1、1/33Preliminary Study of Advanced Technologiestowards 6G Era:QITs20212/33Executive SummaryWith the large-scale commercialization of 5G in 2021,the global industry has witnessed astarting of exploration and research on the 6th generation(6G)communication systems.6G willbuild a new type of network tha

2、t is intelligently and efficiently interconnected between humans,machine and things.On the basis of greatly improving the network capability,it has new functionssuch as endogenous intelligence,multi-dimensional perception,digital twin,endogenous networksecurity and so on.With the in-depth research o

3、n 6G network and key technologies,its integrationand application with Quantum Information Technologies(QITs)will become the focus in thefuture.In 6G era,the importance of cybersecurity in mobile communications is expected to riseexponentially.Quantum cryptography has emerged as a potential solution

4、for safeguarding criticalinformation because it is impossible to copy data encoded in a quantum state.In the first part,thiswhite paper gives an overview of Quantum Secure Communication.Starting with enablingtechnologies of quantum key distribution(QKD),standardization activities for QKD and itsnetw

5、orking technologies are presented,followed by implications of QKD for 6G.In particular,two typical applications scenarios are introduced.One is the quantum encryption system that willbe applied to the construction of Winter Olympics Smart Park and Xiongan New Area.The otheris in Xiongan quantum comm

6、unication pilot,where a quantum communication trunk linebetween Beijing and Xiongan will be deployed,and a quantum key distribution platform will beintroduced to provide security keys for customers in the fields of Internet of things,Internet ofvehicles,smart energy,smart government and so on.The pr

7、ovisions of a many-fold increase in the 6G communication system performancealongside with rich diversity of innovative services call for a revolutionary promotion ininformation processing capability.In this regard,the emerging Quantum Machine Learning(QML)has attracted significant attention due to i

8、ts information processing paradigm by combining theestablished benefits of quantum mechanism and machine learning.In the second part,followed bypreliminary knowledges of machine learning(ML)basic paradigms and their application insolving problem across different layers of in communication systems,an

9、d quantum tools,thiswhite paper presents examples to get insight into the research of QML.3/33Table of ContentsExecutive Summary.21 Introduction.42 Quantum Secure Communication.62.1 Enabling Technologies for Quantum Secure Communication.62.1.1 Overall Picture.62.1.2 Types of QKD.62.1.3 The Needed Op

10、toelectronic Components of QKD and the Low-CostImplementation.82.2 Standardization Activities for QKDN.102.2.1 ITU-T.112.2.1.1 ITU-T Study Group 11.112.2.1.2 ITU-T Study Group 13.112.2.1.3 ITU-T Study Group 17.142.2.2 ETSI ISG-QKD.162.2.3 ISO/IEC JTC 1/SC 27.172.3 Implications for 6G.182.3.1 State-o

11、f-the-art of QKD in 5G.182.3.2 Integration of 6G and QITs.192.3.3 Typical Application Scenarios of QKD.203 Quantum Machine Learning(QML).233.1 Machine Learning for Communication Systems.243.2 Quantum Tools.263.3 QML for Communication Systems.273.3.1 Quantum-enhanced Machine Learning.273.3.2 Machine

12、Learning of Quantum Systems.283.3.3 Quantum Learning Theory.284 Reference.30Acknowledgement.32Abbreviation.324/331IntroductionThe scope of this annually revised white paper is to introduce quantuminformation technologies(QITs)with the aim of taking advantages of their powerfulinformation processing

13、capabilities to fulfil stringent demands of communication andcomputing envisaged by 6G systems.Our previous version in 2020 present theoverview of QITs from the perspectives of QITs&Quantum Internet and QTIs forClassical Signal Processing,respectively.The version of 2021 will further introducefrom t

14、wo benefits expected from QITs to communication systems,i.e.,securecommunication and enhanced information processing capability.Chapter 2.Quantum Secure CommunicationIn 6G era,the importance of cybersecurity in mobile communications is expected to riseexponentially.Quantum cryptography has emerged a

15、s a potential solution for safeguarding criticalinformation because it is impossible to copy data encoded in a quantum state.Chapter 2 gives anoverview of Quantum Secure Communication.Starting with enabling technologies of quantumkey distribution(QKD),standardization activities for QKD and its netwo

16、rking technologies arepresented,followed by implications of QKD for 6G.In particular,two typical applicationsscenarios are introduced as deploying quantum encryption system and deploying quantumcommunication trunk line in providing security keys for customers in the fields of Internet ofthings,Inter

17、net of vehicles,smart energy,smart government and so on.Chapter 3.Quantum Machine Learning(QML)The provisions of a many-fold increase in the 6G communication system performancealongside with rich diversity of innovative services call for a revolutionary promotion ininformation processing capability.

