《中国移动研究院:2022年基于数字孪生网络的6G无线网络自治白皮书(英文版)(52页).pdf》由会员分享,可在线阅读,更多相关《中国移动研究院:2022年基于数字孪生网络的6G无线网络自治白皮书(英文版)(52页).pdf(52页珍藏版)》请在三个皮匠报告上搜索。
1、A Digital Twin Network Approachfor 6G Wireless Network AutonomyWhite Paper(2022)China Mobile Research Institute(CMRI)ContentsPreface.11.Current Situation and Vision of Network Operation and Maintenance Optimizationof Operators.32.Self-Intelligence Network Signposts and Inspirations.42.1 Signposts.42
2、.2 Challenges.62.3 Inspirations.73.6G Network Autonomy Based on Network Digital Twin.93.1 Basic Concepts.103.1.1 Digital Twin Network.103.1.2 Three Contents.123.1.3 Five states.133.1.4 Double closed loops.153.2 Technical Features.163.3 Network Architecture.193.3.1 End to End Architecture.193.3.2 Dat
3、a Plane.213.3.3 Intelligent Plane.233.4 Key Technologies.243.4.1 Data Acquisition and Analysis Techniques.253.4.2 Data Enhancement Techniques.263.4.3 Pre Verification of Data and Knowledge Collaboration Driven.283.4.4 Knowledge Graphs and Graph Neural Networks.283.4.5 Simulation Serviceability.303.4
4、.6 Correction techniques for pre-verification results.313.5 Whole life cycle autonomy of network.323.5.1 Continuous planning.333.5.2 Virtual and real connection.343.5.3 Combination of prevention and cure.353.6 Case description.353.6.1 Optimization of beam weight for large-scale antennas.353.6.2 Inte
5、lligent deep RAN slices.373.6.3 Federated scheduling of multi-dimensional resources.394.Summary and Outlook.41Abbreviations.44Writing unit and staff.44References.45China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper1 1PrefaceAt present,all walks of life are using
6、 the advanced cloud platform technologies and networkconnection services to digitalize and automate the transformation of serviceto improve theservice agility and the flexibility.In the field of mobile communications,although operators havebeen exploring,researching and deploying applications in aut
7、omation of network management andservice provision for many years,however,it cannot solve the problems of high network energyconsumption,complex multi-standard interoperation,high operational cost and low efficiency.Atthe same time,as the network evolves in the direction of programmable,software-dri
8、ven,service-oriented architectures,the complexity and scale of network operation and maintenance(OAM)have reached an unprecedented height.The introduction of new services and technologiesalso puts forward more stringent requirements for the agility of network operations,and operatorsurgently need a
9、more comprehensive,intelligent,scalable and affordable network automatedOAM system.The automation of network OAM has different granularity,which can be the automation oftask,function or process,or the automation of network and service life cycle management.Atpresent,the level of 5G network OAM autom
10、ation of is low,most of which rely onprogram-solidified expert rules and automatic scheduling flows.In some scenarios,the 5Gnetwork OAM still need to rely on manual operations.Network OAM automation based onintelligent means is still fragmented and plug-in.Fragmented refers to a use-case drivenappro
11、ach to achieve a higher degree of automation and less manual intervention for certainfunctions,forexample,basestationself-startinginSON,neighborhoodrelationshipself-optimization,PCI self-optimization,MRO,etc.Plug-in refers to the collection andsummary of relevant data to the network management or re
12、lated platform for training,and themodel is sent to the corresponding network element to generate the intelligence required for OAM.This chimney-typeautomated system and R&D model can enhance the automation level ofnetwork management to a certain extent under the existing network structure,but due t
13、o thelimitations of the existing network structure,the difficulty of ensuring the validity and real-timenature of data,and the difficulty of interoperability and sharing of data between different vendors,the efficiency of network automation is low and the effect is difficult to meet expectations.In
14、the future,the 6G network will construct a brand-new automated network OAM systemthrough the network digital twin,and realize the high-level autonomy of the whole network lifecycle.The digital twin network is a network system consisting of physical network entities andtheir twin digital networks tha
15、t can be mapped in real time.The digital twin entity of the networkis the dynamic modeling or mirror copy in digital space of the real network entity.The digitaldomain generates perceptual and cognitive intelligence through rich historical and real-time dataas well as advanced algorithmic models.It
16、can continuously optimize and simulate the optimalChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper2 2state of physical networks,issue the corresponding OAM operations in advance,correct thephysical network automatically,and solve the network element or network
17、 fault in advance toachieve the effect of cure future disease.Then it can form a closed loop through the collection ofcorrected data to evaluate the operation and maintenance results.Through this closed-loopinteraction between digital and physical domains,cognitive intelligence,and automatic OAMoper
18、ations,networks can quickly recognize and adapt to complex and dynamic environments,andrealize the autonomy of planning,building,maintaining,optimizing and curing the whole lifecycle of the network.China Mobiles Digital Twin Network(DTN)White Paper,released in September by 2021,sets out the concept
19、and definition of Digital Twin Network(DTN),and gives the referencearchitecture of DTN,key enabling technologies,the capability classification system and typicalapplication scenarios 1.On the basis of this,this paper further researches and explores the digitaltwin network for 6G wireless network aut
20、onomy,introduces the related basic concepts of 6Gwireless network autonomy,clarifies the technical features,designs the network architecture,plans the key technology system,and illustrates the whole lifecycle autonomy of 6G wirelessnetworks based on digital twin networks through specific case studie
21、s.Finally,the key technicalproblems which need to be further studied and solved are put forward.China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper31.Current Situation and Vision of NetworkOperation and Maintenance Optimizationof OperatorsThe 5G network has great
22、ly improved the communication quality,but the endless newnetwork services and the continuously expanding network scale have brought many challenges tothe OAM and optimization of the 5G network,and the complexity of the network OAM has beenincreasing.The deployment of innovative technologies is also
23、becoming more difficult.In the traditional 2C scenario,there are some problems,such as high terminal powerconsumption,limited energy-saving effect of the current network,low proportion of 5G terminalson-off,low ratio of 5G shunting and resident,and complex optimization and maintenance of theadjacent
24、 areas of the system.In the 2B scenario,the base station equipment product libraryplanned to industry private network is not perfect,and it cannot satisfy the changes of coverageand scenarios flexibly.At the same time,the 2B service exclusive maintenance system is notperfect,and network maintenance
25、staff skills cannot meet the cross-layer and cross-domainmaintenance capacity needs.In addition,the construction cost of the private network is high,andthe 2B service application scenarios are quite different.Scenarios with large uplink and largedownlink rate classes,low delay control classes,and hi
26、gh reliability requirements will lead tohigher networking cost.The terminal problems can also affect service stability.With the development of cloud computing and virtualization technology,the traditionalnetwork has begun to shift to software and programmable,showing the new characteristics such asc
27、loudification of resources,on-demand design of services,orchestration of resources,etc.,whichmakes network management and OAM more complex.Due to the lack of an effective simulation,prediction and verification platform,it is difficult to shift from the existing scheduled maintenanceto the predictive
28、 OAM.And network optimization operations have to be directly applied to thecurrent network infrastructure,which leads to high cost and risk of network optimization.On theother hand,due to the high reliability requirement,it is difficult for the network operators to usethe current network environment
29、 directly for the research of network innovation technology.TheR&D cycle of new network technologies is long and the deployment is difficult.Facing the future,the mode of network communication,the type of service carried,theobjects served by the network and the type of equipment connected to the net
30、work will present amore diversified development trend,which makes the network highly dynamic and complex,andrequires the network to be more flexible,scalable and responsive to requirements,becoming anautonomous network with self-optimization,self-evolution and self-growth capabilities.Theself-optimi
31、zation network predicts the trend of the future network state in advance,intervenes thepossible performance deterioration in advance,continuously optimizes and verifies the optimalChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper4state of the physical network i
32、n the digital domain,and issues the corresponding OAM operationsin advance to correct the physical network automatically.The self-evolution network analyze andmake decisions on the evolution path of network architecture and functions based on artificialintelligence,including the optimization and enh
33、ancement of current network elements and thedesign,implementation,verification and implementation of new elements.The self-growthnetwork identifies and forecasts different service requirements,automatically arranges anddeploys various domain network functions,generates end-to-end service flows to me
34、et the servicerequirements,automatically expands the capacity of the sites that are short of capacity,andperforms automatic planning,hardware self-on,software self-loading operations for areas that donot covered by the network.2.Self-Intelligence Network Signposts andInspirationWith the development
35、of 5G era,many new industries and scenarios will emerge,newapplications will be deployed,and new network technologies will be introduced and expanded inthe future.How to implement the OAM of the increasingly complex large-scale networkefficiently and continuously introduce new technology rapidly and
36、 iteratively is a commonproblem for the industry.Facing the communication network with the largest number of customersin the world,the richest services and the largest network,China Mobile network OAM isaccelerating the transformation and upgrading of digital intelligence,strivingto build acloud-net
37、work integrated,highly automated and intelligent network system,and consolidating thefoundation for digital and intelligent transformation in all walks of life2.China Mobile hasreleased“China Mobile Automated Driving Network White Paper”in 2021,which defines thescenario classification standard for p
38、rocesses,enhances the level of network OAM autonomy in astep by step and iterative manner.The paper provides inspirations and thought for realizing 5Gand even the 6G network autonomy.2.1 SignpostsThe self-intelligence network requires the automation and intelligentization of thenetwork.It combines t
39、he intelligentization and automation technologies such as AI withnetworks to achieve the network predictability and operational autonomy,aiming to construct theautomatic and intelligent OAM ability of the whole life cycle of communication network.The self-intelligence network hierarchical framework
40、divides the network autonomycapability into six levels of L0 L5.The self-intelligence network hierarchy from L0 to L5meansdifferent network characteristics and capabilities.At the same time,based on the guiding principleof the framework of TM Forum self-intelligence network,and combined with the act
41、ual needs ofChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper5network OAM and management evaluation,the characteristics of each level of network autonomycapability are described from the angle of guiding the implementation of the IT system.Thecorrespondence and
42、 evolutionary path 2 of network autonomy capability and its characteristicsare shown in the following figure:Fig.2.1-1.Self-intelligence network classification and evolution pathIn line with the concept of self-intelligence network,China Mobile plans the digitalizationand intellectualization transfo
