1、 Contents 1 6G Network Design Principles.1 2 6G Network Architecture.2 3 Thoughts on 6G Network Architecture.4 3.1 Network Control Unit(NCU).6 3.2 Network Packet Processing Unit(NPU).7 3.3 Network Data Unit(NDU).8 3.4 Network Intelligence Unit(NIU).9 4 Key Features of the New 6G Network Architecture

2、.11 4.1 Simplified Architecture.11 4.2 User data-centric.14 4.3 Native network AI.15 4.4 Distributed Autonomy.18 5 New Architecture and Services.20 5.1 Immersive XR Services.20 5.2 Intelligent Multi-Access Services.22 5.3 Autonomous Private Network Services.23 6 Prospects and Initiatives.25 Acronyms

3、.26 References.28 Foreword With the large-scale commercialization of 5G network,6G technology research has been systematically carried out in the industry.The 6G network architecture is an important research field,and an important part and technical means for the implementation of the 6G vision.Orie

4、nted to 2030,China Telecom put forward the 6G vision1 of Intelligent connectivity everywhere,communication and sensing in virtuality and reality and new services represented by immersive XR and Metaverse put forward new demands on the network.Cloud-network convergence is an inevitable trend for the

5、6G network.The integration and penetration of DOICT technologies bring new ideas into network architecture design.This white paper expresses thoughts of China Telecom and ZTE on the 6G network architecture based on research on the 6G network architecture.This white paper is divided into the followin

6、g chapters:Chapter 1,6G network design principles,which describes the ideas and basic principles of network architecture design.Chapter 2,6G network architecture,which describes the logical functions of the 6G network at different layers.Chapter 3,thoughts on 6G network architecture,which describes

7、the logical functions and working mechanisms of the 6G network at the network function level.Chapter 4,key features of the new 6G network architecture,which describes the core capabilities of the new architecture that are different from those of 5G.Chapter 5,new 6G architecture and services,which ex

8、plains how the new architecture supports 6G service scenarios such as immersive XR,Metaverse,and autonomous private network.Chapter 6,prospects and initiatives,which puts forward the issues and cooperation initiatives for further research on the 6G network architecture,China Telecom and ZTE are will

9、ing to cooperate with industry partners to promote the research on the 6G network architecture.1 1 6G Network Design Principles In the future,the general idea of 6G network architecture evolution is to improve mobile network service capabilities and fully upgrade user experience by focusing on user/

10、customer requirements,guiding by cloud-network convergence.The research ideas are as follows.Meet the requirements of new services and scenarios,and carry out 6G network architecture research oriented to connection+.Service scenarios such as immersive XR,holographic communication,man-machine interac

11、tion,and machine-machine interaction raise higher requirements on functionalities and performance of the network.These services need not only basic connection services,but also edge computing service resource,high-precision environment and object sensing services and network AI services.This white p

12、aper believes that the 6G network will provide integrated services with the capabilities of aggregate connection,sensing,intelligence and computing.Absorb the essence of technological innovation and integration,and consider technological advancement and implementable capabilities.Oriented towards ne

13、w service requirements,the 6G network architecture should be designed in accordance with the evolution trend of mobile communications and the development trend of DOICT integration,absorb the essence of emerging technologies,evaluate the impact of service-based wireless network,space-air-ground inte

14、gration,and ISAC on the 6G network,analyze the feasibility of introducing emerging technologies such as AI,cloud computing,big data,and block chain,comprehensively clarify the connotation and scope of the 6G network architecture.Focus on end-to-end systematic design to avoid network isolation.The gu

15、arantee of service experience requires end-to-end technical guarantee of the network,any segment failure will cause the whole system failure.Achieving the overall functional and performance requirements of the network requires end-to-end systematic design.In terms of key technical solutions and prot

16、ocol design,end-to-end horizontal coordination of terminals,wireless networks,and core networks are required,and also need the cross-layer and cross-domain vertical coordination of the operational management domain,bearer network domain and cloud-network resource domain.Enhance the flexibility desig

17、n of 6G networks and build a more open industrial ecosystem.6G networks should be more flexible.Cloudification and service-based are important technical measures.The 6G network supports on-demand orchestration and deployment,on-demand service loading,and on-demand traffic scheduling,which enhance th

18、e network adaptability to services.Especially for customized network requirements from different 2 industries,flexibility and agility are crucial,and are prerequisites for introducing more participants and building a more open ecosystem.The 6G network architecture design should fully consider the re

19、quirements in the 2B field.In the process of digital economy development and industrial digital transformation,5G networks play an important role.Facing the future,6G network services will have broad prospects in the 2B field,and will play an increasingly important role in digital transformation.Pra

20、ctice shows that the networks of 2B and 2C have significant differences in terms of service features,network deployment requirements and management solutions.Different industries and enterprises have different network requirements.Therefore,it is necessary to strengthen the overall planning of 2B an

21、d 2C in network architecture design.2 6G Network Architecture From the perspective of the overall network,the 6G network framework is a hierarchical architecture with three-layer and four-plane,see Figure 2-1.Figure 2-1 Hierarchical Architecture of the 6G Network Based on the idea of cloudification

22、and service-based architecture,three layers consist of the cloud-network resource layer,network function layer and application enabler layer from bottom to top.1.Cloud-network resource layer:It is the infrastructure and resources of 6G network deployment,mainly used for the bearer network functions.

