1、1/64Contents1.Preface.22.Progress of Relevant Standard Organizations.32.1.3GPP.42.2.IMT-2030 Promotion Group.52.3.Next G Alliance.52.4.Hexa-X.62.5.6GANA.73.6G Data Service Scenarios and Requirement Analysis.73.1.Sensing Data.73.2.AI Model Training Data.103.3.AI Model Data.113.4.Data Services Based o

2、n DistributedArchitecture.143.5.User-Centered Data Management and Control.153.6.Data in Integrated Satellite-Terrestrial Communication Scenarios.173.7.SOE Data Acquisition.183.8.Self-Generated Data Services.203.9.Data Benefits.213.10.Summary of 6G Data Requirements.234.Definition and Framework of th

3、e 6G Data Plane.244.1.Definition.244.2.Necessity of the Data Plane.264.3.6G Data Plane Framework.275.Key Technologies of 6G Data Plane.345.1.Data Bearer and Transmission Protocol.345.2.Air Interface Data Plane.365.3.Core Network Data Plane Functions and Architecture.395.4.Core Network Data Plane Tra

4、nsmission Protocol.405.5.AI Model Data Compression Technology.425.6.Distributed Data Technology.445.7.Service Interface Technology Based on Semantic Graph.465.8.Coordination between Data Services and Other Services.526.Data Plane Prototype.546.1.Data Plane Prototype 1.546.2.Data Plane Prototype 2.55

5、7.Summary and Outlook.578.References.599.Abbreviations.6110.Contributors to the White Paper.642/641.Introduction5Gsupportsthreescenarios:enhancedMobileBroadband(eMBB),Ultra-ReliableLow-Latency Communications(URLLC),and massive Machine Type Communication(mMTC).Exposure of capabilities and events is s

6、upported through the network exposure function(NEF)orcommon API framework(CAPIF).Application functions(AFs)can obtain the data of the 5Gsystem via NEF/CAPIF,which promotes cross-layer innovation between the network and AF.Asthe volume of data generated by mobile networks continues to grow,the need f

7、or data collectionand data processing has become increasingly prominent.Correspondingly network services haveevolved from communication service to multidimensional services of communication,sensing,and artificial intelligence(AI).Traditional methods of data collection and data processing are notadap

8、ted to new changes,and it is difficult to meet the additional data requirements.This isbecause 5G networks do not have a unified data management framework.As a result,multipletypes of data cannot be integrated or coordinately managed,which may increase the complexityand costs of data governance.Focu

9、sing on data pipelines,5G networks do not fully explore orutilize the value of data.Based on the existing methods,data collection and data processingrequirements of various types of data cannot be met.The data value of communication,sensingand AI cannot be explored.Moreover,rights and interests of d

10、ata cannot be guaranteed.There isnot a unified data management and control framework for data quality management to cannotguarantee the legality,authenticity,and integrity of data.Therefore,5G networks cannot meet therequirements of data regulatory the expectation of privacy and security of users.In

11、 the aspects ofnew services,it is difficult for 5G networks to efficiently collect,transmit,process and analyzelarge amounts of data in mobile networks,such as sensing data,AI model,etc.Figure 1-1 Life Cycle Management of Data in Mobile Network3/64Data has become one of the production factors in the

12、 digital society.6G is an importantinfrastructure of digital society.The data of the 6G system is inevitably an important part of thisproduction factor.The Framework and overall objectives of the future development of IMT for2030 and beyond 1,released by International Telecommunication Union-Radio C

13、ommunicationSector(ITU-R),proposes six major scenarios,including Immersive Communication,MassiveCommunication,Hyper Reliable and Low-Latency Communication,Ubiquitous Connectivity,AIand Communication(AIAC),and Integrated Sensing and Communication(ISAC).This meansthat 6G is a mobile communication syst

14、em that goes beyond communication services.In additionto traditional data transmission pipelines,6G introduces internal data of mobile network,such assensing data in ISAC and AI data in AIAC,etc.Data providers or data consumers of mobilenetwork internal data include user equipment(UE),radio access n

15、etwork(RAN)nodes,andnetwork functions(NFs)of core network(CN)and AFs.In contrast to user data in traditional datatransmission pipelines,the 6G system is tasked with managing the entire life cycle of internal data,including generation,security and privacy,coordination of collection,transmission,proce

16、ssing,quality management and data service.Therefore,a data plane2 is introduced to enable a unified and efficient life cyclemanagement of internal data of the 6G network.For example,the data plane provides datarequired by the sensing function or NEF,so that the 6G system can provide the sensing serv

17、ice ornetwork exposure services.The 6G Data Plane(DP),which operates parallel to the control plane(CP)and user plane(UP),is not constrained by the transmission requirements of signaling or userdata.This allows the functions and configurations of DP protocol can be optimized to provide abetter soluti

18、on to the aforementioned needs.This avoids fragmented solutions for individual usecases.The DP will provide a unified data management framework for the 6G system,enablingdata capabilities.As a result,the 6G system will be able to integrate and manage multiple types ofdata,ensure data security and tr

19、ustworthiness,break data silos,improve the efficiency of datagovernance,protect rights and interests of data.In this White Paper,contributors illustrate ourinitial views and latest achievements on the scenarios and requirements of 6G data service,thedefinition and framework of DP,key technologies,an

20、d prototypes,hopefully contributing to 6Gdevelopment.2.Progress of Relevant Standard OrganizationsThis chapter describes the 6G DP research progress of multiple standard organizations andresearch institutions around the world,including 3rd Generation Partnership Project(3GPP),IMT-2030(6G)Promotion G

21、roup,Next G Alliance,Hexa-X,and 6G Alliance of Network AI(6GANA).These standard organizations have done researches on the data collection requirements,challenges,solutions,and technology trends of 6G.4/642.1.3GPPBased on data requirements of each use case,several standards of 5G networks have beenst

22、udied and standardized to support data collection and data analysis.For example,3 is used tocollect data for the network data analytics function(NWDAF),which is the AI function of CN.Inaddition,there are standards of NEF/CAPI45 for data collection of network exposure.Thereare LTE Positioning Protoco

23、l(LPP)6 and NR Positioning Protocol A(NRPPa),which are used totransmit positioning data for the location management function(LMF).There are minimization ofdrive test(MDT)7 and quality of experience(QoE)used for wireless network optimization andmanagement.To meet data collection requirements of the N

24、WDAF,Release 17 introduces the DataCollection Coordination Function(DCCF)to the 5G network to collect data from NFs anddistribute results requested by NFs 2.The DCCF prevents data providers,such as the Access andMobility Management Function(AMF)and Session Management Function(SMF)from handlingmultip

25、le subscriptions to the same data and sending multiple notifications containing the sameinformation due to no coordination between data consumers.Except the NWDAF,5G CN NFs,such as the AMF and SMF,as main network elements(NEs)of the communication network,mainly provide communication services instead

26、 of data.However,the NWDAF generally needs toobtain a large amount of data for big data analysis.Repeated reporting of a large amount ofidentical data decreases the performance of 5G CN NFs.The NWDAF can subscribe orunsubscribe data from the DCCF through the Ndccf interface.If the DCCF has not previ

27、ouslycollected the data requested by the NWDAF,it can use the service-based interface(SBI)to collectthe data from NFs or collect data through the messaging framework.Subsequently the data can betransmitted to the NWDAF through either SBI or messaging framework.For the optimization and management of

28、RAN,the network management function can sendan MDT or QoE request to the RAN node,which triggers the RAN node to configure MDT orQoE data collection to the UEs.The MDT or QoE data reported by the UE is sent to the tracecollection entity(TCE)or measurement collection entity(MCE)for analysis through t

29、he servicingRAN node.The network management function optimizes configurations of RAN based on theself-organizing network(SON).The LPP is used to exchange positioning control information and positioning data betweenthe UE and the network.Generally,the amount of positioning data is not large.Therefore

30、,the LPPis carried on the CP protocol stack.In other words,the positioning protocol stack consists of theLPP,non-access stratum(NAS),Radio Resource Control(RRC),Packet Data ConvergenceProtocol(PDCP),Radio Link Control(RLC),Medium Access Control(MAC),and PhysicalLayer(PHY).Based on positioning measur

31、ement data transmitted by the LPP,the LMF estimatesthe location information of UEs and provides it to 5G NFs orAFs.5/642.2.IMT-2030 Promotion GroupIn 6G Network Architecture Vision8,IMT-2030 Promotion Group proposed that the 6Gnetwork architecture requires a kind of data functions which are differen

32、t from the traditional userplane.The data functions systematically address the challenges of managing and monetizingnon-traditional user plane data in the 6G network.From the perspective of NF layer,6G datafunctions consist of data orchestration and control,data processing and forwarding,and datasto

33、rage.Based on the capability reported by data processing and forwarding nodes andrequirements of data service,the data orchestration and control node selects the data source nodes,processing and forwarding nodes.Based on orchestration,the selected nodes form a data bearer toprovide data services.The

34、 data processing and forwarding node provides data services as required,such as data collection,data preprocessing,data storage,data privacy protection,security andtrustworthiness,data analysis,data sharing,and data forwarding.The standardization of data lifecycle management can improve data compara

35、bility and reusability,which includes processes suchas data privacy,security and trustworthiness,data coordination,generation,and collection,datastorage,data transmission,data processing,data service,and data quality management,rights andinterests management,etc.In 6G Wireless System Design Principl

36、es and Typical Features9,IMT-2030 PromotionGroup proposed that efficient data governance,a unified data collection method,and data lifecycle management should be considered at the beginning of 6G system design.Native data is oneof the principles for 6G wireless system design.The design of native dat

37、a should ensure datasecurity and privacy,improve efficiency of data collection,transmission,and storage.The designof native data should also enhance data sharing and reusability.Native data builds an open andunified data life cycle standard to support all processes of ubiquitous heterogeneous dataci

38、rculation.With cost-effective,efficient,and reliable data services,it enables 6G networks andassociated industries to process and analyze various types of data accurately and systematically,soas to make better decisions and provide higher quality services.In 6G Data Service Architecture Research10,I

39、MT-2030 Promotion Group proposed 6Gdata services and the data plane architecture of the 6G network,and described the function oforchestration and control,processing functions and so on.In 5G,there are mature protocol stacksof CP and UP to support commercial deployment and applications.The existing p