43、rmation of network OAM,strengthens the construction ofautomation and intelligentization capacity,and sets the overall goal of reaching L4 in 2025 at thelevel of network OAM autonomy.L4 means that the network is highly intelligent,can completethe intelligent workflow of automatic perception,analysis,
44、decision and making execution for thetarget scene,and can form a complete automatic closed-loop process without human intervention.Among them,the ability of automatic perception makes the whole process of data collection,processing,association,sharing,storage and management automatic to realize high
45、 efficiency andstandardization of data management.Theautomatic analysis ability is to realize the intelligentanalysis and modeling of the network,to improve the generalization and generality of the model.The automatic decision-making and execution ability is to have certain credible evaluation andau
46、tonomous decision-making ability,and allow the algorithms to evaluate the quality of decisionsindependently and execute them automatically in the network,so as to realize the security andcredibility of the algorithm.The network can provide rich platform-level,distributed computingservice,support the
47、 automation of information management and control and process interaction onthe interface,have a certain intentional interface capability,and automatically generate rules orpolicies according to user intent requirements.China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy Whi
48、te Paper6The goal of 6G network is to realize L5 level self-intelligence network.L5 means that thewhole process of the network is intelligent,the network has the full intention management ability,there is no need to set service rules artificially,and it can customize strategy according to theintelli
49、gence of service scenario and automatically iterately evolve,so as to truly realize that thenetwork changes with the business.L5 level self-intelligence network aims to provide fullyautomatic,zero-wait,zero-touch,zero-failure innovative network services and ICT services toconsumers and vertical indu
50、stry customers,and build a communication network with self-service,self-repairing,self-optimization,self-healing capabilities.2.2 ChallengesThrough the practice of 5G self-intelligence network,we find that different from other simplesystems,the mobile communication system has the characteristics of
51、high systematization,highcomplexity,high dynamicity and high reliability,which brings many challenges to dataacquisition,algorithm development and application deployment.In the data aspect,the current network data is closed,and it is difficult to obtain the depthdata within the network element.In th
52、e current network system,whether it is the 2/3/4G networkor the newly built 5G network,the distinction between thedata used for network elementself-optimization and network O&M management,the functional coupling relationship betweenthe network elements and the network management equipment and the sy
53、stem architecture havenot changed substantially.The type richness and statistical precision of the internal data of thenetwork element are much better than the data reported to the southbound network managementby the network element,which in turn is better than the data that can be obtained by theno
54、rthbound network management equipment of the operator.However,due to the consideration ofthe privacy of the device,the consumption of the device processing resources and the transmissionresources,a large amount of data within the network elements is not open to the operators,and itis only used for t
55、he algorithm R&D of interfaces within the network element and between networkelements of the same manufacturer.As a result,operators cannot get a complete,real-time andprecision perception of the real state of the network,and the management and control of thenetwork can only be supported by the surf
56、ace statistical data of various longer periods.At thesame time,there are some problems of poor data quality in the available data,such as partial datamissing,unbalanced sample set and lack of data annotation,etc.Due to the limited data storageresources,most of the data stored in the network are half
57、-year historical data with poor timeliness.In the aspect of algorithm,due to the high dynamic of the network environment,thedistribution of the original data of network characteristics also shows a high dynamic.Its oftennot true in real networks that the precondition used by academia for algorithmic
58、 innovation,independent and distributed data.Modeling based on non-independent same distribution andnon-independent different distribution data is a common problem to be solved in the industry,China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper7which needs to be
59、abstracted into scientific problems and make full use of network knowledge.Onthe other hand,due to the high reliability guarantee of the mobile communication system,thealgorithm and the network autonomous decision made by it must be safe and credible tointerface with the current network operations s
60、ystem.At present,only after the algorithm is put online or the decision is executed,the network performance can be judged by the networkperformance statistical indices,which has the risk of network performance deterioration.In orderto guarantee the credibility of the algorithm,it is necessary to bre
61、ak through the interpretabilityof the algorithm.Although there has been some research and breakthrough in the academic world,there is no systematic and automatic solution to solve the black box of the nonlinear modelrepresented by the deep neural network.The selection of simple models with better st
62、ructuralinterpretability may result in performance loss.Finally,there are a lot of repetitive developmentand optimization in the algorithms for different network autonomous scenarios.The centrallydeployed AI open platform can reduce the R&D cost of algorithms to a certain extent,but it stillcannot s
63、olve the chimney model of R&D.The algorithm cannot be reused and migratedefficiently among nodes,regions and autonomous environments.In the aspect of application deployment,the network autonomous scenario of mobilecommunication network is multitudinous and complex,most applications need to concatena
64、temany production links,and the correlation analysis between the links mainly depends on thehuman operator.In the field of network OAM alone,11 types of scenarios,37 core competenciesand more than 1300 sub-processes are sorted out,and the analysis work is time-consuming andlabor-consuming.At the sam
65、e time,the R&D mode of many AI applications is still the high-costand low-efficiency chimney mode.The algorithm is deployed on the server connected to thenetwork management equipment or network element equipment,which displays plug-in mode.The generation,deployment,evaluation and iterative optimizat
66、ion of intelligence in networkautonomous scenario rely heavily on manual work,which affects the improvement of networkautonomous level.2.3 InspirationsTowards the goal of achieving level L4 self-intelligence networks by 2025 and full level L5self-intelligence networks in the future,the network needs
67、 to achieve a deep openness of data interms of perceptual automation,and realize the automation of the whole process of dataacquisition,data Processing,data association,data sharing,data storage and data management.Inthe aspect of analysis and decision automation,the network needs to support the tru
68、sted andinterpretable framework of the algorithm,and needs to support autonomous decision-makingwithout human intervention.In the aspect of application execution,the network must be able torealize the self-generation of autonomous scenarios and autonomous targets,and support theintelligentization of
69、 the whole process of the targets and scenarios.In the aspect of networkChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper8architecture,it should have complete autonomous closed loop,platform-level and distributedcomputing power.The network needs to support the
70、management information and processinteraction needed to perform automation on the interface,and has the ability of purposiveinterface.There are four key technology needs to be highlighted:1.Deep data opennessAs mentioned above,a large amount of data within the network element is not currently opento
71、 the operators,which leads to the operators inability to achieve a comprehensive,real-time andsophisticated perception of the true state of the network.Operators control of the network canonly stay within the scope supported by various long-term statistical data at the surface level,andcannot achiev
72、e a high level of autonomy,which has become the primary problem to be solved for6G network autonomy.Considering the privacy protection of the equipment supplier,it is not a feasible solution todirectly expose all the data inside the equipment.So what data will be further opened?How is itopen to oper
73、ators?It should not only protect the internal privacy inside the device,but alsorepresent the state of the network element to meet the needs of the network autonomous scenario.In this regard,in addition to the new standardized data types,we also need to think about morecost-effective acceptable tech
74、nical solutions.2.Data value increasesAt present,although a large amount of data can be obtained by means of soft and hard mining,road surveying,MDT,MR,and network management data extraction,there are problems such aspoor data quality,low value density,low acquisition efficiency and poor timeliness,
75、which notonly waste a lot of network storage and transmission resources,but also cannot support thegeneration of network intelligence.How to make the data exactly match the requirements of thenetwork autonomous scenario in every phase of the whole life cycle,such as collection,processing,storage,kno
76、wledge transformation and application,so as to avoid the waste ofresources and increase the value density of data.It is the key problem to be solved in the processof 6G network autonomy.In order to achieve the above goal,6G network needs to enhance the ability of data analysisand value mining,to acc
77、urately perceive massive heterogeneous data,to actively push and collectdynamic data on demand,and to avoid data redundancy.We can use the AI technology to minedata value,and improve the response speed of data service by cloud-edge-end distributed storageand strategy optimization of different value
78、data.Through model training and knowledgereasoning,scenario dynamic adaptation is carried out to realize the intelligent orchestrationinvocation of the data services and the intelligent adjustment of the configuration parameters.3.Autonomous requirements generationAt present,the method of discoverin
79、g problems artificially and solving them by AI is alwayslimited by the limitation of artificial cognition,and the intelligentization ability and potential ofnetwork cannot be maximized.At the same time,relying on manual stovepipe problem-solvingChina MobileA Digital Twin Network Approach for 6G Wire
80、less Network Autonomy White Paper9methods often produces conflicting results between different network operations optimization usecases,for example,the improvement of the coverage performance of a cell results in the increaseof the disturbance and the decrease of the service experience index,which l
81、eads to thereciprocating and inefficient optimization work.6G-oriented,more abundant network scenarioswill have different requirements for network architecture,functions and services,and theconfiguration-based approach will not be able to meet and adapt them to the maximum.Facing theunknown new indu
82、stry and new demand in the future,it is even more unrealistic to manuallydiscover and summarize the need for network autonomy.All of the above show that 6G networkneeds the technical means of auto-sensing and auto-excavating network autonomous demand,auto-generating and auto-dispatching network auto
83、nomous use cases,so as to avoid the conflictbetween use cases,and ensure the best effect of overlay implementation,minimize humaninvolvement.4.Low cost trial and optimizingIn order to avoid negative impact on network performance and users service experience,theOAM optimization decisions in current n
84、etworks are usually evaluated and validated by experts ortested for a long time before implementation.The introduction of a new feature,for example,typically requires lab performance testing,field performance testing,and on-line FOA testingbefore upgrading current network equipment for up to a year.
