版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
1/18
TableofContents
1.Abstract 2
2.O-RANArchitecture 2
3.6GVisionanddesigntarget 3
4.Keytechnicalpillarsandconsiderations 4
4.1Networkarchitectureconsiderations 4
4.2ServicebasedRAN 6
4.3AI 8
4.3.1Cross-domainAIcollaboration 10
4.3.2LargeModel 11
4.4Spectrumsharing 12
4.5SustainabilityandEnergysaving 14
5.Forward-Looking 16
Reference 16
Abbreviation 17
Authors 18
2/18
1.Abstract
O-RANALLIANCEhasbeenfoundedin2018byAT&T,ChinaMobile,DeutscheTelekom,NTTDOCOMOandOrange.Sincethen,O-RANALLIANCEhasbecomeaworld-widecommunityofmobilenetworkoperators,vendors,andresearch&academicinstitutionsoperatingintheRadioAccessNetwork(RAN)industry.Themissionistore-shapetheRANindustrytowardsmoreintelligent,open,virtualizedandfullyinteroperablemobilenetworks.O-RANleveragesmostofthephysicalfeaturesdefinedin3GPP,whichmaintainsaunifiedandhealthyecosystem.O-RANspecificationssplitsthenetworkentitiesanddefinestheinterfacestofacilitatethemulti-vendorjointlydevelopandinteroperatetesttheproducts.
ITUdefinedIMT-2030Frameworkandrelatedtimeline,andtheindustriesinitializedthe6Gstudyaccording.O-RANalsokickedoffthe6GstudyinnGRG,whichistoformulatethe6Grelatedviewsbeforestandard.Beyondtheadvancedfeatures,O-RAN’sflexiblearchitecturecouldprovidesomeuniqueadvantagesforfuture6Gnetworks,whichincludesprogramablearchitecturefornetworkintelligence,service-basedRANdesign,sufficientnetworkpoweroptimization,flexiblespectrumsharingandetc.
Thiswhitepaperbrieflyintroducessomekey6GtechnicalpillarsbasedonO-RANnGRGdiscussion.Tofacilitatethereadertounderstandthetechnicalissuesandconsiderations,theO-RANarchitectureisintroducedinsection2.WealsoprovideforwardlookingforO-RANin6Geraattheendofthewhitepaper.
2.O-RANArchitecture
BelowistheO-RANarchitectureoverviewdefinedbyO-RANalliance[1].O-RANleveragesthe3GPPdefinedinterfaceandalsodefinessomenewinterfacesasitsplitstheRANfunctionsintoO-CU,O-DU,andO-RU.
3/18
Figure1O-RANArchitectureoverview[O-RAN]
Withinthearchitecture,RANIntelligentController(RIC)isthelogicalfunctionstoenablethecontrolsinanearrealtimeornon-realtimemanner.Thenetworkcontrolfunctionsaresplittedintothetwoentitiesbasedonrequiredtimescale.E2istheinterfaceconnectingtheNRRICandO-CU,andmostofthenetworkintelligentfunctionsareconnectedviaE2.O-DUandO-RUaretworemarkableentitiestorepresentthenetworkopenness.ThereareseveralsplitoptionsbasedonthesupportedfunctionsonO-RU.OperatorscouldprogramtheRICfunctionswithdifferentAPPs,andmultipleAPPscouldflexiblyenablethedifferentnetworkfunctions.
3.6GVisionanddesigntarget
ITU-RdefinestheIMT-2030Framework[2],whichincludestheusagescenariosandcapabilitiesof6G.Thisframeworkrecommendationisoneofthemostimportantguidancefor6Gandwouldbereferredasdesignguidancefor3GPPandotherSDOtospecifythe6Gstandard.
UsagescenariosofIMT-2030areenvisagedtoexpandonthoseofIMT-2020(i.e.eMBB,URLLC,andmMTCintroducedinRecommendation
ITU-RM.2083
)intobroaderuserequiringevolvedandnewcapabilities.InadditiontoexpandedIMT-2020usagescenarios,IMT-2030isenvisagedtoenablenewusagescenariosarisingfromcapabilities,suchasartificialintelligenceandsensing,whichpreviousgenerationsofIMTwerenotdesignedtosupport.