40、rotocol stacksdecouple functions between planes,and achieve modular,virtualized,and software-based functionmanagement.The 6G data plane also requires complete,flexible,and scalable protocol stacks tosupport data forwarding and control,data service exposure,data agent management and control,etc.2.3.N

41、ext GAlliance6/64In the 6G Technologies for Wide-Area Cloud Evolution11,the Next G Alliance proposedthat in addition to communication services,computing and data planes with dedicated computingand data management functions may be introduced to cellular networks.The 6G Technologies forWide-Area Cloud

42、 Evolution 11 drives network architecture design and introduces new functionsin terms of 6G network computing services to support distributed computing and tightinterworking between the distributed cloud and 3GPP protocols.Therefore,changes in the control,management,and data planes need to be studie

43、d to adapt to the distributed computing process.According to the report content,the data plane is similar to the user plane of existing protocols.Itis possible to meet computing service requirements by enhancing the data bearer.In the AI-native Wireless Networks12,the Next G Alliance proposed that c

44、ompared with theapplications of AI/machine learning(ML)in 5G,AI/ML in 6G should have the following features:Data collection will be at various layers within the network.In addition to near-real-time mannerof AI/ML applications of 5G,AI/ML applications for 6G will operate in the real-time manner.AI/M

45、L will be blended into the design of 6G and the transceivers may be designed to be AI-nativeat the beginning of 6G.Therefore,it is necessary that AI/ML be entrenched in the design of radiolayers with interfacing to AI and data-collection frameworks.These interactions need a strongemphasis on securit

46、y and privacy.This AI native method ensures rapid evolution of wirelesstechnologies,which may be partial independent of standards cycles.2.4.Hexa-XIn the Draft foundation for 6G system design13,Hexa-X proposed that the data collectionand AI frameworks are pervasive functionalities of the 6G system.T

47、he data collection frameworksupport several different types of data and information to be collected from multiple domains andlayers of the network and moved within the network for analysis.Data transmission andintegration will take place considering privacy and ownership concerns of all stakeholders

48、.Coupling data across applications and networks will provide the opportunity to improve thenetwork performance or enable network aware applications.The data collection frameworkcollects the data required by the network management function and supports real-time control andoperations to process requi

49、red data.In the Foundation of overall 6G system design and preliminary evaluation results14,Hexa-X proposed that AI needs novel architectural elements that enable for privacy aware datacollection and learning.Data-driven network control units and UE aggregation units areintroduced to support data sh

50、aring between the network and UEs,so as to improve control anddecisions.Based on data privacy requirements,UE-oriented data collection and learning enableUEs to take advantage of network information along with on-device contextual information(useractivity,intent,and usage patterns)to assist the netw

51、ork in connectivity decisions,i.e.,to improvethe connectivity QoE.7/642.5.6GANAIn the 6G Data Service-Concept and Requirement15,6GANA pointed out that the 6Gnetwork provides new capabilities,such as native AI,native sensing,and native security in thecontext of integrated communication,sensing,and co

52、mputing.Based on the informationtransmission capabilities of traditional mobile communication networks,the new capabilitiesenhance the data production and consumption capabilities of the network and make the 6Gnetwork a platform for information and data circulation.Efficient management of data carri

53、ed onthe 6G network is a key technology of the 6G network.Therefore,based on the trusted dataserviceframeworkproposedintheWhitePaperon6GDataServiceConceptandRequirements15,the 6G network will introduce an independent data plane.The 6G data planebuilds architecture-level unified and trusted data serv

54、ices to clarify the data source,description,collection,processing,storage,application,and privacy protection.3.6G Data Service Scenarios and RequirementAnalysis6G data services indicate that the 6G system provides data resources to internal functions,such as the UE,RAN,and CN.The collected data can

55、be used by network capability exposurefunctions(such as the NEF or CAPIF)to provide data network capabilities or event to externalfunctions of the 6G system,such as AFs.In addition,the collected data can be used by the sensingfunction to provide sensing services to AFs or NFs,or be used by the AI fu

56、nctions to provideassistance information for the control and optimization of networks.With data as the core element,6G data services aim to fully utilize the value of 6G data,so asto break through the boundaries of single-dimensional mobile services and promote integratedinnovation of services.The d

57、ata of 6G networks indicate the data generated or obtained by the 6Gsystem instead of the data in the traditional user plane.It includes the data generated by the UEs,RAN,CN,and network management functions during the operations of communication services.It also includes sensing data(e.g.,sensing me

58、asurements)and AI data(e.g.,AI model)generatedby UEs,RAN,CN and network management functions during the operations of new services(e.g.,sensing services,AI services).The shareable data obtained by the 6G networks from third partiesis also included,such as various types of sensor information(includin

59、g temperature,humidity,andenvironment)and geographic information system(GIS)information.This chapter describesscenarios of 6G data service and analyzes potential requirements.3.1.Sensing Data3.1.1.Description8/64Sensing data in the 6G system is the data describing the physical world status obtainedt

60、hrough radio waves or other sensors during physical world exploration.Sensing data mainlyincludes native sensing data,external sensing data,and multi-modal integrated sensing data.Native sensing dataThe 6G mobile communication system has a sensing function,that is,native sensing.Nativesensing data i

61、s the data generated during native sensing of the 6G mobile communication system.It includes native sensing measurement data obtained by the UE or RAN node based on airinterface signal measurement,and native sensing results.Generally,the native sensingmeasurement data includes two types.One is the d

62、ata obtained during sensing measurement toassist in data communication,such as the data measured for sensing the channel environment.Theother is the data measured purely for sensing the target object or environment,such as the signalstrength and arrival time measured during positioning,and the RF si

63、gnal data measured duringranging,speed measurement,and imaging.Through native sensing,the 6G system obtains thesensing results,such as the position,speed,and imaging of the sensing target.External sensing dataExternal sensing data is the data that the 6G system obtains from external third-party IoTs

64、ensors or GIS,which includes intermediate measurement data and sensing results.The data is notinvisible data that is transparently transmitted on the network.It needs to be processed by thesensing module of the 6G system to generate sensing measurement intermediate data or evenusable sensing results

65、.For example,the 6G system connects to automatic recognition equipment,such as barcode recognition,image recognition,and radio frequency identification(RFID)toobtain target information,and connects to various types of sensors to obtain physical information,such as biomass,chemical,heat,pressure,temp

66、erature,sound,light,electricity,and vibrationinformation.The sensors mainly include mechanical sensors(such as the displacement sensor andlevel sensor),geometric sensors,force sensors(such as the pressure sensor and speed sensor),thermal sensors(such as the temperature sensor),optical sensors(such a

67、s the image sensor andinfrared/ultraviolet sensor),electromagnetic sensors(such as the electric field sensor and voltagesensor),acoustic sensors(such as the sound surface wave sensor and ultrasonic sensor),raysensors,humidity sensors,gas sensors(such as the gas composition sensor and gas concentrati

68、onsensor),ion sensors(such as the PH sensors),physiological sensors,and biochemical sensors.Multi-modal integrated sensing dataCompared with native or external sensing data,multi-modal integrated sensing data focuseson integrated processing of these two types of data.In the 6G system,multi-modal sen

69、sing dataincludes native and external sensing data,which can be processed separately or generate newsensing data or sensing results through integrated sensing.Multi-modal integrated sensing9/64incorporates native sensing and external sensing.It integrates data from multiple sensing channelsfor under

70、standing and processing.Various types of sensing data collaborate,complement,revise,and enhance each other to generate better sensing results than single types of sensing data.Forexample,in an indoor environment covered by the land mobile communication system andInternet of Things(IoT)terminals,cell

71、ular positioning data can be used together with data ofexternal IoT sensing devices,such as Wi-Fi,Bluetooth,Zigbee,and UWB positioning data togenerate multi-modal integrated sensing data.3.1.2.Potential RequirementsIn the 6G system,various types of sensing are available,and the sensing data is exten

72、sive.The new and optimized sensing function causes a huge amount of data in the land mobilecommunication system.In addition to native sensing,the 6G system needs to process and transmitexternal sensing data and multi-modal integrated sensing data.Specifically,the functional requirements on the 6G sy

73、stem are as follows:1)Data collection:During native sensing,the 6G system uses signaling to measureparameters transmitted through radio waves and RF channels and collects sensing datafrom various types of IoT sensing devices over different interfaces.2)Data processing:The 6G system processes the col

74、lected sensing data.Through AImodel training,computing,and other processing methods for sensing data obtainedfrom various sensing channels,usable sensing results are obtained.The usable sensingresults assist in decision making and execution of communication or other functions inthe 6G system,or are

75、used as input parameters to assist some applications.Duringmulti-modalintegratedsensing,datafrommultiplesensingchannelscanbecollaboratively processed to ensure better sensing performance.3)Data transmission:In the 6G system,massive volume of sensing data needs to betransmitted.The intermediate measu

76、rement data needs to be transmitted to network orcomputing nodes for integration and calculation,so as to obtain sensing data or sensingresults.For example,during cellular positioning,measurement data,including thesignal strength and signal arrival time between a UE and multiple RAN nodes needs tobe

77、 transmitted and integrated to obtain the location information of the UE.The performance requirements on the 6G system are as follows:1)Due to massive volume of sensing data in the 6G system and different sensing datarequirements of various types of applications,the 6G communication system needs toh

78、ave good transmission performance,including the bandwidth and latency.Accordingto Study on Scenarios and Requirements of 5G-Advanced Integrated Sensing andCommunication16,the sensing data rate is about 1 kbps to 10 Mbps.2)To calculate and make decisions on massive volume of sensing data,the 6G syste

79、mneeds to have strong computing power and better performance for AI models and10/64various algorithms.3)Sensing data involves a large amount of user privacy because it comes from physicalstatus information about people,device,and environment related to ToB or ToC users.The 6G system needs to meet ap

80、plication requirements and ensure user security andprivacy when performing sensing.3.2.AI Model Training Data3.2.1.DescriptionOn the 5G CN,the NWDAF is mainly used for prediction,such as UE location predictionand NE load prediction in AI analysis scenarios,such as data and model training and modelin