85、On the other hand,the effect ofdecision after implementation can only be known through statistical or road test related networkperformance indicators,and the iterative optimization cycle is long,the cost is high.To achievethe goal of complete self-intelligence for 6G networks,the network needs to ha
86、ve the ability ofautomatic evaluation,high-efficiency closed-loop,agile iteration,so it needs low-cost trial anderror and high-efficiency optimization technology.3.6GNetworkAutonomyBasedonNetwork Digital TwinAfter years of development,the digital twin technology is becoming more and more perfect,and
87、 it is becoming the new grasp of national digital transformation,the new direction ofmultinational enterprise business layout and the new focus of global information technologydevelopment.The digital twin technology provides a new idea and solution for the research of the6G network.By constructing t
88、he digital twin network,6G can realize the autonomous networkwith self-optimization,self-evolution and self-growth capabilities,to meet the aforementioned keytechnical requirements.The 6G wireless network autonomous system based on the digital twinnetwork needs to introduce new concepts,design new n
89、etwork architectures and constructcorresponding key technology systems.China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper103.1 Basic ConceptsDigital twin is defined by professor Grieves.It consists of three parts:physical objects,virtual objects,and the flow of
90、information between physical objects and virtual objects.Subsequently,the term Digital Twin was formally introduced in a NASA technical report anddefined as a system or aircraft simulation process that integrates multiple physical quantities,scales,and probabilities.In recent years,with the continuo
91、us development of the digital twintechnology,its application in aerospace,intelligent manufacturing,smart city and other fields hasbeen relatively mature.Digital twin is becoming a new focus of national digital transformation,anew direction of business layout of multinational enterprises,and a new f
92、ocus of globalinformation technology development 5.Although the definition and connotation of the digital twin technology have not been agreedbetween academic and industrial circles,there is a preliminary consensus on its typicalcharacteristics.The first characteristic is bi-directional precise mapp
93、ing,which means that thedigital twin technology can realize the full presentation,accurate expression and dynamicmonitoring of physical objects in twin contents.The second characteristic is real-time,whichrequires a full real-time connection between physical objects and twin contents.The twin conten
94、tsare the representation of the physical objects which changes with the time axis,and the mappingof the real-time states of the physical objects.The third characteristic is extensibility.The digitaltwin technology has the capabilities to integrate,add and replace digital models and can beextendedfor
95、 multi-scale,multi-physicalquantity,multi-levelmodelcontent.Theforthcharacteristic is whole life cycle.The digital twin technology can run through the entire productlife cycle,including design,development,manufacturing,service,maintenance and even scraprecycling.3.1.1 Digital Twin NetworkThe digital
96、 twin technology provides a new idea and solution for realizing 6G networkautonomy,which is constructing the digital twin network by digital twin of network itself.DigitalTwin Network is a real-time interactive mapping network system which is composed of physicalnetwork entities and their twin digit
97、al networks.The twin digital network elements of the physicalnetwork elements can be constructed by means of data acquisition and simulation,and then thedigital twin contents of the network elements and the digital twin content of the network areformed in the digital domain.In this system,various ne
98、twork management and applications canuse the digital twin contents of the network to analyze,diagnose,simulate and control the physicalnetwork efficiently based on data and models.As the digital mirror image of the physical network facilities,the digital twin content of thenetwork has the same netwo
99、rk elements,topology and data as the physical network,which canrealize the fine replication of the whole process of the network and equipment.It provides aChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper11digital verification environment close to the real netw
100、ork for optimizing operation and policyadjustment of network OAM.Therefore,compared with the traditional simulation platform,the AImodel and the pre-verification result of the network-based digital twin content have higherreliability.On the other hand,the digital twin network also records and manage
101、s the behavior ofthe digital twin content of the network,supports its tracing and playback,and thus can completethe pre-verification without affecting the network operation,greatly reduces the cost of trial anderror.In addition,the digital twin network has the capability to build and expand itself,a
102、nd can becombined with AI technology to explore new service requirements that have not yet beendeployed to the current network and verify their effectiveness in the digital twin network,thusrealize the network self-evolution.China Mobiles Digital Twin Network(DTN)White Paper 1 puts forward the overa
103、llstructure of three layers,three domains and double closed loops.On this basis,this paperproposes that resource objects in digital twin networks will have three contents and five statesfor 6G wireless network autonomy,and the digital twin network can be continuously optimizedthrough the double clos
104、ed loops.The complete logical architecture design is shown in thefollowing figure:Fig.3.1.1-1.three layers,three domains and double closed loops architecture for Digital TwinNetworksNote:The network twin content refers to the digital twin content of the network,and networkplan content refers to the
105、digital plan content of the networkChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper123.1.2 Three Contents6G wireless network autonomy objects can theoretically be a variety of granular networkresources,such as functions,services,base stations,chips,boards,spec
106、trum,power,and so on,which depend on the specific network autonomy scenario.These resource objects will have threecontents in the digital twin network:the physical entity,the digital twin content and the digitalplan content.The physical entity is the physical object itself.For hardware,it is the for
107、m of hardwareitself,such as base station boards,antennas,chips,etc.For software,it is the carrier of software,such as mirror files,various forms of software code,etc.The digital twin content is the digital object twinned by physical objects.For example,twinning the network element generates the digi
108、tal twin content of the network element,andtwinning the network generates the digital twin content of the network.By collectingmulti-dimensional measurement data on physical entities,and modeling physical processes ofphysical entities,we construct images of physical entities in the digital domain.In
109、 the digital twinnetwork,the digital twin content is synchronized with the physical entity state.Considering thecomplexity and cost,the digital twin content does not need to reproduce the physical entity 100%in an all-round way,but selects the state attributes and processes to be tracked for modelin
110、gaccording to the requirements of the network autonomous scenario.The digital plan content is a digital model which is generated by the network after planningthe expected state of the physical object at some time in the future.It represents the optimizationof physical objects for the future.The netw
111、ork adjusts the physical entity according to the data inthe digital plan content,which makes it approach the goal of network autonomy.The digital plancontent may include adjustable configuration parameters of physical entities,loadable softwarefunctions,optimized connection relationships,and so on.T
112、he physical entity and the digital twin content of the object are commonly mentioned in thedigital twin technology,while the digital plan content is a new concept based on the characteristicsof network autonomy.The digital plan content represents the decision-making content and resultsof the digital
113、 domain in the network autonomous scenarios,which is the aim of keeping the digitaltwin content in synchronization with the physical object,and reflects the level of decision-makingintelligence of the digital twin network.It is an important link in the network autonomous closedloop.Currently,the con
114、struction of digital twin systems in various industries is mainly concernedwith the real-time and accuracy of synchronization between physical objects and their digital twincontents in the upward direction,while in the downward direction,the delay,accuracy andperformance jitter of the digital plan c
115、ontent implemented into the physical object also directlyaffect the effect of network autonomy.The relation of three contents is that the digital twincontent is the mapping of the physical entity state.According to the digital twin content andrelated data,the network generates the digital plan conte
116、nt,and optimizes the physical entityChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper13according to the digital plan content.The following diagram shows the relationship between thethree contents:Fig.3.1.2-1.The relationship of three contents in the digital twi
117、n network3.1.3 Five statesDifferent from the separate plan,construction,maintenance and optimization stages in thecurrent network,6G network will realize high-level autonomy in the whole life cycle throughdigital twin,that is,all states of network equipment from delivery to withdrawal are managed in
118、the digital twin network.In order to distinguish the different relationships and technicalrequirements between devices at different stages of the life cycle and the twin digital network,andin order to be compatible with current network devices that do not support digital twin,this whitepaper propose
119、s five states of the design scheme:the initial state,planning state,service state,twinstate and energy-saving state.Fig.3.1.3-1.The relationship between the three contents and the five-statesChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper14Figure 3.1.3-2.The
120、transformation between the five statesThe initial state is the state in which the resource object has not entered the network or hasbeen removed from the network.The object does not exist in the digital domain or physicaldomain of the digital twin network.At this point,the object may exist in the pr
121、oduct library as aninventory resource for network operations,or it may be removed from the network as a backupresource to make up for lost time.If the object has ever run on the network,its correspondinghistorical run data is preserved.The planning state means that the object only exists in the plan
122、 of digital twin network,thatis,the digital twin network has planned the digital plan content of the object,but the physicalobject has not entered the network or is disconnected from the network due to the fault.The digitalplan content may include deployment location,specification parameter value,pa
123、rameter value,physical property value,connection relation and so on.Physical objects are initially configuredaccording to a digital plan content after network initiation.The service state is the state that an object enters after it has started the network and isinitially configured according to the
124、digital plan content.With the initial configuration dataobtained,the object can run normally on the network and provide services.In this state,if thedigital twin network needs to plan the object continuously,the object will enter the twin state,otherwise(if the computing resource is limited,or the o
125、bject is a static object)will stay in theservice state,the plan content is no longer continuously updated,and objects run in the networkbased on the most recent plan configuration data.The twin state.The object has a digital twin and a digital plan in the digital twin network.The physical object is
126、kept in sync with the digital twin content,and is continuously optimizedand adjusted according to the latest digital plan content.In this state,the digital twin content anddigital plan content are connected with physical objects for continuous optimization.Energy-saving state.The object exists in th
