軟銀:AI-RAN - 電信基礎設施邁向人工智能時代 AI-RAN - Telecom Infrastructure for the Age of Al_第1頁
軟銀:AI-RAN - 電信基礎設施邁向人工智能時代 AI-RAN - Telecom Infrastructure for the Age of Al_第2頁
軟銀:AI-RAN - 電信基礎設施邁向人工智能時代 AI-RAN - Telecom Infrastructure for the Age of Al_第3頁
軟銀:AI-RAN - 電信基礎設施邁向人工智能時代 AI-RAN - Telecom Infrastructure for the Age of Al_第4頁
軟銀:AI-RAN - 電信基礎設施邁向人工智能時代 AI-RAN - Telecom Infrastructure for the Age of Al_第5頁
已閱讀5頁,還剩82頁未讀 繼續免費閱讀

下載本文檔

版權說明:本文檔由用戶提供并上傳,收益歸屬內容提供方,若內容存在侵權,請進行舉報或認領

文檔簡介

AI-RAN:Telecom

?SoftBankCorp.

Contents

ExecutiveSummary 4

1.AI-RANVision:ShapingtheFutureofTelecom 5

1.1TelecomChallenges:BalancingMassiveCapitalInvestmentswithROI 5

1.2Opportunities:TransformingNetworkInfrastructurethroughAI-RAN 6

1.3FromCostCentertoPro?tCenter 7

2.RANEvolution:FromdRAN,vRAN,CloudRAN,OpenRANtoAI-RAN 7

2.1KeyDevelopmentsinRANEvolution 7

2.2AI-NativeNetworks:TheRoleofAIinRANTransformation 9

2.3AI-RANDe?nitions 9

3.HistoryofSoftBank'sAI-RANR&D 10

3.1EarlyResearchandAI-RANDevelopment 10

3.2ApplicationsofSoftBankAI-RANResearch 11

3.3PartnershipsandCollaboration 12

4.gRAN:GPU-basedAI-RANArchitecture 13

4.1KeyCharacteristicsofgRAN 13

4.2TheArchitectureofgRAN-basedAI-RAN 14

4.3gRANCaseStudy:NVIDIAAIAerial 15

5.IntroductionofAITRASbySoftBank 17

5.1KeyFeaturesofAITRAS 17

5.2KeyComponentsofAITRAS 17

5.3AI-NativeOrchestration 19

5.4EdgeAI 20

5.5KeyBene?tsofAITRAS 21

6.AITRASEvaluation 22

6.1OutdoorTestbedforAITRAS 22

6.2AITRASPerformanceEvaluation 24

6.3SoftBank’sL1EnhancementsinAITRAS 25

7.AI-and-RANVirtualizedInfrastructureinAITRAS 26

7.1SoftBankAI-and-RANApproach 26

7.2HardwareandResourceManagement 26

7.3AITRASAI-and-RANOrchestrator 27

7.4AgenticAI-ServerlessAPIPoweredbyNVIDIAAIEnterprise 28

7.5MeetingHighAvailabilityandPerformanceStandards 30

7.6SustainabilityandEnergyEfficiency 30

?SoftBankCorp.

