《全球6G技術大會：2024年ICDT融合的6G網絡4.0（英文版）（98頁）.pdf》由會員分享，可在線閱讀，更多相關《全球6G技術大會：2024年ICDT融合的6G網絡4.0（英文版）（98頁）.pdf（98頁珍藏版）》請在三個皮匠報告上搜索。

1、1/97Executive SummaryThe development of information technology is accompanied by the continuous emergenceand integration of new technologies.As information and communication technology(ICT)evolvestowards 6G,it further integrates with big data and artificial intelligence technologies,presentingthe ch

2、aracteristics of ICDT(Information,Communication,Big Data Technology)integration.TheICDT integrated 6G system will become an end-to-end information processing and service system,representing an mobile information network that combines communication networks,sensingnetworks,and computing networks.New

3、technologies and functionalities have become candidatesolutions for 6G design,validated at various levels,including theoretical,simulation,andprototype stages.This white paper classifies and analyzes potential 6G technologies and solutions throughthree aspects:the new trends in ICDT integration,wire

4、less air interface technologies in ICDTintegration,and network aspects of ICDT integration.It proposes corresponding technical systemsuggestions.The new trends in ICDT integration encompass communication and computationtechnology,semantic communication technology,and AI large model technology.Commun

5、icationand computation integration technology is based on core technical capabilities such as thecoexistence of communication and computation resources,functional integration,and serviceintegration,enabling the evolution of wireless access network computing and the upgrade ofgeneral intelligence ser

6、vices.Semantic communication technology focuses on the meaning andsignificance of transmitted information,aiming to overcome the limitations imposed by Shannontheory on communication systems.AI large models,as one of the most prominent scientificdirections,continuously enhance the intelligent autono

7、mous capabilities of communicationnetworks.ICDT integrated wireless air interface technology includes integration of sensing andcommunications,air interface AI,multiple access methods,encoding and modulation,massiveMIMO,and near-field technology.Novel air interface technologies are essential for con

8、structing a6G network with strong connectivity,computing power,intelligence,and security.Currently,thesetechnologies have completed the design of technical and prototype solutions and are undergoingtesting.In terms of the network architecture of ICDT integration,service-oriented networks,nativeAI,di

9、gital twin networks,multidimensional networking,and user-centric networks are keyresearch directions.The new network architecture helps propel the ICDT integrated 6G networkstowards a new generation of mobile information networks with sensing,computation,andintelligence,providing comprehensive cover

10、age across air,land,and space.The research on ICDT integrated 6G technology is gradually transitioning from isolatedtechnical research towards systematic construction and industry consensus.This article advocatesfor strengthening interdisciplinary research and collaboration in the field of informati

11、on andcommunication technology,driven by demand,to accelerate joint tackling of basic theories andbottleneck technical issues and foster collaborative innovation.Simultaneously,emphasis shouldbe placed on the transformation of interdisciplinary innovation towards application and services,nurturing i

12、ndustrial ecosystems at the application layer,and incubating new businesses andapplications.Finally,further efforts should be made to enhance international cooperation,with2/97countries working together to solidify the foundation of 6G and promote its global development.ContentsExecutive Summary.11.

13、Introduction.42.New Trends in ICDT Integration.62.1 Integrated Communication and Computing.62.2 Semantic Communication.152.2.1 Overview.152.2.2 Key Technologies.162.2.3Applications and Challenges.192.3 Large Model and Wireless Communication.212.3.1 Current Research Status of Large Models at Home and

14、 Abroad.222.3.2 Empowering the Communication Industry with Large AI Models.232.3.3 Intelligent Endogenous Communication Network.26References.273.ICDT-integrated Wireless Air Interface.293.1 Integrated Sensing and Communication.293.1.1 Typical Application Scenarios.293.1.2 Key Technologies and Challe

15、nges.313.2AI-nativeAir Interface.333.2.1 Typical Application Scenarios.333.2.2 Key Technologies and Challenges.353.3 MultipleAccess.373.3.1 Typical Application Scenarios.373.3.2 Key Technologies and Challenges.393.4 Code Modulation.443.4.1 Typical Application Scenarios.443.4.2 Key Technologies and C

16、hallenges.453.5 Ultra Massive MIMO.493.5.1 Typical Application Scenarios.493.5.2 Key Technologies and Challenges.503.6 Near-field Technology.613.6.1 Typical Application Scenarios.613.6.2 Key Technologies and Challenges.65References.693/974.ICDT-integrated Network.734.1 ICDT-integrated Network Archit

17、ecture.734.1.1 Overview of 6G Network Architecture.734.1.2 3-Entity,4-Layer,and 5-PlaneArchitecture.744.2 Network as a Service(NaaS).754.2.1 Technical Concept.754.2.2 Typical Application Scenarios and PotentialAdvantages.754.2.3 Key Issues and Challenges.774.3 NativeAI.784.3.1 Technical Concept.784.

18、3.2 Typical Application Scenarios and PotentialAdvantages.784.3.3 Key Issues and Challenges.804.4 Digital Twin Network.824.4.1 Technical Concept.824.4.2 Typical Application Scenarios and PotentialAdvantages.834.4.3 Key Issues and Challenges.854.5 Multi-dimensional Networking.864.5.1 Technical Concep

19、t.864.5.2 Typical Application Scenarios and PotentialAdvantages.874.5.3 Key Issues and Challenges.914.6 UCAN.924.6.1 Technical Concept.924.6.2 Typical Application Scenarios and PotentialAdvantages.934.6.3 Key Issues and Challenges.94References.945.Summary and Outlook.95Acknowledgement.974/971.Introd

20、uctionThe rapid development of information technology has always been the drivingforce for social progress.In recent years,the seamless integration of information,communication,AI,big data,system control,and sensing technologies has ushered ina new era in 6G technology.At the Global 6G Conference in

21、 November 2020,theICDT Integrated 6G Network 1.0 white paper was officially released,regarding 6G asan end-to-end information processing and service system.Its core function isexpandedfromsimpleinformationtransmissiontoinformationcollection,information computing,and information application,providing

22、 users with morepowerful endogenous capabilities in multiple dimensions,including communication,computing,sensing,intelligence,and security.As time progresses,6G technologyundergoes continual evolution.The ICDT Integrated 6G Network 2.0 white paper wasreleased at the Global 6G Conference held in Mar

23、ch 2022.This version focuses on thenetwork capability,architecture,air interface,terminal,and industry that integratesensing,communication,and computing while proposing a new 6G solution.Additionally,in March 2023,the ICDT Integrated 6G Network 3.0 white paper wasreleased,primarily focusing on detai

24、ling the ICDT-integrated 6G technology system,ICDT-integrated wireless communication and networking,as well as ICDT-integratedarchitecture and functions.Over the past year,there has been a rapid advancement in global 6G technology,extending its scope and positioning to reflect a new trend of integra

25、ting big data,AI,sensing,information,and communication technologies.Emerging technologies suchas integrated sensing and communication,integrated communication and computing,AI-assisted communication,and semantic communication have attracted significantinterest.The academic and industrial circles are

26、 collaboratively advancing thesetechnologies in theoretical research,prototype experimentation,and standardizationinitiatives.In June 2023,the International Telecommunication Union(ITU)released aproposalfortheframeworkandoverallobjectivesofInternationalMobile5/97Telecommunications technology for 203

27、0 and beyond,aiming to elaborate on thedevelopment goals,trends,typical application scenarios,and technical capabilities of6G.The release of this proposal signifies a definitive establishment of the globalconsensus and guiding principles for future communication technologies.InSeptember 2023,the Nex

28、t Generation Mobile Networks(NGMN)Alliance also issueda statement outlining its position on 6G,emphasizing the technologys positioning,innovation,services,operations,principles,and spectrum considerations.The releaseof this statement highlights the global consensus on 6G technology and offersindustr

29、y-level guidance to drive its innovation and development.In October 2023,theIMT-2030(6G)Promotion Working Group formed the Semantic Communication TaskForceintendingtoadvancetheresearchandstandardizationofsemanticcommunication technology in 6G.Overall,the initiatives undertaken by theseinternational

30、organizations and working groups have laid a robust groundwork forstandardizing 6G technology,driving its research and development forward,andfosteringglobalcollaboration.Theglobal6Gdevelopmentisshowingitscharacteristics of cross-border integration and multi-layer breakthroughs.Building on the lates

31、t advancements in 6G technology,the 6G working group ofthe FuTURE Forum has led the development of the ICDT Integrated 6G Network 4.0white paper.This version provides a comprehensive overview of the current statusand future prospects of ICDT-integrated 6G technology from three dimensions:newtrends i

32、n ICDT integration,ICDT-integrated wireless air interface technology,andICDT-integrated network architecture.It also provides a concise overview of thetechnical concepts,application scenarios,key issues,and challenges of various newICDT integration technologies,and delves into their development tren

33、ds.This whitepaper aims to offer insights and proposals for advancing 6G technology research,supporting its continued development.6/972.New Trends in ICDT IntegrationThe ICDT integration is undergoing a significant shift with three key trends:theintegration of communication and computing,semantic co

34、mmunication,and theintegration of AI large models into communication systems.While traditionalcommunication systems continue to improve,this integration with sensing,big data,and AI will fundamentally reshape network capabilities and architecture,paving theway for a future of intelligent,efficient,a

35、nd reliable communication.Figure 2-1 ICDT Integration Trends2.1 Integrated Communication and ComputingThe integration and coexistence of wireless communication and computing forma new technological field that has emerged due to the digital economys transition intothe Era of Computing Power.This tran

36、sformation is propelled by various factors,including technological evolution requirements,business development needs,and thetransformationandupgradingofnetworkinfrastructure.Relyingoncoretechnological capabilities such as the integration and coexistence of communicationand computing resources,functi

37、onal integration and coexistence,and serviceintegration and coexistence,radio access network(RAN)evolution into computingpower and the upgrade of integrated communication,computing,and intelligentservices can be achieved.This promotes efficient aggregation,flow,and sharing of7/97resources and factor

38、s,providing solid support for the growth of the digital economy.With the rapid onset of the digital era,characterized by ubiquitous sensing,connectivity,and intelligence,there is a growing societal demand for computingresources.Consequently,the construction of computing networks has emerged as asign

39、ificant national strategic initiative.As a critical national information infrastructure,RANs possess a large number of base stations.The exploration and utilization of theiridle computing power hold significant economic value and represent the final link inachieving network-wide computational upgrad

40、ing.The integration and progression ofcommunication and computing power in RANs has emerged as a significant growthtrend in advancing the efficiency and service enhancement of information andcommunication infrastructure.From a technological standpoint,in recent years,the integration of technologiesl