127、e network,but its digital plan content is emptyand there is no digital twin content.When certain resource objects are not needed in the planningscheme of the digital twin network,they can be turned off to save energy.For hardware modules,shutting down an object is equivalent to a power-off operation
128、,and for software entities,shuttingdown an object is equivalent to de-instantiating.When the object reappears in the planningscheme in the future,it will enter the planning state.From this,it can be seen that the planning state is a transitional state,and the objectstemporarily in the planning state
129、 will be transferred to the service state,or the twin state or theenergy-saving state after being connected to the network,and the equipment on the network willmainly switch between these three states.For devices that do not support digital twin,they willswitch between the service state and the ener
130、gy-saving state.For devices that support digital twin,when they are in the twin state,the technical requirements for the interaction between physicalobjects and twin digital objects are the most.The correspondence between three contents andfive states is shown in Table 3.1.3-1.China MobileA Digital
131、Twin Network Approach for 6G Wireless Network Autonomy White Paper15Table 3.1.3-1.The relationship between three contents and five statesThree contents Five statesInitial statePlanning stateService stateTwin stateEnergy-savingstateDigital plancontentNoYesYes,staticYes,dynamicNoDigital twincontentNoN
132、oNoYes,dynamicNoPhysical entityNoNoYesYesYesNote:For the digital plan content and the digital twin content,the mark No means nocurrent digital plan contents or digital twin contents,but there may be historical digital plancontents or digital twin contents.3.1.4 Double closed loopsIn order to achieve
133、 the 6G high-level network autonomy,the network should have completeautonomous closed loops in architecture,as well as platform-level and distributed computingpower in order to realize hierarchical and muti-domain multi-level closed loop autonomy.Theautonomous closed loop mentioned in standardizatio
134、n and industry organizations generallyincludes four links:observation,analysis,decision-making and implementation.In the digital twinnetwork,the digital twin content synchronizes the physical network state to realize observation,the digital plan content corresponds to the decision result,and the phy
135、sical entity is theexecution object.For the analysis part,because the digital twin network has the digitalnetwork environment with high fidelity,it can verify the decision-making effect and provide thepossibility for closed-loop optimization of decision-making in the digital domain,and makes theanal
136、ysis itself a closed-loop process.Therefore,the autonomous closed loop of digital twinnetworks is double closed loops.The inner closed loop is the process of simulation verification and iterative optimization forthe generation of the digital plan content.Digital twin networks contain all kinds of fu
137、nctions orcomponents needed for network life cycle autonomy,such as all kinds of plan,simulation tools,intelligent models needed for iterative optimization,etc.The ultimate goal is to make the nexttime plan of the network based on the digital twin content in advance,that is,to generate thedigital pl
138、an content of the network.The problem solved by the inner closed loop is how togenerate the theoretically better digital plan content in the digital domain for the next timenetwork state.The outer closed loop means that after the digital plan content is sent to the physicalnetwork,the new state of t
139、he physical network is synchronized to the digital twin body,and thedigital twin network evaluates the effect of network autonomy in the digital domain.According toChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper16the gap between the inner closed-loop and the
140、goal,the function and parameter of the innerclosed-loop are analyzed and optimized,so that the inner closed-loop can generate more effectivedigital plan content,thus approaching the goal of network autonomy.Due to various technicalfactors,the realization of the goal of network autonomy cannot be ach
141、ieved by only one planningprocess.The problem solved by the outer closed loop is the inevitable deviation between the twindigital network and the physical network caused by technical factors.For example,due to theerrors introduced by data acquisition and transmission technology,the digital twin cont
142、ent of thenetwork are not consistent with the real state of the network.Due to the insufficient type ofcollected data and the suboptimal prediction algorithm,the prediction of the network state at thenext moment is inaccurate.Due to the insufficient accuracy of the simulation technology,theimplement
143、ation effect of the planned configuration or new features is not verified,etc.3.2 Technical FeaturesSection 2.3 summarizes the key technology requirements for high-level autonomy of 6Gnetworks based on the 5G self-intelligence network practice,which includes deep data openness,data value density inc
144、rease,autonomous demand self-generation,and low-cost trial and erroroptimization.To meet these technical requirements,this white paper presents four technicalcharacteristics of digital twin networks:1.The models in digital twin networks can be divided into data class models,simulation classmodels an
145、d intelligent class models.Both standardized and non-standard models coexist.The model in the digital twin network can be used in every link of double closed loops.Themodel of the digital twin content and the digital plan content is the frame structure of them.Themodel of digital twin content is com
146、bined with real-time data to generate digital twin contents.The digital twin network plans the related contents according to the model of the digital plancontent,and generates the digital plan content.A data class model is a set of data attributes thatare dependent on data collection and represent t
147、he network history and current state,and is mainlyused to track the dynamic changes of network states.In order to support the deep openness of data,we need to develop more comprehensive data class model standard according to the specificrequirementsofnetworkautonomousscenarios.Consideringthedifferen
148、ceofvendorimplementation,non-standard data class models can be used in combination with standard dataclass models if they can bring performance gain.On the other hand,the digital twin contentgenerated only by the data class models cannot carry out the trial-and-error and optimization ofnew features
149、or decision-making actions effectively.The simulation class model that can simulatethe physical process and function of the network is also needed,and the intelligent class model ofthe input-output mapping relationship should be established.The three can be selected orcombined according to the deman
150、d.For the simulation class models,the digital twin content of thereal physical process or function of the network can be generated by extracting the latest value ofChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper17the configuration data from the data class mod
151、els.For the intelligent class models,based on thetraining samples extracted from the data class models,the mapping relation among the networkelements can also be used as the function digital twin.The three models need to provide astandardized interface to achieve model interconnection and replacemen
152、t.The implementation ofthe models can be diverse.2.Self-generation and self-analysis of network autonomy requirements can realize the datatwin content and the digital plan content which are customized and dynamically generated.In order to support the full self-intelligence network,the digital twin n
153、etwork can analyze andpredict the network state continuously based on the digital twin content of the network,automatically discover the network governance requirements and scenarios,and generate networkautonomous use cases.The digital twin network can be automatically resolved into thecorresponding
154、 network autonomous use cases for the upper-level intention input by the operator.The digital twin network can quickly locate the root cause and implement the cure plan for thefault and alarm.All these depend on the accurate description of network states and datarelationship by the digital twin cont
155、ent of the network.In order to guarantee the performance andreduce the cost of data acquisition and transmission,the digital twin content and the digital plancontent of the network should be on-demand customized and dynamically generated.Among them,on-demand customization refers to the selection of
156、model classes(data class models,simulationclass models,intelligent class models)of digital twin contents and digital plan contents accordingto the requirements of network autonomous scenarios and use cases,the construction of the modelof digital twin content and digital plan contents,the parameters
157、configuration of digital twincontents and digital plan contents,and the maximization of data volume and frequency of virtualreal synchronization.Dynamic generation means that the models and contents of network digitaltwin contents and digital plan contents can be dynamically increased or decreased a
158、ccording to thechanges of the network,and dynamically adjusted according to the frequency of virtual and realsynchronization.,The contents can be the data attributes contained in the data class model,thestructure of the intelligent model,etc.All of these require digital twin networks with nativeinte
159、lligence to accurately identify high-value data and models in different autonomous scenariosand network environments.3.Digital Twin content and digital plan content model based on parallel delivery andreal-time data acquisition can automatically build and expand of digital twin contents and digitalp
160、lan contents of networks.In order to support the whole life cycle autonomy of the network,the management of digitaltwin contents and digital plan contents of the network should be synchronized with the physicalnetwork automatically without human participation.The equipment supplier develops the digi
161、talmodels(digital twin content and digital plan content models)of the product simultaneously at thedesign and development stage.The model is used to adjust and optimize the product offline untilthe product meets the delivery standard.The equipment supplier delivers a digital model of theChina Mobile
162、A Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper18product to the operator along with the physical entity of the product.When the product isdeployed in the network,the digital twin network will automatically acquire the digital model ofthe product,expand the digital twin c
163、ontent of the existing network according to the topologicalconnection relation,then generate its digital twin content through data collection,and generate thedigital plan content though digital domain optimization.In the parallel delivery digital model,thestandardized model part can be obtained by t
164、he digital twin network,while the non-standardizedmodel part may still be hidden inside the product.Therefore,it is necessary to establish themapping relationship between the non-standard model and the standard model.4.Design the simulation scenarios automatically.Orchestrate the simulation workflow
165、 ondemand.Optimize the simulation performance.In order to make the network have the ability of automatic pre-evaluation of effects,efficientclosed-loop and agile iteration,the digital twin network should be able to design simulationscenarios automatically,orchestrate simulation workflows automatical
166、ly and evaluate andoptimize simulation performance.Simulation scenarios refers to the network and service scenarios,user distribution,effect impact objects,network performance indicators and so on,which need tobe verified under specific network autonomy scenarios and use cases.Due to the cost ofsimu
167、lation and verification,digital twin networks need to resolve the network scenarios,verification indicators and simulation performance requirement of simulation pre-verificationaccording to the target of autonomous scenarios,and construct the required simulation scenarios,orchestrate the workflow of