TheusagescenariosofIMT-2030includeImmersiveCommunication,HyperReliableandLow-LatencyCommunication,MassiveCommunication,UbiquitousConnectivity,ArtificialIntelligenceandCommunication,andIntegratedSensingandCommunication.
4/18
Figure2IMT-2030UsageScenarios[2]
O-RAN’snetworkarchitectureprovidesmostflexibilitybysplittingRANfunctionsanddefiningstandardinterface.Asthe6Gusagescenariosaredoubledcomparedto5G,theO-RAN’sflexibilitywouldbeagoodfoundationforfurtherinnovation.Insection3ofthepaper,wediscussedseveralhighlightedtechnicalpillarsfor6Gdesignandanalyzethechallengesandpotentialsolutions.
4.Keytechnicalpillarsandconsiderations
4.1Networkarchitectureconsiderations
6Gnetworkbridgesthephysicalanddigitalworlds.Anincreasingnumberoftrafficwilloccurontheedgeofthe6Gnetwork.Thepotentialfeaturesoffuture6Gnetworkareintelligence,programmabilityandresourcepooling.
Intelligenceisthekeyenablertechnologyfor6Garchitecture,andnativeAIhasarousedmoreattentionfromacademiaandindustry.InordertoachievethenativeAI,therelatedinterface(e.g.,E2)andprocedure(e.g.,AI/MLflow)shouldbeconsideredin6Garchitecture.Thesub-section3.3describesthenativeAIindetail.
Inthecontextof6G,theintegrationofnativeAIneedsanefficientandconvenientapproachtoincorporateAIelementsseamlessly.Inaddition,thedifference6Gservicerequiresdifferentnetworkresources.Therefore,programmabilityemergesasapromisingsolutiontodrivethedevelopmentof6Garchitecture.
5/18
Programmabilityencompassesthreekeycomponents:Parameter,Data,andAlgorithm.Programmableparameterfacilitatestheseamlessadaptationofparametersof6Gnetworkthroughaprogrammableframeworkandgeneralinterface.ProgrammabledatainvolvestheconstructionofdatasetsforAIalgorithmtrainingandtheexplorationofdatarelationshipswithinthenetworkfunctions.Additionally,datacanbesecurelyprovidedtothirdpartiesthroughrelevantsecuredmethods.Theprogrammablealgorithmsdefinetheinputandoutputdataformatwithdifferentscenarios.ThenetworkfunctionsandthirdpartiescanembedorreplacetheAIalgorithmsviaaprogrammableframeworkunderthesubjectsoftheaforementionedinputandoutputandasecuritycheck.Tofacilitatethisprocess,aprogrammableframeworkisrequiredtodeployandmanagethesealgorithmseffectively.Theframeworkshouldencompassacomprehensivesetofprogrammableinterfacesandfunctionmodules,enablingseamlessintegrationandoperation.Additionally,theprogrammablealgorithmensuresthatthe6Gnetworkdynamicallyadaptstovariousscenarios'requirements.Forinstance,ifconsumersseekhighthroughputfromtheRAN,theAIalgorithmincorporatestheRANslice.Similarly,forconsumersprioritizingQualityofService(QoS),theAIalgorithmintegratestheQoSoptimization.
ToenabletheimplementationofprogrammableRAN,itisessentialtoprogressivelyopenthetraditionallyclosedprotocolstackwithintheRAN.Thisinvolvesenhancingthefunctionalityattheprotocolstacklevel,andstandardizingandgeneralizingthenewlyopenedinterfaces.InthecontextoftheongoingevolutionofnativeAI,programmableRANcatalyzesadvancementtowardamoreopenandintelligentRANvision.Byembracingprogrammability,theRANcaneffectivelyadapttodynamicnetworkrequirements,fosterinnovation,andleveragethefullpotentialofnativeAI.