41、ike communication,cloud computing,AI,and digital twin has emerged as aninevitable trend in the evolution of next-generation networks.Research into keytechnologies like AI endogeneity and integrated sensing and communication hascommenced for 6G,bringing forth fresh demands on the computing power of R

42、ANs.Futurenetworkdesignsmustholisticallyaddressthecombinedneedsofcommunicationandcomputingtofacilitatetheintegrateddevelopmentoftechnologies and maximize their advantages.From a business expansion standpoint,emerging services like cloud gaming,XR,and the Internet of Vehicles(IoV)have integrated dema

43、nds for deep edge computing,localdataoffloading,anddeterministicmobileconnectivityassurance.Simultaneously,addressing diverse industry digitization demands,such as industrialquality inspection and remote control,is facilitated by cloud-based base stationsoperating on a universal platform.By harnessi

44、ng the inherent computing power andintegrating it with intelligent capabilities,a deep integration of communication andcomputing power can be achieved.This not only maximizes system efficiency andperformance,but also meets industry needs for cost-effective,agile deployment,andtailoredsolutions,there

45、bycontributing tothe countrysstrategicvision forempowering the future of 6G.8/97Looking at the demand for upgrading network infrastructure efficiency,by theend of 2023,the total number of 5G base stations in China has reached 3.377 million,representing the heaviest investment and the largest proport

46、ion of the 5G networks.Thanks to its unique tidal effect,RANs possess abundant dynamic residualcomputational power,forming a blue ocean of computational value.Similar torenewable energy,it represents a novel economic resource with immense potentialeconomic value,yet requires innovative technology fo

47、r efficient utilization.Likewise,upcoming 6G networks will constitute a vast computing power resource pool,contributing substantial computing power.Presently,the integrated communication and computing scenario primarilytargets typical 6G use cases like network intelligence,edge computing/AI,andsensi

48、ng networks.Network intelligence plays a crucial role in the integration andcoexistence of wireless communication and computing.By consolidating thecomputing power of base stations,it enables the implementation ofAI-driven networkoperation and optimization,effectively supporting the enhancement of h

49、eavy-load cellperformance,intelligent detection of faults and potential hazards,and other specificdemands.Furthermore,leveraging intelligent technologies alongside local businessidentification at base stations ensures the assurance of critical business experiences,such as payment applications within

50、 hospital environments,and ensures seamless userexperiences for hospital-owned mobile apps.Additionally,applications like basestation AI optimization and AI-driven base station/network coordination are pivotalscenarios for the integration and coexistence of wireless communication andcomputing.The in

51、tegration of computing,networks,and services based on base stationcapacity expansion represents a key aspect of the integration and coexistence ofwireless communication and computing.One aspect targets 2C services,catering tolocal computing needs of ultra-low latency and high-bandwidth applications

52、bysharing base station computing power,for example,XR local service rendering andmetaverse.The second aspect focuses on 2B,providing integrated communication andcomputing services in localized settings to achieve flexible and agile service9/97deployment while effectively reducing the overall costs o

53、f industry applications.Typical scenarios include industrial vision AI detection,(Internet of Vehicles)intelligent collision prediction,(UAV)intrusion detection,etc.Both domestic and international enterprises,along with telecom operators andindustry organizations,have undertaken research on the inte

54、gration of wirelesscommunication and computing technologies.Domestic companies like Huawei andZTE begin by analyzing top-level business operations and scenarios to abstractvarious task/application requirements.Then,they break down these requirements intonetwork and computing functions,prioritizing r

55、esearch in areas such as computingand network orchestration,AI/sensing/computing KPI definitions,and capabilityopenness.Their main focus involves contributing technical solutions to CCSA andIMT2030(domestic promotion group)in China,actively conducting technicalvalidations,and promoting enterprise-de

56、veloped intelligent,computing,and platformproducts and solutions.Foreign companies such as Ericsson,Nokia,and Intelprimarily focus on the advancement of integrated communication and computing.They emphasize the integration of mature scenarios,includingAI and edge computing,intothistechnology.Theseco

57、mpaniesareactivelyresearchingintegratedcommunication and computing within industry organizations like NG Alliance andHexa-X.In 2023,China Mobile proposed a network architecture and technology systemfor the integration and coexistence of wireless communication and computing in theindustry.According t

58、o China Mobile,the design of a network architecture for theintegration and coexistence of wireless communication and computing must transitionfrom independent functionality and isolated resources to the integration andcoexistence of communication and computing.This involves moving away fromsiloed de

59、sign towards network platformization and shifting from single capability todiversified services.These changes are necessary to fulfill the demands of 6G-AIintegration development,integrated sensing and communication,and multi-elementintegration.10/97Figure 2-2 Reference Network Architecture of the I

60、ntegration and Coexistence of WirelessCommunication and ComputingThe reference network architecture as shown in Figure 2-2 comprises three mainlayers:infrastructure,network function,and management and orchestration.1)The infrastructure layer relies on massively distributed base stations andterminal

61、devices to offer virtual resources encompassing connectivity,computing,data,and models.2)The network function layer builds upon the new computing resource elementsfrom the infrastructure layer,enhancing traditional user and control plane functions todeliver computational execution and control functi

62、ons for integrated sensing andcommunicationservices.Thecomputationalexecutionfunctionisprimarilyresponsible for processing computing data,such as AI model inference and sensingcomputing.Meanwhile,the computational control function focuses on sensingwireless computing power,computing task demands,joi

63、ntly scheduling andcontrolling communication and computing resources in real time,and ensuringclosed-loop monitoring to guarantee QoS for communication and computing.3)Building upon the traditional operational and management functions of theRAN,the management and orchestration layer introduces the m

64、anagement andorchestration of wireless computing resources,joint management and orchestration ofcommunication and computing,and management and orchestration of wirelesscomputing services.The management and orchestration layer provides both localized,11/97integrated communication and computing servic

65、es and interfaces with the computingnetwork brain to incorporate wireless-side computing power into the broader network,thereby delivering integrated communication and computing services.IMT 2030(domestic promotion group)advocates for 6G to facilitate ubiquitousintelligence encompassing computing,AI

66、,and storage.This entails the ability forcomputing power sensing,scheduling,and sharing to cater to business needs.It alsoemphasizes the integration of wireless communication and computing resources,functions,and services.System design considerations must revolve around three keyaspects:the joint op

67、timization of computing and communication resources,supportfor the discovery,sensing,and control of distributed computing resources,and theprovision of computing services to terminals on the RAN side.In October 2023,theResearch Report on 6G ISAC System Design was released.It analyzed potentialcomput

68、ing key performance indicators and addressed critical technical challenges,including collaboration between sensing and computing nodes,high-performanceterminalcomputingservices,communication-computingcollaboration,andsensing-computing collaboration.When discussing the next-generation wireless networ

69、k,the NG Alliance,a 6GindustryorganizationintheUnitedStates,highlightedtheintegrationofcommunication and computing.It involves the following aspects:Deep integration of communication and computing:achieves more efficient datatransmission and processing.Virtual and software-defined network functions:

70、are proposed to achieve theintegrationofcommunicationandcomputingthroughvirtualizationandsoftware-defined approaches.Unified communication and computing protocols:are developed to facilitate datatransmission and processing for a variety of new applications,ultimatelyimproving network efficiency and

71、reliability.Application of AI and machine learning:achieves intelligent analysis andprocessing of network data through the use of AI and machine learningtechnologies.12/97Additionally,it released the white paper titled 6G Technologies for Wide AreaCloud Evolution,outlining research directions includ

72、ing joint orchestration ofcommunication and computing,dynamic orchestration of computing resources,capability customization,and data management.The goal is to leverage advancedtechnologiesand protocolsto drive the adoption of networkvirtualization,software-defined networks,AI,and machine learning,ai

73、ming to achieve enhancedefficiency in data transmission and processing.Simultaneously,the alliance will worktowards establishing standardized communication and computing protocols that caterto the requirements of diverse new applications,ultimately enhancing the user serviceexperience.The 6G flagshi

74、p research organization Hexa-X,under the European Commission,has outlined and delved into the projected applications and demands for the year 2030.The focus is on investigating how the 6G platform can effectively meet future needsby integrating communication,sensing,network computing,intelligence,an

75、d spacemapping into new digital services.Additionally,they have published the Joint DesignScheme of AI-Driven Communication and Computing to lay the groundwork throughearly-stageresearchonsystemarchitecture,scenarioapplications,andkeytechnologies,followed by advancing the proposedschemes into the gl

76、obalstandardization and coordination phase.According to industry research,the integration and coexistence of wirelesscommunication and computing needs to be achieved across three dimensions:resources,functions,and services.Figure 2-3 Three Major Technical Directions for the Integration and Coexisten

77、ce of WirelessCommunication and Computing(I)Regarding resources,the key challenge is how to maximize the value ofubiquitous RAN infrastructure to adequately fulfill service and technological13/97advancement requirements.Upon integrating computational tasks into the RAN,diverse service applicationsex

78、hibit substantial variations in their demands for computing power across differentcomputing hardware.Furthermore,the unified hardware platform must concurrentlysupport wireless networks and diverse business applications,presenting entirely newchallenges in real-time sharing,allocation,and management

79、 of heterogeneouscomputing resources.The integration of communication and computing resources tosupportreal-timeallocationofcomputingresourcesencompassesthreekeytechnologies:1)wireless computing power measurement designed to meet diversescheduling demands;2)multidimensional computing power model abs

80、traction,incorporating factors such as latency and computing capability;3)a lightweight cloudplatform capable of evolving into a real-time scheduling management platform forcomputing resources.(II)At the functional level,the primary challenges involve addressing featuressuch as space-time variations

81、 and fragmentation of computing power within highlydynamic wireless network environments,alongside constraints on communication andcomputingresources.Thecentralissuepertainstoachievingintegratedcommunication and computing services while ensuring their reliability.The concept of functional integratio

82、n and coexistence involves utilizing wirelesscommunicationprotocolsasthefoundationtointegratecomputingandcommunicationprocesses,enablingreal-timeprecisecontroloverbothcommunication and computing.As communication and computing services becomemore integrated within base stations,a growing array of AI

83、capabilities andcomputing services are being implemented within these stations.Consequently,theseAI capabilities and computing services are placing elevated requirements on both thecontrol plane and user plane of RANs.Research into key integrated communicationand computing technologies will therefor

84、e center on bolstering both the control anduser plane.Enhancing the control plane primarily aims to facilitate real-time sensing of14/97computing resources and tasks within the RAN,ensuring closed-loop quality ofservice for computing tasks.Key technologies encompass(1)design of wirelessfunctions,int

85、erfaces,and processes that support integrated control of communicationand computing;(2)coordinated scheduling of communication and computingresourcesundermulti-dimensionalobjectivesincludingend-to-endlatency,computing quality of service,and energy efficiency,alongside various communicationand comput