168、 each scenario,manage the resources required for simulationvalidation,evaluate and optimize simulation performance(E.G.simulation accuracy,simulationtime)on demand.Figure 3.2-1 illustrates the digital twin network,the digital twin content of the network,thedigital plan content of the network,the dou
169、ble closed loops,the relationship among the models,and the corresponding positions of the four technical features.The number in the circle representsthe technical feature number.China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper19Fig.3.2-1.The relationship betwe
170、en the basic concepts of the digital twin network3.3 Network ArchitectureCompared with 5G,6G has the new digital twin network of the native intelligence plane,thedata plane and virtual-real interaction,and realizes a high level of autonomy in the whole lifecycle of 6G network through the interaction
171、 and fusion of native intelligence and the digital twinnetwork.Fig.3.3-1.The schematic diagram of the 6G digital twin network,the intelligence plane and thedata plane3.3.1 End to End ArchitectureThe 6G digital twin network will present a hierarchical cross-domain architecture thatcombines centralize
172、d and distributed architectures.The digital twin content,the digital plancontent and the basic functions of digital twin in local domain are used to support the networkautonomy in the local domain.The end-to-end digital twin platform stores the end-to-endnetwork-level digital twin contents,digital p
173、lan contents and basic functions of digital twin builtbased on the digital twin contents of each domain to support the end-to-end network-levelautonomy requirements.The multi-layer,cross-domain,closed-loop autonomy architecturefacilitates the generation of digital twin contents and digital plan cont
174、ents on demand nearby,eases the pressure on the data acquisition and transmission,protects the equipment internal dataprivacy,and timely supports network autonomous requirements within the scope of different scale,and takes into account the real-time nature of network element twins and the end-to-en
175、d integrityof cross-domain twins.The architecture can support optimization of various applications andservices based on digital twin contents at the edge of the network,as well as end-to-end globalservice optimization.At the same time,it is also easy to integrate with other architectures(such asthe
176、6G native AI network architecture 6).As can be seen from Figure 3.3.1-1,the digital twinChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper20content,the digital plan content and basic functions of the digital twin content constitute anend-to-end digital network s
177、ystem.Fig.3.3.1-1.The end-to-end architecture diagram of the twin digital networkIn order to support the main technical characteristics of the digital twin network,the designof basic functions of digital twin and its influence on network architectures are analyzed asfollows:1.Generate and analyze ne
178、twork autonomy requirements:the network itself discovers therequirements for plan,construction,maintenance and optimization by analyzing theupper-layer intention and network status.For non-real-time network autonomous usecases,this function can be implemented on the management side.Control plane sup
179、portis needed for autonomous use cases with high real-time requirements.2.Construct,arrange and adjust digital twin content and digital plan content models:selectmodel types,construct model structures,orchestrate and combine models according tothe requirements of autonomous scenarios.For more static
180、 autonomous scenarios,modelconstruction can be implemented on the management side.For highly dynamicautonomous scenarios,the model may need to be adjusted quickly according to thechanges of network states,so the support of the control plane is needed.To reduceresource overhead,high-value data and mo
181、dels in different autonomous scenarios andnetwork states need to be identified and adjusted online.3.Generate and update network digital twin content:combining real-time data acquisitionand model generation and updating network digital twin contents.Because digital twincontents need to track the sta
182、te changes of the real network,it has high requirements forreal-time and accuracy of data acquisition and synchronization in autonomous scenarioswith strong dynamics,and needs the support of control plane.To reduce resourceoverhead,it is necessary to identify high-value data in different autonomous
183、scenariosand network states and adjust data collection and synchronization policies online.4.Generate and implement the digital plan content of the network:based on the goal ofnetwork autonomy and the digital twin content of the network,through iterativeChina MobileA Digital Twin Network Approach fo
184、r 6G Wireless Network Autonomy White Paper21optimization algorithms and effect verification,generate the digital plan content withbetter regional performance and implement it into the physical network.For autonomoususe cases implemented on the management side,the real-time requirements of thisfuncti
185、on are not strong.The cases can be achieved by trying multiple optimizationalgorithms and carrying out sufficient modeling and simulation.For autonomous usecases implemented on the control plane,they need the support of high real-timetechnology.To sum up,the digital twin network will flexibly choose
186、 appropriate architecture,construct,orchestrate and adjust various models,and generate,update and implement digital twin contentand digital plan content according to the performance requirements of network autonomousscenarios.In this process,the digital twin network will strongly depend on the high
187、performancedata and intelligence capability.The current chimney-type data collection means and plug-inintelligent provision methods can no longer support them.Systematic,native data and intelligentcapability systems in 6G networks are required,that is,the addition of data planes and smartplanes is n
188、ecessary.3.3.2 Data PlaneConsidering the data challenge in network self-intelligence practice and the demand of digitaltwin networks for high-performance data services,6G will add data plane to the networkarchitecture 7.Data elements in the data plane will cover the internal and external data of the
189、network,including service data,user data,network data,perception data,external data,resourcelayer data,etc.Basic data services include data acquisition,data preprocessing,data storage,dataaccess,data sharing and coordination,etc.Basic data service has the following technicalcharacteristics:support o
190、f reliable authentication,authorization,access,efficient data storage andmanagement,dynamic data acquisition on demand,data pre-processing and aggregation,ability toopen to foreign trade and injection,etc.Figure 3.3.2-1 shows the logical functional architecture ofthe 6G network data plane.China Mobi
191、leA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper22Fig.3.3.2-1.The schematic diagram of 6G new data plane functional architectureThe 6G digital twin network invokes data plane infrastructure services to generate,store,access,and transport its digital twin contents,digita
192、l plan contents,and various models.In order torealize the digital twin content and the digital plan content of on-demand and dynamicallygenerated network,the data plane needs to be combined with the new intelligence plane of 6Gnetwork 7.The close combination of AI and data services will promote the
193、evolution of dataservice capabilities in data collection,processing,storage,knowledge transformation,applicationand other aspects to form self-growth data services.Through the combination of AI and dataservices,AI provides network awareness and intelligent strategies/algorithms for data collection,p
194、rocessing and flow,and provides knowledge association for data applications.The self-growthdata service has the following characteristics:accurate perception of massive heterogeneous data,active push and dynamic on-demand collection,avoiding data redundancy and improving dataanalysis and value minin
195、g capabilities.Data value mining is carried out by AI means,and theresponse speed of data services is improved by cloud-edge-end distributed storage and strategyChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper23optimization for data of different values.Through
196、 model training and knowledge inference,scenario dynamic adaptation is carried out,and intelligent orchestration and adjustment ofconfiguration parameters of data services are realized.Data service capability can be improved from data collection,data middle processing anddata application to realize
197、self-growth of data service 8.The digital twin content and the digitalplan content of the network can be used as a kind of data application.Fig.3.3.2-2.Data service self-growth3.3.3 Intelligent PlaneAt the design stage,6G considers deep integration with AI.Different from the AI functionsuperposition
198、 and plug-in in 5G,6G native AI controls and orchestrates computing power,dataand models end-to-end,and supports deep integration of technologies in connection,computing,data,AI and other technologies of different fields at the architecture level.It supports on-demandorchestration of AI capabilities
199、 to wireless,transport,host,core,and cloud to provide the basiccapabilities required for intelligence for high-level network autonomy and diversified servicerequirements.6 proposed the intelligent plane architecture of 6G native AI.6G network autonomy is one of the driving forces of the native AI.Th
200、e demand forintelligence in digital twin networks will be obtained through the intelligent surface of 6G.Therequirements includesnetwork autonomous demand generation and resolution,iterativeoptimization to generate digital plan content,the generation and update of digital twin networkintelligent mod
201、els,the on-demand dynamic adjustment of digital twin content,digital plan content,their models,etc.These requirements will become AI use cases in the intelligent plane 8,whichwill be satisfied by calling various AI services of the network(including AI training,AIverification,AI reasoning and AI data
202、).In this process,the performance requirements of networkautonomy will be resolved or mapped into the quality requirements of each AI service(QoAIS),which will be satisfied through the QoAIS evaluation and guarantee mechanism of the intelligentplane.Figure 3.3.3-2 shows the functional design of the
203、intelligent plane.China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper24Fig.3.3.3-1.AI lifecycle workflow orchestrationFig.3.3.3-2.The schematic diagram of function design of the 6G intelligent plane3.4 Key TechnologiesBased on the technical requirements of 6G net
204、work high level autonomy mentioned in 2.3,this section briefly introduces the technologies that may meet the above requirements.Facing thedeep openness of data,it is necessary to carry out the research on data collection,analysis andextension.In order to improve the value density of data,the knowled
205、ge graph can be used toanalyze data in a deeper level.The self-generation of autonomous requirements and low costChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper25trial-and-error optimization can be realized by pre-validation of model performance based onreinf
206、orcement learning and simulation microservitization.In addition,in order to ensure thestability of the digital twin network,the correction technology of the digital twin network will alsobe introduced in this section.3.4.1 Data Acquisition and Analysis TechniquesThe construction of digital twin netw
207、ork is inseparable from data collection.The datacollected from terminals,base stations,the core network and the network management in wirelesscommunication network is multi-source heterogeneous data.After standardized pre-processingoperations such as cleaning,classification,association and construct
208、ion,heterogeneous data ishighly aggregated to form a basic data warehouse.The basic data warehouse can maintaindifferent data sources,on this basis,the data can be further analyzed,such as correlation analysis,clustering,native factor extraction and knowledge graph construction,etc.,to establish a t
209、opiclibrary oriented to specific topics/services.Based on the correlation analysis of native factors incommunication network,data warehouse can provide various characteristic data sets for variousintelligent network optimization application scenarios.The logical architecture of the dataacquisition a