18、In this regard,the emerging QML has attracted significantattention due to its information processing paradigm by combining the established benefits ofquantum mechanism and machine learning.Chapter 3 starts with the concepts of QML on a highlevel and then discusses machine learning(ML)basic paradigms

19、 and their application in solving5/33problem across different layers of in communication systems.Followed by preliminaryknowledges of quantum tools,Chapter 3 presents examples to get insight into the research ofQML.Consequently,QMLforcommunicationsystemscanbeobtainedbyMLforcommunication system being

20、 synergy with quantum speedup.6/332Quantum Secure Communication2.1Enabling Technologies for Quantum Secure Communication2.1.1Overall PictureQuantum secure communication means combing the secret key generated from quantum keydistribution(QKD)device with existing symmetric encryptor.The distribution p

21、rocess of thesecret key is guaranteed by law of quantum mechanics.Figure 2.1 Quantum secure communication,a system view2.1.2Types of QKDThe types of QKD can be categorized by the behavior of transmitter and receiver,also theusage of physical degree of freedom.Based on the behavior of transmitter and

22、 receiver,the QKDtypesareprepare-and-measure,twotransmitterstoonecommonreceiver(Measurement-device-independent(MDI)QKD,twin-field(TF)QKD),onecommonentanglement-based transmitter to two receivers(Entanglement based QKD),shown in Figure 2.2.7/33Figure 2.2 Types of QKD in terms of the behavior of Tx an

23、d RxThe prepare-and-measurement QKD is most commercially matured one and it can be furtherdivided into two types:DV-QKD and CV-QKD,as shown in Table 2-1.Table 2-1 DV-QKD and CV-QKD,a comparisonDiscrete Variable QKD(DV-QKD)Continuous Variable QKD(CV-QKD)Maximum Baud rate at 1.25Ghz forproductMaximum

24、Baud rate at 10Ghz recordBased on single photon detectionDegree of freedom:polarization,timebin+phase,frequencyDark fiber preferred,good at high losschannelCo-existencewithdatacommunicationpossible,lowtolerance.Relatively simple post-processingRecordfromUniv.Geneva:6.5bps69.3dBMaximum Baud rate no m

25、ore than100Mhz for productMaximum Baud rate around 1GhzrecordBased on coherent detectionDegreeoffreedom:In-phasecomponent and quadrature of EM fieldDark fiber is not a must,good at lowloss channelCo-existencewithdatacommunicationpossible,hightoleranceComplex post-processingRecordfromBUPT&PKU:6.2bps3

26、2.45 dB8/33Phys.Rev.Lett.121,190502(2018)Phys.Rev.Lett.125,010502(2020)2.1.3The Needed Optoelectronic Components of QKD and the Low-CostImplementationIn Figure 2.3,The three typical QKD systems using photons physical degree of freedom arelisted:DV-QKD Polarization,DV-QKD Time-Phase,CV-QKD Transmitte

27、d Local Oscillator(TLO).The source of high cost comes from the usage of Lithium niobate modulator,electricpolarization controller,fiber-based Asymmetric MZI and single photon detector.Thanks to therapid progress of silicon photonic chip and III-V material photonic chip development recent years,the Q

28、KD can benefit from low-cost device.In Figure 2.4,an example is shown how the traditionway of modulating the intensity and polarization of the quantum signal carrier can be shrink into asmall device.Figure 2.3 Cost issue with QKD system9/33Figure 2.4 Shrink the size of tradition component into a sma

29、ll device for QKDWithcompactsiliconphotonicschipandIII-Vcomponents(Figure2.5)andapplication-specific integrated circuit(ASICs),the full optoelectronic functions can be packagedinto a standard C form-factor pluggable(CFP)size module that is widely used in traditionaloptical communication industry,whi

30、ch implies standard and cost-effective QKD Tx module andQKD Rx module are feasible in the near future.Then the QKD functions can be realized via CFPQKD module with on-board computation electronics,this will benefit the implementation ofquantum secure communication system in terms of size,cost and fl

31、exibility(Figure 2.6).Figure 2.5 Compact III-V material based single photon detector10/33Figure 2.6 The future of standard CFP QKD module2.2Standardization Activities for QKDNQKD and its networking technologies have attracted a lot of interest in multiple SDOs,e.g.,ISO,IEC,ITU,IEEE,IETF,ETSI,as show

32、n in 2.7.The status of Quantum Key DistributionNetworks(QKDN)standardization in different SDOs will be briefly reviewed in the followingsub-clauses.11/33Figure 2.7 QKDN standardization timeline2.2.1ITU-TITU-T is the first SDO to standardize QKD as a network since 2018.At the time of thisreports publ

33、ication,ITU-T Study Groups 13 and 17 had cumulatively initiated 18 work items onthe network and security and aspects of QKD networks,respectively.2.2.1.1ITU-T Study Group 11At the time of this reports publication,SG11 had initiated 1 work items on QKDN for study,as listed in Table 2-2.Table 2-2:QKD

34、related work items in ITU-T SG11QReferenceTitleTypeStatusQ2/11Q.QKDN_profrQuantum key distributionnetworks ProtocolframeworkRecommendationUnderdevelopment2.2.1.2ITU-T Study Group 13At the time of this reports publication,SG13 had adopted 5 standards on QKDN,including12/33the QKDN overview(Y.3800),fu