Resourcepoolingplaysanimportantrolein6Garchitecture.Theresourceisstillheterogeneous,itconsistsofcommonanddedicatedresources.Generalresourcesarecommon,standardizedhardware(i.e.,industrialserversbasedonX86orARMCPUs),anddiversifiedhardwarechipswithscalability,includingaccelerationandclockresourcechips,andgraphicsprocessingunit(GPU)forAImodeltraining.ForRAN,high-speedprocessingandalargenumberofdedicatedresourcesarerequired,suchasFieldProgrammableGateArray(FPGA)forcodingandencoding.TheclockresourcesareappliedtofulfillsynchronizationaccuracyamongnetworkelementsandUEs.Dedicatedresources(e.g.,ASICchips)providespecializedservicesforasmallnumberoffacilitieswithlargecapacityandultra-high-performancerequirements.
6/18
Figure3ProgrammableRAN
4.2ServicebasedRAN
Historically,RANarchitecturewasmainlydesignedtoguaranteetheconnectionservicefortraditionalToCbusiness,usingarelativelyclosedprotocolbutwiththeperformanceadvantagesofspecialization.Asmorescenariosandservicesintroducedin5Gand6G,andITtechnologiesisintegratedinmobilenetwork,RANarchitectureneedtoevolvetoprovidemoreflexible,adaptablenetwork.O-RANisdefiningamoreopenarchitecture,buildingaunifiedcloudplatformforRAN,standardizingmoreopeninterfaces,andintroducinganintelligentfunction.TheimplementationofcurrentcloudRANonlychangestherunningplatformforsoftwareinsteadofchangingsoftwarearchitectureoftheRAN.ThisRANarchitectureisnotcloud-friendlyandcannotmakefulluseoftheadvantagesofcloud-native.
Cloud-RANisthefirststep,RANsystemcanbefurtherevolved.SBAin5GCoreNetworkcanbeusedasreference.ThegoalofService-basedRANisachievingafullycloud-nativearchitecturebyrebuildingRANfunctionsintocombinableandreusablenetworkservicesandusingunifiedinterfacewithRANinternalservicesandCN.
Theadvantagesoftheservice-basedRANinclude:
1)Flexibleandelasticdeploymentofnetworkfunctions,rapidupgradingandexpansionofnetworkcapabilities,enablingmorebusinessscenarios;
2)Bringnewend-to-endnetworkinteractionwayswithoutreducingtheimpactof
7/18
cross-domainnewfunctionsintroductiononexistingservices;
3)Moretimelyandmulti-dimensionalopeningofwirelessnetworkcapabilities;
4)IntegratedmanagementandorchestrationwithCNservices,reducingthecomplexityofnetworkoperationandmaintenance,improvingthenetworkofadaptabilitytonewbusinesses.
Figure4Theconceptofservice-basedRAN
Thedesignoftheservice-basedRANarchitectureneedstoconsiderthefollowingaspects:
ServiceGranularity
ThegranularityforRANserviceswhichisrebuiltformtheoriginallyRANfunctionsiscrucial.Thesmallerthegranularity,themoreflexibleitis,butitmaybringperformanceandefficiencyissues.Theimplementationofthe5Gcorenetworksservice-basedarchitectureincludestwolevels:NFandNFServices.NFscommunicatewitheachother,andinternalNFservicescansharedatabaseswhichreducescomplexityandisalsodifferentfromMicroservicesarchitecture.Consideringtheinternalfunctionalcorrelationsandcomplexity,RANcaninitiallyberebuiltinasimilarwaytotheCoreNetwork.
Service-basedRANfunctions
5GRANcanbefunctionallydividedintocontrolplaneanduserplane,andthereisalsotheconceptofseparationofcontrolplaneanduserplane,butinthedeploymentlayer,itstilladoptsasinglemode.Thecontrolplanemainlyincludesfunctionssuchasconnectionmanagement,sessionmanagement,mobility,andmeasurement,andtheuserplaneincludestheprocessingofdatapackets.Therearedifferentconsiderationsforondifferentfunctionalplanes.