210、nd analysis technology is shown in the figure.Fig.3.4.1-1.The logical architecture of data acquisition and analysis technology in wirelesscommunication networkThe data collection technology covers multiple processes of data collection.In the currentnetwork,DPI(Deep Packet Inspection)collection techn
211、ology is mainly used to completeChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper26collection,timestamp,de duplication and other functions.The obtained signaling data is parsedinto bit-level data by signaling decoding technologies,and the service data is recons
212、tructed andreshaped by DPI and DFI(Deep/Dynamic Flow Inspection).It is translated and identified fordifferent protocols and applications.These data can be used to associate backfill and generateinterface data through user information and other analysis techniques.In 6G network,more dataneeds to be c
213、ollected to meet diversified requirements,such as,the data collection supported bynew interfaces such as MANO,SDN-C,and Vswitch under SDN networking.Data preprocessing refers to data processing,time partitioning,structured processing ofunstructured data,ETL warehousing and other operations on the co
214、llected original data of wirelesscommunication network,including associated backfilling of data,data cleaning,data conversion,data encryption and data loading.After data acquisition,pre-processing and statistical index extraction,the basic datawarehouse is obtained 9.The basic data warehouse is the
215、preparation area of the basic subjectdatabase.It usually tries to retain the most original and complete characteristics and attributes ofdata,classifies network data according to the data model,and performs full or incremental updateprocessing on source data 10.Data processing technology including t
216、he construction of the knowledge graph of data,theassociation rules analysis of data,the clustering and native factor extraction processing of data,thecomplete and consistent analysis of network data at the levels of subject-oriented,characterizingthe various data involved in each classification and
217、 the relationship between the data,and theconstruction of the feature data set,for example,quality of experence(QoE)feature data set,wireless network optimization feature data set,etc.Data warehouse also has on-demand customization function 11,which can further processdata and extract characteristic
218、 data according to various communication scenarios and researchneeds of communication personnel based on the analysis results of the association rule analysismodule in the subject library,such as quantitative representation of nodes and correlation degreebetween two nodes.In addition,feature data ca
219、n be evaluated according to feature extractionefficiency,feature sensitivity,fit degree and other indicators to further filter the data,screen outfeatures corresponding to specific KPIs and meet the requirements,and build customized featuredata sets.3.4.2 Data Enhancement TechniquesThe digital twin
220、network uses real wireless network status data to train the virtual scenario,and can carry on the data augmented on the real data,can be used to simulate the virtual scene of amore comprehensive,is able to provide more variety of training data,then preliminariesverification network key performance i
221、ndicators or models,and can achieve better performance,more robust decision-making configuration or models.China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper27Take large-scale MIMO weight optimization as an example.On the one hand,the weightcombination space of
222、large-scale MIMO is huge,and the number of parameters increasesexponentially with the increase of base stations,which is limited by time cost and money cost.It isdifficult for the real wireless network to collect all the beam pattern data samples under thecombination of user distribution(such as use
223、r position information,user DOA information)andantenna weights(such as beam azimuth,dip angle,horizontal beam width,vertical beam width,etc.).On the other hand,frequent adjustment of large-scale MIMO weights may adversely affectthe performance of wireless networks.Data augmentation technology can be
224、 used to combine thephysical model of wireless communication with real wireless network data to augment the data ofthe beam pattern,and generate more virtual beam pattern scenes under user distribution,channelenvironment and antenna weights,which can be used to guide the intelligent weight optimizat
225、ionalgorithm design of antennas.Based on this,the mass MIMO antenna weight configurationscheme is pre-verified and iterated to improve the timeliness and robustness of mass MIMOweight optimization.Conditional Generative Adversarial Networks(CGAN)are one of the key technologies toachieve beam pattern
226、 expansion 12.Generative Adversarial Networks(GAN)are often appliedto computer vision and text generation by adding additional conditional information to exert somecontrol over the generated data.Due to the generation adversarial structure of CGAN themselves,and the randomness and controllability of
227、 the input noise,CGAN have a certain applicationprospect in the field of communication,which is mainly reflected in two aspects:one is thevirtual-real adversarial learning,which adapts dynamically to the network environment throughthe self-adaptive adjustment of CGANs own structure.The second is for
228、 data augmentation,through the input noise and controllable conditions to achieve the purpose of augmenting the dataset.The beam pattern expansion module based on CGAN uses the user position distribution andantenna weights as conditions to generate a module to expand the beam pattern,and at the same
229、time,the discriminant module outputs the discriminant results of the augmented sample and thereal sample.The effective and controllable data augmentation is realized through the virtual-realantagonistic learning of the generated module and discriminant module.Fig.3.4.2-1 The beam pattern extension m
230、odule based on CGANChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper283.4.3 Pre Verification of Data and Knowledge Collaboration DrivenAt present,intelligent algorithm driven by data and knowledge synergistically is one of theresearch hotspots.In the field of m
231、achine learning,especially data-driven methods such as deeplearning and reinforcement learning,by combining traditional theoretical knowledge withdata-driven algorithms,the limitations of traditional theoretical models and data-driven algorithmscan be effectively solved,such as the complex ity of tr
232、aditional theoretical models and the demandfor data quantities of data-driven algorithms.Therefore,in the digital twin technology,thecombination of the two can be considered to reduce the data requirements of the algorithm 13.Meanwhile,communication theory models can also be used to guide the design
233、 of reinforcementlearning algorithms based on deep neural network to reduce computational complexity andimprove algorithm performance.Intelligent algorithms co-driven by data and knowledge can beused in the pre-validation module of digital twin networks.As shown in Figure 3.4.3-1,in the KPIpre-verif
234、ication module of large-scale MIMO antenna weight tuning,user position distribution,antenna weight scheme and corresponding beam pattern can be taken as input,and reinforcementlearning based on the deep neural network can be used to estimate the parameters required in theKPI model established based
235、on traditional theoretical knowledge(for example,unknownparameters such as signal intensity,interference intensity and noise estimated by deep neuralnetwork),and further KPI pre-validation results are obtained according to KPI models(such asShannon channel capacity formula,etc.),and the pre-validati
236、on performance is improved throughdata and knowledge co-drive.Fig.3.4.3-1.The KPI pre-validation module3.4.4 Knowledge Graphs and Graph Neural NetworksDigitaltwin network terminal nodes are characterized by densitization,dynamics,regularization of business requirements,etc.In order to effectively cl
237、arify the internal factorcorrelation between these changes,The development of intelligence theories such as data mining,knowledge graphs and machine learning has made it possible to collect,analyze,cluster,andanalyze the correlation relationships of internal factors in massive data,which can effecti
238、velyclarify the correlation relationships between these changes,so as to improve the effectiveness,versatility,and intuitiveness of digital twin networks 14,15.China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper29For example,knowledge graphs can be used to get th
239、e correlation of native factors ofwireless communication network protocol.First to build knowledge graphs,the first step need todefine the mapping of the type of entity and attribute,the second step requires the definition of therelationship between different entities,the third step defines entities
240、 and relationships to write forthe form of a triple(The triple is a general representation of knowledge map consisting of the twoentities with the semantic connection relationships and the relationships between entities).Finally,the required knowledge graph is formed according to the obtained triple
241、s.In association rulesbased on the wireless communication network protocol based native factors of knowledge graphs(through the acquisition of the data fields with the internal algorithm to calculate the correlation),by defining node sparse representation of vector and find out the similarity betwee
242、n the nodecosine similarity so as to realize the update of knowledge graph figure structure,the combinationof knowledge graphs new topological structure associated with the node analysis.This methodcompletes the calculation of the correlation degree of each node and the representation andlearning of
243、 the feature vector,which provides a technical support for the deep inference andmining of the correlation between nodes.The implementation process of this method is shown inFigure 3.4.4-1.Fig.3.4.4-1.The construction of knowledge Atlas and the flow of association analysis methodFurther,if we make f
244、ull use of wireless network big data and artificial intelligence to buildwireless network meta-model(a pre-training model with multi-scene migration ability)andChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper30meta-algorithm(an algorithm that can guarantee per
245、formance in multi-scene and multi-taskenvironment by applying the meta-model to different scenarios),it can also form digital twinnetworks with strong generalization ability that can be migrated to a variety of environments.However,with the introduction of knowledge graphs,the deep learning method u
246、sed by thedigital twin network faces challenges.Traditional deep learning methods have achieved greatsuccess in feature extraction based on Euclidean spatial data.However,the data in knowledgegraphs is generated from non-Euclidean spatial data,and the performance of traditional deeplearning methods
247、in processing non-Euclidean spatial data is still unsatisfactory.This is becausethe data nodes and topology in the knowledge graph are not regular and different data nodes arenot independent.By introducing the graph neural network 16,the model can take into accountthe scale,heterogeneity and deep to
248、pological information of input data,which shows convincingand reliable performance in mining deep effective topological information of knowledge atlas,extracting key complex features of data and realizing rapid processing of massive data.3.4.5 Simulation ServiceabilityThe current simulation platform
249、 has specialized simulation functions,and it is difficult fordifferent models to interact with each other.Typical simulation characteristics such as simulationmodeling,resource scheduling and system management are not supported enough,so it is difficultto serve as an efficient pre-verification envir
250、onment for digital twin networks 17.For the6G-oriented digital twin network,it is necessary to increase the research on the service ofsimulation platforms,and form a complete set of service simulation support platforms forcross-domain collaboration.The 6G-oriented digital twin network simulation sho