81、ference.In such scenarios,a small-scale neural network model,such as the recurrent neuralnetwork(RNN),convolutional neural network(CNN),or long short-term memory(LSTM)is used.Generally,the model contains thousands to hundreds of thousands of parameters.The MTLFneeds to collect tens to hundreds of me

82、gabytes of model training data to train a model.The 6Gnetwork has more extensive AI analysis scenarios,such as object recognition and trajectoryplanning based on sensing data and network policy control based on network data.Driven bydiverse scenarios,the 6G network requires larger-scale and more com

83、plex models for inferenceand analysis.Academic research results show that the amount of data required to train a model isproportional to the model size.To train a model with 100 million model parameters,at least 8 GBdata is required for pre-training,and other specific service data is required for mo

84、del fine-tuning.To facilitate further clarification,this section uses positioning as an example to explain thescenarios and requirements of AI model training data.More and more service use cases requirehigh-precision positioning.However,it is difficult to achieve high-precision positioning in manyap

85、plication scenarios due to issues,such as multipath,Non-Line-of-Sight(NLOS),and indoorcoverage and synchronization.Traditional positioning methods cannot achieve the requiredpositioning accuracy.AI positioning is required to effectively utilize UE information in multipledimensions,such as frequency,

86、time,and spatial domains and discover other information hidden inchannels,thereby improving the positioning performance in NLOS or synchronization errorscenarios.To achieve this,the network needs to obtain UE-related data from the UE or RAN node,such as channel measurement information in time domain

87、 and channel impulse response(CIR)formodel training of the AI positioning service.Massive volume of training data is required for AI model training.Traditional data reportingthrough the UE signaling plane or user plane is not suitable for the training data.This is becausetransmission of a large amou

88、nt of training data occupies many bandwidth resources,training datadoes not need to be transmitted in real time,and training data must be fully reused,that is,trainingdata may be reused in multiple scenarios and even AI model training of various services.11/64Therefore,the 6G system needs to collect

89、 a large amount of training data based on the data planeto conduct model training for the AI positioning service.3.2.2.Potential RequirementsIn the existing 5G system,CN NEs can collect UE data through the signaling or user plane.Data collection through the signaling plane indicates that the UE repo

90、rts data to the network usingNAS signaling,RRC signaling,or other methods.Data collection through the user plane indicatesthat the data collection application function(DCAF)collects data from the UE based on the AFevent exposure(EVEX)and sends it to the network.The 6G system needs to support thefoll

91、owing functions:1)Collects UE data as network AI model training data(such as the CN AI positioningmodel).2)Transmits UE training data in offline mode.3)Reuses the collected training data to train multiple AI models of a single service or AImodels of multiple services.3.3.AI Model Data3.3.1.Descripti

92、onAs a core component in the AI field,AI models have achieved great success in academicresearch and had a significant impact on actual applications.AI models are trained with input databased on a learning algorithm to learn patterns and make predictions or decisions.AI modeltraining aims to generali

93、ze the models to unseen data to adapt to diverse tasks and achieveintelligent functions.AI models are widely applied in image processing and natural languageprocessing(NLP).(1)Image processing:In the 1990s,classic computer vision methods,such as edge detectionand color histogram,dominated the image

94、processing field.However,these methods havelimitations in processing complex scenarios and abstract features.In the early 2010s,large-scaledatasets,such as ImageNet,emerged and the GPU performance improved greatly.The era of deeplearning arrived,and deep neural network models developed significantly

95、.Based on LeNet-5,AlexNet won the ImageNet Large Scale Visual Recognition Challenge(ILSVRC)with a deepnetwork structure and large-scale GPU parallel computing.This marked the rise of deepconvolutional neural networks(CNN)in the image processing field.Subsequently,a large numberof CNN structures emer

96、ged,such as VGGNet,GoogLeNet,ResNet,MobileNet,and DenseNet.They have excellent performance in visual fields,such as image recognition and target detection.12/64Leveraging depthwise separable convolution,linear activation,global average pooling(GAP),andother technologies,MobileNet significantly reduc

97、es the model size and computational complexityand ensures model accuracy.Therefore,it is suitable for real-time image recognition andclassification tasks of mobile devices and embedded systems.In the 2020s,self-supervisedlearning,large-scale pre-training,and cross-modal learning emerged.Researchers

98、began to uselarger-scale data and more complex tasks to pre-train models and fine-tune specific tasks to ensurebetter performance ofAI models,such as CLIP and DALL-E.(2)Natural language processing(NLP):Before 2010,the N-gram model and hidden Markovmodel(HMM)were the two most important AI models in t

99、he NLP field.They used statisticallanguage modeling but were restricted by limited context.With better word embedding,two majorneural networks,that is,RNN and LSTM better capture semantic information and improve theperformance of AI models in the deep learning era.At the end of the 2010s,the Transfo

100、rmermodel used the attention mechanism to enhance long text processing.In addition,two large-scalepre-training models BERT and GPT have emerged,which significantly improved the performanceofAI models in multiple NLP tasks.Since the 5G era,3GPP standard organizations have introduced AI functions in t

101、he CN,network management,and access network domains.With version iteration,the AI functions covermore and more scenario cases and communication function entities.Use the NWDAF used for theCN AI function as an example.In 3GPP standards,about 10 use cases are defined,which describestatistical and pred

102、iction methods for user service experience,terminal exception,networkperformance,and user behavior3.Starting from Release 18,the access network domain hasbegun to study AI-based air interface feature improvement.Identified use cases with gains includeAI-based CSI feedback,beam management,and positio

103、ning17.Although AI model technologies develop rapidly,there are still many challenges for applyingthem in practice.One major challenge is collaborative AI model training with low overheads andprotected privacy.UEs and IoT terminals generate a large amount of data related to user features,and the dat

104、a can effectively drive AI model training.However,centralized data collection isrestricted due to privacy and security issues.Federated learning(FL)is a distributed AI modeltraining framework that allows user devices to collaboratively train neural networks withoutsharing data sample.This solves con

105、tradiction between intelligence and privacy.However,privacyprotection of FL has trade-offs.FL training requires repeated transmission of models during thetraining process.AI model convergence generally requires training with hundreds or eventhousands of rounds of communication,and each round involve

106、s several or even dozens of users.Therefore,the communication overhead for training a model is huge.In addition,FL requiresusers to download the global model and upload the updated local model.This two-waycommunication also increases the communication overhead on the UE side.Beside of traffic orpowe

107、r consumption,the huge communication overhead significantly diminishes users enthusiasm13/64for participating in AI model updates.In addition,transmission of a large amount of AI modelsoccupies excessive network bandwidth,affecting user experience of other services.Table 3.3.1compares the communicat

108、ion overheads between traditional centralized learning and FL.Table 3.3.1 Communication Overheads of Traditional Centralized Learning and FL(Parametersare represented using 32-bit floating-point numbers.)Model SizeDataset SizeCentralizedLearningFederated LearningCNN(2.22MB)MNIST(45MB)Uplink：45MBUpli

109、nk:21.7GBDownlink:21.7GB(1000 rounds)CNN(4.08MB)CIFAR-10(147MB)Uplink：147MBUplink:19.9 GBDownlink:19.9GB(500 rounds)LSTM(3.13MB)Shakespear(61.38MB)Uplink：61.4MBUplink:15.3 GBDownlink:15.3GB(500 rounds)The other major challenge is rapid deployment of AI models on mobile devices.Many AImodels with goo

110、d performance have deep network layers.For example,the original ResNet50model has 25 million parameters,and the pretrained language model GPT-3 has 175 billionparameters.Even the lightweight MobileNet model of the minimal size(i.e.,the 0.25MobileNet128 model used for facial attribute classification)

111、,designed specifically for deviceswith limited computing resources and power consumption,has 200,000 parameters.The enormousnumber of model parameters presents several challenges for rapid deployment of AI models onmobile devices:(1)The traffic required for downloading AI models is huge,which increa

112、ses userscosts.(2)Mobile devices usually have weak computing power,hence the AI model inferencelatency is long and service quality is low.3.3.2.Potential RequirementsRequirements for AI model deployment,distributed training,and joint inference include:1)Reduced communication overhead for distributed

113、 AI model training on wirelessnetworks:To achieve this,there are two paths.That is,reduce the communicationoverhead in each epoch and reduce the number of epochs.However,these two pathsusually conflict with each other.For example,sparse and quantized AI models canreduce the communication overhead in

114、 an epoch.However,the errors introduced usuallyslow down AI model convergence and increase the number of epochs required for modelconvergence.To reduce the communication overhead for distributed AI model training,local data distribution of devices,important hyperparameters that affect AI model14/64c

115、onvergence,and AI model compression and device scheduling schemes need to beconsidered.2)Lightweight AImodeldesign:Modelpruning,quantification,weightsharing,knowledge distillation,and other methods can be used to simplify AI models,reduceusers use costs,shorten the AI model inference latency,and imp

116、rove service quality.However,model compression decreases model accuracy.Therefore,the lightweightmodel technologies need to balance the model size and inference accuracy based on thestorage and computing capabilities of mobile devices.3)Cloud-edge AI model inference:The goal is to effectively utiliz

117、e resources of cloud andedge devices and improve the efficiency,response speed,and scalability of the systemwith guaranteed system performance.In this mode,some inference tasks are completedon the cloud,and others are executed on the edge devices.Another task allocationmethod is model splitting.Comp

118、lex models may be split into multiple parts.Some partsare executed on the cloud,and others are executed on the edge devices.To ensureefficient cloud-edge inference of AI models,it is important to determine where theinference tasks are executed.The inference tasks can be allocated based on the taskna

119、ture,real-time requirements,device resources,and other factors.3.4.Data Services Based on DistributedArchitecture3.4.1.DescriptionNew services,such as extended reality(XR)have higher requirements on networkcapabilities,such as edge computing and ultra-low latency.The centralized network architecture

120、has defects,such as complex architecture,large data volume,prone to attacks,and poor datasecurity,which drive the 6G network to evolve into a distributed network architecture.However,the distributed network architecture faces challenges in data exchange and collection betweennetwork nodes.In the dis

121、tributed network architecture,different network nodes are customized for differentscenarios or users.Data scattered on multiple network nodes is of different types,cannot beexchanged,and may be redundant.If the data service model in the existing centralized networkarchitecture is still used,the data