Forthecontrolplane,theservitizationcanrebuildtheexistingcontrolplanefunctionsintofinergrainedservicesaccordingtothedegreeofcoupling,anddifferentservicescanbecombinedandflexiblydeployedindifferentscenariosandregionsondemand.Forexample,inthescenariooftheInternetofvehicles,themobilitymanagementserviceissuitableforcentralizeddeploymenttooptimizethemobilityexperience.Atthesametime,theservice-basedfunctionsofthecontrolplanecanrealizedirectaccesstotheCoreNetworkcontrolplane,reduceunnecessarysignalingforwarding,andtheinteractionwithothercorenetworkservicescanbechangedfromserialinteractiontoparallelinteraction,optimizingthesignalingprocessofthecontrolplane.The
8/18
optimizationofsignalingprocesseshelpsimprovenetworkperformance,suchasdelayandefficiency.Besides,forextremerequirementsofspecificservices,italsohelpsRANandCNintegratesattheedge,simplifyingdeploymentcomplexityandimprovingperformance.Finally,forthemorecomplexfunctionalconfigurationandparameterconfigurationofthefuturenetwork,theservice-basedcontrolplanecanbeexecutedandupdatedatasmallergranularitywithoutaffectingtheoperationofotherservices.
Fortheuserplane,thetraditionalmobilecommunicationprotocolsallfollowtheOSIhierarchicalprotocoldesignconcept.Eachlayerreceivesspecificservicesprovidedbythelowerlayerandisresponsiblefortheupperlayer.Theupperandlowerlayersinteractwitheachotheraccordingtotheinterfaceagreement,andthesamelayerinteractswitheachotheraccordingtotheprotocolagreement.Theproblemofthislayereddesignconceptisthattheprotocolandservicemodelarefixed,andflexiblecross-layersignalinginteractionandcross-layerfunctioncombinationcannotberealized.Thediversifiedcharacteristicsoffutureapplicationswillbringmoredifferencesindatapacketprocessing,suchassmalldatapacketsforindustrialcontrol,whichrequirehigherreliabilityandneedtoutilizethePDCPreplicationfunctioninuserplane;ImmersiveinteractiveapplicationshavedifferentprocessingrequirementsforI-frame,P-frame.TheuserplaneneedsfunctioncombinationandarrangementforthenewQoSguarantee.Inadditiontothecurrenttypesofexistingapplications,sensing,AIandothernewapplicationshavealsobroughtnewdatapacketmodels,requiringtheuserplanetobeabletomatchtheprocessingofdifferentdatapackets,aswellasforwarding.Theservice-baseduserplanehasadvantagesinflexiblecombination,deployment,andrapidupdate.Forscenarioswithdifferentuserdatapacketprocessingrequirements,theservice-baseduserplanecanbepreferred.
Besides,thenewservicessuchasAI,computing,sensingwillbeprovidedbythefuturewirelesssystem,ontheonehand,thiscanenabletheenhancementoftheexistingfunctionalplane,suchasintroducingcontrolfunctionsforsensingandcomputingpowerandintroducingnewuserpacketprocessingmodeinuserplane.Ontheotherhand,RANmayalsointroducenewfunctionalplanes,suchasdataplane,bringingnewfunctionalinteractiveways,thatwillraisemoredemandsonnetworkflexibilityandrapidupdate.Service-basedarchitecturehascertainadvantagesintheseaspects.
Service-basedinterface
Atpresent,theRANandtheCoreNetworkinteractthroughthepoint-to-pointN2interface.Forservice-basedRAN,aservice-basedN2interfacecanbeconsidered,andtheRANisstillanindependentwhole,RANservicescaninteractwitheachotherthroughaninternalefficientinterface.Thisapproachisrelativelyeasytoimplementandcanbeadoptedduringtheinitialphaseofserviceorientation.TheotherwayistouseaconsistentinterfacebetweenRANinternalservicesandthecorenetwork,RANservicesandcorenetworkservicesareinapeerpositionandcanachievedirectinteraction,thisapproachhasmoreadvantages,butatthesametimewillbringmoreissuesrelatedtonetworksecurity,ecologicalchange.
4.3AI
ArtificialIntelligence(AI)hasbeenproposedasoneofthemostpowerfultechnologiesthat
9/18
improvessystemperformanceandenablesnewfeaturesinthewirelesscommunicationnetwork,byanalyzingthedatacollectedandautonomouslyprocessedthatcanyieldfurtherinsights.