23、The resource types include various types of computing(CPU,GPU,FPGA,etc.),storage,network,spectrum,and 3 various dedicated hardware.Network resources mainly refer to the IP bearer network,implementing the connection among cloudified network functions.2.Network function layer:It includes the network f

24、unctions that should be supported by the 6G network.It provides the connection+network service,which can be further divided into the control plane,user plane,data plane and intelligence plane.3.Application enabler layer:It aggregates network service capabilities and general application service compo

25、nents,providing services for applications or peripheral ecosystems through capability exposure and application enabler frameworks to implement unified application enabler management.The operation and management of cloud-network runs through all layers.With the evolution trend of cloud-network conver

26、gence,in addition to inheriting the current network management,service provisioning and billing functionalities,the O&M of 6G networks will also introduce new operational functions,including the intelligent orchestration and scheduling function,digital twin network management function,network sharin

27、g management based on blockchain and other technologies.Four Plane is an extension of the mobile core network description by 3GPP,including the control plane,user plane,data plane and intelligence plane.To meet the functional requirements of the connection+,the logical functions of four-plane in the

28、 network function layer are as follows:Control plane:The control plane implements unified control of network connection services,intelligent services,service resource and sensing services.As the center of network control,the control plane works closely with other planes to implement integrated contr

29、ol of multiple access,authentication,mobility management,session management,policy control,computing resource allocation and management.User plane:The user plane supports network programmability and flexibly defines data processing policies.The functionalities include tunnel management,data flow ide

30、ntification,service sensing,deterministic communication,data encapsulation,data forwarding and traffic steering,processing and forwarding user data,environmental object sensing data,and AI task data.Data plane:The data plane is further separated from the service logic to reduce the tight coupling be

31、tween data and service processing.In the network,an independent data plane is introduced to implement unified management of various data,and provides exposure services to the control plane,user plane,and intelligence plane through standard interfaces.Various data includes static data and dynamic rea

32、l-time data,such as user subscription data,user status data,network status data,connection data and resource utilization data.Intelligence plane:The intelligence plane is the intelligent component of the 6G network,supporting the comprehensive intelligence of both core network and access network.The

33、 intelligence plane serves not only the intelligence of the 6G network itself,but also the 4 intelligence requirements of users and service applications.The intelligence plane provides network AI-related functionalities,including data modeling,model training,inference and decision,knowledge graph,fe

34、edback and evaluation.In the overall framework of the three-layer four-plane network,the network function layer is the key to network architecture evolution.This white paper focuses on a prospective discussion of the network function layer.3 Thoughts on 6G Network Architecture Based on the current n

35、etwork,6G network needs to consider more about supporting multi-mode heterogeneous and full-scenario access,including how to meet the network customization requirements of users in various industries;how to design the network architecture that cannot be completely solved in the existing network.And

36、it is necessary to consider how new services such as the XR,Metaverse and man-machine interaction can obtain better network services.Therefore,this white paper puts forward the goal of network architecture evolution:simplified,native network AI,distributed autonomy with user data-centric,and propose

37、s DDAA(Data-driven Distributed Autonomous Architecture).As shown in Figure 3-1,the network evolution architecture is simplified into four network units,including NCU(Network Control Unit),NPU(Network Packet Unit),NIU(Network Intelligence Unit),and NDU(Network Data Unit).These network units are inter

38、connected by using two buses:SBI and data channel.The four network units respectively correspond to the control plane,user plane,intelligence plane,and data plane of the network function layer in the hierarchical 6G network architecture.In this way,the existing network with a large number of NFs and

39、 complex signaling can be optimized,and the new network can be flexibly customized,thereby enabling service innovation in the humans,machines,things and virtual world in the future.5 Figure 3-1 Data-driven Distributed Autonomous Architecture Network Control Unit(NCU):As a network control center,it i

40、s responsible for driving tasks,and implementing coordination of tasks such as connection,computing,intelligence and sensing.Network packet processing Unit(NPU):Inherits the basic functionalities of data forwarding on the existing user plane,introduces programmability of the user plane,flexibly defi

41、nes the processing logic of the user plane,and evolves to functionalities such as computing&network service sensing,deterministic communication,and autonomous domain interconnection.Network Data Unit(NDU):Unified and aggregated data plane not only includes UDR/UDSF,but also integrates network data a

42、nd computing data,and provides standardized signaling interaction modes and data access channels to solve the decoupling problems of different vendors in the existing data architecture design.Network Intelligence Unit(NIU):Provides functionalities such as AI task scheduling,inference and decision,mo

43、del training,knowledge and model repositories 2.It implements cross-layer and cross-domain intelligence among multiple nodes through distributed learning and collaboration,assists internally in implementing autonomous network management and intelligence services,and provides flexible and customized

44、AIaaS4 for external applications.The dual-bus architecture is used for interconnection of network units.One is the service-based bus for signaling interaction,which inherits the existing service-based design concept,develops the service-based scope from control plane of the core network to the user-

45、plane(i.e.NPU),and further extends to the wireless access side which including RAN-CP and RAN-UP to implement end-to-end service-based in the mobile network.The 6 other is a new data bus proposed in this white paper,that is,a data channel.Different network units interact with the NDU through the dat

46、a channel to obtain and process data efficiently.Data interaction among different network units is also transmitted through the data channel.3.1 Network Control Unit(NCU)NCU aggregates functional entities in the traditional control-plane network,and constructs and manages user-level digital twins,dy

47、namically presents PUE in the form of DUE digitalization.A single DUE provides complete capabilities,instead of being supported by multiple NFs combinations.The DUE capability can be stronger than the PUE capability.The full user data capability of the DUE can drive the network implement the corresp

48、onding functionalities.The NCU includes four core parts:basic function,extended function,access adaption and digital twin UE,see Figure 3-2.Basic function:Inherits the existing basic network capabilities,including mobility management,session management,authentication and charging,and implements corr

49、esponding signaling interaction.Extended function:On the basis of providing basic capabilities,driven by new future services,there are more capabilities for options:such as enhanced deterministic connectivity,computing&network collaboration,integrated positioning and sensing integration,and expand n