122、 quality and use efficiency may be low.Therefore,it brings newdemand for data service under distributed network architecture.At the same time,data services inthe distributed network architecture also have the following problems:(1)In the distributednetwork architecture,processes such as UE handover

123、and service continuity guarantee involvedata exchange and collection between multiple network functions across network nodes.Theprocesses are complex and have long latency.(2)Data scattered throughout the network may beacquired and collected multiple times.Data from different locations,such as sensi

124、ng data,AI data,15/64network behavior,and status data may eventually be converged in one location and be used bynative AI and digital twins for optimizing the network performance and improving user experienceor be shared by multiple applications.In traditional intelligent analysis and application ex

125、posurescenarios,the NWDAF,NEF,or other network functions obtain data from the AMF,SMF,userplane function(UPF),and other network functions through service-based interfaces(SBIs).Thissignaling transmission and decentralized data collection method may lead to low data collectionefficiency,poor data qua

126、lity,repeated data collection,inconsistent data structures,and other issues.(3)When data is synchronized across network nodes,direct transmission of the original data mayresult in large data volume,long transmission latency,high bandwidth pressure,and even dataprivacy leakage.Therefore,data needs to

127、 be pre-processed or masked before being transmitted.3.4.2.Potential RequirementsWith rapid development of the distributed network architecture and technologies such asnative intelligence and digital twin network,mobile communication data management and controlface new challenges and have potential

128、new requirements on the 6G data plane architecture,datacollection mechanism,and related key technologies.1)The 6G data plane needs a new distributed data service architecture with unified datarepresentation and management mechanisms and on-demand data synchronization andinteroperation,to support eff

129、icient collaboration between different generations ornetwork nodes and solve issues,such as scattered data,heterogeneous data,and dataislands.2)The 6G data plane needs to efficiently obtain distributed data and share it reliably.Toachieve this,the 6G data plane can standardize data collection interf

130、aces,simplify thedata collection process,correctly parse data collection requests,quickly orchestrate datacollection paths,transmit any topology of multiple data sources and data consumers,transmit data between NFs across network nodes at a high rate,perform datapre-processing and in-network process

131、ing based on the intrinsic data relationship,reduce the amount of data transmitted,enhance data privacy protection,and improvedata service quality.3)The 6G data plane needs distributed data storage mechanisms,distributed datatransmission protocols,and distributed data security management technologie

132、s toimprove storage and collection efficiency of massive volume of data,reduce thetransmission latency of large amount of data,and ensure data security and reliabilitythroughout the life cycle.3.5.User-Centered Data Management and Control16/643.5.1.DescriptionIn the recommendation released by ITU-R

133、in 2023,ITU-R proposed that the 6G networkshould provide user-centered communication,user-based service quality,and consistent userexperience in different locations to develop user-centered networks.In addition to users in thetraditional mobile communication network,the users here may be network ser

134、vice consumers,such as network tenants.To meet users personalized requirements or ultimate requirements insome scenarios,the network needs to comprehensively analyze user data,such as capabilities,requirements,locations,and resource status,so as to provide high-quality network services thatfocus on

135、users real-time locations and dynamic requirements.On the 5G network,users sessioncontext,authentication and subscription data,policy data,and other data are scattered on multipleNFs,such as the AMF,SMF,unified data management(UDM),UPF,and policy control function(PCF).As a result,the network cannot

136、comprehensively sense users or dynamically adjust theperformance based on users.In addition,to ensure consistent user experience in different locations,the network must be able to migrate user data and network data quickly.On the 6G network,users may serve as network service consumers and provide se

137、nsing datato the network,such as environmental data collected by cameras in public areas.In this scenario,users need to flexibly control whether the data they provide can be shared to ensure data privacy.In addition,the network needs to flexibly assign different operation permissions and transmissio

138、npriorities to different data consumers to meet requirements in diverse service scenarios of the 6Gnetwork.3.5.2.Potential RequirementsUser-centered data management and control require UEs to provide more incremental dataand the network to integrate and transmit different data as required.1)6G UEs,s

139、uch as mobile phones and cameras in public areas can collect sensing data forthe network.The 6G data plane needs to manage and control this type of data by userand allows users to flexibly control their data and data service capabilities.2)The 6G data plane needs to collaboratively sense and intelli

140、gently integrate usersmulti-modal data.Based on real-time and other requirements,different data has differenttransmission priorities.The network can comprehensively analyze multi-modal data toobtain new knowledge,so as to accurately sense users real-time statuses andrequirements,intelligently predic

141、t the user and network statuses at the next moment,adjust the network-side policy in advance,and improve user experience.17/643.6.Data in Integrated Satellite-Terrestrial CommunicationScenarios3.6.1.DescriptionThe 6G-oriented integrated satellite-terrestrial network(ISTN)deeply integrates theterrest

142、rial communication system and satellite communication system to complement advantagesin coverage and capacity.In 3GPP Release 15,research on the non-terrestrial network(NTN)wasstarted.In 3GPP Release 17,the first NTN standard for the transparent forwarding mode wasformulated.In 3GPP Release 18,the C

143、N UPF could be deployed on the satellite network.In 3GPPRelease 19,the NTN standard for the on-board processing mode was supported,and RAN nodesand some CN functions could be deployed on the satellite network.When the access network,bearer network,CN,and other functions are deployed on the satellite

144、 network as required,low-earth orbit(LEO)satellites move at high speeds relative to the Earth,and it takes about 130minutes for an LEO satellite to orbit around the Earth once.When the network provides userservices,satellites or beams may be switched due to mobile satellites and users.As a result,da

145、taavailability on the satellite network is dynamically changed with high-speed movement of thesatellites.Movement of a satellite may cause users it serves to switch the group.RAN node,CNcontrol plane,and other context data of multiple users need to be exchanged and synchronizedbetween satellites to

146、ensure reliable network status migration.In addition,user plane data needs tobe migrated without loss to ensure service continuity.Satellites are in a special space environment.Data collected by satellites may contain invalid data due to environmental interference.If all theenvironmental data collec

147、ted by a satellite is directly transmitted to the terrestrial network forprocessing in satellite data backhaul scenarios,bandwidth resources on the feeder link areseriously wasted.In addition,different types of data on satellites have different real-timerequirements and need to be transmitted as req

148、uired.3.6.2.Potential RequirementsIn integrated satellite-terrestrial communication scenarios,entities that control data andcontrol functions need to be fully decoupled and data collected by satellites needs to be processedbefore being transmitted.1)Satellites move dynamically.Based on the service d

149、uration of ephemeris and other dataused to analyze satellite data,the 6G data plane can intelligently migrate RAN node andCN control plane context data of multiple users between satellites and migrate userplane data between satellites without loss to ensure service continuity.For example,thedata pla

150、ne migrates control plane data and user plane data at user or group granularity.This prevents high-cost data migration by session.As processing and data are decoupled,18/64the data plane can reliably migrate the network status when the satellites movedynamically,improving the user group switching ef

151、ficiency.2)The 6G data plane needs to deploy data processing functions on the network edge.Forexample,the on-the-fly processing function can be used to process the original datacollected by satellites locally.With the function,the data plane can remove invalid data,integrate data,and transmit the da

152、ta to the terrestrial network for further analysis.Thisimproves the performance of the data backhaul service.3.7.Data Collection for SOE3.7.1.Description6G performance indicators include not only higher capacity,lower latency,and lower packetloss rate but also better user experience.As shown in the

153、following figure,we should pay moreattention to extremely poor user experience and reduce the probability of extremely poor userexperience.Figure 3.7-1 Probability Distribution Function of User ExperienceThe existing network-centered data collection and optimization solution aims to improve theovera

154、ll or average performance at the cell level or network level.As shown in the following figure,3GPP protocols related to network-centered data collection and optimization include SON,MDT,etc.With the evolution release by release,the solution of traditional network-centered datacollection and optimiza

155、tion has gradually evolved into collaboration between the network and UEs.For example,in L1/L2-triggered mobility(LTM)of Release 18,the network topology can beexposed to the UE.That is,the UE determines whether the source cells and target cells belong tothe same DU based on the node ID provided by t

156、he network.Then,the UE determines how toprocess data at the RLC layer and PDCP layer during handover.The UE assistance information(UAI)allows the UE to report its preferences,including the discontinuous reception(DRX)19/64configuration,carrier aggregation(CA)configuration,and RRC status.During condi

157、tionalhandover(CHO),the UE detects configurations of multiple candidate RAN nodes configured bythe source RAN node and the handover trigger conditions.If the handover trigger conditions aremet,the UE hands over to the target RAN node through random access.Figure 3.7-2 Network-Centered Data Collectio

158、n and OptimizationIf the UE reports more and more data(such as real-time and accurate user experienceinformation)to the network,user privacy may be exposed.Mobile phones on the market haveoptimized user experience based on the implementation of UEs(such as video pre-caching insubways and other scena

159、rios).This indicates that UEs have basic capabilities of self-optimizingexperience(SOE).SOE collects network data based on customized requirements of each UE.To optimize userexperience,the UE can use implementation solutions,provide user preferences to the network,ordetermine configurations within t

160、he allowed range configured by the network.This ultimatelyreduces the probability of extremely poor user experience and ensures consistency of high-qualityuser experience.For example,a paging failure is one of the reasons leading to the failure calledvoice service.In this scenario,the UE may not kno

161、w whether a paging failure occurs.Eventhough the UE knows that a paging failure has occurred through a notification of application layer,the UE does not know when and where the network sent the paging message.As a result,the UEcannot discover and analyze the cause in time.In this scenario,the networ

162、k may not know wherethe UE is when the paging failure occurs and whether the paging failure is caused by coverage,abnormal UE status,or other reasons.In conclusion,it is difficult to detect poor user experience and analyze the cause in time bydepending only on network data or UE data.Therefore,it is

163、 impossible to provide solutions toensures consistency of good user experience.To ensure a better user experience on the 6Gnetwork,the networks and UEs can collaborate based on features or use cases.3.7.2.Potential RequirementsAccording to description in the user experience optimization scenario,the

164、 UE needs to obtainnetwork data and provide user preferences to the network.1)The UE collects required CN data and RAN data based on the requirements of user20/64experience optimization.For example,the paging failure event,tracking area identity(TAI)and cell ID involved in paging,and time for sendin