86、ing constraints.Regarding user plane enhancement,the user plane mechanism with a verticallyenclosed pipeline design can no longer meet the requirements for computing data.Theuser planes ubiquitous openness,dynamic routing,and integrated service capabilitieswill become the direction of evolution.To m

87、eet the demands of applications likeAI-native air interface and real-time rendering,it is crucial to develop a datatransmission framework for computing services that offer flexibility,efficiency,andopenness.Key technologies include diverse data offloading and fully connecteddynamic routing.(III)At t

88、he service level,the primary challenge lies in achieving collaborative,dynamic,on-demand orchestration and openness of wireless communication,computing,and intelligent resources,functions,and services.The primary research goal of the orchestration service is to establish a distributed,hierarchicalin

89、tegratedorchestrationsystemforwirelesscommunicationandcomputing,enabling agile localized orchestration capabilities for RANs.Keytechnologies encompass 1)joint orchestration of wireless communication andcomputing functions,interfaces,and procedural frameworks;2)modeling computingservice intentions,in

90、tention translation,and service(local)openness technologies.The integration and coexistence of wireless communication and computingtechnology represent the deep integration of DOICTs multi-domain resources andcapabilities.It serves as a critical foundation for future integrated services such asinteg

91、rated sensing and communication,integrated communication and intelligence,and integrated communication and computing.This technology effectively enhancesend-to-end efficiency in communication infrastructure and stands as one of the core15/97technologies driving the evolution of RANs toward the integ

92、rated communication,computing,and intelligence paradigm of 6G.2.2 Semantic Communication2.2.1 OverviewThe design of traditional communication systems relies on Shannons classicinformation theory,utilizing bit-based metrics to assess network performance-including bit error rate and transmission rate-

93、while focusing on resolving thetechnical challengesassociatedwithaccuratelytransmittingbits/symbols.Ascommunication technology advances rapidly,system capacity has now approached thetheoretical limit proposed by Shannons theory.Furthermore,with the deep integrationof communication technology and AI,

94、intelligent terminals can now comprehendtransmission tasks and scenarios,autonomously executing instructions.Consequently,task-driven communication systems prioritize achieving deep semantic-level accuracyover shallow bit-level precision.Nevertheless,traditional communication overlooksthe content an

95、d meaning of transmitted information,posing challenges in meeting thediverse and efficient transmission needs of multi-modal tasks in future 6G networks.Semantic communication is a new type of communication paradigm that focuseson the transmission of meanings and significance,rather than the symbols

96、 themselves.It emphasizes the accurate transmission of the meaning of the transmitted symbolsand how the received meanings can influence system behavior as intended 6.Semantic communication integrates users information needs and task semantics intothe communication process.By extracting,coding,trans

97、mitting,and reconstructingsemantic features,it greatly enhances communication efficiency and user experience,fulfilling future communication demands,and consequently drawing widespreadattention from global academic and industrial circles.Presently,studies on semantic communication are primarily cate

98、gorized into twotypes.One type is rooted in fundamental theoretical research of classical informationtheory,with a primary emphasis on developing semantic information theory 7.For16/97example,semantic entropy is defined using logical probability rather than statisticalprobability.Drawing inspiration

99、 from Shannons information theory,semanticinformation theory incorporates concepts such as semantic entropy,semantic ratedistortion,and semantic channel capacity.These elements aim to address thechallenges associated with measuring semantic information,semantic coding anddistortion,and maximizing se

100、mantic communication traffic.Despite some attention,there is currently no comprehensive and effective mathematical framework availableto describe semantic information theory.The other type involves designing AI-based systems,with a primary focus onalgorithm design and optimization for semantic featu

101、re extraction,coding,andtransmission.In recent years,as data processing capabilities have advanced,semanticfeature extraction methods empowered by deep learning have found successfulapplications across various types of information sources.For example,semanticfeatures of text information are extracte

102、d using natural language processingtechniques,whereas semantic features of image information are extracted usingcomputer vision and pattern recognition techniques.Additionally,in semanticcommunication systems driven by various tasks,extensive research has beenconducted on reinforcement learning-base

103、d scheduling and transmission policies.AsAIadvancesrapidly,semanticcommunicationpresentsextensivepotentialapplications in fields like multimedia communication,augmented reality,andimmersive communication.2.2.2 Key TechnologiesFigure 2-4 shows the system model of semantic communication.In contrast to

104、traditional communication,semantic communication incorporates a knowledge baseand replaces the traditional source coding module with semantic coding.The semanticknowledge base is deployed at both the transmitter and receiver,representing theshared common knowledge of both parties.The semantic encode

105、r is employed toextract semantic features associated with underlying transmission tasks.The semanticdecoder is employed to reconstruct corresponding semantics based on the received17/97features.In semantic communication,apart from the physical channel,the concept ofa virtual semantic channel is intr

106、oduced.Semantic noise arises from discrepancies inknowledge bases or semantic mismatches at both the transmitter and receiver.Leveraging knowledge bases,semantic communication selectively transmits the mostpertinent semantic features for a given task,rather than transmitting all raw data.Thisallows

107、for the removal of a significant portion of redundant data,thereby enhancingnetwork efficiency.Figure 2-4 Semantic Communication System ModelThe semantic communication system mainly encompasses the following keytechnologies:Joint Semantic-Channel Coding:With traditional source coding and channelcodi

108、ng technologies nearing their theoretical limits,there has been extensive researchon joint source-channel coding technology.Studies indicate that integrated designoutperformsseparatemodulardesign.Thefundamentalconceptofjointsource-channel coding involves allocating more bits to source coding in high

109、signal-to-noise ratio(SNR)conditions to enhance system transmission efficiency.Conversely,in low SNR conditions,allocating more bits to channel coding serves tocounteract the detrimental effects of noise.In semantic communication systems,semantic encoders replace traditional source coding modules an

110、d use deep neuralnetworks(DNN)to extract semantic features of input information,which can reducecommunication traffic.As shown in Figure 2-5,joint semantic-channel coding,inspired by joint source-channel coding,has received widespread attention and isconsidered one of the most important technologies

111、 in semantic communication.By18/97incorporating the SNR into the channel features during DNN training,jointsemantic-channel coding enables the extraction of semantic features in noisyenvironments.Thanks to the rapid development of deep learning algorithms,jointsemantic-channel coding has been applie

112、d in various sources 8,9,such as text,images,and audio,especially significantly enhancing stability in low SNR conditions.Figure 2-5 Joint Semantic-Channel Coding ModelSemantic-based Modulation and Transmission:In a semantic communicationsystem powered by deep learning,the output signal of the encod

113、er is a continuousfloating point number.For the sake of simplicity,some studies explore the directtransmission of continuous output signals through analog modulation,eliminating theneed to discretize them into constellation symbols.However,such modulationassumptions are too idealistic,making it diff

114、icult to deploy them in practice due tohardware limitations.Compared with analog modulation,digital modulation has lessimpact from noise and higher capacity.Therefore,semantic-based digital modulationand transmission technology becomes crucial.Currently,the primary method involvesquantizing continuo

115、us signals,mapping them into constellation symbols,anddiscretely transmitting them.In some studies,a uniform quantization method isutilized to equidistantly map the output of DNNs into symbols,while the other studiesinvolve non-uniform quantization.In this context,an additional neural network isempl

116、oyedtomapcontinuoussignalsintopotentialconstellationsymbols.Subsequently,appropriate symbols are selected based on probability,enabling robusttransmission in semantic communication systems 10.19/97Resource Allocation in Semantic Communication Systems:Wireless resourceallocation is commonly conceptua

117、lized as a constrained mathematical optimizationproblem,aiming to achieve equilibrium among various factors,such as energyefficiency,transmission delay,and throughput,to attain optimal system performance11.It is no longer enough to pursue traditional bit-based optimization goals,like biterror rate m

118、inimization,to ensure semantic communication quality.Thus,an essentialfield of research is developing new optimization goals to assess the significance ofsemantics.The new indicators of semantic communication include semantic rate,semantic spectral efficiency,and semantic Quality of Experience(QoE).

119、For example,prior research has maximized semantic spectral efficiency in single-modal texttransmission systems and optimized the semantic QoE in multi-modal communicationsystems 12.2.2.3Applications and Challenges1)Can semantic information theory be analogized to classical informationtheory?Classica

120、l information theory offers a comprehensive and general mathematicalframework,along with associated theoretical limits,and has achieved significantsuccess in the design of communication systems.Inspired by it,numerous researchershave sought to advance semantic information theory.Specifically,they ha

121、ve proposedthe concept of semantic entropy by replacing statistical probability with logicalprobability.By introducing distortion among semantic features,they have exploredthe semantic rate distortion theorem.Although significant efforts have been made inthe field of semantic information theory,ther

122、e is still no universally accepteddefinition for semantic entropy or semantic channel capacity.In contrast to traditionalbit-level communication systems,semantic communication systems are goal-directedand task-oriented.Therefore,there may not be a universal mathematical framework todescribe the sema

123、ntic communication system of multi-modal data.In our view,semantic entropy and semantic channel capacity should be linked to the underlyingtransmission tasks and background knowledge bases.Furthermore,they requireadaptation to diverse environments.20/972)Can existing measurement indicators evaluate

124、the performance of semanticcommunication systems?Appropriate measurement indicators are the foundation for designing wirelesssystems.For example,they play a crucial role in designing the loss function andselecting parameters during the training of DNN models.Despite initial efforts toexplore semanti

125、c measurements of diverse data sources such as text,images,andaudio from various perspectives,including accuracy and timeliness,there remains aneed for more appropriate and systematic performance evaluation standards before thepractical implementation of semantic communication systems.In addition to

126、 usingtraditionalobjectiveindicatorstomeasuresystemperformance,semanticcommunication systems also require subjective performance indicators to evaluatehuman experience.Furthermore,the existing semantic measurements are tailored forspecific data sources,emphasizing the necessity of developing univers

127、al measurementstandards(similar to the bit error rate in traditional wireless communication)forevaluating the performance of systems integrating multi-mode or multitaskingnetworks.3)How do we address the issue of excessive computing overhead induced byAI?IMT-2030 proposes that integrated AI and comm

128、unication is one of theanticipated use scenarios for 6G.Consequently,there is an increasing demand toincorporateAIintowirelesscommunicationsystems.CurrentsemanticcommunicationsystemsarepredominantlypoweredbyDNNs,necessitatingadditional computing resources and potentially resulting in significant com

129、putingoverhead.Therefore,there is a subtle balance between network performance andcomputing power during the practical implementation of semantic communicationsystems.Fortunately,the computing power required for semantic communication isavailable on some new terminal devices.Furthermore,edge intelli