50、ew capabilities based on full DUE data as required.Access adaption:The basic principle is to be compatible with legacy and new user terminals,the NCU provides different adapters to flexibly implement access adaption to wireless and satellite,etc.In this way,new and legacy terminals can achieve plug-

51、and-play,user awareness,so that the new-generation network capability does not depend on the maturity of terminal chips.Digital Twin UE:The DUE is the smallest management unit,it has own data repository.The NCU dynamically loads UE-side data,network data,computing data,and core data on demand.The PU

52、E interacts with the DUE to drive the DUE to process network tasks.7 Figure 3-2 Logical Functions of Network Control Unit 3.2 Network Packet Processing Unit(NPU)In accordance with the separation design principle of control and forwarding,the NPU implements service enhancement on the basis of the exi

53、sting UPFs,including service-based interfaces with the control plane,data network,and wireless interaction interfaces.The NPU inherits the functionalities of routing and forwarding of UE service data,action,and policy execution from traditional user-plane,it focuses on the evolution of user-plane pr

54、ogrammability,computing&network service sensing,deterministic communication,and autonomous domain interconnection,and implements on-demand and scenario-based network customization to meet the requirements of real-time mobile services such as drone,Internet of Vehicles,and brain computer interface co

55、llaboration in the future,see Figure 3-3.Figure 3-3 Logical Functions of the Network Packet Processing Unit User plane programmability:To reduce forwarding performance restrictions caused by general hardware,use the programmable switching chips,smart NICs,and DPU heterogeneous chips to implement eff

56、icient data processing and power 8 consumption reduction.In addition,the programmable technologies is used to implement the programmability of communication network protocols,including protocol independence and flow table programmability.This improves the flexibility and exposure of the user plane,i

57、ntegrates required network functionalities into the user plane,removes useless protocols as required,and simplifies the user plane.Computing&network service sensing:In order to realize computing&network collaboration,the future network has the capability of sensing and scheduling of computing,networ

58、k,and storage resources 5.Build a cognitive user plane,and add the capability of service resource sensing and service resource routing.The integrated scheduling of computing&network resources can be implemented based on the identified QoS of the computing&network to meet the consistency of mobile lo

59、w-latency service experience.Deterministic communication:Integration of terminals,networks,clouds and services,and to achieve network information measurable and visible information from the NPU to the wireless side and edge computing side,including the latency,jitter and packet loss rate.In this way

60、,deterministic connections are always provided in real time to avoid abnormal services caused by UE mobility,network congestion,and computing node failures,and guarantee reliable and high-quality operation of real-time mobile services.Autonomous domains interconnection:In a distributed network,the N

61、PUs of different autonomous domains are interconnected.A VPN tunnel is established between the NPU in the private network and the NPU in the public network.Users in the private network can access external networks through the NPU as the proxy in the public network.3.3 Network Data Unit(NDU)The const

62、ruction principle of network data unit is the network applications are separated from user data.The data of each autonomous domain can be stored separately in the local area.Based on the concept of data development and design according to new application requirements,network data units classify,desi

63、gn,store,converge,and process data.The data capabilities of network data units are exposed through API interfaces on the data channels,which can provide basic data information for NIU to perform intelligent network analysis,and the network data units are core logical units in the 6G network architec

64、ture.Cross-domain data is shared through data channels.At the same time,based on data transmission cost and data security considerations,multi-domain data also supports cross-domain federated learning.9 User data:End-to-end user data,including UE access capability data,low/medium/high speed marking

65、data,realtime trajectory data,current network capability,ongoing tasks,etc.Core data:Core static user data in the 6G network,including user subscription data,service policy data,and service subscription data,etc.Network data:Network slice management data,operational data,and processing capability da

66、ta of each node in the network,for example,network slice management,load status of each node in the network,and network SLA data.Computing data:It is mainly the data of the future network where users are consuming computing resources and the services built on top of them,including general computing

67、resource data,heterogeneous computing resource data,and service resource data.On the basis of network data,develop a series of network management functionalities,see Figure 3-4.At the same time,the SBI and data channels are enabled on the NDU for the NCU,NPU,and NIU.Figure 3-4 Logical Functions of t

68、he Network Data Unit 3.4 Network Intelligence Unit(NIU)The NIU is used to implement the functions of the intelligence plane,and is a basic unit for processing intelligent tasks in the network.Internally,the AI technologies are applied in the network to improve the intelligence of the communication n

69、etwork.In addition to the basic load analysis and congestion prediction,it can predict the resource deployment and distribution requirements of the NCU and NPU,analyze the service tailoring and customization requirements to make the network simple,elastic and intelligent.Besides,it provides flexible

70、 AI network services(AIaaS)for the 3rd party applications,for example,assists applications in network performance analysis and prediction,network route 10 selection,and service QoS configuration.As shown in Figure 3-5,NIU includes three main functions:AI service function,intelligent storage function

71、,and distributed coordination function.Figure 3-5 Logical Functions of the Network Intelligent Unit AI service function:Based on the data sensing and preprocessing provided by the NDU,AI service function mainly provides model training,inference and decision in the full-cycle AI process.The model tra

72、ining layer assists in inference,decision and data processing,which generally includes basic learning such as LSTM(Long Short-Term Memory)and RNN(Recurrent Neural Network).In the future,it will focus more on distributed learning such as federal learning,transfer learning,and multi-agent reinforcemen

73、t learning,and is equipped with model evaluation capabilities and environments for evaluating the learning outcomes.On one hand,the inference&decision layer is used for network and service related AI inference tasks,for example,network resource status prediction,and can also form final decisions or