165、g the paging messages.Due tothe requirement for timeliness of data,the network data that the UE collects over the airinterface includes real-time data and historical data.When the UE collects network data,the data consumers include the UE modem and AF.The AF may be deployed on the UEor a third-party

166、 server authorized by the UE.2)The UE can provide user preferences to the network,such as resource configurationsrequired for user experience optimization,to assist the network in optimizing userexperience.The UE can also determine configurations within the allowed rangeconfigured by the network.In

167、this case,configuration parameters of the UE can bequickly adjusted based on real-time and accurate user experience information to avoidpoor user experience.3.8.Self-Generated Data Services3.8.1.DescriptionNew capabilities,such as sensing and AI computing introduced to the 6G network bring alarge am

168、ount of new data,such as sensing data and AI data.Multiple NEs,including the CN,RANnode,and UE are involved in generation,termination,processing,transfer,storage,and other linksof these new types of data.Each NE may have multiple data functions in data services.Underdifferent network requirements,da

169、ta may be exchanged between the CN and RAN node,betweenthe RAN node and UE,between the CN and UE,inside the CN,inside the RAN node,or betweenRAN nodes.Data services may only include several links among data collection,data processing,data analysis,data storage,and data transfer.6G data services are

170、large and complex.Dataservices that rely on manual selection of data sources,NE functions,data processes,and datatopologies cannot meet the requirements of agile service response and rapid service rollout.6Grequires self-generated data services to quickly adapt to service requirements in diverse sce

171、narios.AI development,especially large model development makes self-generated data services possible.Self-generated data services are agile and efficient data services that use AI methods toautomatically select data-related NEs,functions,processes,and topologies and intelligentlyorchestrate the data

172、 process.Integrated with AI,the self-generated data services can automaticallyselect data collection sources based on network requirements,the air interface environment,andNE capabilities,can automatically clean,process,and refine data based on the function of thecollected data,and can automatically

173、 plan data transfer and aggregation paths based on the datavolume,latency,computing capability,and scenario requirements.Take RAN node sensing as an21/64example.Low-frequency RAN nodes are suitable for large-scale low-precision sensing,andhigh-frequency RAN nodes are suitable for small-scale high-pr

174、ecision sensing.Data services needto properly use the sensing capabilities of different RAN nodes.In data services,sensing dataneeds to be processed flexibly,effective data and long-term environment features need to beextracted from historical data,and real-time data needs to be pre-processed to red

175、uce redundantdata.Sensing data transfer and aggregation needs to consider the function of data aggregationwithin RAN nodes,data sharing between RAN nodes,and RAN node data integrated by the CN indata services.Non-self-generated data services are rigid and have low adaptability.Self-generated data se

176、rvices can automatically select sensing RAN nodes and transfer appropriatesensing data to where needed,providing personalized,timely,and accurate data support for RANnode sensing.3.8.2.Potential RequirementsWith massive heterogeneous data,diversified data services,extensive application scenarios,and

177、 complex and large-scale network,the 6G network urgently requires data services that supportintelligent data supply and automatic data process orchestration to meet endless and ever-changingrequirements in the future.The network has the following requirements for self-generated dataservices:1)Automa

178、tically select and orchestrate NEs,functions,processes,and topologies based onnetwork requirements,and automatically adjust NEs,processes,and topologies based onnetwork changes.2)Automatically manage the data service life cycle.3)Provide data service capabilities and data exposure capabilities.3.9.D

179、ata Benefits3.9.1.DescriptionData benefits indicate the rights and interests of different roles(such as data generators,dataowners,data processors,and data users)of data subjects clarified under laws,regulations,andprivacy protection.Data subjects,such as users(including ToC and ToB users),operators

180、,anddevice vendors play different roles in data collection,data processing,data transmission,dataanalysis,and other links.Correspondingly,they have different data benefits.The data benefits mechanism appropriately evaluates data subjects value in data services andexchanges data value within a safe a

181、nd controllable range.This promotes proactive data sharingand exchange,enhances data abundance,and improves the data value.22/64Operators can use rewards(such as point rewards)to make ToC users more willing to enablethe data reporting function.On the 5G network,SON and MDT can only measure and colle

182、ct dataon several network features of specific users.If users initiative is stimulated on the 6G network,more users will provide a larger amount and wider range of data.For example,in addition to morecommunication feature data,users can provide more environmental sensing data and AI modeltraining da

183、ta.In addition,industries and enterprises attach great importance to data security and hope tocontrol enterprise data.With data benefits protected,enterprises can control the data exposurescope and exchange data value,and enterprise users are willing to proactively share enterprisedata.Device vendor

184、s CN and RAN nodes usually process,transmit,analyze,and store data.Databenefits can protect user/customer privacy.Driven by interests,more and more new NEs,such as the reconfigurable intelligent surface(RIS)wall glass and home/enterprise-built RAN nodes will emerge.These new NEs can make upfor the d

185、eficiencies of the 6G network at specific stages,improve network efficiency,and reducenetwork costs.On the 6G network,operators can provide network services to users and allocateand control data benefits.Data benefits properly present the value of all network subjects,and thesubjects are willing and

186、 have opportunities to participate in network construction and sharing,promoting network development.Therefore,the 6G network needs to use data benefits to evaluateand measure the value of network subjects in a safe and trustworthy manner and exchange thevalue between the network subjects.3.9.2.Pote

187、ntial RequirementsOn traditional networks,data benefits are unclear.As a result,network subjects are notproactive,and the data value cannot be maximized.Data benefits promote network participation,ensure data sharing and value transfer,and achieve multi-party collaboration and valuemaximization in t

188、he future 6G wireless system.Therefore,data services require the data benefitsmechanism.Requirements on data benefits include:1)Data benefits subjects and their data benefits need to be standardized and clarified.2)Functions related to data benefits need to be added before,during,and after services.

189、Functions before services include authentication,certification,and authorization.Functions during services include benefits exchange,task allocation,collaborativecontrol,process supervision,billing rewards,and fault penalties.Functions after servicesinclude value evaluation and secure transaction se

190、ttlement of benefits.3)The network needs to support various mechanisms,processes,and modes related to databenefits.Data benefits mechanisms include data benefits allocation,value assessment,23/64and transaction settlement.Data benefits processes include benefits classification,benefitsinformationexc

191、hange,benefitschanges,andbenefitsdeletion.Thesubscription-publish and request-response modes can be used in data benefits processes.For example,the network publishes data benefits information,including benefitsallocation,value measurement,value exchange,and trustworthiness,or the UE sendsrequests to

192、 trigger the network to allocate and modify data benefits.3.10.Summary of 6G Data RequirementsBased on the preceding 6G service scenarios and requirement analysis,the 6G system needstwo categories of data,as described in Table 3.10-1.One is new data,including sensing data andAI data that matches 6G

193、new scenarios and requirements.The other is standardized data that isseparated from 6G system functions,including user-level data or NE data on the network to meetrequirements,such as distributed architecture and integrated satellite-terrestrial communication.Compared with the 5G system,the 6G syste

194、m needs to support collaborative data collection andmore automated data collection,including network data acquisition by UEs,data benefits,andself-generated data services.In addition,the 6G system needs a unified data life cyclemanagement mechanism to improve the efficiency of transferring massive v

195、olume of network dataand avoid standard overheads of fragmented solutions for single use cases.Table 3.10-1 Summary of 6G Data Plane DataData TypeData ProviderData ConsumerData VolumeSensing data(such assensing measurementdata)UE,RAN node,andthird-party sensorCN,RAN node,andUEThe data is related tos

196、ensing use cases,therate is about 1 kbps to10 Mbps.NetworkAIdata(such asAI models)UE,RAN node,CNNF,and AFCN,RAN node,andUEThe data is related toAI models,and theAImodeldataforbeam management isapproximately 10 KBto 100 MB17.User-level data on thenetwork(such as usersubscription data andRAN node and

197、CNNFCN,RAN node,andUEThe data is related tothe number of users.24/64context information)NEdata(suchaspagingdataandnetwork load)RAN node,CN NF,andnetworkmanagement functionNetworkmanagementfunction,CN,RANnode,and UEThe data is related tothe number of cellsand users.The data plane should have function

198、s,such as data collection,data processing,data storage,and data collaboration.The data plane can process data on any node and allocate data processingtasks in a centralized manner or based on processing capabilities of different nodes.This ensuresrefined collection,real-time processing,efficient tra

199、nsmission,distributed collaboration,andelastic storage of data throughout the data life cycle,thereby improving data utilization and dataservice capabilities of the 6G network18.For example,based on different sensing collaborationmethods,including different multi-point sensing methods in system netw

200、orking and differentmodes of sensing,the data plane needs to support cross-domain data collaboration betweendifferent domains on the 6G network,between the UE and network,and between internal andexternal functions of the 6G network.4.Definition and Framework of the 6G Data Plane4.1.DefinitionIn defi

201、nitions of the control and user planes in existing standards,the plane usually has thefollowing features:End-to-end connectivity:The UE,RAN,and CN support end-to-end interaction through thepeer-to-peer protocol layers.Isolation:The functions of different planes are isolated.Each plane focuses on a c

202、lass offunctions and procedures.Collaboration:Different planes collaborate to achieve complete functions and services,suchas communication services.From the perspective of air interface,different planes may share several protocol layers.However,the configurations of protocol layer of each plane are

203、different.Meanwhile,optimizedfunctions can be introduced based on different characteristics of protocol planes,the differentiatedrequirements of each plane also can be met.Therefore,the design satisfies both functionalmultiplexing among different planes and the differentiated requirements of each pl

204、ane.From theperspective of CN,each plane includes different network functions with their respectiveresponsibilities.Based on the above planes and functional features,the 6G network may introduce25/64a data plane in addition to the control plane and user plane.In the 6G system schematic diagramshown

205、in Figure 4.1-1,the control plane,user plane,and data plane collaborate to support keynative features,such as native intelligence,native computing,and native security,therebyproviding external services.External services include super communication services(e.g.,NTNand almost zero power IoT),basic in

206、formation services(e.g.,ISAC),and integrated computingservices(e.g.,AIAC).Figure 4.1-1 6G System DiagramThe data plane is responsible for the data life cycle from data generation to data consumptionin the mobile communication network.The data life cycle management includes data generation,data colle