3

AI-RAN:TelecomInfrastructurefortheAgeofAI

8.AITRASAIApplications 31

8.1TheShifttoComputing-CentricArchitecture 31

8.2UseCasesfortheAITRASAI-on-RAN 31

9.StrategicBusinessModelsandRevenueGeneration 35

9.1DemandForecasting,CustomerSegmentation,andBusinessModels 35

9.2AITRASAI-and-RANforNewRevenueGeneration 37

9.3TCOAnalysis 37

10.CaseStudy:AI-RANTCOAnalysis 37

10.1AI-RANDeploymentSimulationinUrbanArea,Tokyo 37

10.2RegionalPeakTrafficVariations 38

10.3ROIAnalysisofAI-RANwithNVIDIAGB200-NVL2 39

11.Conclusion 41

11.1ChartingtheFutureofTomorrow’sNetworks 41

11.2Long-TermVisionandSustainableGrowthStrategies 42

References 44

Acknowledgment 45

Glossary 45

4

AI-RAN:TelecomInfrastructurefortheAgeofAI

ExecutiveSummary

SoftBank’sAI-RANinitiativeaimstorevolutionizethetelecomindustrybyintegratingArti?cialIntelligence(AI)intoRadioAccessNetwork(RAN),transformingtraditionalnetworksfromcostcentersintointelligent,revenue-generatingplatforms.Withmobiledatatrafficcontinuouslygrowing,AI-RANisexpectedtomeetthedualchallengesofrisinginfrastructurecostsandintensifyingmarketcompetition.Thisapproachisexpectedtoenabletelecomoperatorstooptimizenetworkperformance,reducecosts,andcreatenewrevenuestreamsthroughAI-enabledservices.

AI-RANmaybeimplementedleveragingasoftware-de?ned,GPU-poweredarchitecturecalledgRAN(GPU-basedRAN).Thisadvancedarchitecturesupportshigh-performancenetworkoperationsbyutilizingtheparallelprocessingpowerofGPUs.gRANenablesreal-timedataprocessing,intelligentresourcemanagement,andscalablemulti-tenantoperations.AsthesameplatformsupportsbothnetworkandAIworkloads,gRANoffersunparalleled?exibility,enablingseamlessintegrationofRANservicesandAI-nativeapplicationssuchasautonomousdriving,real-timerobotics,andedgecomputing.

SoftBank’sAI-RANproduct,AITRAS,exempli?estheconvergenceofAIandtelecominfrastructure.AITRASintegratesRANandAIworkloadsintoasingle,AI-nativecomputingenvironment,offeringcarrier-gradeRANfunctionalitywithenhancedscalabilityandefficiency.Thesystemsupportsmulti-tenantoperations,enablingnetworkproviderstorunAIservicesalongsidetraditionalnetworkfunctions,creatingnewrevenueopportunities.AITRASispoweredbyNVIDIAGH200GraceHopperSuperchip,whichenablereal-timeAIinferenceandnetworkmanagementwithoptimalpowerefficiency.

Fieldandlaboratoryevaluationshavecon?rmedAITRAS’sabilitytodelivercarrier-gradestability,higherenergyefficiency,andcost-efficientoperations.Inurbantrials,thesystemsuccessfullysupportedhigh-densitytrafficscenarios,whilelabtestscon?rmedthatitspowerconsumptionwascomparabletothatofcurrentRANsystems,despitehandlingsigni?cantlyhigherworkloads.ThisbalancebetweenperformanceandsustainabilitypositionsAI-RANasatransformativeforceintelecominfrastructure.

Toaccelerateindustryadoption,SoftBankplayedaleadingroleinestablishingtheAI-RANAllianceincollaborationwithmajortechnologypartnerssuchasNVIDIA,Arm,Ericsson,Nokia,Samsung,andT-Mobile.Thisallianceisfosteringinnovationthroughcollaborativeresearchanddevelopmentactivities,advancingAI-RANtechnologieswhilealigningwiththeglobalstandardssetbyorganizationslike3GPPandO-RANAlliance.

SoftBankenvisionsaphaseddeploymentroadmapforAITRAS,SoftBank’sAI-RANproduct,beginningwithOver-the-Airpilotina?eldareain2024,followedbycommercializationby2026.

5

AI-RAN:TelecomInfrastructurefortheAgeofAI

1.AI-RANVision:ShapingtheFutureofTelecom

ThevisionofSoftBankAI-RANR&DistorevolutionizetelecommunicationsbyintegratingAIintothecoreofRANinfrastructure,transformingtraditionalRANintointelligent,adaptive,andrevenue-generatingplatforms.