74、suggestions based on inference results,for example,providing the best network route.Intelligent storage function:The intelligent storage function consists of a complete model repository and knowledge repository.The knowledge repository implements knowledge management and knowledge services.The model

75、 repository implements model query,addition,update and deletion.The intelligent storage function manages the data obtained in the AI service process,the trained models,and generated knowledge to improve the learning efficiency and intelligence level,thus to achieve the dual-drive of data and knowled

76、ge.Models and knowledge will also be used by edge nodes to be close to users for real-time needs,etc Distributed coordination function(AIaaS):The distributed coordination function is used to support cross-layer and cross-domain coordination among between multiple NIUs,or between a NIU and another do

77、main.Model training,inference,and decision 11 are jointly completed by means of data or model parameter interactions.The distributed coordination function specifically includes functions such as data and knowledge sharing,model training collaboration,inference and decision assistance,and computing s

78、cheduling.The 6G network intelligence architecture is composed of standalone NIUs and built-in NIUs.The standalone NIU is a special functional unit providing AI services.According to the differences in AI capabilities,service scenarios,coverage areas,and computing resources,it can include aggregated

79、 NIUs,regional NIUs,service NIUs,and edge NIUs.It supports AI services,intelligent storage,and distributed coordination on demand.The built-in NIU adds AI capabilities for NCU,NPU,NDU and other network units,and mainly provides data sensing and processing,model training,inference and decision.In add

80、ition,the built-in NIU can obtain data or knowledge,inference or decision support from the standalone NIU.NIU is an intelligent enabler of the network,which will change from connection-based intelligence to task-driven intelligence,and more widely matches intelligent application scenarios.NCU,NPU,ND

81、U,other NIUs,and other domains can send intelligent analysis tasks to the NIU through enabled interfaces.After receiving the intelligent tasks,NIU coordinates connections,data,model algorithms,computing resources to complete AI tasks.4 Key Features of the New 6G Network Architecture Based on the tho

82、ughts about the new 6G network architecture proposed in the previous chapter,this chapter further describes the key features of the new architecture.4.1 Simplified Architecture Based on the service-based design,the new architecture achieves a simplified network architecture through three main aspect

83、s:architecture simplification,interface simplification and procedure simplification,as shown in Figure 4-1.12 Figure 4-1 Simplified network architecture 1.Simplified architecture In addition to the basic NFs,there are more and more NFs with the support of new functions and scenarios.Although service

84、-based interfaces are used,the procedure design and data information still rely on each other 23.Fault location or new service deployment involves multiple NFs and interoperability interfaces,bringing complexity.The new architecture implements the reconstruction and aggregation of NFs of the same ty

85、pe,simplifies the functionalities of each NF on the existing control plane by functionality classification.The NCU integrates some functions into a single network unit,such as the AMF,SMF,NSSF,NEF,and logic processing parts of PCFs and UDMs,etc.The static data of UDM and PCF and UDR data are process

86、ed by the NDU in a professional way.The functionalities of UPF and NWDAF are optimized as the NPU and NIU respectively.The new architecture directly reduces more than forty NFs to about four network units to implement simplification of function types and flexible customization of services.In additio

87、n,the networking hierarchy is reduced and the topology structure is simplified.In addition to the significant reduction in the number of NFs,the simplification of the network architecture is reflected in the following:both the public networks and private networks use the same design architecture,fac

88、ilitating network interconnection,elastic scalability,and reducing the difficulty of O&M as well;the internal services of the network unit can be dynamically tailored to adapt to new scenarios such as ISAC and avoid affecting external interfaces;the NCU provides a unified service discovery mechanism

89、 through service mesh,facilitating flexible service complementary and migration;by introducing the DUE,NCU designs a user-based processing mechanism to enhance user-based sensing and service capabilities;the NCU supports adaptation to various heterogeneous access such as satellite access,fixed acces

90、s,mobile access,and WIFI access,and backward compatibility to the 4G/5G,providing capabilities in accordance with the unified DUE mechanism internally to reduce the complexity of multi-network coexistence,unifies network services,and simplifies the overall network planning,design,and O&M.2.Simplifie

91、d interfaces 13 The interface simplification is reflected in the number of service-based interfaces.Through the aggregation of NFs,the number of interfaces will be greatly reduced.There are a large number of existing 5G NFs,and at least more than 80 logical interfaces2.Through the design of network

92、units such as NCU,the interface between the original NFs becomes an internal implementation,and external services are designed only according to the interworking requirements.The number of interfaces can be reduced to about ten.In addition to reducing the number of interfaces,new professional and st

93、able data channels are created between network units,and a dual bus-driven architecture is formed together with SBI service interfaces to separate data access from signaling interaction.On one hand,signaling interaction meets real-time,customized and flexible service requirements.On the other hand,i

94、n terms of data,the original service-based interface is changed into a unified data access interface to achieve large throughput and efficient data query and processing.The two types of interactions do not affect each other.They can be processed asynchronously and executed concurrently,and finally a

95、chieve the simplification and efficiency of service interfaces and data interfaces.3.Simplified procedures The procedure simplification is reflected in the number of interactions among NFs,that is,the number of message steps.Through NF aggregation,especially those with strong dependence,the informat

96、ion interactions among original NFs is converted into the processing of micro services within the network unit,and therefore,the procedure is simplified.For example,the procedures of session processing and policy acquisition with more interactions between AMF,SMF,and PCF in 5G networks6 are implemen

97、ted internally by the same NCU in the new architecture,and usually do not involve external interactions.Only in an interworking or mobility scenario,an operation such as necessary context transfer is performed with other NCUs as required.It is estimated that the number of steps in the registration,s