251、uld have the following capabilities:(1)On-demand service organization:During the operation of the platform,it can refer to the specificrequirements and provide corresponding simulation support,without human interaction,so as torealize the self-configuration function of SON autonomous network.(2)Univ
252、ersal network access:Design a lightweight communication mechanism,such as Restful API,to allow cross-end accessof all kinds of networks and devices.(3)Resource pool construction:Standardized methods andprocesses are used to integrate various resources into a resource pool to serve multiple users.Dif
253、ferent physical and virtual resources can be dynamically allocated and reallocated according touser requirements.(4)Service linkage:different simulation components can be linked accordingto needs and combined to form standardized interfaces and more diverse standards of services.(5)Improved reusabil
254、ity:Through the micro-service mechanism,well-defined service interfaces areutilized to provide more precise services in smaller modules,thus facilitating interface reuse.Current research on micro-service simulation architecture 18 indicates that the modelingand simulation architecture based on micro
255、-service can provide a series of flexible and pluggablemicro-service simulation technology components,and each micro-service component realizes asmall and highly reusable function,which can be assembled and linked flexibly on demand forChina MobileA Digital Twin Network Approach for 6G Wireless Netw
256、ork Autonomy White Paper31different scenarios.Simulation tools and simulation scheduling platforms are deployed in thecloud.Users can submit and manage simulation tasks through the cloud application platform,andquickly obtain flexible,reliable and secure simulation services.The architecture can be d
257、ividedinto five layers,with the application portal layer as the input design,the application service layeras the business provision,the data exchange layer as the resource schedule,the resource layer asthe model encapsulation,and the physical network layer as the basic support,forming a functionalcl
258、osed loop of micro-service simulation 19.Using container virtualization technology anddistributed system architectures as the infrastructure base,cloud migration of the existingsimulation platform is realized based on the principle of DevOps.Through the linkage ofsimulation components of microservic
259、es,the communication between three contents and thetransformation between the five states of the resource objects in the digital twin network arerealized.3.4.6 Correction techniques for pre-verification resultsIn the long-term operation of the digital twin network,there may be errors in data acquisi
260、tionand transmission,big difference between predicted data and real data,difference between twinenvironmentandrealenvironment,etc.Thesewillmakethedifferencebetweenthedecision-making and the expected decision-making of the digital twin network,and even have anegative impact on the real network.These
261、differences may arise from each link of the digital twinnetwork,such as data acquisition,state prediction,decision-making generation,pre-verification ofdecision-making effect,etc.The differences in these modules can be corrected separately.Forexample,the training data of the state prediction module
262、is updated online to correct the deviationof the prediction model,or to correct the deviation of the module which directly affects the finaldecision of the twin network layer,that is,the decision effect pre-verification module.Fig 3.4.6-1.The decision pre-verification result rectification processIn
263、the large-scale antenna beam weight optimization scenario,the correction module can bedesigned to correct or optimize the pre-verification result of the decision-making effect.When theChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper32weight of the antenna beam
264、 in the decision is assigned to the physical network,the network statewill be updated and synchronized to the digital twin content,that is,the network state after thedecision is implemented,which is reported by the digital twin content to the correction module inthe twin network layer.At the same ti
265、me,the correction module collects the information ofpre-verification environment(including base station parameters,building parameters,userlocation,etc.)and the decision of weight configuration d.Then according to these data,thecorrection model of RSRP pre-verification performance r is generated to
266、correct the r,so that it ismore in line with the real effect of the wireless network after the weight configuration decision isimplemented.The twin network layer can evaluate multiple candidate weight allocation decisionsbased on the performance of pre-verification,and can further determine the caus
267、e ofpre-verification performance bias,such as data acquisition and transmission error,state predictionerror,network digital twin simulation error.The specific calculation method 13 is as follows:Denote1tdas expert weight allocation decision for storing history of the deviation correctionmodule,1tdra
268、s RSRP performance of the expert decision,1tuas reinforcement learningalgorithm weight allocation decision,1turas RSRP performance of the algorithm,1tsasprediction of the next user distribution state,1tsas state of the real user distribution at the nextmoment of the acquisition,),(11ttudbestas the o
269、ptimal weight configuration,and1tras thereal RSRP of this weight decision.For example,through a neural network algorithm,a model thatobtains corrected pre-validation results is trained with historical state-decision-pre-validationresult pair and state-decision-network performance pair.With the input
270、 of states,decision andprediction performance,we can get the pre-verification performance after correcting deviation:Pre-verification performance of expert decision making:),(11111tdttdsrdsfrtDTtPre-verification performance of the reinforcement learning algorithm decision making:),(11111tuttusrusfrt
271、DTt3.5 Whole life cycle autonomy of networkNetwork life cycle generally includes plan,construction,maintenance and optimizationstages.The current network autonomy level is low,and the network life cycle of each stage isseparated from each other.More work needs to be done manually offline.Resource an
272、d time costChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper33are high,and OPEX is high as well.6G will take over the whole life cycle of the network throughthe digital twin network.Based on the three contents,five states and double closed loops,the technical p
273、roposals in each stage is optimized and verified on-line,forming a high level ofclosed-loop autonomy and greatly reducing human consumption.In the digital twin network,thelife cycle of the network will go through three stages:continuous planning,virtual-realdocking and fault self-healing.Fig 3.5-1.T
274、he comparison between 5G network OAM management and 6G digital twin networkautonomy3.5.1 Continuous planningIn the digital twin network,the digital domain is constantly synchronizing the current state ofthe network,predicting the future state of the network,and planning the parameters(digital planco
275、ntent)of the network in advance for the next moment.Therefore,the traditional networkplanning and optimization work will be realized by the generating digital plan content in thedigital twin network,which is divided into generating long-term digital plan content andgenerating short-term digital plan
276、 content according to the time scale.For example,when thecoverage,capacity and service requirements change,based on wireless resources and equipmentcapabilities,the design of network or site solutions(including regional network planning,blindsite planning,heating site planning,site expansion)belongs
277、 to the generating digital plan content.In order to make the network performance indicators meet the predetermined requirements(suchas coverage,capacity,mobility,access,user experience,load,PCI,ANR,energy saving and otherindicators),the optimization of parameter configuration and software upgrade pe
278、rformed belong tothe generating short-term digital plan content.The following diagram illustrates a commonprocess framework for the digital plan content generated by digital twin networks:China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper34Fig 3.5.1-1.The proces
279、s for the digital plan content generated by digital twin networks.3.5.2 Virtual and real connectionThe network construction stage mainly depends on the manpower,and the networkautomation ability mainly manifests in the equipment self-start aspect.Take the base stationequipment as an example,the base
280、 station self-start makes the base station or the hardware moduleenter the normal working state on the planning site.Except for manual installation and power-up,all other processes are automated,including self-activation(software,configurations,licenses),self-check,self-authentication,self-configura
281、tion,and so on.The existing network alreadysupports the self-check,self-authentication,self-activation processes.In the 6G digital twinnetwork,the physical entity and the digital domain docking should be completed in this process,the self-configuration of basic software and wireless parameters shoul
282、d be completed byconnecting the pre-generated digital plan content,and a two-way synchronous connectionshould be established to construct the digital plan content and digital twin content,complete thetransformation from planning state to service state or twin state.The following diagramillustrates p
283、rocess of virtual-real docking in a 6G digital twin network:Fig 3.5.2-1.The virtual-real connection process in the 6G digital twin networkChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper353.5.3 Combination of prevention and cureThrough the continuous planning
284、of the digital twin network,the risk of some networkperformance deterioration can be avoided,and some potential faults can be cured ahead of time.However,some of the types of faults caused by external emergencies,such as cell retreatment,link failure,equipment power failure,etc.,cannot be avoided an
285、d can only be repaired after thefact.Therefore,in the self-healing aspect,the digital twin network requires both active preventionand passive remediation.On the other hand,due to the low probability of network failure,it isvery difficult to obtain fault samples in current networks,and the effectiven
286、ess of self-healingschemes is difficult to verify.Digital twin networks have the capability to generate virtual networkscenarios and pre-verify the effects,which can help to solve the traditional problems of few failuresamples and difficult to verify the effectiveness of solutions.Fig 3.5.3-1.The fa
287、ult self-healing process of the 6G digital twin networkIn addition to continuously optimization of the network digital plan content in the faultscenarios to achieve self-healing goals through double closed loops,the digital twin networkwill also build a knowledge base for fault identification and se
288、lf-healing through the virtual faultlaboratory.Real fault knowledge is generated by collecting the digital twin content and the digitalplan content under the real fault scenarios.Implement cures,collect samples,and generate virtualfault knowledge by artificially creating faults on twin digital netwo
289、rks.At the same time,theexpert experience of OAM personnel needs to be combined to provide sufficient information forfault detection,prediction,root cause location,prevention and treatment.3.6 Case descriptionThis section will illustrate the concepts,characteristics,architectures and technologies of
290、 thedigital twin network in real-world application scenarios though three network optimizationexamples,as well as the potential performance advantages of these cases based on digital twinnetwork.3.6.1 Optimization of beam weight for large-scale antennasIn order to meet the challenges of more flexibl
291、e beam configuration,more accurate userrequirements and more complex optimization scenarios,5G wireless networks need tointelligently optimize the beam weights of large-scale antennas and generate custom beams forChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Pape
292、r36scenarios according to geographical features and user distribution.At present,there are manykinds of heuristic optimization methods,such as genetic algorithms,particle swarm algorithms,moth algorithms and so on,but the existing simulation techniques can only evaluate part of theperformance of the