3GPPintroducedanewlogicalfunctionentity,namedNWDAF,tothe5GCtoprovidemultipletypesofnetworkdataanalyticservices.Thenetworkdataanalyticservicesinclude:
ObservedServiceExperiencerelatednetworkdataanalytics,toprovideaverageofobservedServiceMoSand/orvarianceofobservedServiceMoSindicatingserviceMOSdistributionforservicessuchasaudio-visualstreamingaswellasservicesthatare
notaudio-visualstreamingsuchasV2XandWebBrowsingservices;
NFLoadAnalytics,toprovidetheaverageloadoftheNFinstance;
NetworkPerformanceAnalytics,toprovidethebasestationstatusinformation,resource
usage,communicationperformanceandmobilityperformanceinanareaofinterest;
UErelatedanalytics,toprovidetheUEmobilityanalytics,UEcommunicationanalytics,expectedUEbehavioralparametersrelatednetworkdataanalyticsandabnormalbehaviorrelatednetworkdataanalytics;
UserDataCongestionAnalytics,toprovidecongestionexperiencedwhiletransferringuserdataoverthecontrolplaneoruserplaneorboth;
QoSSustainabilityAnalytics,toprovidetheQoSchangestatisticsorlikelihoodofaQoSchangeforananalyticstargetperiodinacertainarea.
InRAN,3GPPalsoconductedseveralstudiesontheAI-enablednetwork.InRelease17,3GPPconductedastudyonAI-enabledRANintelligence,whichdefinedareferencefunctionalframeworkandidentifiedasetofhigh-levelprinciplestoguidethestandardswork.ThestudyonAI-enabledRANintelligencefocusedonthreeusecases:
NetworkEnergySaving,tooptimizetheenergysavingdecisions(e.g.,cellactivation/deactivation)bypredictingtheenergyefficiencyandloadstateofthenextperiod;
LoadBalancing,toprovidehigherqualityuserexperienceandtoimprovesystemcapacitybybasedoncollectionofvariousmeasurementsandfeedbacksfromUEsandnetworknodes;
MobilityOptimization,toreducetheprobabilityofunintendedeventsassociatedwithmobility,topredictUElocation,mobilityandperformance,andtosteertraffictoachieveefficientresourcehandling.
InRelease18,3GPPconductedastudyonAIforNRairinterface,toexplorethe3GPPframeworkforAIforair-interfacecorrespondingtoeachtargetusecaseregardingaspectssuchasperformance,complexity,andpotentialspecificationimpact.ThestudyonAIforNRairinterfacealsoadoptedausecasecentricapproach,focusingonthreeselectiveusecases,namelyCSIfeedbackenhancement,beammanagementandpositioningaccuracyenhancement.InRelease19,3GPPconductedastudyonAIformobility,toimprovehandoverand/orRRMperformancebypredictingcelllevelmeasurement,handoverfailure/radiolinkfailure,andmeasurementevents.
10/18
IntheO-RANarchitecture,theintroductionoftheRIChasbeenanimportantdevelopment,makingitpossibletointroduceAIbasedsolutionstoawidelyusecases.EnablingAIdrivennetworkingrequiresaparadigmshiftinthearchitecturalblueprint.Inthe6G,therearethreeimportantfeaturesofAIneedtobeconsidered,namely,nativeAI,crossdomainAI,andnetworklargemodel.
NativeAIreferstoembeddingAIintofunctionalitiessupportedbyvariousnodes/endpointsandinterfaceswithinanetworkarchitecture[8].ConsideringthefourkeycomponentsofAI,i.e.,computingpower,data,AIalgorithmsandfunctionalities,anativeAInetworkshouldbewithahybridcentralized/distributedAIarchitecture.ThecentralizedAIentitiesrunfororchestration,managing,deployingandcontrollingallthedistributedAIentities,e.g.,ontheSMOplatformthatinteractswithotherdomain-specificAIentities.ThedistributedAIentitiesrunforservingfunctionsofthelocalnetworkandreceivingcommandsfromthecentralizedAIentities,e.g.,ontheCN,TN,BSandUErespectively.