98、ession establishment,and handover procedures is reduced from 25+to 8+,20+to 4+,and 30+to 8+respectively.The number of message interactions is greatly reduced,and thereby the procedure is simplified.In summary,the new 6G architecture is expected to achieve a 90%reduction in the number of NFs,an 80%re

99、duction in the number of service-based interfaces and a 70%reduction in the number of message steps,ultimately achieving a simplified network architecture,as shown in Table 4-1.14 Table 4-1 Comparison of Network Architecture Simplification 5G 6G Simplified architecture AMF,SMF,PCF,NRF,NSSF,UDM.40+NC

100、U,NDU,NIU,NPU.4+Simplified interfaces N1,N2.N88 80+/10+Simplified procedures Registration 25+Registration 8+Session 20+Session 4+Handover 30+Handover 8+4.2 User data-centric The ultimate service object of the network is users.The design of the existing network architecture and the organization of da

101、ta are aimed at providing users with connections through the interconnection of NFs.The new architecture will highlight user-based services,evolve to user data-centric,and implement rapid iterative development of functionalities around user data.As shown in Figure 4-2,the new architecture is mainly

102、reflected in two aspects:user data-driven and digital twin UE.Figure 4-2 User data-centric 1.User data-driven The existing user data is deployed in a centralized mode,and separates computing from storage.For example,subscription data,policy data,application data,and structured data can be stored in

103、the UDR,and unstructured data such as UE contexts can be stored in the UDSF2.However,in fact,due to deployment requirements,technical implementation,and vendor interoperability and other issues,most of the data is still scattered in NFs such as UDMs,PCFs,AMFs,SMFs,and NSSF,presenting the characteris

104、tics of data dispersion and diversification.15 The new architecture further classifies and separates data from the users perspective through the NDU,including core data such as user subscription and policies,network data,and computing data.The aggregation processing and analysis of several types of

105、data can also implement on-demand customized network functionalities,such as connection enhancement,computing&network collaboration,positioning and sensing integration.The unified and standard data processing mode of NDU makes it easy to simplify data processing logic by using professional database

106、technologies.2.Digital Twin UE Existing NFs are divided in accordance with functional logic,for example,access and mobility management,session management,and policy control.Services are invoked among NFs to provide UEs with data connections.The procedure mechanism and data processing are designed ba

107、sed on different functionalities required for connection,which is not helpful for user-dimension management,and it is difficult to directly improve user experience.NCU enhances user service design,dynamically constructing a user-level digital twin DUE in the NCU to associate with the PUE,and impleme

108、nting signaling processing and management based on users.The main advantage is that the DUE can pull full amount of data and provide external capabilities around data to avoid data inconsistency.DUE data aggregation requires only one copy,reducing the amount of duplicate data and addressing the issu

109、es of wasted storage resource and high consumption of network resources,thus improving overall performance.In addition,the DUE has AI capabilities to reduce data collection,thereby reducing network overhead for AI services to enable AI evolution.In summary,a user data-centric network architecture is

110、 designed based on the aggregated analysis of user data and the processing mechanism of user-level digital twin.Tasks are driven based on user service requirements to achieve a holistic coordination of connectivity,computing and capabilities.4.3 Native network AI The AI native features of the 6G net

111、work architecture are mainly reflected in the optimization and enhancement of both the networking and the intelligent service capabilities.1.Distributed AI native architecture The existing 5G network mainly uses centralized standalone AI(e.g.,NWDAF)7.Each NF(e.g.,AMF,SMF)lacks AI capabilities and de

112、pends on the centralized analysis of standalone AI,which may also bring some problems as a result,for example,centralized AI needs to collect a large amount of data,which seriously consumes communication network resources;centralized processing,analysis and feedback is more difficult to 16 guarantee

113、 latency and other performance requirements under large-scale networking;AI analysis often requires computing resources,and centralized AI analysis requires a high aggregation of computing resources,making it more difficult to efficiently utilize multi-node and edge resources.The intelligent 6G netw

114、ork architecture presents the characteristics of the AI native and distributed collaborative,that is,the hybrid model of standalone AI+built-in AI.As shown in Figure 4-3,the NCU,NPU,and NDU and other network units,on one hand,have their own data processing and AI analysis capabilities through the bu

115、ilt-in NIU service.At the same time,they form distributed learning and inference frameworks together with the standalone NIU,and dynamically allocate the data,capabilities and resources of each NIU on demand.Together,they build multi-node collaborative AI native networks.In addition,the built-in DUE

116、 of the NCU can also act as a proxy for the user to independently create AI tasks,monitor and guarantee services,and dynamically migrate networks or services in real time.The advantage of the architecture is that in line with the principle of AI analysis as close to the data source as possible,the c

117、entralized standalone NIU can collect data as needed,or transmit model parameters in a distributed learning manner,thus greatly reducing the consumption of communication resources.The multiple NIUs collaboration focuses on distributed,multi-node,and cross-level service collaboration of large-scale n

118、etworks,it makes full use of the data,capabilities and local computing resources of standalone and built-in NIUs,and flexibly processes them in accordance with the performance requirements such as timeliness and model accuracy,and finally achieves the swarm intelligence coordination with comprehensi

119、ve and secure information and complementary and enhanced features.Figure 4-3 Distributed AI native architecture 2.Intelligent service capability enhancement Considering that the current network AI has issues such as inflexible data processing,17 weak distributed learning capability,lack of decision-

120、making and feedback mechanism,lack of knowledge accumulation,and simple collaboration,the 6G AI native architecture will focus on improving the capabilities of data native,distributed learning,autonomous decision-making,knowledge storage,and coordinated exposure.Data native:The NIU works with the ND