207、ction coordination,data security and privacy management,data processing,data qualitymanagement,and data benefits.In addition,the data plane provides data services to other layersand planes.The data plane is an independent logical functional plane.It is logically parallel to thetraditional communicat

208、ion plane(that is,the control plane and user plane in the 5G system).Asa general public infrastructure,the data plane effectively manages the whole life cycle of internaldata of 6G new services,such as sensing,computing,and intelligence.Broadly speaking,the dataplane provides data services for funct

209、ions in specific service areas.The 6G data plane logicallyincludes the control of data plane and the data of data plane.The control of data plane includes theestablishment and release of bearer and so on.The data of data plane includes data transmission,data processing and so on.In the standardizati

210、on process of 6G,separation between control anddata of data plane,and integration between the control of data plane and existing control plane26/64need to be considered.4.2.Necessity of the Data Plane4.2.1.Traditional Subscription Data Migration Brings CompatibilityChallengesTraditional subscription

211、 data includes user subscription data,policy subscription data,andservice subscription data.Smooth and rapid migration of subscription data is a key link in networkevolution but faces challenges due to inconsistent user data formats.Manufacturers adopt privateimplementations,and user data storage NE

212、s have different data formats.In addition,networkupgrades lead to changes in logical networking,and services with different data formats indifferent areas need to be migrated and integrated.Therefore,the 6G network needs to optimizethe storage method and performance of user data storage NEs to provi

213、de a unified user view.4.2.2.New Data on the 6G Network Brings Challenges in Full-ProcessProcessingNative intelligence,ISAC,and other visions of the 6G network lead to new types of data onthe 6G network,including AI data and sensing data.These types of data need to be collected,stored,processed,anal

214、yzed,and transmitted on the network,and pose new challenges,includingreal-time and fine-grained data collection,reliable data storage and fast indexing,dataoptimization during data processing and transfer,differentiated QoS in data transmission,andtransmission of any topology.The existing 5G CN does

215、 not support or supports only some of thepreceding functions.Therefore,the 6G network needs to introduce data-related functions toprocess new data,such asAI and sensing data to provide trusted data services.4.2.3.ConclusionThe 6G network will transfer from user-oriented single data forwarding to dat

216、a productionand consumption oriented to all factors.The existing network only serves as a data transmissionpipeline and cannot meet the preceding requirements.Therefore,more data-related functions arerequired.These new data functions have different performance and mechanisms from thetraditional cont

217、rol and user planes.The control and user planes mainly require high reliability,short latency,and other key performance indicators(KPIs).These new data functions requirehighly reliable storage,high-speed reading,and extremely reliable transmission.Therefore,thesenew data-related function sets are de

218、signed into an independent functional plane,namely the dataplane.27/64The data plane introduced on the 6G network prevents the control plane from bearing a largeamount of non-signaling data,and prevents user plane transmission from being terminated insidethe mobile network.In addition,the data plane

219、 improves the transmission efficiency of networkinternal data,ensures data reusing,enhances cross-domain data collaboration,and can be easilyexpanded to meet new requirements.As a new plane in the 6G network architecture,the data planeaims to achieve efficient and secure iteration of massive volume

220、of user data,reliable migration ofthe network status,and introduce data services in a trusted way.This plane focuses on efficientand reliable data storage,and standardizes data service functions to decouple data services basedon data service orchestration and other controls.Based on control plane en

221、hancement,the dataplane supports coordinated global control of data services and mobility management,as well asservice control and execution of the data plane,ensuring full-process processing of 6G diversedata.4.3.6G Data Plane FrameworkThe 6G data plane framework is generalized,evolvable,and backwa

222、rd and forwardcompatible and provides unified and service-oriented data services that are independent of eachother but mutually support each other and are for different service domains.The data plane needsto unify data concepts and functions and establish association between data to support applicat

223、ions.With unified data function sets,data services are transferred from a monolithic architecture to anormalized architecture.Figure 4.3-1 Framework of 6G Data PlaneThe 6G data plane framework is divided into a unified set of management function,a unifiedset of processing function,a unified set of e

224、xposure function,and a unified set of trading function.Functions of these four sets are mapped to specific NEs such as the CN,RAN,and UE in the dataplane.The unified set of management function is mapped to the network policy formulation andtask orchestration management network elements of the core n

225、etwork;the unified set ofprocessing function is mapped to the data collection and analysis network elements in the core28/64network or access network,base stations,and UEs;the unified set of exposure function is mappedto the capability exposure network elements of the core network and access network

226、;the unifiedset of transaction function is mapped to the network management and operation network elements.The specific information of the four functional sets is as follows:Unified set of management function:Provides data service orchestration and data lifecycle management functions.Data service or

227、chestration generates data services ondemand and supports centralized and distributed deployment and control.Resourcesupport for data services needs to be considered in data service orchestration.Data lifecycle management provides comprehensive management from data generation todestruction.Unified s

228、et of processing function:Provides data collection,data preprocessing,dataanalysis,dataset extraction,data storage,and data forwarding functions.Data collectionnot only includes how to acquire data but also the requirements for the format of thecollecteddata.Datapreprocessinginvolvesprocessessuchasd

229、atacleaning,transformation,integration,normalization,and masking of raw data.Data analysisidentifies data clues,relationships between data,and discovers data value.Based on dataanalysis,dataset extraction retrieves contextualized data and typical datasets withmeaningful knowledge according to data f

230、eatures.In addition,data preprocessing alsoincludes data storage and data routing and forwarding policy formulation functions.Unified set of exposure function:Provides functions such as data exposure,datasetexposure,and data service exposure.Data exposure includes the exposure of originaldata and th

231、e exposure of preprocessed data,which is called basic data exposure here.Dataset exposure is the exposure of contextualized datasets and feature datasets.Dataservice exposure allows third parties to customize and orchestrate new data services.Unified set of trading function:Provides data rights and

232、interest division,data serviceauthentication,data value evaluation,and data security transaction functions.Data rightsand interest division clarifies the rights and benefits of each participating entity in thedata service.Data service authentication ensures the security of data access byidentifying

233、and authorizing data users.Data value assessment involves the monetizationof data assets for data asset transactions.Data security transactions require technologieslike blockchain to ensure the secure and fair circulation of digital assets.The functional sets within the 6G data plane framework can i

234、nteract through the data bus toform data services that control all elements and the entire life cycle of the business.The 6G dataplane framework can interact with the outside world to provide data service support forcommunication,sensing,computing,and intelligent businesses while improving data serv

235、icecapabilities.29/644.3.1.LogicalArchitecture and Functions of 6G Data PlaneThis section explores the potential logical functions and architecture of the data plane.Currently,the logical architecture and functions of the data plane mainly focus on terminalfunctions and network element functions,wit

236、hout presenting Operation Administration andMaintenance(OAM)and Business Support Systems(BSS).Therefore,management functions andtransaction functions related to OAM and BSS are not presented.Then,considering research fromtwo dimensions,one is the logical application layer decoupled from specific net

237、work elements,including the basic functions required by the data application layer and the interactionrelationships between various functions.The second is the mapping between the basic functions ofthe aforementioned data application layer and UE,RAN,and CN.Data plane protocol support isrequired for

238、 interaction between different peers,including UE-RAN,UE-CN,RAN-CN,RANfunction-to-function,and CN function-to-function.As shown in the logical architecture diagram of the data plane in Figure 4.3-2,the data planefunctions include data control,data privacy and security,data storage,data transmission,

239、dataprocessing,data provision,and data consumption.The mapping relationship between the logicalfunctions and the terminal,radio access network,and core network is shown in Section 4.3.3.Thissection does not distinguish between terminals or network elements.The specific explanations ofthe aforementio

240、ned logical functions are as follows:Data control function(DCF):Supports data collection coordination,data serviceconfiguration(such as data packet size and interval),data transmission configuration(such as establishing,modifying,and releasing data plane transmission channels),dataprocessing configu

241、ration(such as data preprocessing or data analysis configuration),data quality management,and data rights and other full life cycle managementconfiguration.Data privacy and security function:Supports authentication,authorization,accesscontrol,and other privacy and security mechanisms initiated by ot

242、her data functions.Data storage function:Also known as the data repository function,it supportspersistent storage and retrieval of data collected by the data plane.Data transmission function:Supports forwarding and transmission of data plane dataaccording to DCF configuration.Data transmission funct

243、ion can be UE,base station,orcore network node,etc.According to the different peers of data transmission,data planetransmission includes UE and RAN,RAN and CN,UE and CN,RAN internal functions,or CN internal functions and other multiple data transmission protocols.Data processing function:Provides da

244、ta preprocessing and data analysis according todata service requirements.30/64Data provider:Supports generating and providing required data according to DCFconfiguration.Data consumer:Supports sending data requests and receiving data responses.If thedata consumer is an external network function,it n

245、eeds to obtain data based on networkcapabilities such as network capability openness.Figure 4.3-2 LogicalArchitecture of 6G Data Plane4.3.2.6G Data Life cycle ManagementThe 6G data plane is responsible for the full life cycle management of internal data in mobilenetworks,including data generation,da

246、ta collection coordination,data security and privacymanagement,data processing,data quality management,and data rights.In the data qualitymanagement phase,the focus is on data effectiveness and availability,while in the data rightsphase,the focus is on matching the rights between data providers and

247、consumers.The time cyclesof each phase may vary.For example,the rights brought by data may persist long after dataconsumption/service completion.Additionally,data providers may only have rights for a shortperiod,but the data obtained by data consumers remains valid for a long time.Therefore,the 6Gda

248、ta plane must consider the characteristics of each phase in life cycle management.31/64Figure 4.3-3 Life Cycle Management of DataBased on the 6G data plane framework,the entire lifecycle from data generation to dataconsumption within the mobile communication network can be realized through the data

249、planefunctions.As shown in Figure 4.3-3,data life cycle management includes the following parts:Data collection coordination:Based on the data requirements of various functions,redundant data collection is avoided through coordination.Suitable data providers areselected based on collected data infor

250、mation,and factors such as the capabilities of dataproviders(e.g.,UE,RAN,CN).This enables the configuration of data provisionfunctions to generate and provide data as needed.Data privacy and security management:Protects user privacy through user authorization,authentication,data masking,data encrypt