293、se technical solutions,such as RSRP at the downlink receiver.It is impossibleto simulate and evaluate more indicators,so it is impossible to fully pre-verify the performance ofthe technical solution and accurately evaluate the effect of the beam configuration scheme appliedin the current network.The
294、 resulting decisions may affect network performance.The intelligent antenna weight optimization technology based on network digital twincombines the physical model of wireless communication with the real wireless network data tobuild the digital twin content of the network,and integrates the expert
295、experience as the guaranteeof the weight decision.The performance of the decision is pre-verified to further ensure theperformance of the decision,based on which the weight assignment scheme of the large-scaleMIMO antennas is iteratively optimized.The principle of intelligent antenna weight optimiza
296、tiontechnology based on network digital twin is shown in the figure.According to the current network state and corresponding network optimization requirements,the technology collects real wireless network data including network environment information,channel state information and user state informa
297、tion,etc.on demand,and combines data models,simulation models and intelligent models to build the twin network layer including the user stateprediction model,the decision-making generation model,the decision-making selection modeland other mainly modules.Among them,wireless mobile networks,including
298、 base stations andterminals,constitute the physical entity of digital twin networks.The twin data of mobile networksin the digital domain,including user location information,the control beam RSRP received by theuser,and base station parameters(such as base station location,system bandwidth,carrierfr
299、equency band,base station transmit power,antenna parameters,air interface channel model),etc.,constitute the attributes in the digital twin model of the network.The beam weights of thearea to be optimized by the network are combined as the attributes in the digital plan contentmodel.The inner loop o
300、f the digital twin network can predict the state of the network at the nextmoment(user position information),and then make the beam weight decision.The combinationof expert knowledge(mapping between user location distribution and beam weights)and artificialintelligence(reinforcement learning)is used
301、 to generate the weight assignment decision.Theexpert knowledge guarantees the lower limit of the beam weight decision applied to the physicalnetwork,and the digital twin network chooses one with better coverage of the beam weightsgenerated by expert knowledge and reinforcement learning as the effec
302、t verification object.Then,the digital twin network can pre-verify the coverage of the network in the pre-verificationenvironment,and iteratively optimize the weight decision to generate the final digital plan content.The digital twin network sends the antenna weight assignment instruction to the re
303、al wirelessnetwork to optimize the network coverage performance.The outer loop of digital twin networkChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper37can estimate the distance from the coverage target and analyze the cause of the error according tothe covera
304、ge performance of real network feedback,train the correction model and correct thecoverage performance of the digital twin network pre-verification,and optimize the reinforcementlearning decision generation algorithm and the expert experience base.Fig 3.6.1-1.The schematic diagram of intelligent ant
305、enna weight optimization of based onnetwork digital twinFor the future of 6G wireless networks,higher frequency applications(such as millimeterwave,terahertz radiation,etc.)will lead to smaller coverage and more intensive deployment ofwireless sites.The need for large-scale antenna beam weight optim
306、ization for inter-cellcooperation will become more prominent.Meanwhile,the beam weight optimization of ultra-largescale antennas without cellular structure,as well as the joint optimization of intelligent reflectorand the beam weight of base station antenna may become the typical scenarios of the 6G
307、 wirelessnetwork autonomy.For these scenarios,the technology solution based on the digital twin of thenetwork can be used to obtain reliable and better performance antenna weights,reducing the costand complexity of network optimization.3.6.2 Intelligent deep RAN slicesThe intelligent RAN slice based
308、 on network digital twin aims at mining slice configurationexperience,capturing short-term environmental change characteristics and achieving efficient andreliable slice resource management.The traditional wireless access network(RAN)slicing schememainly adopts the slicing configuration method based
309、 on the optimization model,which isconstrained by the specific service quality and the total resource quantity,and aims at maximizingthe network performance or the operator revenue.The optimal slice resource allocation scheme issolved by classical iterative algorithms.However,the slice resource mana
310、gement method based onChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper38the optimization model relies too much on the prior traffic model and the computationalcomplexity increases significantly with the increase of network size 19,20.In order to realize enviro
311、nment-self-adaptive,low-complexity and high-performanceintelligentRANslice,miningsliceconfigurationexperienceandcapturingshort-termenvironmental change characteristics to achieve high-efficiency and high-reliability slice resourcemanagement,it is necessary to explore the digital twin framework of re
312、al-data-driven networkslicing system,study the intelligent RAN slicing system based on the network digital twin and thetwin enhancement technology of near real-time interaction with real environment.So that theslicing scheme can adjust actively and adaptively to adapt to the environment quickly unde
313、r thecircumstance that the distribution of user traffic is difficult to predict and the network scale keepsgrowing.Fig 3.6.2-1.The sketch map of intelligent RAN slice based on network digital twinFigure 3.6.2-1 is a schematic of the intelligent RAN slice based on network digital twin.Thedata warehou
314、se is composed of the data collected in the real environment and preprocessed.Itprovides training samples to the digital twin prediction module and the pre-verification module,atthe same time,the training data is augmented by GAN,which is used to match and correct digitaltwin contents.The digital tw
315、in prediction module updates the slice pre-configuration scheme withthe predicted future network state output.The pre-verification module interacts with the slicereconfiguration module in the digital twin content and iteratively optimizes the performance of thereconfiguration module of digital twin
316、contents until convergence.Then,the slice reconfigurationmodule interactively fine-tunes with the real world to quickly adapt to the real world.In the slicereconfiguration algorithm,the slice pre-configuration space is used as the input of dynamic slicereconfiguration module.Then the greedy strategy
317、 is used to select the slice configuration action,China MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper39and the network environment feeds back the statistical information of the slice satisfying rate,spectrum efficiency and packet loss rate in the slice window,to
318、calculate the network reward andfurther update the utility and number of slices configuration actions.The architecture includes the physical entity,digital twin contents,digital plan contents,theinner closed loop and the outer closed loop.The physical entity includes peripheral equipmentsconsisting
319、of base stations,user devices and other peripherals,software and hardware systems,andso on.Digital Twin contents are the twins of wireless networks in digital domain,including thedigitalization of user-side features and base station-side features.The digital plan contents are thecombination of slice
320、 resource blocks in the region to be optimized.In the two closed loops,theinner loop is composed of the digital twin prediction module,the slice pre-configuration module,reconfiguration module and the pre-verification module.They form a closed loop for iterativetuning until that the prediction modul
321、e,the slice pre-configuration module,and the reconfigurationmodule converge.The outer loop is composed of the pre-configuration module,re-configurationmodule and data warehouse which actually act on the actual system.In the outer closed loop,thedigital twin contents are corrected according to the re
322、al data collected by the data warehouse andthe slicing effect of the network feedback.The correction includes error analysis of thepre-verification module and the prediction module,network fitting and generalization erroranalysis,and fine-tuning based on real samples.For the future 6G wireless netwo
323、rk,Cloud RAN is the general trend.From the point of viewof the service capability,6G wireless network will be reconfigured into more fine-grained serviceRAN.The protocol stack will turn microservice into the function module,and will break throughthe traditional protocol design idea.The calling relat
324、ion between the function and the function isno longer restricted by the upper and lower layer protocol relation.The function module can beflexibly called and dynamically combined into the required RAN stack according to the serviceperformance requirements.Therefore,the RAN slice of 6G wireless netwo
325、rk includes not only theslice of physical resources,but also the slice of the protocol stack function to realize the deepRAN slice.In this wireless autonomous scenario,the slice configuration optimization scheme canbe extended based on the use case scheme,using the network digital twin to generate a
326、nd optimizein advance.3.6.3 Federated scheduling of multi-dimensional resourcesThe amount of data brought about by emerging technologies,applications and scenarioscontinues to grow,and there is a more urgent need for computing power and networks in all walksof life.Increasing the overall scale of co
327、mputing power has become a focus of common concern inthe industry.Building computing power networks has also become a national strategy.The 6Gcomputing power network puts forward higher requirements for efficient allocation andscheduling of wireless communication resources such as time-frequency res
328、ources,spaceresources,future computing resources,cache resources and so on.Moreover,the large-scaleChina MobileA Digital Twin Network Approach for 6G Wireless Network Autonomy White Paper40computing power network has a large resource cost.Following the principle of sustainabledevelopment,it is neces
329、sary to take energy-saving measures for computing power network.Improving resource scheduling accuracy and reducing overhead such as redundancy is one of theimportant ways to save energy,which can be achieved through the digital twin of the network.By analyzing and reasonably predicting service traf
330、fic,user distribution,channel stateinformation and other data,energy-saving strategies are generated independently by the networkon the premise of ensuring sufficient communication resources to guarantee the performance ofthe network in all aspects.The strategy can perform channel shutdown,symbol sh
331、utdown,carriershutdown,base station switch and other operations on the base station,and complete thescheduling of communication resources such as time slot,frequency,and space.It can reduce theenergy consumption of the base station with lower load and effectively improve the energyutilization rate.D
332、igital Twin Network divides resource objects into three forms:the physical entity,digitaltwin contents and digital plan contents.In this scenario,the physical entity consists of the6Gwireless network including macrobase stations,microbase stations,edge computing nodes andusers.Digital twin contents
333、realize the twin of 6G wireless networks in the digital domain throughthe system-level simulation platform.The base station parameters(base station locationcoordinates,coverage cell radius,carrier frequency,load power,antenna parameters,etc.),userlocation information and service distribution state are modeled.Digital plan contents are networkresource allocation schemes and switch operation schemes