121、U to ensure data reliability and privacy.It can sense,collect,process,and store massive,polymorphic,and heterogeneous data,and enhance data-related AI analysis to implement data native.Distributed learning:Because 6G network security and privacy requirements are higher and services are gradually ext

122、ending to the far edge,centralized learning methods will face problems such as data privacy,communication overhead,and service timeliness.It is necessary to focus on multi-node collaborative learning methods of distributed AI architecture.For example,federated learning 8,multi-agent reinforcement le

123、arning9,and swarm learning10 should be supported among multiple NIUs and among NIUs and DUEs.Autonomous decision-making:The trends of changeable 6G networks,service complexity,and elasticity promotes the introduction of the AI technologies to enhance the decision-making capability of networks and se

124、rvices.The decision functionality of NIU works with other network units to support self-adaptive multi-agent coordination,dynamic adjustment,and closed-loop autonomous decision-making services.Knowledge storage:The key to efficient and sustainable 6G operation is to pay attention to knowledge accumu

125、lation,storage,and flexible application.Through the construction of the knowledge repository,NIU supports knowledge-driven AI learning methods such as transfer learning 11 and continuous learning 12,to achieve both data and knowledge-driven network intelligence.Coordinated exposure:Due to the data a

126、nd terminal heterogeneity and large variation in service and performance across domains and levels of the 6G network,6G AI native network requires cross-layer and cross-domain NIU coordination to better support AIaaS services,mainly including a full range of AI service coordination such as inference

127、 and decision,model training,computing resources,data and knowledge.In short,multiple NIUs work together to build the Intelligence plane helping 6G networks to meet the requirements for elastic,simple and customized services to create an intelligent and simplified network.18 4.4 Distributed Autonomy

128、 The existing network mainly uses the networking mode:relatively centralized networking in 2C public network and dispersed networking in 2B customized network.In the future,with the emergence of new scenarios and the raising of extreme requirements for service experience,independent private networks

129、 will increase in various industries due to factors such as security,service scope,capabilities and performance.There is a need for separation and autonomy as well as a need for complementary enhancements through interconnection between networks.In addition,there is a high risk of security and relia

130、bility due to the very large area of impact of failures for a relatively centralized network deployment,therefore,a distributed construction mode is urgently required.The new architecture will change from relatively centralized and dispersed mode to the Intra-domain autonomy and Inter-domain interco

131、nnection mode.Each autonomous domain is similar to a local area network,and when a failure occurs,it is limited in scope and does not affect the overall network.In addition,industrial manufacturing,V2X,XR immersive experience,and other scenarios need to use the autonomous domain to make users,networ

132、ks and services as close to each other as possible in terms of space,so as to achieve better guarantees,such as high bandwidth and low latency,and obtain better service experience.Multiple autonomous domains can also be interconnected through the backbone network to form a wider area network,dependi

133、ng on the expansion of coverage or access requirements,as shown in Figure 4-4.Figure 4-4 Distributed autonomy 1.Intra-domain autonomy A certain number of RANs,infrastructure of a certain scale,and distributed core network can build an independent autonomous domain within a certain space range.A sing

134、le domain may include a complete set of NCU,NPU,NDU and NIU.Alternatively,19 there may be only an NCU,an NPU,and a partial NDU,where the NDU and the NIU may leverage the NDU and the NIU capabilities of the upper-level network through negotiation among multiple NCUs.The NCU is a centralized control n

135、ode.Intra-domain RANs,NPU,NDU,and NIU are registered in the NCU,and mutual discovery is supported through the NCU to implement configuration-free and plug-and-play for RANs;NPU can also do plug-and-play when performing performance expansion is performed.2.Inter-domain interconnection The purpose is

136、to expand the network coverage.The traditional design concept is to expand the network coverage as wide as possible.The distributed architecture expands the network coverage through inter-domain interconnection.When the network needs to expand the coverage or access capability,the original network i

137、s basically not affected,for example,there is no need for interoperability between multiple NFs,etc.Each NCU has a unique ID,which is used for interconnection.On the basis of interconnection among different NCUs,new capabilities can be further negotiated.For example,negotiation between a private net

138、work and a public network,some private services are processed internally,when the internet service is used,a VPN transmission tunnel is established between a private network NPU and a public network NPU through negotiation among NCUs,that is,the internet can be accessed via the public network NPUs.I

139、n terms of security,inter-domain security and mutual trust may be authenticated by using a block chain or other distributed security technologies,so as to perform interconnection for the transmission of data,signaling,etc.3.Inter-domain migration When a user is moving and is about to cross the edge

140、of an autonomous domain,inter-domain migration needs to be implemented.By monitoring the moving trajectory of a user,the DUE in the NCU can recommend the next proper autonomous domain to the user.The recommendation should guarantee users current service continuity.For example,the bandwidth,latency,a

141、nd computing required by the service of the next autonomous domain need to be considered.If the service is more important,it can be negotiated between NCUs before inter-domain migration,e.g.whether the user requirements can be met in the new domain,and if so,the service and network are migrated conc

142、urrently.Besides,during the migration,a user can maintain multi-domain access until both a new domain and services are ready,and then perform seamless handover.To sum up,the new architecture adapts to differentiated network requirements through distributed network technologies,enabling intra-domain

143、autonomy and inter-domain interconnection.20 5 New Architecture and Services The 6G network provides greater bandwidth,lower latency and more comprehensive access modes,more convenient architecture,and more intelligent network capabilities,and will bring more diversified new services and new modes,s

144、uch as new service capabilities oriented to immersive XR,Metaverse and other services,multi-access services oriented to space,air,ground and sea,distributed autonomous private networks oriented to all fields and industries and other 2C and 2B application scenarios,laying a solid foundation for diver