Figure1.AI-RAN:AIandRANintegration

1.1TelecomChallenges:BalancingMassiveCapitalInvestmentswithROI

Thetelecomindustryisfacingsigni?cantcapitalexpenditurepressuresduetorapidlyevolvingtechnologiesandincreasingdatademands.TheGSMA'sTheMobileEconomy2024report

1

revealsthatintheglobalmobilemarket,totaloperatorrevenuesareprojectedtogrowfrom$1.11trillionin2023to$1.25trillionby2030,representingamodestcompoundannualgrowthrate(CAGR)of1.74%.However,totalcapitalinvestmentsthrough2030areestimatedat$1.5trillion,exceedingtotalsingle-yearrevenues.Thishighlightsacriticalchallengefacedbyoperatorsworldwide.

Foremostamongthesechallengesisthesubstantialinvestmentcostassociatedwith5Gnetworkdeployment.Newinfrastructurerequirements,suchastheutilizationofhigherfrequencybandsandthemassdeploymentofMIMOantennas,necessitatesigni?cantfunding.Additionally,theimpactofincreasedtrafficfromgenerativeAIapplicationslikenewlyemergingLargeLanguageModels(LLMs)oninfrastructuremustalsobeconsidered.Thesupplyofequipmentforthisinfrastructureiscurrentlydependentonafewspeci?cvendors,makingitdifficulttoreducecostsandencouragecommoditization.Additionally,therapidproliferationofIoTdevicesandthegrowingpopularityofhigh-de?nitionvideostreamingnecessitatecontinuednetworkcapacityexpansion.Meanwhile,intensepricecompetitionin

1GSMA,TheMobileEconomy2024Report

:/solutions-and-impact/connectivity-for-good/mobile-economy/wp-

content/uploads/2024/02/260224-The-Mobile-Economy-2024.pdf

6

AI-RAN:TelecomInfrastructurefortheAgeofAI

telecomservicesmakesitincreasinglydifficulttorecoverinvestmentsthroughtraditionaldataservicefeemodels.Furthermore,theemergenceofOver-The-Top(OTT)providerswhooperatewithouttheirowncommunicationsinfrastructureimpactstelecomoperators'pro?tability.

AccordingtoGSMA'sTheMobileEconomy2024report,despitediscussionsaboutapotentialslowdowningrowth,monthlyglobalmobiledatatrafficperconnectionsawasigni?cantincreasefrom10.2GBin2022to12.8GBin2023,representingthelargestabsolutegrowthsincedatatrackingbeganin2016.Lookingforward,itisprojectedthattotalmobiledatatrafficwillgrowatanaverageannualrateof23%between2023and2030,andexceed465exabytes(EB)permonthbytheendofthedecade.Thisnetworkresourcestrainisforcingtelecomoperatorstomakesubstantialcapitalinvestments.Consequently,dependingontheirrevenuemodels,operatorsfacetheriskofbeingunabletorecovertheirincreasinginvestmentcosts,presentingacriticalmanagementchallenge.

Concurrently,pricecompetitionfortelecomserviceshasintensi?ed,makingitchallengingtorecoupinvestmentsthroughtraditionaldatacommunicationfeerevenuemodels.

Inthiscontext,telecomoperatorsareconfrontedwiththechallengeofimprovinginvestmentefficiency.Speci?cally,theyfacetwokeyissues:reducinginfrastructuredevelopmentcostsandcreatingnewrevenuestreams.Thisnecessitatesnotonlymoreefficientoperationandgreatercostreductionsinnetworkinfrastructure,butalsothedevelopmentofvalue-addedservicesandtheestablishmentofnewbusinessmodelstosecureadditionalrevenuesources.

1.2Opportunities:TransformingNetworkInfrastructurethroughAI-RAN

AI-RANpresentsauniqueopportunitytofundamentallytransformnetworkinfrastructure,makingitmoreadaptable,efficient,andcapableofsupportingnewAIservices.ByleveragingAI,telecomoperatorscanoptimizenetworkoperationsinrealtime,improveresourceutilization,andintroducenewrevenue-generatingopportunities.