130、gence enables rapiddata exchange by distributing computing resources to the network edge,and cloudintelligence offers supplementary computing power.In addition,the computing powernetwork(CPN),as a new architecture,can integrate various computing resources andfacilitatetheircollaborativesharingamongu

131、sers,therebyintroducingnewpossibilities for enabling semantic communication.21/974)Privacy and security in semantic communication systemsMaintaining information privacy and security represents a critical concern acrossdiverse wireless communication systems.As semantic communication systemstransmit o

132、nly specific essential data(semantic features)and rely on the recipientsknowledge base for decoding,they are considered a method capable of providingprivacy and security protection to some extent.Nevertheless,since only essential datais transmitted,the eavesdropping of semantic information could hav

133、e profoundlyserious implications.In semantic communication systems,which encompass bothphysical and semantic channels,securing communication involves encryptingsemantic features during transmission and ensuring alignment between knowledgebases.Encrypting semantic features can protect privacy,but inc

134、orporating additionalsecurity coding may lead to higher communication overhead.Therefore,it isnecessary to dynamically manage the trade-off between data security and transmissionefficiency depending on various transmission tasks.5)Application and standardization of semantic communication in 6GSemant

135、ic communication is seen as a significant enabler of augmented realityand immersive communication in the context of 6G,facilitating the connectionbetween the virtual and physical worlds.As wireless communication technologyadvances,IMT-2030 strives to enhance and enrich immersive experiences,extendco

136、verage,and enable intelligent collaboration.Immersive communication in 6Gentailsstringentrequirementsforreal-timequalityandcapacity.Semanticcommunication can extract semantics from various signals,including action,gesture,and voice inputs.Removing redundant information reduces the volume of transmit

137、teddata while maintaining seamless data interchange.Semantic communication canalleviate downlink pressure and catalyze the implementation of augmented reality andimmersive communication technologies.2.3 Large Model and Wireless CommunicationThe rapid advancement ofAI,driven by big data and cloud com

138、puting,has led tothe emergence of numerous large AI models,such as ChatGPT.These models offer22/97highly intelligent human-machine interaction experiences and incredibly creativecontent generation capabilities,revolutionizing peoples work and lifestyles.This hasrapidly ignited a global enthusiasm fo

139、r large models,signaling a new era inAI.Communication systems form the backbone of modern information society andare increasingly required to be more efficient,stable,and intelligent.As 5G wirelesscommunicationnetworksadvance,thearchitectureofwirelessnetworksiscontinuously evolving,accompanied by an

140、 increasing variety of terminals and serviceapplications.For a long time,there have been many technical problems incommunication that are difficult to model accurately or solve efficiently usingtraditional methods.The demonstrated potential of large AI models has led towidespread belief in their bro

141、ad applicability in areas such as natural languageprocessing,computer vision,and multi-modal fields,as well as their ability toadvance intelligent autonomous network capabilities within communication networks.In addition,the rapid development and broad application of large AI models also needto be s

142、upported by a communication network with large bandwidth,low delay,andhigh reliability.Therefore,researching the integration of large AI models withwireless communication is of significant importance.2.3.1 Current Research Status of Large Models at Home andAbroadOriginating from natural language pro

143、cessing,large models refer to machinelearning models with considerable parameter scales and complexity.Typically,theseare abbreviated as Large Language Models(LLM)when mentioned.In terms ofparameter scale,large AI models have undergone three stages:pre-training models,large-scale pre-training models

144、,and ultra-large-scale pre-training models.The numberof parameters has advanced from the scale of billions to trillions,marking significantbreakthroughs.Based on modal support,large AI models can be categorized intonatural language processing,machine vision,and scientific computing.Depending ontheir

145、 application fields,large models can be categorized as either general-purpose orindustry-specific.General-purpose large models exhibit powerful generalizationabilities,resembling general education,whereas industry-specific large models23/97adjust the models with industry knowledge,similar to profess

146、ional education.Currently,major institutions abroad that release LLMs include OpenAI,Anthropic,Google,and Meta,with model parameter scales mainly in the tens ofbillions and hundreds of billions.Up to now,the top GPT large models abroad includeChatGPT,Claude,Bard,and Llama.Among them,after Google rel

147、eased the latestnative multimodal large model Gemini,Bard was officially renamed Gemini.On a global scale,prominent institutions that have released large models includeOpenAI,Anthropic,Google,and Facebook.To date,leading large AI models abroadmainly include ChatGPT,Claude,Bard,Llama2,HuggingChat,and

148、 PaLM 2,but newlarge models continue to emerge.For example,on December 6,2023,Googleofficially launched the Gemini large model,which is inherently equipped to supportmulti-modal processing,enabling it to handle various types of data.In this global competition,China is also staying abreast.Domestical

149、ly developedlarge models are rapidly progressing,transitioning from the confines of research labsto real-worldtesting.Notable examples include iFlyteksSpark,TencentsHunyuan,Alibabas Tongyi Qianwen,Huaweis Pangu,Baidus Ernie Bot,andChina Mobiles Jiutian.Data indicates that by the end of November 2023

150、,more than300 large models have been introduced to the Chinese market.2.3.2 Empowering the Communication Industry with LargeAI ModelsLarge AI models can serve as tools to enhance information and communicationservice capabilities.For example,massive data from communication networks can beutilized to

151、train large communication network models,and the robust natural languagecapabilities of large models can be employed to enhance operational service functionslike intelligent customer service.Moreover,the operation and provision of services bylarge models like ChatGPT necessitate substantial computin

152、g power and robustnetwork infrastructure.This,to some extent,drives the development of integratedcomputing and networking,thereby advancing the evolution of information andcommunication technology.Large models are pre-trained and learned on massive unlabeled data,and then24/97domain-specific models

153、are obtained by fine-tuning a small amount of labeled data.Therefore,the robust natural language comprehension and generation abilities of largemodels enable their widespread application in the field of communication.Thefollowing provides several typical application scenarios of largeAI models in th

154、e fieldof communication:Figure 2-6 Application of Large Models in the Field of Communication(1)Intelligent customer serviceTraditional human-operated customer service deals with a high volume ofcustomer inquiries daily,encountering a vast array of information needs.Thechallenge lies in swiftly ident

155、ifying user requirements and delivering precise servicesefficiently.Large AI models enable intelligent optimization of communicationservices,enhancing the quality and efficiency of customer service in various areassuch as voice assistants,smart recommendations,and personalized offerings.Thisaddresse

156、s the increasing demand for personalized experiences,enriches userinteractions,and facilitates a deeper understanding of customers,ultimately boostingenterprise competitiveness and profitability.For example,by integrating large AImodels into the customer service hotlines of telecommunications operat

157、ors,thesystem can autonomously recognize speech content and discern the users emotionalstate to provide suitable automated responses,thereby alleviating the workload of25/97customer service staff.(2)Intelligence developmentAs communication technology evolves,numerous new communication devicescontinu

158、ally emerge,necessitating extensive research and testing.This process requiresa significant investment of manpower and time.Large AI models can aid in analyzingrequirements and designing prototypes during the development of software or systemsinthecommunicationfield.Thisassistancehelpsdevelopmenttea

159、msandstakeholders gain a deeper understanding of the systems workflow and functions,facilitating early issue detection and requirement improvements.Additionally,largemodels facilitate virtual testing of communication devicesto validate theirperformance and stability,enabling rapid iterative improvem

160、ents and expeditingproduct introduction and promotion.This serves as a valuable reference during theplanning and design of new communication products,leading to time savings in plandevelopment and increased efficiency.(3)Network optimizationIn response to communication networks like core networks of

161、 data services,circuit-switched core networks,and radio access networks,the growing diversity ofprocesses and services is steadily increasing the pressure on network operations.Leveraging large AI models allows for the detection of problems like networkcongestion and performance issues,the simulatio

162、n of communication network,andthe evaluation of performance indicators,thus offering optimization suggestions.Inaddition,large AI models can forecast trends by analyzing historical data andmonitoring outcomes,thus balancing network loads to ensure optimal user networkexperience.For example,proactive

163、 prediction and resource allocation can be appliedto high-traffic areas and periods.At the same time,an intelligent shutdown of certainbase station facilities can be enacted in low-traffic areas and periods,optimizingoperational costs and maintaining the communication networks peak performance.(4)Ne

164、twork O&MIn traditional communication network O&M,personnel must rely on manualinspections and data analysis to gather network status information,resulting in lower26/97efficiency.Introducing a large AI model at the operational level of network O&Menables real-time,efficient monitoring of network st

165、atus information to identifypotential faults and issues.It can quickly identify abnormalities,perform faultdiagnosis,and make forecasts by analyzing log data,alarm information,and networkequipment performance indicators.Employing large model technology,for example,improves log text comprehension in

166、anomaly detection and makes template-based logparsing and data generation possible.In the end,this improves network stability andreliability by making it easier to do duties like alleviating alarms,identifyingabnormalities,anticipating defects,and diagnosing problems.It also makes it possibleto take

167、 the proper maintenance and repair measures.2.3.3 Intelligent Endogenous Communication NetworkHuman society is about to enter an era of intelligence,and one of the keyfeatures of 6G networks will be the deep integration of AI technologies to establishAI-native networks.Therefore,6G networks must inc

168、orporate AI seamlessly from theoutsetin theirarchitecturaldesign,comprehensivelygrasptheperformancerequirements and challenges confronting 6G networks,and devise a networkarchitecture tailored to meet the requirements of the next generation of wirelesscommunication systems.Presently,research on the

169、architecture of AI-native 6G networks,encompassingsystem structure,functions,and processes,is steadily progressing.CICT Mobile hasintroduced an AI-native 6G network system with a one-superbrain,multi-entity,andatomic framework 13.Similarly,China Mobile has outlined a comprehensive3-entity,4-layer,an

170、d 5-plane architecture for 6G,examining the 6G networkarchitecture from spatial,logical,and functional perspectives 14.Moreover,ChinaTelecom and ZTE have collaborated on a 3-layer and 4-plane intelligent architecturefor 6G networks,encompassing cloud network resources,network functions,application e

171、nablement,as well as control,user,data,and intelligence planes 15.Additionally,the IMT-2030(6G)Promotion Group has proposed a system frameworkgeared towards AI-native systems,offering a comprehensive analysis of the27/97requirements for such architecture across functions,services,deployment,control,

172、and execution perspectives 16.The deep integration of large AI models and communication networks will makeintelligence the gene of the network,thus constructing a new type of AI-basedcommunication network,promoting people to accelerate towards an era of intelligentinclusiveness,truly realizing the b

173、eautiful vision of Intelligent Connection ofEverything and Digital Twin.References1 IMT-2030(6G)推進組：2023 年 6G 通感融合系統設計研究報告2 Nan Li,Qi Sun，Xiang Li，Fengxian Guo，Yuhong Huang，Ziqi Chen,Yiwei Yan，MugenPeng,“Towards the Deep Convergence of Communication and Computing in RAN:Scenarios,Architecture,Key Te

174、chnologies,Challenges and Future Trends”，China Communications，2023,Vol.20,Issue(3)3 Qi Sun,Nan Li,Chih-Lin I(Fellow,IEEE),Jinri Huang,Xiaofei Xu,Yuxuan Xie,“IntelligentRAN Automation for 5G and Beyond”,IEEE wireless communications,20234Yuhong Huang,Nan Li,Qi Sun,Xiang Li,Jinri Huang,Ziqi Chen,Xiaofe

175、i Xu,Chih-Lin I,“Communication and Computing Integrated RAN:A paradigm shift for Mobile Networks”,IEEENetworks,2024.5 Guo,Fengxian,Mugeng Peng,Nan Li,Qi Sun,Xiang Li,“Communication-ComputingBuilt-in-Design in Next Generation Radio Access Networks:Architecture and Key Technologies”，IEEE network 2023.