35、nctional requirements(Y.3801),functional architecture(Y.3802),key management(Y.3803),control and management(Y.3804)and initiated 17 workitems on QKDN for study,as listed in Table 2-3.Table 2-3:QKD related work items in ITU-T SG13QReferenceTitleTypeStatusQ16/13Y.3800Overview onnetworks supportingquan

36、tum keydistributionRecommendationPublished(2019-11)Q16/13Y.3801Functionalrequirements forquantum keydistribution networkRecommendationPublished(2020-07)Q16/13Y.3802Quantum keydistribution networks-FunctionalarchitectureRecommendationPublished(2021-04)Q16/13Y.3803Quantum keydistribution networks-Key

37、managementRecommendationPublished(2021-03)Q16/13Y.3804Quantum KeyDistributionNetworks-Controland ManagementRecommendationPublished(2021-01)Q16/13Y.3805Quantum KeyDistributionNetworks-SoftwareDefined NetworkingControlRecommendationUnderdevelopmentQ6/13Y.3806Requirements forQoS Assurance ofthe Quantum

38、 KeyDistribution NetworkRecommendationUnderdevelopmentQ16/13Y.Sup70ITU-T Y.3800-series-Quantum keydistribution networksSupplementPublished(2021-09)13/33QReferenceTitleTypeStatus-Applications ofmachine learningQ16/13Y.QKDN_BMQuantum KeyDistributionNetworks-Businessrole-based modelsRecommendationUnder

39、developmentQ16/13Y.QKDN_frintFramework forintegration of QKDNand secure storagenetworkRecommendationUnderdevelopmentQ16/13Y.QKDN-iwfrQuantum keydistribution networks-interworkingframeworkRecommendationUnderdevelopmentQ16/13Y.QKDN-ml-fraQuantum KeyDistributionNetworks-Functionalrequirements andarchit

40、ecture formachine learningRecommendationUnderdevelopmentQ6/13Y.QKDN-qos-faFunctionalarchitecture of QoSassurance forquantum keydistribution networksRecommendationUnderdevelopmentQ6/13Y.QKDN-qos-genGeneral Aspects ofQoS(Quality ofService)on theQuantum KeyDistribution NetworkRecommendationUnderdevelop

41、mentQ6/13Y.QKDN-qos-ml-reqRequirements ofmachine learningbased QoS Assurancefor quantum keydistribution networksRecommendationUnderdevelopmentQ16/13Y.QKDN-rsfrQuantum keyRecommendationUnder14/33QReferenceTitleTypeStatusdistribution networks-resilienceframeworkdevelopmentQ16/13Y.supp.QKDN-roadmapStan

42、dardizationroadmap on QuantumKey DistributionNetworksSupplementUnderdevelopmentThe structure of work on QKDN standardization in SG13 is illustrated in Figure 2.8.Figure 2.8:QKDN standardization work items in SG132.2.1.3ITU-T Study Group 17SG17 established a new Question,Q15/17,Security for/by emergi

43、ng technologies includingquantum-based security,approved by TSAGs September 2020 meeting.The Q15/17 terms ofreference are available at 1.At the time of this reports publication,SG17 had adopted 3 standards on QKDN and QRNG,including QKDN security framework(X.1710),key combination and confidential ke

44、y supply15/33(X.1714)and QRNG architecture(X.1702),and initiated 10 work items on QKDN for study,aslisted in Table 2-4.Table 2-4:QKD related work items in ITU-T SG17ReferenceTitleTypeStatusX.1702Quantum noise random numbergenerator architectureRecommendationPublished(2019-11)X.1710Security framework

45、 for quantumkey distribution networksRecommendationPublished(2020-10)X.1714Key combination and confidentialkey supply for quantum keydistribution networksRecommendationPublished(2020-10)XSTR-SEC-QKDSecurity considerations for quantumkey distribution networkTechnical ReportPublished(2020-03)X.1712Sec

46、urity requirements and measuresfor QKD networks-keymanagementRecommendationUnderdevelopmentX.sec_QKDN_AAAuthentication and authorization inQKDN using quantum safecryptographyRecommendationUnderdevelopmentX.sec_QKDN_CMSecurity requirements and measuresfor quantum key distributionnetworks-control and

47、managementRecommendationUnderdevelopmentX.sec_QKDN_intrqSecurity requirements forintegration of QKDN and securenetwork infrastructuresRecommendationUnderdevelopmentX.sec_QKDN_tnSecurity requirements for QuantumKey Distribution Networks-trustednodeRecommendationUnderdevelopmentTR.hybsec-qkdnTechnical

48、 Report:Overview ofhybrid security approachesapplicable to QKDTechnical ReportUnderdevelopmentThe structure of work on QKDN standardization in SG17 is illustrated in Figure 2.9.16/33Figure 2.9:QKDN standardization work items in SG172.2.2ETSI ISG-QKDETSI initiated the industry specification group(ISG