OneofthekeyopportunitiesofferedbyAI-RANisitsabilitytoshiftfromastatic,hardware-dependentnetworkarchitecturetoadynamic,AIandsoftware-drivenapproach.AIallowsforintelligentdecision-makingatthenetworkedge,enablingreal-timeresponsestotrafficconditions,userdemand,andservicerequirements.Thislevelofadaptabilityensuresthatnetworkswillalwaysoperateatpeakefficiency,providebetterqualityofservice,andsuppressenergyconsumption.

Furthermore,AI-RANopensthedoortonewserviceofferingsthatwerepreviouslynotfeasible.Forexample,advancednetworkslicing,enabledbyAI-drivenresourcemanagement,allowsoperatorsto

7

AI-RAN:TelecomInfrastructurefortheAgeofAI

createcustomizedend-to-endvirtualnetworkstailoredtothespeci?cneedsofdifferentcustomersegments,suchaslow-latencyconnectionsforgamingandLLMinferencingandhigh-reliabilitynetworksforenterprisemissioncriticalapplications.Thisabilitytoofferdifferentiatedservicesnotonlyenhancescustomersatisfactionbutalsocreatesnewrevenuestreamsforoperators.NewEdgeAIinferencingservicesarealsopossibleonthesameAI-RANinfrastructure.

1.3FromCostCentertoProfitCenter

AI-RANisseenasastrongapproachtoenhancingthereturnoninvestmentinnetworkinfrastructurefortelecomoperators.OneofAI-RAN'skeyfeatures,multi-tenancy,notonlyutilizesRANresourcesforhigh-throughputbroadbandcapacity,wirelessqualityimprovement,andnetworkoptimizationbutalso?exiblyallocatesresourcesforedgecomputinginfrastructuresthatsupportAItrainingandinferencing.Thismulti-purposecapabilityenablesoperatorstoimprovemobilenetworkqualitywhilecreatingnewrevenueopportunities.

ByadoptingAI-RAN,telecomoperatorscanmaximizethepro?tabilityoftheirnetworkinvestmentsandestablishsustainablegrowthmodels.Thistransformationconvertstraditionalnetworkinfrastructurefromacostcenterintoapro?tcenter,enablingoperatorstoachievesustainablegrowththroughnewbusinessmodels.

Figure2.AI-RANrede?nestelecombusiness

2.RANEvolution:FromdRAN,vRAN,CloudRAN,OpenRANtoAI-RAN

2.1KeyDevelopmentsinRANEvolution

TheRANlandscapehasundergonearemarkableevolution,transitioningfromtraditionalhardware-centricmodelstomoreadvanced,AIandsoftware-basedarchitectures.ThisevolutioncanbecharacterizedbytheprogressionfromDistributedRAN(dRAN)toVirtualizedRAN(vRAN),CloudRAN(C-

8

AI-RAN:TelecomInfrastructurefortheAgeofAI

RAN),OpenRAN,andultimatelyAI-RAN.

2.1.1EvolutionfromdRANtovRAN,C-RAN,andOpenRANtowardAI-RAN

dRANrepresentsthetraditionalRANsetup,whereradiounits,distributedacrosssites,arecloselycoupledwithbasebandunitsforsignalprocessing.Thissetupoftenleadstoincreasedcostsincludingsiteassets,inefficientresourceuse,anddelaysinserviceevolution.

vRANemergedasaresponsetothesechallengesbyvirtualizingbasebandfunctions.WithvRAN,networkfunctionscouldbeseparatedfromdedicatedhardwareanddeployedoncommercialoff-the-shelf(COTS)hardware,enhancing?exibilityandscalability.

C-RANfurtheradvancedthisconceptbycentralizingbasebandprocessinginacloudenvironment.Thecentralizedprocessingreducedhardwarerequirementsatindividualsites,allowingbetterpoolingofresourcesandcentralizedmanagement.Itimprovedefficiencybutrequiredarobustbackhaultomanagelatencychallenges.