176、6 Z.Qin,F.Gao,B.Lin,X.Tao,G.Liu,and C.Pan,“A generalized semantic communicationsystem:from sources to channels,”IEEE Wireless Commun.,vol.30,no.3,pp.1826,Jun.2023.7 J.Bao,P.Basu,M.Dean,C.Partridge,A.Swami,W.Leland,and J.A.Hendler,“Towards atheory of semantic communication,”in Proc.IEEE Netw.Science

177、Workshop,Jun.2011,pp.110117.8 H.Xie,Z.Qin,G.Y.Li,and B.-H.Juang,“Deep learning enabled semantic communicationsystems,”IEEE Trans.Signal Process.,vol.69,pp.26632675,Apr.2021.9 Z.Weng and Z.Qin,“Semantic communication systems for speech transmission,”IEEE J.Sel.Areas Commun.,vol.39,no.8,pp.24342444,Au

178、g.2021.10 Y.Bo,Y.Duan,S.Shao,and M.Tao,“Learning based joint coding modulation for digitalsemantic communication systems,”in Proc.IEEE WCSP,Nov.2022,pp.16.11 Z.Ji and Z.Qin,“Energy-efficient task offloading for semantic-aware networks,”in Proc.IEEE ICC,Jun.2023,pp.35843589.12 L.Yan,Z.Qin,R.Zhang,Y.L

179、i,and G.Y.Li,“Resource allocation for text semanticcommunications,”IEEE Wireless Commun.Lett.,vol.11,no.7,pp.13941398,Jul.2022.13 S Bao,W Sun,H Xu.A Native Intelligent and Security 6G Network ArchitectureC.2022IEEE/CIC International Conference on Communications in China(ICCC Workshops),2022:28/97395

180、-400.14 中國移動.中國移動 6G 網絡架構技術白皮書R.2022.15 中國電信研究院,中興通訊.6G 網絡架構展望白皮書R.2023.16 IMT-2030.面向 6G 網絡的智能內生體系架構研究報告R.2022.29/973.ICDT-integrated Wireless Air Interface3.1 Integrated Sensing and Communication3.1.1 TypicalApplication ScenariosThe evolution of mobile communication from traditional functions to s

181、ensingcapabilities is a pivotal element in underpinning the digital transformation acrossdiverse sectors and facilitating the robust growth of innovative 6G services.For years,wireless sensing,encompassing positioning,motion detection,and imaging,has beenan independent technology evolving alongside

182、mobile communication systems.Until5G,positioning stood as the sole sensing as a service(SaaS)exclusively offered bymobile communication systems,primarily catering to active devices like terminalsand vehicles transmitting signals.6G aims to deliver a more comprehensive range ofsensing services,encomp

183、assing the ability to sense both active devices and passiveobjects 1.Integrated sensing and communication will enable mobile communicationsystemoperatorstooffermanynewservices.Theseservicesencompasshigh-precision positioning,tracking,biomedical and security imaging applications,synchronized mapping

184、for automated construction of complex indoor or outdoorenvironment maps,pollution or natural disaster monitoring,gesture,and activityrecognition,defect and material detection,along with numerous other services 2-4.Hidden object imagingObject detectionSecurity checkFigure 3-1 Typical Sensing Applicat

185、ionsIn turn,these services will offer a variety of service application scenarios forfutureconsumersandverticalindustries.Usecasesacrossvariousapplications/industries(vertical industries,consumers,and public services)arecategorized into four functional groups:high-precision positioning and tracking,3

186、0/97synchronous imaging,mapping,and positioning,enhanced human perception,andgesture and activity recognition.Furthermore,it is noteworthy that apart from usingcommunication signals,6G networks introduce sensing as a novel service.Thesensing results can further aid in communication 5 and positioning

187、,subsequentlycontributing to optimizing communication network performance.Integrated sensing and communication in smart life has diverse applications,ranging from enhancing smart home features such as home control,securitymonitoring,and behavior tracking to healthcare functions like respiratory and

188、heartrate monitoring,terahertz imaging,and spectral examination.Furthermore,it extendsto cultural and entertainment uses,encompassing extended reality(XR)experiences.In terms of smart manufacturing,integrated sensing and communication can drive theadvancement of industries such as smart factories,UA

189、Vs,and IoVs.Integratedsensing and communication in a smart society can support environmental monitoring(like rainfall/atmospheric humidity measurement,pollution gas detection,and airquality monitoring),public security(like security checks and electronic fences),andurban management.Figure 3-2 Applica

190、tion Scenarios of Cooperative ISACAs integrated sensing and communication develop in the future,applicationscenarios will progress along two dimensions:breadth and depth.Broadening thescope of application scenarios involves diverse exploration into integrated sensing andcommunication,resulting in an

191、 extensive array of scenarios that provide moreopportunities for technical research,resonate with a wider spectrum of industry31/97partners,and enhance the technologys impact on the industry.Addressing the depth ofapplication scenarios primarily involves refining optimal settings and selecting andex

192、aminingscenariosbasedontechnologicalfeasibilityandurgentscenariorequirements.3.1.2 Key Technologies and ChallengesIn the context of 6G,both communication and sensing services must take intoaccount channel models.This presents a substantial challenge to the methodology ofchannel modeling.Before 5G,th

193、e random channel modeling method,with its lowcomputing complexity and ease of standardization,had been the dominant approachfor evaluating wireless communication.This method was widely used in variousprojects and standards,including 3GPP-SCM,WINNER-I/II,COST 2100,andMESTIS 6,and proved adequate for

194、assessing communication performance.However,given the diverse nature of sensing applications,it is challenging forstatistical models to accommodate the varied requirements of these applications.Regarding air interface design,the signal design for sensing and communicationprimarily focuses on joint w

195、aveform design 7-8.The primary challenge is theconflicting key performance indicators(KPIs)for communication and sensing.Inparticular,communication aims to maximize spectral efficiency,whereas sensingprioritizes parameter estimation resolution and accuracy in optimal waveform design.According to the

196、 current technological level,there is still room for waveform designto strike a balance between excellent communication and sensing performance.Like maintaining communication continuity,ensuring continuity is also essentialfor sensing services.This can be achieved through the switching of sensing mo

197、desand(or)sensing nodes.The primary challenges in switching and selecting sensingnodes revolve around the design of air interface signaling.This necessitates a clearsignaling interaction process and specific signaling content,including the exchange ofsensing configuration information,sensing-related

198、 data,and(some)sensing resultsbetween the source and target nodes.32/97From an algorithmic perspective,high-precision sensing algorithms facilitatecooperative ISAC.In practical networks,it is necessary to optimize the design ofsensing algorithms by considering factors such as algorithm performance,r

199、esourceoverhead,andcomplexity,toachieveoptimalcommunicationandsensingperformance.Furthermore,introducing non-line-of-sight(NLOS)environmentsbrings about sensing errors.Further investigation is required to attain high-precisionsensing in NLOS conditions.When designing integrated sensing and communica

200、tion systems,a key emphasislies in the shared utilization of baseband and RF hardware.The integrated sensing andcommunication hardware effectively reduces combined power consumption,systemsize,and information exchange delay between the two systems,while also fosteringmutual benefits in distortion ca

201、libration and compensation for both sensing andcommunication systems.However,it is important to note that due to the differingevaluation criteria and algorithms for communication and sensing,there are distinctdifferences in hardware requirements.As a trade-off,it is important to consider howthe dist

202、ortion parameters of traditional communication hardware can affect sensingperformance.In cooperative ISAC,leveraging information acquired from sensing hardware incommunication to enhance performance may be a key benefit of the integratedsensing and communication system.In particular,sensing hardware

203、 acquiresenvironmental information through echoes to establish a more deterministic andpredictable propagation channel.The environmental knowledge from sensinghardware not only enhances the accuracy of channel estimation in mmWave but alsoprovides beam alignment information or reduces link blockage,

204、thus leading to asignificant reduction in overhead.The mobile communication network presentssignificant opportunities and advantages for collaborative sensing.Collaborationamong multiple nodes,such as base stations and UEs,is an effective approach foraccessing spatial gain and reaping the joint proc

205、essing benefits of cooperative ISAC.When multiple network nodes function as a unified sensing system,the schedulingandintegrationofmulti-dimensionalinformationenableshigh-precisionand33/97super-resolution sensing of specific targets within a designated area.Maximizingcollaborative reception gain thr

206、ough the integrated processing of multi-dimensionalinformation from multiple collaborating nodes forms the core challenge of multi-nodecollaboration.The primary research challenges in this context revolve aroundachieving optimal sensing-communication integration results at minimal cost throughsynchr

207、onization,joint processing,information transmission,and compression,as wellas network resource allocation3.2 AI-nativeAir Interface3.2.1 TypicalApplication ScenariosThe AI-native air interface can be applied to two typical scenarios.Firstly,AIlearning can enhance air interface performance by impleme

208、nting end-to-end AI fortailored air interface optimization,integrating sensing capabilities,and advancing theair interfaces self-evolution.Secondly,integrating intelligence depth into wirelessnetworks to enable large-scale distributed training and real-time edge inference via thewirelessairinterface

209、deliversadaptableintelligencecapabilitiesandhigh-performanceAI services for diverse application scenarios.Figure 3-3 AI-based Wireless End-to-End Communication System34/97Figure 3-4 Intelligent Wireless Resource ManagementAI empowers the air interface to achieve peak performance,commonly applied ini

210、ntelligent signal processing and wireless resource management.Intelligent signalprocessing:Leveraging AI techniques allows for a shift from the current modularoptimization approach in the signal processing chain of communication systemsphysical layers to achieve end-to-end optimization of transmitte