The6GwirelessnetworkwillnativelyintegratecommunicationcapabilitieswithAI.Ontheonehand,end-to-endAImayleveragemassiveamountsofdataproducedbyairinterfacesandnetworkstooptimize6Gnetworksandofferconsumerscustomizednetworkservices.Ontheotherhand,asthecomputingpowerofinfrastructureandterminaldevicesenhances,futurenetworkswillbeabletoofferadistributeddeploymentenvironmentforAI,deliveringmoreflexibleandreal-timeAIservicesatthenetworkedgeforusers.Firstly,itisessentialthatsupportforAIbetakenintoconsiderationfromthebeginningwhendesigningnetworkarchitecture.Thisconsiderationmustensuretheseamlessintegrationoftraditionalcommunicationinteractions,whilethemetricsfortrainingandinferenceofAIareconvergedintothecontrolanddataflow.CollaborationwithinAIisalsoacriticalfactortotakeintoaccount.ThisincludescooperationbetweencentralizedanddistributedAIdeployment,cooperationbetweenlargenetworkmodelsandotherspecializedmodels,andthecross-domainAIcollaborationamongRAN,CN,andmanagementsystems.Hence,itisimperativetodesignefficientAIcollaborationmechanismsfromtheperspectivesofAIorchestrationandmanagement,datainteraction,distributedlearningalgorithms,andcomputingpowerscheduling.Lastbutnotleast,theproblemofAIsecurityhascontinuouslypresentedamajorobstacletotheuseofAItechnologies,requiringprotectionsintrustworthyAI,datasecurity,andprivacytoguaranteethedependabilityandsecurityof6GAIapplications.
4.3.1Cross-domainAIcollaboration
CrossdomainAIreferstocollaborationandintegrationofAI-enabledfunctionalitiesacrossdifferentdomains,wherethedomainscanmaptonetworksdomains(e.g.,RAN,CN,TN,networkapplications,networkdigitaltwins)orotherdomains[8].InordertoenablecoordinatedAIcapabilitiesacrossdifferentnetworkdomains,thecentralizedAIentities(e.g.,ontheSMOplatform)shouldhandletheend-to-endAImanagementandorchestrationcapability,suchascross-domaindataarrangementandmapping,AItaskidentificationanddecomposition,mappingAItaskswithcomputingresources.
Figure3providesapotentialarchitecturefornativeandcross-domainAI,wherethecentralizedAIentityislocatedinthemanagementdomain.ForE2Eintelligentscenario,across-domainAImanagementfunctionshouldbeaddedintheSMOasacentralizedAIentityto
11/18
coordinatetheAIcapabilitiesfromotherdomains.ThismoduleisrequiredtohandleAIserviceorchestration,networkcomputingresourcemanagement,modelstorageandmanagement,cross-domainAIlifecyclemanagement,andotherrelatedfunctions.
Figure5Nativeandcross-domainAInetworkarchitecture.
Forthenativeandcross-domainAInetworkarchitecture,thecollaborationcontrolbetweendifferentnetworkdomainsisanewchallenge[9].Firstly,thecollaborationcontrolfunctionlocatedinthecentralizedAIentitywilldecomposeintentsintoservicerequirementsonevolvednetworkdomains,wheretheservicerequirementswillaffectconnectionrequirements,AIalgorithmrequirements,datarequirementsandcomputingrequirements.Basedontheservicerequirements,thedistributedAIentitylocatedinthenetworkdomainwilldecomposetheservicerequirementsintothenetworkfunctionrequirements,connectionrequirementsandresourcerequirements.Secondly,toprovideamorereal-timemanagementcapabilities,theserviceswithhighreal-timerequirementsandlowcomplexitywillbeprocessedbythedistributedAIentities,whiletheserviceswithlowreal-timerequirements,largeareasandhighcomplexitywillbeprocessedbythecentralizedAIentity.Therefore,thecollaborativecontrolmethodsshouldbeconsidered,e.g.,federatedlearning,splitlearningandtransferlearning.