OpenRANbuildsuponthevirtualizedandcloud-basedapproachesbyintroducingstandardizationandinteroperability.ItdisaggregatesRANcomponents,allowingoperatorstomixandmatchsolutionsfrommultiplevendors,breakingvendorlock-in,reducingcosts,andencouraginginnovation.Thisopennesssupportsgreater?exibilityandadaptabilityinnetworkdeployments.

AI-RANintegratesAIcapabilitiesintoRANoperationsoveracommonacceleratedinfrastructureandsorepresentsthegreatestadvance.ByprovidingAIandRAN,AIforRAN,andAIonRAN,operatorscanmovebeyondmereconnectivityandmakenetworksmoreintelligent,self-optimizing,andproactive.

2.1.2DriversofTransformation

Thekeydriversbehindthesetransformationsincludecostoptimization,performanceoptimization,improving?exibilitywithsoftware,enhancingoperationalefficiencyandcapturingnewmonetizationopportunities.TraditionalRANsolutionsrequiresigni?cantcapitalexpenditure(CAPEX)forspecializedhardware,whiledRANalsofacedscalabilitylimitationsandhighoperationalcosts.Movingtovirtualizedandcloud-basedsolutionsaddressesthesechallenges,allowingoperatorstominimizecostsandfullyutilizethescalabilitypotentialofthecloudinfrastructure.OpenRANandAI-RANtakethesebene?tsfurtherbyenabling?exibilitythroughopeninterfaces,greateroperationalefficiency,andnewmonetizationmethodsbasedonAIservices.

9

AI-RAN:TelecomInfrastructurefortheAgeofAI

2.2AI-NativeNetworks:TheRoleofAIinRANTransformation

AIisplayingatransformativeroleintheevolutionofRANbyprovidingadvancedtoolstooptimizeperformance,automateresourceallocation,andwillultimatelytransformhowmodernnetworksoperate.

ImprovingSpectralEfficiency,OptimizingPerformanceandResourceManagement:AIisfundamentallychanginghowRANresourcesaremanagedbymakingoperationsmoreadaptiveandefficient.IntraditionalRANsetups,resourceallocationandmanagementrequiremanualcon?guration,makingithardtoreacttouserdemands.AI,however,enablesalevelofdynamicadaptabilitythatwaspreviouslyunachievable.Forexample,AI-nativemodelscanautomaticallyallocatebandwidthbasedonreal-timeusagepatterns,manageinterferencemoreeffectively,andensureoptimalloadbalancingacrossthenetwork.Thisimprovesspectralefficiencyandoptimizesperformanceandmakesbetteruseoftheavailableinfrastructure.

FromReactivetoPredictiveModels:Oneofthemostsigni?cantcontributionsofAItoRANisitsabilitytoconvertnetworksthatmerelyreactintothosethatcanpredict.Traditionally,networkmanagementrespondstoissuesonlyaftertheyoccur.AIchangesthisparadigmasitspredictivecapabilitiesallownetworkstoanticipateproblemsandtakepreventiveaction.Machinelearningalgorithmscananalyzevastamountsofnetworkdatatoidentifypatternsandpredictpotentialfaultsorcongestionpointsbeforetheyimpactservicequality.Thisnotonlyimprovesreliabilitybutalsohelpsminimizedowntimeandoperationalcosts.

UnlockingRevenuePotential:AI-nativenetworksaretransformingRANassetsfromtraditionalcostcentersintorevenue-generatingcenters.GenerativeAIintroducesnewuserexperiencesbyprovidingEdgeAIinferencinganddynamicresourceallocation,leadingtobetterservicequalityandhighercustomersatisfaction.GenerativeAIalsoidenti?esopportunitiesformonetizingnetworkcapabilities,suchasofferingpremiumservices,targetedadvertising,andedgecomputingsolutions.ThisshiftnotonlymaximizesthevalueofRANinvestmentsbutalsopositionsnetworksasstrategicassetsdrivingpro?tability.