211、rs and receivers 13.Specificusecasesencompasswirelesschannelmodeling,channelestimation/prediction/feedback,signal processing optimization,intelligent transceiverprocessing,as well as positioning and sensing 14,15,16.AI-based wirelessresource management:Can rapidly respond to complex and dynamic link

212、environments,resource features,and service requirements,intelligently handling themulti-user,multi-target,high-dimensional optimization decision problems of complexlinks.Specific use cases include MAC layer scheduling,MIMO pairing,powercontrol,MCS selection,RRC layer switching,and load balancing 17,

213、18.(a)Distributed AI Model Inference Service(b)Distributed AI Model Training ServiceFigure 3-5 Air Interface-enabled Efficient Distributed AI Services35/97Air interface-enabled efficient distributed AI services provide diverse supportcapabilities for AI via the wireless air interface,leading to more

214、 effective andreal-time AI training and inference.Typical applications encompass distributed AImodel inference and AI model training services.Distributed AI model inferenceservice:Leveraging the combined capabilities of communication and computing,thisserviceprovidesreal-time,high-precisionmodelinfe

215、rencetousersthroughcollaborative modeling.This enables terminals with limited computing power toaccess high-performance AI services.High-precision model inference services withreduced latency can be provided by leveraging the computing and model resources ofedge network nodes through the wireless ai

216、r interface 19.Distributed AI modeltraining service:In large-scale distributed model training,it facilitates high-speedcommunication and efficient scheduling of data and model parameters amongdistributed intelligent agents via the wireless air interface.This supports swift modelintegration and distr

217、ibution while ensuring the protection of model training and userprivacy.Terminals utilize local data to train models,which are then uploaded to thenetwork via the wireless air interface.The network combines and disseminates thesediverse terminal models,enabling collaborative user learning 20.3.2.2 K

218、ey Technologies and ChallengesThe key features of the AI-native air interface include data driven,continuouslearning,dynamic wireless environment,and corresponding key technologiesincluding dataset construction,online learning,and distributed learning.Dataset construction forms the foundation and pr

219、erequisite for AI methodologies.A typical wireless AI research dataset may comprise sub-datasets like channel data,environmental data,empirical data,user profile data,and pre-trained model data.Sub-datasets are collected in real time during wireless communication processes.Additionally,these data ca

220、n aid various communication tasks,contributing to thenetworks increased intelligence,even through pre-trained model training andinference 21.Online learning enables both the transmitter and receiver to process the36/97transmission signals based on neural networks and adapt to changes in thetransmiss

221、ion environment.The transmitter can then complete relevant optimizationsby receiving corresponding feedback information.The wireless channel between thetransmitter and receiver is a crucial feature that sets AI-native air interface apart fromtraditional AI.Designing feedback signals for online learn

222、ing of the AI-native airinterface necessitates careful attention to the features of the wireless channel.Distributed learning technologies,exemplified by Federated Learning,enable thetraining of global models through distributed methods while ensuring the privacy ofuser data.In wireless networks,the

223、 highly dynamic nature of connection topology dueto user mobility poses new challenges for the deployment and utilization ofdistributed learning technologies.The key to successful wireless distributed learninglies in designing a highly scalable and robust distributed learning approach thatcapitalize

224、s on features like user mobility in wireless networks.AI-native air interfacealso encounters a series of challenges in theoretical,architectural,and technologicalresearch as well as practical implementation.For example,at a foundationaltheoretical level,the guidance on building the theoretical frame

225、work for training anddeploying AI-native air interface models is crucial for optimizing overhead andperformance in communication,computing,and storage.The evaluation criteria andindicators serve as benchmarks for assessing the efficacy of AI-native air interfacealgorithms,providing essential foundat

226、ions for future research and standardization ofAI-native air interfaces.It is integral to scientifically evaluate the performance ofAI-native air interfaces,delineating the learning capacity indicators ofAI models(e.g.,model capacity)and establishing methodologies and theoretical frameworks to asses

227、sthe convergence,learning speed,training costs,as well as the performance limits andreliability of model inference for wireless AI algorithm models.Furthermore,thequality of datasets forms the cornerstone of AI-native air interface research,significantly influencing AI algorithm design and performan

228、ce AI algorithmsdesigned with specific datasets often struggle with limited generalizability acrossdifferent scenarios,necessitating thorough research into techniques for improvinggeneralization.At the level of system architecture and technical solution research,we37/97explore the optimal technology

229、 for achieving peak performance of the AI-native airinterface.We aim to discover innovative ideas and methods for air interface protocoldesign and implementation,as well as identify the most effective technical solutionsto manage extensive resource overhead in areas such as communication,computing,a

230、nd storage within the system.Additionally,our focus is on realizing the architecturaldesign of the AI-native air interface.In practice,we encounter challenges related tohigh energy consumption and limitations in chip capabilities induced by AIcomputing.Furthermore,we aim to address the communication

231、 and computingresource overhead associated with learning and training in wireless dynamicenvironments.Additionally,we seek to establish universal training methods toaccommodate the presence of numerous heterogeneous models within the network.3.3 MultipleAccess3.3.1 TypicalApplication ScenariosIn Jun

232、e 2023,the ITU released a proposal outlining six major usage scenarios for6G as part of the future development framework and overall objectives 22.In theMassive Communication scenario,there are heightened requirements for connectiondensity.This progresses from 2,000 connections per square kilometer

233、in 4G,tomillions of connections per square kilometer in 5G,and further escalates to tens ofmillions or even billions of connections per square kilometer in 6G.Apart from theincreased connection density,there will be new demands in the traffic modelcompared to 5G,such as an increase in the data trans

234、mission frequency of terminals.The challenge that needs to be overcome is supporting communication for massiveterminals while staying within the limits of network signaling and data transmissionresources.Therefore,it is essential to explore new multiple access technology toaccommodate a greater numb

235、er of terminal connections.38/97Figure 3-6 Application Scenarios of Multiple AccessBased on varying service features and indicator requirements,the applicationscenarios for massive terminal communication fall into two categories.One categorypertains to ultra-large-scale connection scenarios,necessit

236、ating support for asignificant number of terminals(e.g.,108devices/km2).This category exhibitsinsensitivity to latency and usually involves the transmission of intermittent smalldata packets,as illustrated by the orange curve in Figure.The other category caters toscenarios supporting a substantial n

237、umber of terminals(e.g.,106devices/km2)whileimposing increased requirements on indicators such as data packet size,end-to-endcommunication latency,and data transmission reliability,as illustrated by the bluecurve in Figure.In 6G,multiple access technology must meet the requirements of theaforementio

238、ned two application scenarios.More specifically,it can be implemented inthe following typical use cases.Figure 3-7 Typical Use Cases of Multiple Access39/97Table 3-1 Features of Typical Use CasesTypical Use CasesFeatureToC Digital TwinWorldPersonalized experience:requires high connection density,sup

239、porting a 106devices/km2connection density.Real-time experience:requires stringent latency to achievemillisecond-level end-to-end communication.Immersive experience:imposes rather high requirements onthe size of transmitted data packets,ensuring that hundreds tothousands of bytes can be transferred

240、within each packet.CriticalConnection-Intensive IoVHigh-density IoV information transmission with massivenessand burstiness.Required low latency and high reliability for informationtransmission.Fastvehiclemovement,frequentcellhandovers,anddiminishededgereliabilitycausedbyneighboringinterference.Ultr

241、a-Low-PowerIoT 23A significant increase in the number of terminals comparedto 5G.Low cost of terminals,passive deployment,and minimalmaintenance operations.Including passive terminals,semi-passive terminals,andactive low-power terminals.3.3.2 Key Technologies and ChallengesIn certain scenarios deman

242、ding extensive connectivity,real-time quality,and lowlatency like ToC digital twin world the transmission of sizable data packetstypically necessitates millisecond-level latency.However,current multiple accesstechnologies struggle to fulfill the connection density needs while also addressingcommunic

243、ationlatencyrequirements.Moreover,forapplicationswithultra-large-scale connections such as ultra-low-power IoT,the likelihood of collisionsand communication latency steadily rises alongside the growing number of connectedterminals.Consequently,existing multiple access technologies face challenges in

244、meeting the connection density demands for these application scenarios.To supportthe typical application scenarios in section 3.3.1,the evolution of multiple accesstechnologies can be approached from various dimensions.(1)Non-Orthogonal Multiple Access(NOMA):enhances the systems capacityfor multi-us

245、er multiplexing by enabling the sharing of time-frequency resources for40/97data transmission among multiple terminals.This facilitates a higher density ofconnections and improved communication among a larger number of terminalssimultaneously.It enables the differentiation of users who access and co

246、mmunicatesimultaneously based on features such as code domain,power domain,and spacedomain.Some typical technologies include Integration of Random Access andMultiple Access Transmission,Multiple Access Scheme Based on Virtual UserSplitting,and MultipleAccess Scheme Based on Resource Hopping.(2)Uncoo

247、rdinated method:In scenarios involving extensive terminal access anddata transmission,employing a coordinated method brings about substantial signalingresource overhead.Constrained resources and specific delay requirements limit themaximum allowable number of connections.To enable simultaneous acces

248、s and datatransmission for a large volume of terminals and enhance connection density,itbecomes crucial to consider uncoordinated access and data transmission methods.Typical technologies include Integration of Random Access and Multiple AccessTransmission,Efficient Connectionless Transmission,Multi

249、-User Encoding andDecoding Schemes Based on ODMA,and Sparse IDMAfor UMA.(3)Integration of random access and multiple access transmission:Traditionalcommunication systems typically involve accessing process before transmitting datafor communication purposes.In scenarios where massive terminals are ac

250、cessing andtransmitting data concurrently,collisions may occur during the random access process,requiring conflict resolution.Integrating the process of random access into multipleaccess transmission eliminates the need for separate access and transmissionprocesses,leading to improved transmission e

251、fficiency and increased connectiondensity.Typical technologies include Integration of Random Access and MultipleAccess Transmission,Efficient Connectionless Transmission,Pattern DivisionRandomAccess,and Sparse IDMAfor UMA.(4)Receiver design:It is essential to minimize receiver complexity whilemainta

252、ining performance to accommodate the substantial terminal access demand.The receiver design should exhibit robust multi-user interference suppression41/97capabilities to elevate system performance and meet the demand for terminalconnections.Multi-antenna technology is employed to improve systemperfo

253、rmance and capacity and increase the number of connections.MassiveMIMO technology is leveraged to facilitate the access of a huge number of terminals.TypicaltechnologiesincludeIterativeReceiverBasedonSparsificationTransformation and Capacity-Optimized and Low-Complexity Iterative Receiver andMulti-U