251、ion/decryption,etc.,ensuring privacy andsecurity throughout the life cycle of data.Data transmission and forwarding:Data transmission in the data plane is flexible andterminates at terminals,radio access networks,and core networks.Data transmission(such as wireless bearer,and message framework)shoul

252、d be able to support massivedata transmission efficiently.Introducing data plane and data bearers to optimizeinternal data transmission in mobile networks is a potential technological path.Data processing:According to data requirements,UE/RAN/CN data plane functions canoptionally implement data coll

253、aboration and knowledge through data processing.Dataprocessing includes data storage,compression,cleaning,integration,format conversion,data analysis.32/64Data quality management:Ensuring data quality is a crucial feature of the 6G data plane.Based on data requirements,data provision functions selec

254、t and process data to meetquality of service requirements such as accuracy,quantization precision,response time,validity period,completeness,redundancy,correctness,and consistency.Data service:Also known as data consumption,it involves providing necessary data todata consumers(e.g.,sensing functions

255、,network capability open functions).Potentialdata consumers include terminals,radio access networks,core networks,and applicationfunctions.4.3.3.An example of 6G System Architecture with DataPlaneThe 6G data plane adopts a unified framework to support full life cycle management of data.Specifically,

256、based on this unified framework,the 6G data plane provides data for differentbusiness entities through processes such as data collection,forwarding,and processing.From thethree aspects of management,control,and execution,the 6G data plane functions can be dividedinto three categories:data service or

257、chestration management functions,data plane managementand control functions,and data plane execution functions.Among them,data service orchestrationmanagement functions belong to the network management domain,responsible for resource andbusiness orchestration of data services,as well as data model m

258、anagement,enabling on-demanddeployment and elastic reconstruction of data plane functions.Data plane management and controlfunctions and data plane execution functions belong to the network function domain,responsiblefor control information interaction and data transmission and processing in the dat

259、a plane,respectively.To present the 6G system architecture more intuitively,based on the above functionalclassification and the 5G system architecture,an example of 6G system architecture with a dataplane is proposed,as shown in Figure 4.3-3.The 6G system architecture includes the core network,radio

260、 access network,and terminals.Based on the 5G service-based architecture,data functions,sensing functions,etc.are introduced.The existing service-based architecture is mainly used forthe interaction between various functions in the core network control plane.Considering theefficiency of data transmi

261、ssion,it may be necessary to enhance service-based protocols orintroduce a messaging framework to support the transmission of large amounts of data within theCN.Similarly,radio access networks and terminals also need to introduce peer data functions tosupport the new features of 6G.33/64Figure 4.3-3

262、 Example of 6G System Architecture with Data PlaneAnalyzing from the CN perspective,data plane functions are added and the existingcontrol plane and user plane are enhanced.For an option,the data plane can independently supportmanagement and control functions of data plane,such as data service reque

263、st processing,accesscontrol,policy management,data management,orchestration of forwarding path of data bearer,aswell as data execution functions such as data transmission and forwarding,in-network processing.For another option,the data plane can cooperate with the enhanced control plane and user pla

264、ne tosupport the whole procedures of data services and achieve collaboration between data services andconnection services.Considering that some management and control functions of data plane(suchas policy control and registration)are related to the existing communication control planefunctions,the e

265、xample enhances existing NFs of control plane to support some management andcontrol functions of the data plane,such as access control,policy control,registration discovery,and data management.Enhancements are made to support the data plane,such as enhanced policycontrol functions and enhanced acces

266、s and mobility management functions.Additionally,ifconsidering the deep integration of the data plane and the control plane,existing control planefunctions such as UDM and UDR may also belong to the data plane through restructuring.Thedata plane execution functions are responsible for executing data

267、 forwarding,storage,andprocessing operations(e.g.,preprocessing and analysis).When data forwarding and dataprocessing are performed at the same node,on-demand in-network processing can be achieved.For example,potential methods include enhancing user plane functions to support data forwarding,storage

268、,and processing functions of the data plane.From the perspective of the UEs and RAN,UE and base stations can support data planefunctions as needed.For example,UE or base stations can register and update the capabilities ofdata plane,and support data operations such as data collection,preprocessing,a

269、nd forwarding forUE data and base station data.Service-based interfaces can be used for the interface between RANand CN,or enhancements can be made based-on the existing point-to-point interfaces.Although34/64there is discussion in the industry about service-based interfaces between UE and the netwo

270、rk,service-based interface between UE and CN or service-based interface between UE and RAN isstill more controversial.Therefore,they are not presented in the example.5.Key Technologies of 6G Data Plane5.1.Data Bearer and Transmission ProtocolThe service data transmitted in 5G networks is mainly user

271、 data of UEs that areconnection-oriented(such as application layer data of UEs).The protocol data unit(PDU)sessionmechanism can only effectively serve wireless communication services.To efficiently support thetransmission and forwarding of new service types associated with ubiquitous sensing,computi

272、ng,intelligence and communication,such as sensing and computing,intelligence and storage,andintelligence and security,the data plane protocols need to be designed in multiple aspects.1)Data service session mechanism:Supports the construction and maintenance oftransmission path topology for arbitrary

273、 data type sessions,including flexible topologytransmission between UE and RAN,UE and CN,the internal transmission within thecore network or RAN.the QoS flow will be terminated in RAN or Core networkfunctions(i.e.will not be delivered to DN).2)Data transmission protocol stack:Efficiently carry and t

274、ransmit new data types withinthe 6G network.For example:computing intelligence-related data that requiresultra-high speed switching.3)Data processing at intermediate nodes:For specific service data such as sensing,intelligence,and digital twins,intermediate network element nodes do not simplyforward

275、 but require computing and processing,such as complex data packet parsing andprocessing.Currently,intelligent computing and processing are primarily centralized in the cloud,whichmay lead to several issues:1)Lack of user behavior data in specific geographic or business areas.Uploading largeamounts o

276、f data from UE or RAN to the cloud for processing consumes significant airinterface resources and transmission resources,and longer interaction delays.2)Increased network resource consumption and energy overhead due to multiple networkelement involvement in data transmission.With the proliferation o

277、f network intelligence,the demand for such use cases continues togrow.To improve prediction accuracy and reduce overhead,intelligent and sensing data handled35/64by the data plane will also be transmitted between RAN and UE,with RAN and UE performingtasks such as model training and inference.The fol

278、lowing use case scenarios can be considered:Data collection scenario:The base station collects data from the UE or the UE collectsdata from the base station.The base station or UE can train or infer the model based onthe collected data.Model transmission scenario:The base station or UE is responsibl

279、e for model training.After the training is completed,the base station transmits the model to the UE or the UEtransmits the model to the base station.The collected data varies depending on the specific use case and the reporter and can includeparameters such as RSRP,SINR,RI,PMI,RSRQ,coordinates,servi

280、ng cell identifiers,movementspeed,packet loss rate,and RB utilization rate.The volume of collected data depends on the datatypes and reporting cycle,while the size of the model depends on the number of its parameters.The aforementioned data is generated by RAN or UE and transmitted between them.Thet

281、ransmission of this data will use RAN and UE as data terminations,and the control functions anddata plane execution functions,data connection establishment process,QoS configuration,andtransmission methods related to the data plane is not define in the current 5G system.Control and Execution Functio

282、ns Corresponding to the Data PlaneTo support local data transmission between UE and RAN,corresponding control andexecution functions need to be added to the data plane.Control functions include parameterconfiguration,connection control,data collection,model inference,and model training,whileexecutio

283、n functions include data collection,data processing,and data transmission.These controland execution functions can be independent functional entities or deployed within existingprotocol function layers.Data Service Session Mechanism Corresponding to the Data PlaneIf the transmission is primarily bet

284、ween RAN and UE,the connection procedure between UEand the core network can be more lite,or some of the core network functions can be moved to theaccess network.For example,for QoS rules,before establishing a connection between RAN andCN,CN instructs QoS rules to RAN.During the subsequent transmissi

285、on process between RANand UE,the base station configures UE according to the instructions from the core network.Thiseliminates the need for the AMF to send QoS Profile signaling to RAN during sessionestablishment.Furthermore,the RAN can take on some of the functionalities of the SMF(SessionManagemen

286、t Function)and UPF(User Plane Function),generating PDR(Packet Detection Report)and using the PDR to filter downlink data packets.For UEs QoS processing,QoS Rule can alsobe sent to UE directly via RAN RRC(Radio Resource Control)signaling instead of beingtransmitted to via NAS(Non-Access Stratum).36/6

287、4Radio Bearer and Protocol Stack Corresponding to the Data PlaneIn the current user plane protocol stack,after UEs data is sent to RAN,it is furthertransmitted to DN(Data Network)through UPF.Once the data transmission is terminated in RAN,it can be implemented in multiple ways:1)Use existing SRB(Sig

288、naling Radio Bearer)or DRB(Data Radio Bearer)to carry control or data of the data plane.By configuring the bearer,the datacan terminate in RAN,eliminating the need to forward it to the core network.2)Use special SRBor DRB to carry control or data of the data plane,and the default termination of this

289、 bearer is RANor UE.3)Use a new type of radio bearer and transmission protocol stack that does not rely on theexisting air interface protocol stack architecture,allowing for protocol stack function optimizationand configuration optimization based on data plane characteristics.Network Element Process

290、ing Corresponding to the Data PlaneData processing at intermediate nodes requires real-time or near-real-time processing andanalysis of data during its generation and transmission.Network elements perform deeper sensingand more complex network packet processing,facilitating more efficient packet tra

291、nsmission.Specific actions of processing by intermediate network element nodes include content recognition,feature extraction,information compression,distribution,and forwarding and so on.5.2.Air Interface Data PlaneThe design of data plane protocols is a key factor in determining the efficiency of

292、the dataplane,especially for the air interface protocols of data plane,including the protocol between UEand RAN,the protocol between UE and CN.If the protocol of data plane for controlling and datatransmission is noted as DSAP(Data Service Application Protocol).From the perspective of theair interfa