AI-RANisthusattheforefrontofamoreproactiveandefficientnetworkmanagementapproach,transformingRANintoakeyenablerofintelligentandautonomousnetworkservices.

2.3AI-RANDefinitions

AI-RANreferstotheapplicationofarti?cialintelligencetechnologytotheRAN.Itaimstoimprovemobilenetworkefficiencyandoptimizepowerconsumption,whileenhancingtheutilizationoftheexistinginfrastructure.TheconceptinvolveshostingbothAIapplicationsandvirtualRAN(vRAN)softwareonthe

10

AI-RAN:TelecomInfrastructurefortheAgeofAI

sameinfrastructure,allowingtelecomoperatorstogeneraterevenuefrombothnetworkaccessandAIserviceswithasinglecapitalinvestment.

TheAI-RANAlliancehasestablishedthreeitemstoaddressdifferentaspectsofAIintegrationinRAN:

AI-for-RAN:FocusesonusingAItoenhanceRANperformance.ItexploreshowAIcanimproveoperationefficiency,boostcapacity,andachievekeyperformancetargetsintheradioaccessnetwork.

AI-and-RAN:InvestigateshowtousethesameinfrastructuretorunbothRANworkloadsandAIworkloadssimultaneously.ThegoalistoincreaseresourceutilizationandopenupnewrevenuestreamsfortelcosbyhostingvariousAIapplicationsonthesameplatformsthatrunnetworkfunctions.

AI-on-RAN:AddressessolutionsforrunningAIapplicationsontheradioaccessnetwork.ItfocusesonenhancingRANtoensureitcanhandletheincreasingdemandsofAIandgenerativeAIapplicationswithoutcompromisingkeyfactorslikelatencyandsecurity.

Figure3.ThreeitemstoaddressdifferentaspectsofAIintegrationinRAN

TheseitemscollectivelyaimtointegrateAIintothefabricoftheradioaccessnetwork,transformingnetworksintoself-organizing,self-optimizing,andself-managingsystemsthatcanhandlereal-timechanges,anticipatemaintenanceneeds,andmoreefficientlymanageresources.

3.HistoryofSoftBank'sAI-RANR&D

3.1EarlyResearchandAI-RANDevelopment

SoftBankcontinuestoexplorenewwaystocreatevaluebyintegratingtraditionaltelecominfrastructure

11

AI-RAN:TelecomInfrastructurefortheAgeofAI

withAIamidsttherapidinnovationinAItechnology.Recognizingthatthefullperformanceof5Gremainsunrealizedsinceitsintroduction,SoftBankbeganeffortstoenhance5GthroughAIandMachineLearning(ML).

SoftBankisleadingthedevelopmentofAI-RAN,anewarchitecturethatintegratesAIapplicationsandsoftware-basedRANintoasinglecomputer.AI-RANenhancesthecapabilitiesandqualityofRANwhilealsoprovidingasharedcomputingplatformforAIapplicationsacrossvariousindustries.SoftBankaimstodeployAI-RANequipmentinAIdatacentersdistributedthroughoutJapan,directlyconnectingSoftBankbasestationstotheseAIdatacenterstooffersecureandlow-latencyAIservices.

3.2ApplicationsofSoftBankAI-RANResearch

Inthepast,theprimarystrategyforachievinghigh-speed,high-capacitywirelesscommunicationwastoincreasethefrequencybandsused,asevidencedbythetransitionfrom3GtoLTEand5G.However,theemergenceofAI-RAN,whichcanenhanceuserexperiencewithoututilizingmorefrequencybands,holdsgreatpotentialforeffectivelyutilizingthe?nitepublicresourceoftheradiospectrum.Withthetransitionfrom5Gto6Gnetworksapproaching,theimportanceofAI-RANisexpectedtostrengthenevenmore.