49、)on QKD in 2008.ETSI ISG-QKD haspublished nine specifications on QKD until 2019 and have several work items ongoing as listed inTable 2-5.The previous work mainly focused on QKD link-level issues,including QKD opticalcomponents,modules,internal and application interfaces,practical security,etc.Note

50、that ETSIhas also initiated the study of QKD network architectures recently and the specification of QKDsecurity certification based on common criteria.Table 2-5:QKD related work items in ETSIReferenceTitleStatusGS QKD 002Quantum Key Distribution(QKD);UseCasesPublished(2010-06)GR QKD 003Quantum Key

51、Distribution(QKD);Components and Internal InterfacesPublished(2018-03)GS QKD 004Quantum Key Distribution(QKD);Application InterfacePublished(2010-12)GS QKD 005Quantum Key Distribution(QKD);Security ProofsNOTE Revision in progressPublished(2010-12)GR QKD 007Quantum Key Distribution(QKD);VocabularyPub

52、lished(2018-12)17/33ReferenceTitleStatusNOTE Revision in progressGS QKD 008Quantum Key Distribution(QKD);QKDModule Security SpecificationPublished(2010-12)GS QKD 011Quantum Key Distribution(QKD);Component characterization:characterizing optical components forQKD systemsPublished(2016-05)GS QKD 012Qu

53、antum Key Distribution(QKD)Deviceand Communication Channel Parametersfor QKD DeploymentPublished(2019-02)GS QKD 014Quantum Key Distribution(QKD);Protocol and data format of key deliveryAPI to Applications;Published(2019-02)GS QKD 015Quantum Key Distribution(QKD);Quantum Key Distribution ControlInter

54、face for Software Defined NetworksPublished(2021-03)DGS/QKD-0010_ISTrojanQuantum Key Distribution(QKD);Implementation security:protectionagainst Trojan horse attacks in one-wayQKD systemsUnderdevelopmentDGS/QKD-0013_TransModCharQuantum Key Distribution(QKD);Characterisation of Optical Output ofQKD t

55、ransmitter modulesUnderdevelopmentDGS/QKD-016-PPQuantum Key Distribution(QKD);Common Criteria Protection Profile forQKDUnderdevelopmentDGR/QKD-017NwkArchQuantum Key Distribution(QKD);Network architecturesUnderdevelopmentDGS/QKD-018OrchIntSDNQuantum Key Distribution(QKD);Orchestration Interface of So

56、ftwareDefined NetworksUnderdevelopment2.2.3ISO/IEC JTC 1/SC 27ISO/IEC JTC 1/SC 27 initiated the study period Security requirements,test and evaluation18/33methods for quantum key distribution in 2017.In 2019,the study period was completed,and anew work item ISO/IEC 23837(Part 1&2)was established as

57、listed in Table 2-6.Table 2-6:QKD related works items in ISO/IEC JTC1ReferenceTitleStatusISO/IEC23837-1Security requirements,test and evaluation methods forquantum key distribution Part 1:requirementsUnderdevelopmentISO/IEC23837-2Security requirements,test and evaluation methods forquantum key distr

58、ibution Part 2:test and evaluationmethodsUnderdevelopment2.3Implications for 6G2.3.1State-of-the-art of QKD in 5GIn 5G era,the importance of cybersecurity in mobile communications will rise exponentially.Quantum cryptography has emerged as a potential solution for safeguarding critical informationbe

59、cause it is impossible to copy data encoded in a quantum state.Some mobile operators haveapplied encryption technology using QKD to 5G networks,for example,in April 2021,SKTelecom(SKT)and its subsidiary ID Quantique(IDQ),a Geneva-based leader in quantum-safecryptography,have developed a quantum virt

60、ual private network(VPN)based on the QKD.VPNis a secured communications channel implemented over shared,public networks to connectremote users and machines to a private network.QKD is a secure communication method thatimplements a cryptographic protocol involving components of quantum mechanics 2.In

61、 6G,with the development of technology,it matures day by day.In order to resist the potential impact on the classic cryptography system,256 bits algorithmswill be endorsed to replace the 128 bits algorithms.In 5G,the 128 bits algorithms NR IntegrityAlgorithm(NIA)/NR Encryption Algorithm(NEA)1/2/3 ar

62、e used for the Access Stratum(AS)and Non-Access Stratum(NAS)security protection based on the shared key,meanwhile thecorresponding 256 bits algorithms are already under investigation in 3GPP SA3 and ETSI19/33Security Algorithms Group of Experts(SAGE).The new 256 bits algorithms will probably beintro

63、duced in 6G era.AES-256 will be one of the candidates,even with currently known quantumalgorithms like Grovers,National Institute of Standards and Technology(NIST)believes thatAES 256 keys will still be safe for a very long time and recommends that current applications cancontinue to use AES with ke

64、y sizes 128,192,or 256 bits 3.For asymmetric algorithms,e.g.,Elliptic Curve-Based Certificateless Signatures forIdentity-Based Encryption(ECCSI),RSA,they are widely used in 5G system and Internetservices.NIST has initiated a process to solicit,evaluate,and standardize one or morequantum-resistant pu

