensingresults.Forexample,duringcellularpositioning,measurementdata,includingthesignalstrengthandsignalarrivaltimebetweenaUEandmultipleRANnodesneedsto

betransmittedandintegratedtoobtainthelocationinformationoftheUE.

Theperformancerequirementsonthe6Gsystemareasfollows:

1)Duetomassivevolumeofsensingdatainthe6Gsystemanddifferentsensingdatarequirementsofvarioustypesofapplications,the6Gcommunicationsystemneedstohavegoodtransmissionperformance,includingthebandwidthandlatency.AccordingtoStudyonScenariosandRequirementsof5G-AdvancedIntegratedSensingandCommunication[16],thesensingdatarateisabout1kbpsto10Mbps.

2)Tocalculateandmakedecisionsonmassivevolumeofsensingdata,the6GsystemneedstohavestrongcomputingpowerandbetterperformanceforAImodelsand

10/64

variousalgorithms.

3)Sensingdatainvolvesalargeamountofuserprivacybecauseitcomesfromphysicalstatusinformationaboutpeople,device,andenvironmentrelatedtoToBorToCusers.The6Gsystemneedstomeetapplicationrequirementsandensureusersecurityandprivacywhenperformingsensing.

3.2.AIModelTrainingData

3.2.1.Description

Onthe5GCN,theNWDAFismainlyusedforprediction,suchasUElocationpredictionandNEloadpredictioninAIanalysisscenarios,suchasdataandmodeltrainingandmodelinference.Insuchscenarios,asmall-scaleneuralnetworkmodel,suchastherecurrentneuralnetwork(RNN),convolutionalneuralnetwork(CNN),orlongshort-termmemory(LSTM)isused.Generally,themodelcontainsthousandstohundredsofthousandsofparameters.TheMTLFneedstocollecttenstohundredsofmegabytesofmodeltrainingdatatotrainamodel.The6GnetworkhasmoreextensiveAIanalysisscenarios,suchasobjectrecognitionandtrajectoryplanningbasedonsensingdataandnetworkpolicycontrolbasedonnetworkdata.Drivenbydiversescenarios,the6Gnetworkrequireslarger-scaleandmorecomplexmodelsforinferenceandanalysis.Academicresearchresultsshowthattheamountofdatarequiredtotrainamodelisproportionaltothemodelsize.Totrainamodelwith100millionmodelparameters,atleast8GB

dataisrequiredforpre-training,andotherspecifi