254、ser Encoding Scheme.Table 3-2 Key Technologies for Multiple AccessTechnology NameOverviewIntegration ofRandomAccess andMultipleAccessTransmissionIntegrate the initial access and multiple access transmissionprocesseswithoutrequiringnetworkcoordination.Itemploys a non-orthogonal approach to simultaneo

255、uslyaccomplishrandomaccessandmultipleaccesstransmission.By meeting the error performance requirements for eachterminal and implementing a unified encoding process,itenables extensive resource sharing among terminals.Thismethod eliminates the need for network coordination,conserves network signaling

256、resources,removes constraintson resources based on the number of terminals,and cansupport the access and transmission of a massive number ofterminals.EfficientConnectionlessTransmissionDeeply sleeping nodes can transmit data directly without aprior connection establishment.Upon completing the datatr

257、ansmission,they enter deep sleep without the need torelease the connection.Through the use of data-only,independent multi-pilot,andsparsepilottechnologies,itaimstominimize pilotcollisions,enablingstraightforwardandefficient42/97connectionless heavy-load transmission.Multi-UserEncoding andDecoding Sc

258、hemesBased on ODMAODMAschemes operate independently of repetition(spreadspectrum)and user interleaving.It causes the decodingfactor graph to become ultra-sparse within a very shortperiod by substituting vacancies for repetitions.This enables a notable enhancement in multi-user decodingperformance,wh

259、ile also significantly reducing decodingcomplexity.MultipleAccessScheme Based onVirtual UserSplittingThe combination of rate splitting,NOMA,and SDMAschemes allows for the treatment of certain inter-beaminterference signals as useful signals,while also leveragingthe power differences among multiple u

260、sers within thebeam to effectively cancel inter-beam interference,resultingin performance gains.The multi-beam transmission system of the High AltitudePlatform Station(HAPS)enables a greater number ofconnections and improved SE performance.MultipleAccessScheme Based onResource HoppingIt involves seg

261、menting data packets within a single timeslot,and performing segment encoding to create redundantdata segments.It maps encoded data segments onto distinctchannel resources,generating a multitude of hoppingpatterns as signatures to represent the transmission signalsof a vast number of users.By levera

262、ging redundantsegments,it addresses partial collisions among users.Thereceivercombineschip-levelserialinterferencecancellation(SIC)to enhance collision resolution capabilityat the expense of complexity.Pattern DivisionRandomAccessIt utilizes a pattern superposition pattern-domain pilot,expanding the

263、 pilot competition space to the pattern domain43/97withoutrequiringadditionalphysicalresources.Thisapproach allows the receiver to achieve partial collisionjudgment with low complexity,thereby lowering the pilotcollision probability and enhancing the access success rate.Iterative ReceiverBased onSpa

264、rsificationTransformationItutilizesareceiverframeworkthatemploysaquasi-message passing algorithm(MPA)and features aninformation-preservingfactorgraphSparsificationTransformation(ST).It effectively removes numerous shortcycles within the factor graph,reduces the node degree inthesystem,suppressesinte

265、r-userinterference,anddiminishes the iterative computational complexity of the EPalgorithm.Capacity-OptimizedandLow-ComplexityIterative Receiverand Multi-UserEncoding SchemeOptimizedreceiversincludeMU-OAMP/VAMP,MU-MAMP,aswellasoptimalreceiversbasedonmulti-user encoding principles and constrained cap

266、acity.Sparse IDMAforUMAItintegratescompressedsensingandsparseIDMAsuperimposedcoding,constructingacodebookusinginterleaving patterns,bit repetition counts,and zero-fillnumbers.After being mapped through compressed sensinginto pilots,the codebook is transmitted before the data.Itenhances resistance to

267、 multi-user interference through bitrepetition,diminishesmulti-userinterferencebyincorporating zero elements,distinguishes users via diverseinterleavers,and randomizes multi-user interference.44/973.4 Code Modulation3.4.1 TypicalApplication ScenariosOTFS application scenario 1:high-speed railway sce

268、narioFor railways,continuously improving train speeds is a common goal in globalrailway development.At present,the Beijing-Shanghai high-speed railway hasachievedatestspeedof470kilometersperhour,andtheCR450,a450-kilometer-per-hour high-speed train,will be completed in 2024.At the same time,the Centr

269、al Japan Railway Company has achieved a test speed of 603 kilometers perhour for maglev trains in Yamanashi-ken,Japan.In addition,the pipeline flying car,which can reach speeds of over 1000 kilometers per hour,is also under development.Based on the high-speed development of railways,developed high-s

270、peed railwaycountries worldwide are focusing on the intelligence of high-speed railways.Theintelligence development of high-speed railways requires advanced communicationsystems and standards to provide support,but the high-speed movement of trains inhigh-speed railway scenarios will pose a great ch

271、allenge to the reliability ofground-to-train and train-to-train communication.OTFS application scenario 2:low-earth orbit satellite scenarioLow-earth orbit(LEO)satellite communication is a technology that usessatellites in low-earth orbit to achieve communication.Unlike traditional high-earthorbit s

272、atellite communication,LEO satellite communication satellites are typicallylocated between hundreds of kilometers and two thousand kilometers from the ground.Compared with traditional high-earth orbit satellites,it has the advantages of lowlaunch cost,low communication delay,low transmission loss,an

273、d seamless globalcoverage after networking,and has attracted the attention of many Internet,communication,and aerospace companies around the world.OTFS application scenario 3:air coverage scenarioWith the progress of aviation communication,airplanes are transforming fromthe past isolated islands of

274、information networks into key carriers for realizingglobal interconnection.The emergence of in-flight WiFi allows passengers to access45/97the Internet on airplanes.However,the arrival of the 5G era has broughtunprecedented challenges to air communicationthe demand for massive real-timeInternet data

275、 transmission.This challenge requires communication systems to behighly adaptive,and capable of improving communication quality between aircraftand ground stations or satellites in high-speed mobile environments,ensuringlow-latency and high-reliability transmission of internet data.OTFS application

276、scenario 4:IoVBased on the OTFS-ISAC mechanism,the following IoV functions orapplicationscanbesupported:Accuratelysensingthesurroundingdrivingenvironment,including vehicles,obstacles,road conditions,etc.,to enhance drivingsafety and achieve intelligent driving;accurately sensing the positions and mo

277、tionstates of both receivers and transmitters,providing prior information for channelestimation,beamforming,etc.,to improve communication performance;distributednode collaborative sensing,expanding the range of node sensing,and enhancing theaccuracy and precision of sensing.OTFS application scenario

278、 5:underwater acoustic communicationThe Smart Ocean project is a major project related to the national strategy ofbuilding a maritime power,and with the advancement of the maritime power and theconstruction of the Smart Ocean project,rapid development has been achieved invarious fields such as moder

279、n fisheries,marine observation and monitoring,offshoreoil and gas exploration and development,and marine transportation.Underwateracoustic communication is an important part of the marine communication network.Acousticwavesarecurrentlytheonlyeffectivelong-distanceinformationtransmission carrier unde

280、rwater.However,underwater acoustic(UWA)channelsencounter numerous challenges,including limited bandwidth,substantial delay spread,frequency-selective fading,and vulnerability to Doppler effects.3.4.2 Key Technologies and ChallengesDelay Doppler domain modulation technology for high-speed mobile scen

281、ariosBasic principles of OTFS46/97Orthogonalfrequencydivisionmultiplexing(OFDM)isawidelyusedmodulation technique in wireless communication systems such as 4G,5G,and WiFi.OFDM based on cyclic prefixes can effectively deal with multipath fading and onlyrequires low-complexity frequency domain equalize

282、rs.With the development ofwireless communications,high-speed mobile communication scenarios in complexscattering environments are becoming increasingly abundant,such as IoV,high-speedrailways,and LEO satellite communications.However,for high-speed mobilescenarios,OFDM under high-speed mobility will

283、lose subcarrier orthogonality due tothe influence of Doppler spread,and its transmission reliability will deteriorate.Therefore,the ICDT-integrated wireless air interface requires the development ofinnovative multi-carrier modulation schemes specifically tailored for high-speedmobilescenarios.Inrece

284、ntyears,researchershaveintroducedmodulationtechnologies 24 in the transform domain,such as Orthogonal Time Frequency Space(OTFS),with its fundamental principle diagram illustrated in Figure 3-8.Figure 3-8 OTFS System ModelAdvantages and disadvantages of OTFSAdvantagesRegarding the basic principles o

285、f OTFS,OTFS boasts the following advantagesover OFDM:1OTFS users map data symbols to the Delay Doppler(DD)domain instead of thetraditional Time-Frequency(TF)domain.This allows user data symbols to bespread over all time-frequency domains using unitary transforms such as theSymplectic Finite Fourier

286、Transform(SFFT),achieving full diversity oftime-frequency domain channels and improving transmission reliability.2The channel spreading function in the DD domain can reflect the specificscattering environment.Channel taps with different delays and Dopplers can47/97correspond to different mobile scat

287、terers.Compared with the channel impulseresponse in the time latency domain and the channel transfer function in thetime-frequency domain,the channel spreading function in the DD domain haspotential compactness,sparsity,and stability in high-speed mobile scenarios,which can reduce the signaling over

288、head of physical layer adaptive schemes25.3Delay and Doppler frequency shifts can reflect the distance and velocity ofobjects in the physical world,making OTFS waveforms particularly suitable forintegrating communication and sensing.DisadvantagesHowever,in comparison to the long-standing and widely

289、used OFDM technology,OTFS technology still encounters numerous unresolved issues from its theoreticalfoundations to engineering applications.The current predominant research on OTFS revolves around system designsusing rectangular waveforms,potentially leading to out-of-band leakage issuessimilar to

290、those of OFDM,and facing an error floor when subjected toband-limited filters.Designing the reference signal for OTFS involves considering PAPR,channelestimation algorithms,data detection algorithms,and multiple access methods tostrike a balance between overhead,performance,and complexity.The curren

291、t scheme for integrating OTFS and MIMO fails to fully leverage thepotential advantages of multi-antenna systems.Additionally,a rise in the numberof antennas leads to a marked increase in receiver complexity.Consequently,OTFS technology necessitates adaptable support for diverse multi-antennatechnolo

292、gies.Modeling channels for high-speed scenarios requires significant further researchand development.In terms of standardization,the design of OTFS-based systems needs to becompatible with the OFDM technology in existing communication standardswhile addressing the issues of high-speed mobile scenari

293、os,aiming to maintain48/97its flexibility and inherent advantages as much as possible.OTFS related researchTo address the above challenges,researchers have proposed the OrthogonalDelay Doppler Multiplexing(ODDM)scheme 26 and Zak-OTFS scheme 27for waveform design.The ODDM scheme proposes orthogonal w