65、blic-key cryptographic algorithms.It is intended that the new public-keycryptography standards will specify one or more additional unclassified,publicly disclosed digitalsignature,public-key encryption,and key-establishment algorithms that are available worldwide,and are capable of protecting sensit

66、ive government information well into the foreseeable future,including after the advent of quantum computers.It was planned to get the draft standards onPost-Quantum Cryptography(PQC)available at 2022-2024.This is the most critical issue tostandardize the most stable and secure PQC before deploying t

67、hem into the 6G.Early adoption ofpost quantum algorithms would be both very complex,and yet result in potentially uncertainsecurity guarantees.2.3.2Integration of 6G and QITsThe composition of 6G network requires high-precision data capability,computing capabilityand security,which can be enabled by

68、 quantum technologies such as quantum precisionmeasurement,quantum computing and quantum communication.(1)Quantum computing will help 6G maximize spectrum utilization and improve resourceallocation efficiency.In the 6G era,the wireless industry may re-examine the traditional spectrum allocationmecha

69、nism and further evolve the dynamic spectrum sharing technology.Through the use of20/33Artificial Intelligence,Blockchain and other technologies,more intelligent and dynamic spectrumallocation,control and scheduling can be realized to maximize spectrum utilization.Quantumcomputing will achieve optim

70、al wireless resource allocation and cell planning and improve energyefficiency and spectrum efficiency.(2)Quantum private communication technology ensures network data security andsupports the development of digital economy.Traditional cryptography based on computational complexity will face the thr

71、eat of quantumcomputer attacks in the 6G era.Enhanced cryptography such as quantum key and wirelessphysical layer key will provide a stronger security guarantee for 6G.In the future,6G networkswill rely on lightweight access authentication,quantum key,blockchain and other advancedsecurity technologi

72、es to provide active defense for network infrastructure.2.3.3TypicalApplication Scenarios of QKDQuantum encrypted communication can be applied to protect the data acquisition andprocessing system of infrastructure,ensuring the security of data communication.It can be widelyused in frontier fields su

73、ch as digital twins,smart parks,blockchains and so on.Taking the management and scheduling of the smart park as an example,collect and analyzethe environmental information of the park through sensing equipment(camera,radar),roadsideunitandpositioningreferencestation,andbuildabusinesssystembasedonveh

74、icle-road-human-cloud collaboration,which can realize the efficient and fast management ofpersonnel,materials and equipment in the park.The collected data is closely related to themanagement ability of the park,and its authenticity and integrity can be protected by quantum keydistribution.21/33Figur

75、e 2.10 Data Encryption of Smart Park Based on Quantum Security SystemData transmission with the quantum encryption system is shown in the Figure 2.10.Thequantum key distribution system provides keys for reliable authentication and data encryption ofvideo,pictures,point cloud data,traffic information

76、,location information and other data of thepark.The quantum key distribution system can also change the key according to the specificbusiness requirements,realizing the intelligent management of the park and secure datatransmission.In the future,the quantum encryption system will be applied to the c

77、onstruction ofWinter Olympics Smart Park and Xiongan New Area.For example,in Xiongan quantum communication pilot as illustrated by Figure 2.11,aquantum communication trunk line between Beijing and Xiongan will be deployed,and aquantum key distribution platform will be introduced to provide security

78、keys for customers in the22/33fields of Internet of things(IoT),Internet of vehicles(IoV),smart energy,smart government andso on.The quantum key distribution platform and the service application server can be deployedtogether without changing the original network topology,and the encrypted business

79、is stilltransmitted in the original service channels.Figure 2.11 Quantum Communication Pilot in Xiongan23/333Quantum Machine Learning(QML)It is highly expected that the 6th generation(6G)communication systems will lay afoundation of pervasive digitization,ubiquitous connection and full intelligence.

80、The provisions ofa many-fold increase in the communication system performance and rich diversity of innovativeservices call for a revolutionary promotion in information processing capability.In this regard,theemerging Quantum Machine Learning(QML)has attracted significant attention due to itsinforma

81、tion processing paradigm by combining the established benefits of quantum mechanismand machine learning.In the following,we start with the concepts of QML on a high level andthen discuss machine learning(ML)basic paradigms and their application in solving problemacross different layers of in communi

82、cation systems.Followed by introduction of quantum tools,we present examples to get insight into the research of QML.Consequently,QML forcommunication systems can be obtained by ML for communication system being synergy withquantum speedup.QML is the integration of quantum algorithms within machine

83、learning programs,and thusachieving quantum speedup.Therefore,QML can generate and recognize statistical data patternsthat are beyond the capabilities of computing or machine learning in the classical domain 4.On ahigh level,the concepts of QML can be categorized into the following three tiers 5.Tie

84、r 1.The most common use of the term refers to machine learning algorithms for theanalysis of classical data executed on a quantum computer,i.e.quantum-enhancedmachine learning.While machine learning algorithms are used to compute immensequantities of data,quantum machine learning utilizes qubits and