145、sified life and production in the future with stronger network capability.5.1 Immersive XR Services Immersive XR and Metaverse,as the mainstream communication mode in the future,will bring users an immersive and multi-dimensional sensing experience through interaction with all the senses of users.Im

146、mersive XR and Metaverse have strict requirements for network bandwidth,latency,jitter,reliability,audio and video rendering quality and efficiency,see Figure 5-1.Figure 5-1 New Architecture Supporting immersive XR As the digital twin of users XR terminal,the DUE plays a center role in the network.O

147、n the 21 one hand,network and data requirements are proposed to the NCU and NDU in accordance with service requirements to achieve the coordination of connectivity,computing and capabilities.On the other hand,it is responsible for driving the NDU/NIU to perform various tasks related to services,incl

148、uding AI tasks:data collecting,model training,inference and decision,computing scheduling and other processes,and feeds back data,models,and analysis results,etc.to the NCU as required.The NCU is responsible for resources scheduling and coordination control,implements service requirement perception

149、and flexible customization capabilities,monitors QoE and guarantees QoE.When an XR service is initiated,the NCU requests and schedules the network and computing resources that match the service requirements to ensure that the multi-dimensional sensing and real-time interaction service flows can be t

150、ransmitted reliably at the same time.During the XR service procedure,the NCU drives the NDU/NIU to initiate several tasks(such as a trajectory tracking prediction task,a network status monitoring task,and a video quality monitoring task)in accordance with different DUE requirements,to monitor and an

151、alyze network QoS and service QoE in real time.Through real-time coordination between network resources and service flows,dynamic traffic adaptation between services and networks is implemented,ensuring the best service experience of users.Trajectory tracking prediction task:Tracks the user trajecto

152、ry in real time,and predicts the user trajectory through the NIUs AI capabilities.During the XR service process,NCU prepares network connections and computing resources in advance in accordance with the results of the trajectory tracking prediction task to ensure service continuity when the user mov

153、es,promoting the XR and Metaverse from fixed places to broader spaces.Network status monitoring task:Monitors the network status in real time.Once the network capabilities do not match the service requirements,the NCU immediately sends a new resource allocation request to dynamically adjust the reso

154、urces to ensure user experience.At the same time,the NCU can expose the network status information to the service through the capability exposure interface,so that the service can perceive the network status in real time.When the network is congested,the XR service can rapidly adjust the service cod

155、ecs and transmission traffic models,helping the service quickly adapt to the network status while discarding unimportant data packets,so as to avoid further network congestion.Video quality monitoring task:Monitors the video quality of the XR service in real time.The NCU can accurately sense the XR

156、service requirements,the NPU can identify the features and attributes of various XR service flows,uses the AI capabilities to set the specific QoS to guarantee the transmission of audio and video streams,tactile signal streams,sensor signals,etc.The NCU reduces the complexity 22 of management throug

157、h AI technologies,and supports the consistent experience in the case of high-density XR service flows.AI technologies play an important role in audio and video rendering and multi-sensor integration.As an intelligent network unit,NIU can have a variety of built-in and external AI capabilities,such a

158、s AR marking,XR rendering,audio and video codec,ASR and TTS,etc.The standalone NIU and built-in NIU are combined to form a distributed AI native architecture,and together with deep learning,federal learning,and other AI technologies,provide powerful AI capabilities for immersive XR and the Metaverse

159、.5.2 Intelligent Multi-Access Services Integration Space-Air-Ground-Sea is an important direction for future 5G-A/6G communication 13 providing broadband access or wide area IoT for mountains,deserts,seas and air areas.First,the integrated communication covers the entire domains,so that users can ac

160、cess the network anytime anywhere.Second,the future Space-Air-Ground-Sea communication is fully heterogeneous and multi-mode access,and needs to support multiple access modes such as satellite,aircraft,and aerial platforms(HAPS),drone,ground base station,and fixed network.Network heterogeneity and d

161、iversified transmission environments pose great challenges to the core network convergence.Multi-Access Adapter For multi-mode access,different adapters are used to adapt to different access modes.The adapters are implemented in the form of components or micro services,and can be plug-and-play in th

162、e NCU for adaption and conversion of different protocols,see Figure 5-2.The NCU uses unified DUEs to perform terminal modeling and management,and provides unified authentication and user data management to ensure service continuity.Figure 5-2 New Architecture Supporting Multi-access Convergence 23 S

163、pace-Air-Ground-Sea Integration The development trend is to integrate the air-based network,space-based network,and ground-based network.Through the collaboration of NCU,NDU and NPU,the integrated network can fully cover the space,air,ground and sea,see Figure 5-3.In addition,the future Space-Air ne

164、twork will face many challenges such as a high latency and a fast-moving speed.For example,each LEO satellite serves a user only in minutes at a time,resulting in frequent handover.Therefore,enhanced mobility management capabilities are required to simplify procedures and improve reliability of hand

165、overs.In the new simplified architecture,the number of handover messages is reduced and the handover procedure is shortened through quick synchronization of DUEs master and replica of user data.Figure 5-3 Space-Air-Ground-Sea Integration Network 5.3 Autonomous Private Network Services The distribute

166、d and autonomous domain network mode can well support private networks in various industries.Industry customers can customize networks as needed.For example,a customized deterministic network is used for collaboration among industrial robots,or a customized small core network is used to build its ow

167、n mobile LAN for field exploration.Other temporary private networks in sports events and concerts,small mobile private networks in airplane and ferry scenarios,and campus private networks in enterprise and school scenarios can be customized for the core network as required.A customized private netwo

168、rk has the following features:A private network is an autonomous domain.Similar to a wired LAN,a terminal 24 may access a service in a dedicated network,or may implement mutual access among terminals in the private network,for example,a campus network user accesses a campus network service,or mutual