294、aveforms forODDM signals,addressing both Doppler and delay resolution.Signals can bemodulated across multiple carriers in the Delay Doppler domain and transmittedas time-domain signals.In the field of channel estimation and demodulation,researchers have proposed areference signal design scheme based

295、 on sequence pilot to reduce PAPR,anoff-grid channel estimation scheme based on sparse Bayesian estimation toenhance estimation performance and reduce pilot overhead 28,as well aslow-complexity detection algorithms such as Expectation Propagation(EP)29,Memory Approximate Message Passing(MAMP)30,and

296、low-complexity datadetection policies for large-scale antenna arrays 31,with detection complexityreaching linear order of packet size.In channel modeling,researchers have conducted characterization of measuredchannels in high-speed railway scenarios and pointed out that the time variationof the chan

297、nel spreading function is an important factor to consider in OTFSdesign 32.In channel estimation,researchers have proposed a channel estimation schemebased on sequence pilots to reduce PAPR and an off-grid channel estimationscheme based on sparse Bayesian estimation to lower pilot overhead 28.In the

298、 multiple access schemes,researchers have introduced orthogonal codedomain resources to design an ideal channel spreading function in high-speedrailway scenarios 33 and a grant-free access scheme for weakly compactchannel spreading functions 34.Additionally,they have researched jointpositioning and

299、multiple access schemes 35.Some researchers have alsoproposed the OTFS-SCMA scheme based on user codebooks and signal sparsity36,as well as research on power-domain OTFS-NOMA37.49/97As the research on OTFS progresses,the OTFS scheme tailored for engineeringimplementation is becoming increasingly mat

300、ure.Due to its superior reliabletransmission performance compared to OFDM schemes,the OTFS scheme isexpected to be applied in the next-generation wireless communication systems inhigh-speed mobile scenarios.3.5 Ultra Massive MIMO3.5.1 TypicalApplication ScenariosUltra massive MIMO are primarily depl

301、oyed to enhance spectral efficiency,transmission reliability,coverage extension,and interference suppression in mobilecommunication systems.In the upcoming 6G systems,ultra massive MIMO willcontinue to play crucial role in these areas.Regarding application scenarios,apartfrom diverse uses for covera

302、ge extension,the high spatial resolution of ultra massiveMIMO can also enhance positioning and sensing accuracy in services such as locationtracking and sensing.Ultra massive MIMO are primarily used for coverage extension in the followingapplication scenarios:Wide area coverage:6G supports mobile co

303、verage characterized byultra-large coverage radius and three-dimensional coverage.Ultra-largecoverage radius mainly includes seemingly uninhabited areas such as sea,desert,and forest.Ultra massive MIMO can extend the coverage distance ofa single base station as much as possible,effectively solving t

304、he problem ofwide area coverage.Three-dimensional coverage mainly emphasizes spatialcoverage in the vertical direction,including low-altitude coverage such asUAVs and high-rise buildings,and high-altitude coverage such as civilianaircraft.It can meet the three-dimensional coverage requirements of 6G

305、 byadopting new antenna structures and increasing vertical direction freedom38.50/97Macro cell coverage:Macro cell is a traditional coverage scenario of mobilecommunication systems.It provides high date rate transmission forhigh-user-density and densely populated urban areas through dense cells,indo

306、or to outdoor coverage,etc.The ultra massive MIMO of 6G can increasethe area traffic of macro cell coverage,reduce inter-cell interference,andachieve coverage over longer distances 39.Hotspot coverage:Hotspot has been a traditional coverage scenario since theintroduction of 4G,focusing on high peak

307、data-rate transmission and limitedarea coverage.Hotspots including scenarios such as office buildings,medium to large venues,transportation hubs,and enterprise campuses,characterized by complex wireless environments and high population density.They demand ultra-dense networking technology.However,th

308、e rise in sitedensity leads to heightened inter-cell interference,requiring the integration ofbeamforming technology with Ultra massive MIMO to mitigate interference39.Ultra-large-scale antennas can also be applied to positioning,sensing,and otherservices:The 6G system demands greater precision in p

309、ositioning.Integrated sensingand communication is a key focus of 6G technology.The high spatialresolution capability of massive MIMO antennas enhances the accuracy ofdetection for positioning and sensing.Furthermore,the integration ofmassive MIMO antenna technology into fundamental system operations

310、,including channel state information acquisition,channel detection,and beammanagement,enables a more efficient design for integrated communication,positioning,and sensing 40.3.5.2 Key Technologies and ChallengesTo meet the future requirements of 6G communication systems for spectralefficiency,data r

311、ates,and emerging services and applications,the advancement ofmulti-antenna technology will encompass virtual MIMO,centralized ultra massive51/97MIMO enhancement,distributed MIMO enhancement,novel antenna architecture,and the advancement of multi-antenna technology into new dimensions.In thefollowin

312、g sections,we will delve into these pivotal technologies and their associatedchallenges.Pseudo MIMO technologyThis innovation requires the transformation of the receiver to integrate ahigh-speed reconfigurable antenna system.By quickly reconfiguring the receiveantenna pattern and upsampling the rece

313、ived OFDM symbols,this technology enablesthe transmission of concurrent data streams exceeding the number of RF chains 65.Due to the performance and operation of this technology being similar to asystem with additional“virtual”RF chains,this technology is named Pseudo MIMO.Unlike traditional hybrid

314、analog-digital MIMO systems,pseudo MIMO technologyachievesgreaterspectralefficiencywhilemaintainingcomparablehardwarecomplexity and power consumption.On the other hand,compared with fully digitalMIMO systems,pseudo MIMO systems can attain comparable spectral efficiencywhile markedly decreasing hardw

315、are cost and power consumption.PseudoMIMO technologypresentsasignificantchallengedue to therequirement for rapid reconfiguration of antenna feature patterns at the receiver withina single OFDM sampling interval,necessitating high-performance antenna featureswitching devices.The design of these high-

316、speed switching devices represents afundamental hurdle in the practical implementation of pseudo MIMO technology.Inaddition,due to the transformation of the receiver by pseudo MIMO technology,it isnecessary to design a completely new channel estimation and synchronization methodfor pseudo MIMO techn

317、ology.Centralized ultra massive MIMO enhancementAccording to the theory of massive MIMO,as the size of antenna arraysincreases,the channel vectors of individual users tend to become orthogonal,demonstrating what is known as channel hardening characteristics.Thissuggests the elimination of interferen

318、ce regions among co-scheduled users52/97within the system.Additionally,system-level simulations indicate thatincreasing the number of antennas and RF channels at base stations cansignificantly improve average spectral efficiency,edge spectral efficiency,transmission reliability,and coverage capacity

319、.However,the increase in the number of antennas and RF channels posessignificant challenges for implementing and deploying communicationsystems.Specifically,the volume,weight,and frontal area of antenna arrays,as well as computational complexity,power consumption,and CSI feedbackoverhead,all experie

320、nce a substantial increase.Therefore,specific schemesneed to be developed,such as distributed MIMO,novel antenna arrays andarchitectures,andexpansionintonewdimensionsofmulti-antennatechnology,to address the preceding challenges faced by the centralizedultra-massive MIMO system.Distributed MIMO enhan

321、cementThe distributed deployment of massive MIMO systems offers significantadvantages.Specifically,it can expand the scale of the equivalent antennaarray,enhance signal angular spread,and reduce the correlation of theequivalent MIMO channels,resulting in a greater spatial degree of freedomand improv

322、ed channel capacity.Specifically,advanced signal collaborativeprocessing improves beam coverage and enhances spectral efficiency.Furthermore,adding distributed nodes shortens signal propagation distancesand increases energy efficiency.The spatial redundancy from multi-sitecooperationenhancestransmis

323、sionreliability.Finally,distributedmulti-antenna technology provides a robust air interface transmissionfoundation for user-centric access networks.However,distributed MIMO encounters various technical challenges.Theseinclude the need for research into efficient transmission schemes,signalprocessing

324、 algorithms,and the rational allocation of computational tasksacross different network nodes.Additionally,there is a requirement to53/97achieve time-frequency synchronization and antenna calibration betweensites to fulfill the demands of coherent joint transmission.An elastic andscalable architectur

325、e is essential to accommodate changes in networkservices without necessitating a complete network overhaul.Efficientmobilitymanagementisalsocrucialtosupporttheestablishment,modification,addition,and removal of physical cooperative clusters withindedicated service areas.Furthermore,cost-effective rap

326、id deploymentsolutions are necessary to enable the plug-and-play functionality fordistributed nodes.New antenna array and architecture designDespite utilizing a digital-analog hybrid structure,the continued scaling ofactive antenna arrays remains constrained by factors such as cost,powerconsumption,

327、and equipment complexity.The innovation of antenna arraytechnology will be an important breakthrough for effectively expanding thescale of the antenna array.Given the potential of reconfigurable intelligentsurface technology for extending coverage and transforming the propagationenvironment,it is po

328、ssible to control a large number of low-cost,low-powercontrollablereflector/transmission/leakageelementswithinthereconfigurable metasurface in multi-antenna systems,which enables theachievement of massive antenna beamforming effects.For reconfigurablemetasurfaces,ReconfigurableIntelligentSurface(RIS

329、)41andReconfigurable Holographic Surface(RHS)42 are currently the mostrepresentative types of new antenna structures.As shown in Figure 3-9,the RIS antenna array consists of a large numberof densely arranged metamaterial reflection units and external feedsources.Thisconfigurationmitigatestheextensiv

330、euseofhigh-energy-consuming active devices,thereby facilitating cost reductionand lower power consumption.Additionally,it enables a substantialimprovement in spatial dimensions within multi-antenna systems.54/97Figure 3-9 RIS Antenna Array PrototypeThe RHS antenna array is composed of internally emb

331、edded feed sourcesand numerous densely arranged sub-wavelength metamaterial radiatingelements,as shown in Figure.The feed sources inject electromagneticwaves carrying transmission information(analogous to the referencewave in holographic imaging technology)into RHS,sequentiallyexciting the radiating

332、 elements of the RHS array.This causes theradiating elements to emit the modulated reference waves outward in acontrolled manner.By adjusting the impedance matching level of eachelement through control methods,the amplitude of the radiatedelectromagnetic waves is altered to achieve holographic beamf

333、orming.(a)Principle of RHS Antenna Array Structure(b)Physical Structure of RHS AntennaArrayFigure 3-10 RHS Antenna ArrayDespite thorough research and validation in academia and industry,metasurfaces still face several significant challenges.In particular,when55/97utilizing metasurfaces for signal transmission and beamforming,it is vital toaddress the channels in two segments:from the feed source t