85、 quantum operations orspecialized quantum systems to improve computational speed and data storage done byalgorithms in a program.This includes hybrid methods that involve both classical andquantum processing,where computationally difficult subroutines are outsourced to aquantum device.These routines

86、 can be more complex in nature and executed faster on aquantum computer.24/33Tier 2.Quantum algorithms can be used to analyze quantum states instead of classicaldata.The term quantum machine learning is also associated with classical machinelearning methods applied to data generated from quantum exp

87、eriments(i.e.machinelearning of quantum systems),such as learning the phase transitions of a quantumsystem or creating new quantum experiments.Tier 3.Quantum machine learning also extends to a branch of research that exploresmethodological and structural similarities between certain physical systems

88、 and learningsystems.For example,some mathematical and numerical techniques from quantumphysics are applicable to classical deep learning and vice versa.Furthermore,researchersinvestigate more abstract notions of learning theory with respect to quantum information,sometimes referred to as quantum le

89、arning theory.3.1Machine Learning for Communication SystemsMachine learning technologies are conventionally regarded as comprising of three basicparadigms,i.e.,supervised learning,unsupervised learning and reinforcement learning.Thissection gives a brief introduction of applying the above three para

90、digms in solving a wide range ofproblems across different layers of communication systems,as illustrated in Table 3-1.Table 3-1 Example applications of ML in different layers of communicaiton systemSupervisedUnsupervisedReinforcementPHY andMACChannel modelgenerationSignal processingPower allocation

91、andinterferencecancellationCSI estimationSpectrum sensingMultiple accessLocationBeam managementOn-demand resource(power,radioresources)optimizationTransmission modeselectionAdmission controlNetworkTraffic classificationCachingNetwork anomalydetectionTraffic predictionOptimized routingNetworkstate/pa

92、rametersProactive cachingTraffic predictionand classificationOptimized routing25/33predictionSupervised Learning is a machine learning task to infer a function that maps an input objectto a desired output based on example input-output pairs,where the example input-output pairs arereferred to as labe

93、led training data such as obtained from domain knowledges 6.Supervisedlearning is typically used to solve classification and regression problems.Therefore,supervisedlearning can play an important role in channel model generation,signal processing,and so on.Unsupervised Learning is a machine learning

94、 task that learns patterns from unlabeled data7.Additionally,if a small amount of labeled training data is available,the machine learning taskis called semi-supervised learning.Here we will avoid these details and only take unsupervisedlearning for instance to do discussion.Some of the most common a

95、lgorithms used in unsupervisedlearning include clustering,anomaly detection and approaches for learning latent variable models.In contrast to supervised learning,unsupervised learning has no dependence of labeled data andthus can be applied for clustering or tracking in a fast time-varying environme

96、nt.Unsupervisedlearning can potentially be applied for clustering or pairing of network nodes or endpoints for thepurposes of optimal allocation of various resources,particularly in a vehicle-to-everything(V2X)communication system.Reinforcement Learning(RL)8 is an area of machine learning concerned

97、with howintelligent agents react in an environment with a target of maximizing the reward.The focus of RLis on finding a balance between exploration(of uncharted territory)and exploitation(of currentknowledge)9.As compared to supervised learning,labeled training data is not required forreinforcement

98、 learning.However,partially supervised RL algorithms can combine the advantagesof supervised and RL algorithms.One powerful feature of RL is suitable for dealing with largeenvironments.Reinforcement learning is typically used for solving control and classificationproblems.Conventional and notable RL

99、 algorithms such as Q-learning and multi-armed bandittake as an input the current state of the network and enable the prediction of the next state.Apromising application of RL in communication contributes to scheduling parameters optimizationacross various layers.Additionally,deep learning can be co

100、mbined with RL to facilitate learning26/33long-term temporal dependence sequences in such a way that the accumulation of errors wontgrow very fast 10.3.2Quantum ToolsThis section revisits some principles and concepts associated with quantum mechanism,which will be referred to in the succeeding intro

101、duction of QML.Quantum entanglement 11 is a physical phenomenon that occurs when a groupof particles are generated,interact,or share spatial proximity in a way such that the quantumstate of each particle of the group cannot be described independently of the state of the others,including when the par

102、ticles are separated by a large distance.The topic of quantum entanglementis at the heart of the disparity between classical and quantum physics:entanglement is a primaryfeature of quantum mechanics lacking in classical mechanics.Quantum superposition 12 is a fundamental principle of quantum mechani

103、cs.It states that,much like waves in classical physics,any two(or more)quantum states can be added together(superposed)and the result will be another valid quantum state;and conversely,that everyquantum state can be represented as a sum of two or more other distinct states.Mathematically,itrefers to

104、 a property of solutions to the Schrdinger equation;since the Schrdinger equationis linear,any linear combination of solutions will also be a solution.Quantum simulators 13 permit the study of quantum system in a programmable fashion.In this instance,simulators are special purpose devices designed t