4.3.2LargeModel
AsabreakthroughdevelopmentofAItechniques,NetworkLargeModel(NetLM)haveattractedattentionfromboththescientificcommunityandindustryalike.ComparedwithtraditionalAImodelsthatoptimizenetworksunderpredefinedoperations,theNetLMleveragesgenerativeAIalgorithms,e.g.,generativeadversarialnetworkandtransformer,toautomaticallyandcreativelygeneratecustomizednetworksolutions.Forexample,forthejointcommunicationandsensing,theNetLMcouldhelpingeneratingrelevantrays(e.g.,generatingtherightdistributionofAzimuthandElevationanglesoftheradiofrequencybeamtransmittedoutofthenode)tocaptureandsensethesurrounding.AnotherexampleistheNetLMfordigitaltwin,where
12/18
theNetLMisabletolearnthetailbehaviorofthetraindatasetdistributionwithonlyafewsamples,andgeneratethenewbehaviorbasedonwhatwaslearnedthatisconsistentwithreality.
SinceNetLMusuallyconsistsofbillionsofparameters,itisdifficulttodeploytheNetLMontheedgedirectlyduetothelimitedcomputing,communicationandstorageresources.Inaddition,thedeploymentoftheNetLMinthecorenetworkwillalsocausetremendoustransmissionlatencyduetothehugeanddistributeddatatobecollectedandtrainedinthecloud.Therefore,thedeploymentoftheNetLMwillemphasizethecollaborationbetweenNetLMwithvariousscales,includingthecollaborativetrainingandinference.Astandardizedcollaborationmechanismneedstobedefined,suchasnetworkarchitecturewithnewnetworkelement,large-scaledatadistributedstorageandreal-timeprovisionmechanism,andmodel-basedcollaborativeinterface.
4.4Spectrumsharing
From4Gera,mobileoperatorsjointlydeploytheRANnetworkstoextendthecoverage.ThiscouldlargelyreducetheCAPEXandbecomesthemajortrendswhen5Gcomes.InChinaandotherregions/countries,operatorssharethefrequenciesandcooperateontheRANnetworkconstruction.CMCC&CBNjointlydeploythe5GNRnetworkonbandn28,CT&CUjointlydeploythe5GNRnetworkonbandn79andotherfrequencies.
3GPPspecifiedtheRANsharingmechanism,whichisknownasMOCN,standingforMulti-OperatorCoreNetwork.InaMOCNset-up,oneradioaccessnetworkprovidesaccesstothenetworkofmultipleoperators.Eachoperatorrunsherowncorenetwork,buttheradioaccessnetwork,includingcarriersignals,isthesameforallpartnersinacertainregion.OnedrawbackisonlytheoperatorwhoownstheRANnetworkcouldoptimizethescheduler&configurationbasedonservicecharacters,andotheroperatorswouldnothavesuchflexibility.Iftheotheroperators’newusecases(e.g.XR)aredifferentthantheoriginaloptimization,theyhave
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 电子商务企业区域业务代表面试要点
- 世纪佳缘在线销售面试经验
- 生物医药研发人员的招聘面试技巧
- 新媒体公司在线客服专员的工作流程及规划
- 机构研究报告-消费行业市场前景及投资研究报告:亚洲消费者趋势洞察
- 铁路运输企业区域销售经理的职责与能力要求详解
- 正常分娩护理要点
- 分支行内控制度与操作规程
- 主管的沟通技巧与艺术培训资料
- 高新技术企业在中国的融资策略研究
- 2026年江西信息应用职业技术学院单招综合素质考试参考题库含详细答案解析
- 《2026年》融资租赁岗位高频面试题包含详细解答
- 北京市东城区2025-2026学年高二上学期期末考试化学试卷(含答案)
- 2026年春季学期西师大版三年级下册数学教学计划附教学进度表(2024新教材)
- 统编版(2026)八年级下册历史教材课后问题答案(全册)
- 动物入场查验制度规范
- 2026及未来5年中国宠物殡葬服务行业市场竞争态势及投资前景研判报告
- 中国电子技术标准化研究院:零碳工厂建设现状与发展路径研究
- 大肠病损切除术后护理查房
- 2025年中国医美注射类产品行业发展研究报告
- 股东薪资确认协议书
评论
0/150
提交评论