TheAI-RANdatacenterbeingdevelopedbySoftBankwillallowboth"RANoperations"and"AIapplications"torunsimultaneouslyonthesameserver.Thisadvancementenablestelecomoperatorstosecuretworevenuestreams—RANandAI—withasinglecapitalinvestment.Moreover,byintegratingdifferentservices,operatorscanimprovetheoperationalefficiencyoftheirinfrastructure.Consequently,AI-RANholdsthepotentialtosigni?cantlyimprovethereturnoncapitalinvestmentfortelecomoperators.

CaseStudy:ApplicationofAIforchannelinterpolationinlowerlayersofwirelesscommunication

Indenseenvironmentswithmultiplebasestationsandterminals,radiosignalsareoftendistortedbymultipathfading.Asaresult,conventionalsignalprocessingtechnologymayfailtoaccuratelyestimatewirelesscharacteristics,leadingtolowerthroughput.

Toaddressthis,weappliedAI-nativesuper-resolutiontechnology,originallyusedinimageanalysis,toradiosignalprocessing.SimulationswereconductedtoevaluatethepotentialuplinkthroughputimprovementsbyreconstructingdegradedsignalsusingAI.AftertrainingtheAImodelwithsimulatedradiosignaldatabasedonreal-worldenvironmentalconditionsandtestingitwithuplinksignals,a30%improvementinuplinkthroughputcomparedtoconventionalsignalprocessingtechnologywasobserved(Figure4).

12

AI-RAN:TelecomInfrastructurefortheAgeofAI

Figure4.Comparisonresultofuplinksignals:30%throughputgain

3.3PartnershipsandCollaboration

Figure5.AI-RANAlliancelaunchceremonyatGSMAMobileWorldCongressBarcelona2024

SoftBankisacceleratingthedevelopmentofAI-RANthroughitspartnershipwithNVIDIAandotherindustryleaders,havingbegunthedevelopmentofAI-RANsolutionsonnewhardwaresuchastheNVIDIAGraceHopper200Superchip(GH200),whichiscurrentlyevolvingintotheNVIDIAGraceBlackwellplatform.

TopromotethewidespreadadoptionanddevelopmentofAI-RANtechnology,SoftBankhaspartneredwithindustryleadersincludingNVIDIA,Arm,T-Mobile,Ericsson,Nokia,andSamsungtoestablishtheAI-

13

AI-RAN:TelecomInfrastructurefortheAgeofAI

RANAlliance

2

.SinceitslaunchatGSMAMobileWorldCongressBarcelona2024,thealliancehasgrownto58members(asofDecember2024),encompassingadiversemixoftelecomoperators,semiconductorcompanies,andacademicinstitutionsunitedbythemissionofadvancingRANperformanceandcapabilitiesthroughAIinnovation.

SoftBankbelievesthatAI-RANhasthepotentialtobecomethetechnologythatwillsigni?cantlyimpactnotonlythetelecomindustrybutalsosocietyasawhole.Withtheimminenttransitionfrom5Gto6Gnetworks,theimportanceofAI-RANisundeniable.SoftBankwillcontinuetofocusoncontinuingAI-RANdevelopmenttoinitiateanewparadigmshiftforthe6Gera.

SoftBank'sAI-RANR&Deffortshavethepotentialtorevolutionizetelecomnetworksandcreatenewbusinessopportunitiesacrossvariousindustries.

4.gRAN:GPU-basedAI-RANArchitecture

gRAN,atermintroducedbySoftBank,standsforGPU-basedRANthatoffersanarchitecturefordeployingAI-RAN,consideredthedesirableevolutionarystageofRANfollowingvRAN,cRAN,andO-RAN.TheintroductionofgRANmarksasigni?canttechnologicalleapintheevolutionoftheradioaccessnetwork.ByleveragingthepowerofGPUsinadditiontoCPUs,gRANenhancestheefficiency,scalability,and?exibilityofRANinfrastructures;itsupportsadvancedAI-nativefunctionsandmeetstheever-growingdemandsofAIapplicationsandmoderntelecomnetworks.