105、o provide insight aboutspecific physics problems.Quantum simulators may be contrasted with generally programmabledigital quantum computers,which would be capable of solving a wider class of quantumproblems.Quantum simulators have been realized on a number of experimental platforms,including systems

106、of ultracold quantum gases,polar molecules,trapped-ions,photonic systems,quantum dots,and superconducting circuits.27/333.3QML for Communication SystemsThis section presents examples to get insight into the research of QML from three tiers,i.e.,quantum-enhanced machine learning,machine learning of q

107、uantum systems and quantum learningtheory.Being synergy with the aforementioned machine learning(ML)basic paradigms and theirapplication in the context of communication,we can obtain QML for communication systems.3.3.1Quantum-enhanced Machine LearningQuantum-enhanced supervised and unsupervised lear

108、ningThe work in 14 provides supervised and unsupervised quantum machine learningalgorithms for cluster assignment and cluster finding,which proves that QML can take timelogarithmic in both the number of vectors and their dimension,therefore providing an exponentialspeed-up over classical ML algorith

109、ms.It is described in 15 a quantization method which refersto the process that partially or totally converts a classical algorithm to its quantum counterpart inorder to accelerate learning algorithms.In particular,the quantized routines employed for learningalgorithms that translate into an unstruct

110、ured search task is done by k-medians.Quantum-enhanced reinforcement learningIn quantum-enhanced reinforcement learning,a quantum agent interacts with a classical orquantum environment and occasionally receives rewards for its actions,which allows the agent tolearn what to do in order to gain more r

111、ewards.There are various ways of achieving quantumspeedup.For example,in 16 a quantum agent which has quantum processing capability isprovided in achieving a quadratic speed-up for active learning.Alternatively,the work in 17gains speed-up by probing the environment in superpositions.Furthermore,a g

112、eneral method ofquantum improvements in three paradigms of machine learning is provided in 17.A quantumspeedup of the agents internal decision-making time has been experimentally demonstrated intrapped-ions 18,while a quantum speedup of the learning time in a fully coherent(quantum)interaction betwe

113、en agent and environment has been experimentally realized in a photonic setup28/3319.In order to make the QML for communications a reality,powerful simulators of quantumdevices are required to facilitate development of QML algorithms.However,the quantumcomputers simulators available today can only s

114、imulate a small number of circuits,because thesimulation of a quantum computer on a classical computer is a computationally hard problem.Inparallel,the development of quantum devices(sensors,measurement,etc)with a high degree ofprecision and sensitivity is crucial not only for facilitating the devel

115、opment QML algorithms butalso for exploitation of quantum mechanism concepts and principles.3.3.2Machine Learning of Quantum SystemsThe work in 20 shows an experiment performed to reconstruct an unknown photonicquantum state with a limited amount of copies.In particularly,a semi-quantum reinforcemen

116、tlearning approach is employed to adapt one qubit state,an agent,to an unknown quantum state,an environment,by successive single-shot measurements and feedback,in order to achievemaximum overlap.The experimental learning device herein,composed of a quantum photonicssetup,can adjust the corresponding

117、 parameters to rotate the agent system based on themeasurement outcomes 0 or 1 in the environment(i.e.,reward/punishment signals).3.3.3Quantum Learning TheoryQuantumlearningtheory4pursuesamathematicalanalysisofthequantumgeneralizations of classical learning models and of the possible speed-ups or ot

118、her improvementsthat they may provide.The framework is very similar to that of classical computational learningtheory,but the learner in this case is a quantum information processing device,while the data maybe either classical or quantum.Quantum learning theory should be contrasted with thequantum-

119、enhanced machine learning discussed above,where the goal was to consider specificproblems and to use quantum protocols to improve the time complexity of classical algorithms for29/33these problems.Quantum learning theory is still under development.The fundamental in learning theory is a concept clas

120、s,each of which is usually a function onsome domain.The goal for the learner is to learn(exactly or approximately)an unknown targetconcept from this concept class.The learner may be actively interacting with the target concept,orpassively receiving samples from it.In active learning,a learner can ma

121、ke membership queries tothe target concept c,asking for its value c(x)on inputs x chosen by the learner.The learner thenhas to reconstruct the exact target concept,with high probability.In passive learning,the learnerreceives random examples(x,c(x),where x is distributed according to some unknowndis

122、tribution D.The goal of the learner is to output a hypothesis function h such that h(x)=c(x)withhigh probability when x is drawn according to D.30/334Reference1 https:/itu.int/en/ITU-T/studygroups/2017-2020/17/Pages/q15.aspx2 https:/ Cryptography-FAQs.”NIST Computer Security Resource Center,6 Aug201

123、9,https:/csrc.nist.gov/Projects/Post-Quantum-Cryptography/faqs4 S.J.Nawaz,S.K.Sharma,S.Wyne,M.N.Patwary and M.Asaduzzaman,QuantumMachine Learning for 6G Communication Networks:State-of-the-Art and Vision for theFuture,inIEEEAccess,vol.7,pp.46317-46350,2019,doi:10.1109/ACCESS.2019.2909490.5 https:/en