169、 access among industrial terminal devices in an enterprise.The private network can be connected to the public network.A terminal can be connected to a public network through the private network to implement Internet service access and differentiated charging.The local private network implements the

170、bridging functionality between the terminal user and the users home network,for example,users on the plane can access Internet services.Private networks can be shared.External non-private network users can access the public network and data services through the private network,solving the network co

171、verage problems in some areas.In this case,an enterprise private network needs to implement a bridging functionality between a user and a public network while ensuring its own network security.For example,a public network user temporarily accesses the enterprise private network and is transferred to

172、 the public network.A private network can be connected to multiple operator networks.Users accessing the private network may be subscribers of different operators.Therefore,the private network also needs to implement interconnection and interworking with public networks of multiple operators.Private

173、 networks can be interconnected.A user of another private network may access a private network at the current location to a home private network,for example,a user may roam and access to each other in two friendly private networks.The features of accessing the public network through a private networ

174、k or accessing the private network through the public network are mainly based on the inter-domain interconnection capability of the distributed network,as shown in Figure 5-4.For details,please refer to the chapter of“4.4 Distributed Autonomy”.25 Figure 5-4 New Architecture Supporting Autonomous Pr

175、ivate Network 6 Prospects and Initiatives The network architecture is a key issue for 6G research,and is the basis for the realization of the vision of Intelligent connectivity everywhere,communication and sensing in virtuality and reality.Scenario requirements and the introduction of the DOICT tech

176、nology will affect the 6G network architecture.Based on current research,this white paper proposes a preliminary consideration on the new 6G network architecture.With the evolution of technologies,this white paper will be updated iteratively.At the current stage,the 6G network architecture and key t

177、echnologies are still under discussion.Many problems need to be discussed,and some key technologies need to be verified.End-to-end cross-layer and cross-domain coordination is very important,the cross-domain coordination mechanism,cross-protocol information transfer methods,and even cross-domain sta

178、ndard joint design are required.China Telecom and ZTE are willing to collaborate with academia and industry partners to discuss 6G network architecture and key technologies,address bottlenecks,and jointly promote 6G technology progress and industrial development.26 Acronyms Abbreviation Description

179、2B To B 2C To C 3GPP 3rd Generation Partnership Project AI Artificial Intelligence AMF Access and Mobility Management Function ASR Automatic Speech Recognition AIaaS AI as a Service DDAA Data-driven Distributed Autonomous Architecture DUE Digital User Equipment DOICT Data&Operational&Information&Com

180、munication Technology DPU Data Processing Unit FPGA Field Programmable Gate Array GPU Graphics Processing Unit HAPS High Altitude Platform Station IPv6 Internet Protocol Version 6 ISAC Integrated sensing and communication LAN Local Area Network LEO Low Earth Orbit Satellite LSTM Long Short-Term Memo

181、ry ML Machine Learning NCU Network Control Unit NDU Network Data Unit NIU Network Intelligence Unit NPU Network Packet Unit NF Network Function NEF Network Exposure Function NRF NF Repository Function NSSF Network Slice Selection Function NWDAF Network Data Analytics Function PCF Policy Control func

182、tion PUE Physical User Equipment QoE Quality of Experience QoS Quality of Service RAN-CP Radio Access Network Control Plane RAN-UP Radio Access Network User Plane RNN Recurrent Neural Network 27 SBA Service Based Architecture SBI Service Based Interface SMF Session Management Function SRv6 Segment R

183、outing over IPv6 TTS Text to Speech UDM Unified Data Management UDR Unified Data Repository UDSF Unstructured Data Storage Function UE User Equipment V2X Vehicle-to-Everything XR Extended Reality 28 References 1 China Telecom R&D Institute,6G Vision and Technology White Paper R.2022 2 3GPP TS 23.501

184、,System architecture for the 5G System(5GS);Stage 2(Release 17),v17.7.0,Dec.2022.3 3GPP TS 23.502,Procedures for the 5G System(5GS);Stage 2(Release 17)”,v17.7.0,Dec.2022.4 NGMN.6G Use Cases and AnalysisR,2022 5 Wang Weibin.ODICT-converged network 2030J.2022(01):47-56 6 IMT-2030(6G)Promotion Group.Ov

185、erall Vision and Potential Key Technologies of 6G White Paper R.2021 7 3GPP TS 23.288,Architecture enhancements for 5G System(5GS)to support network data analytics services(Release 17),v17.7.0,Dec.2022.8 Z.Yang,M.Chen,K.-K.Wong,H.V.Poor and S.Cui,“Federated Learning for 6G:Applications,Challenges,an

186、d Opportunities,”Eng.,vol.8,Jan.2022,pp.3341.9 A.Feriani,E.Hossain,Single and Multi-Agent Deep Reinforcement Learning for AI-Enabled Wireless Networks:A Tutorial,in IEEE Communications Surveys&Tutorials,vol.23,no.2,pp.1226-1252,Secondquarter 2021,doi:10.1109/COMST.2021.3063822.10 Warnat-Herresthal,S

187、.,Schultze,H.,Shastry,K.L.,et al.Swarm Learning for decentralized and confidential clinical machine learning.Nature 594,265270(2021).11 F.Zhuang et al.,A Comprehensive Survey on Transfer Learning,in Proceedings of the IEEE,vol.109,no.1,pp.43-76,Jan.2021,doi:10.1109/JPROC.2020.3004555.12 Parisi G.I.,Kemker R.,Part J.L.,Kanan C.,Wermter S.Continual lifelong learning with neural networks:A review.Neural Networks,113(2019),pp.54-71 13 CCSA:Research Report on Star-Ground Convergence Technology for 5G Enhancement and 6G.2021