293、ce,DSAP includes at least DSAP between UE and RAN and DSAP between UE and CN.Due to the need for further research on the relationships between the above two protocols,theDSAP between UE and RAN is temporarily referred to as DSAP1,and the DSAP between UE andCN is referred to as DSAP2.Figure 5.2-1 Pro

294、tocol of Data Plane between UE and RANAs shown in Figure 5.2-1,an example of the air interface data plane protocol stack betweenUE and RAN includes DSAP1,L2,and L1.DSAP is mainly responsible for controlling closelyrelated to data plane functions and data transmission.The functions of L2 and L1 need

295、to consider37/64the relationship with the existing protocol functions of control plane and user plane.The design ofL2 and L1 should be configurable based on the characteristics of DSAP,simplifying protocols andenhancing efficiency.For instance,in the case of L2 and L1 of 5G,an example with minimalim

296、pact is that DSAP1 is over PDCP/RLC/MAC/PHY.Figure 5.2-2 Protocol of Data Plane between UE and CNAs shown in Figure 5.2-2,an example of the air interface data plane protocol stack betweenUE and CN includes DSAP2 and lower-layer transport protocols.In order to focus on the protocolstack between UE an

297、d CN,the internal functions of core network data plane,including datacontrol function,data privacy and security function,data storage function,data transmissionfunction,and data processing function,are not shown in the figure.Here,the core networkfunctions responsible for data plane control or data

298、transmission are temporarily referred to as CNData Plane Function(DPF).When the control and forwarding functions of CN DPF are separated,the function responsible for data forwarding in CN DPF maps to data transmission function in thelogical architecture of data plane in Section 4.3.On one hand,DPF s

299、ends and receives data planemessages and data related to UE based on protocol stack of data plane between UE and CN.Onthe other hand,DPF interacts with other data plane functions(such as data repository function,etc.)and control plane functions(such asAMF,SMF,etc.)based on internal interfaces of the

300、 CN.Next,we will elaborate on the data reliability,integrity assurance,and minimalist protocolstack design of the air interface data plane in detail.Data Reliability and Integrity AssuranceIn 5G networks,service data transmitted originates from the application layer,and theintegrity of the data tran

301、smission is ensured by the application layer.However,new types of datain 6G networks such as sensing and AI data may be generated internally within the network,requiring the 6G network itself to ensure the reliable and complete transmission of data.Forexample,when UE initiates AI-related data servic

302、es,the AI data itself may be larger.Transmittingall data may take a long time,during which dynamic changes in the air interface environment mayprevent successful data transmission to the destination within the specified time.Additionally,during transmission,situations may arise where the receiver no

303、 longer needs the data or thetransmitter cannot transmit the data,leading to the termination of data transmission.Therefore,38/64timely policy control of data services is required to ensure the process of new data services.Thefollowing figure illustrates the data service policy control process.Figur

304、e 5.2-3 Policy Control Procedure of Data ServiceAs shown in Figure 5.2-3,before initiating a data service,the transmitter determines whetherto initiate a new data service based on policies.If the policy conditions are met,the transmitterinitiates a data service request.Upon receiving the data servic

305、e request,the receiver thendetermines whether to initiate a new data service based on policies and replies with data serviceinstructions.Policy assessment may include considerations such as support for new data services,whether data volume thresholds or RSRP thresholds are met,and AI prediction resu

306、lts.It is worthnoting that data services may be triggered by either the transmitter or the receiver.Minimalist Protocol Stack DesignThe layered design of the 5G protocol stack results in the addition of protocol headers at eachlayer,leading to overhead and latency issues during data processing.Moreo

307、ver,when data istransferred between protocol layers,it undergoes multiple copies,increasing memory consumptionand data processing delays.This multi-layered protocol stack approach to data processing andcopying cannot meet the performance requirements of 6G,such as extremely high transmissionrates an

308、d ultra-low transmission latency.Therefore,new data forwarding mechanisms are neededfor the 6G air interface data plane.1)Functional component orchestration:Forming generic functional components andflexibly arranging them based on service types and requirements,with protocol layerheaders added only

309、once.2)Unified data storage and bus based reading:Instead of defining multiple data cachebuffers,a unified data storage area is defined,allowing various functional componentsto directly access the unified data storage area through a data bus without the need formultiple data copies.To summary up,the

310、 aforementioned functional component orchestration,unified data storageand bus based reading can simplify the procedure of data unit headers addition and data copying,thereby enhancing data processing efficiency and enabling the formation of a minimalist protocol39/64stack tailored to the requiremen

311、ts of 6G new data.5.3.Core Network Data Plane Functions andArchitectureTo support the addition of an independent data plane,the control plane needs to be enhanced,starting from the unified control of mobility characteristics and achieving the integration controlof the entire process of mobile networ

312、k data processing.Specifically,a detailed design of the dataplane and enhanced control plane is shown in Figure 5.3-1.(1)Adding Data Plane Execution Function DF:Adding data collection,data processing,and data storage functions to achieve end-to-end data processing.Data Collection Service(DC):Geared

313、towards massive data,supporting the establishment ofa real-time and efficient mechanism for collecting data from multiple heterogeneous sources,enabling fine-grained on-demand data collection.This function can be integrated into other NFs.Data Processing Service(DP):Establishing multi-dimensional,st

314、andardized data processingmodels to efficiently and flexibly perform operations such as cleansing,masking,analysis,andencapsulation,supporting in-network processing during transmission,enabling in-network dataprocessing for 6G.Data Repository Service(DR):Based on the evolution of UDR and UDSF,suppor

315、tinghierarchical and classified storage of data resources such as sensing data,AI model data,trainingand inference data,user data,and personalized policy data,enabling data multiplex and efficientretrieval based on the granularity of data services,achieving unified storage of diverse 6G data.(2)Addi

316、ng Data Plane Service Control and Management Function DMF:Gearedtowards data service tasks,orchestrating data processing service pipelines;building a unified frontend for data storage to achieve efficient storage and read/write of diverse data.Stored Data Management Service(SDM):Based on the evoluti

317、on of UDM,designing aunified front-end interface for various data,balancing storage capacity and efficiency,andmanaging centralized+distributed storage data.Data Service Control Service(DSC):Based on the task requirements of data services,flexibly orchestrating data collection,processing,and storage

318、 service pipelines,efficientlycontrolling the entire data service process.Data Service Register Service(DSR):Based on service-based interfaces,registering data40/64collection,processing,and storage services with DSR to achieve unified management of dataservices.(3)Enhancing the Existing Control Plan

319、e Functions(eAMF,eSMF,ePCF):Achievingtop-level control of data services,including access,QoS management,and policy configuration.Figure 5.3-1 Detailed Design of Data Plane of CN5.4.Core Network Data Plane Transmission ProtocolWith the emergence of new scenarios and services in 6G,the types of data i

320、n the networkhave become increasingly diverse,leading to a significant increase in data interaction volume.Therefore,in 6G networks,large-scale data transmission within the network forms the basis forscenarios such as ubiquitous integration,immersive communication,and integrated sensing andcommunica

321、tion(ISAC).Currently,the signaling interaction and data transmission between network functions in the5G core network are carried out using the HTTP/2 protocol.According to the description in IETFRFC7540,the maximum frame payload of the HTTP/2 protocol is 16M.Although theService-based Architecture(SB

322、A)based on the HTTP/2 protocol can efficiently handle signalinginteraction between network functions,the efficiency of data transmission using the HTTP/2protocol will gradually decrease with the increase in data interaction volume between networkfunctions.Therefore,the 6G Core Network(6GC)needs to e

323、volve based on the current SBAarchitecture to improve data transmission efficiency while ensuring signaling interaction betweennetwork functions.41/64Figure 5.4-1 Transfer Protocol of DP of CNExpanding the data channel based on the SBI bus to form a dual-bus architectureinterconnection.The SBI is ma

324、inly used for signaling interaction between network functions.Thedata channel interface(DCI)is primarily used for the transmission and migration of large amountsof data,providing an efficient data pipeline for scenarios such as network intelligence and ISAC.From the interface analysis,the transmissi

325、on protocols between RAN and the core networkinterface may include existing point-to-point protocols and service-based protocols.For example,RAN evolves towards service-based evolution while retaining the existing N2 and N3transmission protocols with AMF and UPF.Combining the cloudification and serv

326、ice-basedimplementation of RANs intelligent and sensing functions,the design of the SBI and DCIbetween RAN and CN supports data transmission in the data plane of 6G networks.At the sametime,considering the existing N1 between UE and AMF,it may be possible to consider openingup the SBI and DCI betwee

327、n UE and CN as needed in the future,allowing UE to directly requestnetwork services such as data services and computing services.The data channel supports variousprotocols and data transmission methods,such as FTP,RDMA,and Kafka.Negotiations betweenthe requester and transmitter of the data can be co

328、nducted through the SBI to select specific datatransmission methods based on different data types,structures,and transmission requirements.FTP:File Transfer Protocol(FTP)is a standard protocol used to transfer files over acomputer network.It allows users to transfer files from one network node to an

329、other,with advantages such as fast transmission speed,high reliability,and support forlarge-capacity file transfer.Compared to HTTP,FTP does not require metadata in thetransmitted data and does not have excessive chunked encoding,making it suitable fortransferring large files and files with high rel

330、iability requirements,such as machinelearning model files.FTP is essential for scenarios requiring frequent model updates,real-time data sharing,or distributed computing,such as federated learning and modelsharing in 6G networks.RDMA:Remote Direct Memory Access(RDMA)is a technology developed to addr

331、essserver-side data processing latency in network transmission.RDMA technology allowsapplication data from the user side to be directly transmitted to the storage area on the42/64server side,eliminating multiple CPU copy operations during transmission.On one hand,RDMA exhibits characteristics of low

332、 latency and high bandwidth data transmission;onthe other hand,it effectively reduces CPU load.Communication protocols based onRDMA technology mainly include Infiniband(IB),RDMA over Converged Ethernet(RoCE),and iWARP,among which RoCE is the most commonly used RDMAtechnology in Ethernet and has been

333、 deployed in most large-scale data centers.RoCE iscarried over UDP,resulting in lower transmission latency but lower reliability,and itrequires a lossless network environment,making RDMA mainly suitable for scenarioswhere both the transmitter and receiver are in the same data center.Additionally,transmission acceleration methods based on RDMA can significantly speed up AIscenarios such as federate