4.1KeyCharacteristicsofgRAN

ThetransitionfromtraditionalRANarchitecturestosoftware-drivenapproacheswithhigherperformancehaspavedthewayforgRAN.AsRANbecomesvirtualizedandopen,andmostimportantlysoftware-de?ned,itlaysthefoundationforportingRANoverGPU-basedacceleratedinfrastructure,andbringinganewlevelofcomputationalpowerandefficiencywithgRAN.

4.1.1WhyGPUsforvRANEvolution?

GPUsarewell-suitedforhandlingthehighlyparallelprocessingworkloadscommoninmodernRANenvironments.UnlikeCPUs,whichareoptimizedforserialprocessing,GPUsexcelatexecutingintensivematrixcalculationsandmultipletaskssimultaneously,makingthemidealforreal-timeRANdataandsignalprocessing.Thisparallelismisparticularlyimportantinhandlingthecomplexalgorithmsrequiredfor5Gandfuture6Gtechnologies,suchasmassiveMIMO,beamforming,and

2FormoredetailsabouttheAI-RANAlliance’smissionandinitiatives,refertotheirwhitepaperat

/wp-content/uploads/2024/12/AI-

RAN_Alliance_Whitepaper.pdf

14

AI-RAN:TelecomInfrastructurefortheAgeofAI

energyreduction.UtilizingGPUsinadditiontoCPUsallowstelecomoperatorstohandletheseworkloadsmoreefficiently,reducelatency,andimproveoverallnetworkperformanceandefficiency.

4.1.2TheTechnologicalLeapwithGPU-basedvRAN(gRAN)

GPU-basedRANenablesmodernAIserviceslikeLLMinferencingbyprovidingthecomputationalpowerneededforreal-timedataprocessinganddecision-makingattheedge.ThisallowsRANtohandlecomplexAIworkloadsefficiently,reducinglatencyandenhancingresponsiveness.Additionally,GPUsfacilitatedynamicresourceoptimization,suchasadvancedSelf-OrganizingNetwork(SON),byenablingrapidanalysisandadaptationofnetworkresourcestochangingdemands,ensuringoptimalperformanceandreliabilityinRANenvironments.TheseattributesallowgRANyieldthemoredynamicandresponsivenetworkscrucialforsupportingemergingusecasessuchasGenerativeAI/LLMinferencing,augmentedreality(AR),virtualreality(VR),andotherdata-intensiveapplications.

4.2TheArchitectureofgRAN-basedAI-RAN

ThearchitectureofgRANconsistsofseveralkeycomponentsthatworktogethertocreateahighlyprogrammable,intelligent,andhighperformingnetworkenvironment.ThecoreelementsofgRANaretheRadioUnit(RU),DistributedUnit(DU),CentralizedUnit(CU),theintegrationofAIcapabilities,andamulti-tenantanddynamicorchestrator.

RadioUnit(RU):TheRUhandlestheradiofrequency(RF)signals,convertingthembetweenanaloganddigitalformats.Itisresponsibleforcommunicatingwithuserdevicesandservesasthemobileuser’sentrypointtotheRAN.

DistributedUnit(DU):TheDUisresponsibleforlower-layerprocessing,includingreal-timetaskslikescheduling,beamforming,and

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,請聯系上傳者。文件的所有權益歸上傳用戶所有。
  • 3. 本站RAR壓縮包中若帶圖紙,網頁內容里面會有圖紙預覽,若沒有圖紙預覽就沒有圖紙。
  • 4. 未經權益所有人同意不得將文件中的內容挪作商業或盈利用途。
  • 5. 人人文庫網僅提供信息存儲空間,僅對用戶上傳內容的表現方式做保護處理,對用戶上傳分享的文檔內容本身不做任何修改或編輯,并不能對任何下載內容負責。
  • 6. 下載文件中如有侵權或不適當內容,請與我們聯系,我們立即糾正。
  • 7. 本站不保證下載資源的準確性、安全性和完整性, 同時也不承擔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