124、.wikipedia.org/wiki/Quantum_machine_learning6 https:/en.wikipedia.org/wiki/Supervised_learning7 https:/en.wikipedia.org/wiki/Unsupervised_learning8 https:/en.wikipedia.org/wiki/Reinforcement_learning9 Leslie Pack Kaelbling,Michael L.Littman,and Andrew W.Moore.1996.Reinforcementlearning:a survey,in J

125、ournal of Artificial Intelligence Research.4,1(Jnauary 1996),237285.10U.Challita,L.Dong and W.Saad,Proactive Resource Management for LTE in UnlicensedSpectrum:A Deep Learning Perspective,in IEEE Transactions on Wireless Communications,vol.17,no.7,pp.4674-4689,July 2018,doi:10.1109/TWC.2018.2829773.1

126、1https:/en.wikipedia.org/wiki/Quantum_entanglement12https:/en.wikipedia.org/wiki/Quantum_superposition13 https:/en.wikipedia.org/wiki/Quantum_simulator#Trapped-ion_simulators14 S.Lloyd,M.Mohseni,and P.Rebentrost,“Quantum algorithms for supervised andunsupervised machine learning,”,2013.15 E.Ameur,G.

127、Brassard,and S.Gambs,“Quantum speed-up for unsupervised learning,”Machine Learning,vol.90,no.2,pp.261287,2013.16 Paparo,Giuseppe Davide;Dunjko,Vedran;Makmal,Adi;Martin-Delgado,Miguel Angel;Briegel,Hans J.(2014).Quantum Speedup for Active Learning Agents.Physical Review31/33X.4(3):031002.17 Dunjko,Ve

128、dran;Taylor,Jacob M.;Briegel,Hans J.(2016-09-20).Quantum-EnhancedMachine Learning.Physical Review Letters.117(13):130501.18 Sriarunothai,Theeraphot;Wlk,Sabine;Giri,Gouri Shankar;Friis,Nicolai;Dunjko,Vedran;Briegel,Hans J.;Wunderlich,Christof(2019).Speeding-up the decision making of alearning agent u

129、sing an ion trap quantum processor.Quantum Science and Technology.4(1):015014.19 Saggio,Valeria;Asenbeck,Beate;Hamann,Arne;Strmberg,Teodor;Schiansky,Peter;Dunjko,Vedran;Friis,Nicolai;Harris,Nicholas C.;Hochberg,Michael;Englund,Dirk;Wlk,Sabine;Briegel,Hans J.;Walther,Philip(10 March 2021).Experimenta

130、l quantum speed-upin reinforcement learning agents.Nature.591(7849):229233.20 Yu,Shang;Albarran-Arriagada,F.;Retamal,J.C.;Wang,Yi-Tao;Liu,Wei;Ke,Zhi-Jin;Meng,Yu;Li,Zhi-Peng;Tang,Jian-Shun(2018-08-28).Reconstruction of a Photonic QubitState with Quantum Reinforcement Learning.Advanced Quantum Technol

131、ogies.2(78):1800074.32/33AcknowledgementGrateful thanks to the following contributors for their wonderful work on thiswhitepaper:Editors:Chih-Lin I(China Mobile),Xin GUO(Lenovo)Contributors:China Information and CommunicationTechnologiesGroupCorporation(CICT)&NationalInformationOptoelectronicsInnova

132、tionCenter(NOEIC)Yi QIANUniversity of Science and TechnologyBeijing(USTB)Zhangchao MA,Jianquan WANG,Lei SUNAppleShu GUOChina UnicomQi Liu，Meng Song，Chuanxin Zeng，Xingrong XuLenovoXin GUOAbbreviationASAccess StratumASICApplication-Specific Integrated CircuitCFPC Form-factor PluggableCVContinuous Vari

133、able QKDDVDiscrete Variable QKDECCSIEllipticCurve-BasedCertificatelessSignaturesforIdentity-BasedEncryptionIDQID QuantiqueIoTInternet of Things33/33IoVInternet of VehiclesISGIndustry Specification GroupMDIMeasurement-Device-IndependentMLMachine LearningNASNon-Access StratumNEANR Encryption Algorithm

134、NIANR Integrity AlgorithmNISTNational Institute of Standards and TechnologyPOCPost-Quantum CryptographyQITQuantum Information TechnologyQKDQuantum Key DistributionQKDNQuantum Key Distribution NetworkQMLQuantum Machine LearningRLReinforcement LearningSAGESecurity Algorithms Group of ExpertsSDOStandar

135、d Developing OrganizationSKTSK TelecomTFTwin-FieldTLOTransmitted Local OscillatorV2XVehicle-to-EverythingVPNVirtual Private NetworkFuTURE FORUM is committed to cutting edge technologies study andapplications.Controversies on some technical road-maps and methodologies mayarise from time to time.FuTUR

136、E FORUM encourages open discussion and exchangeof ideas at all levels.The White Paper released by FuTURE FORUM represents theopinions which were agreed upon by all participating organizations and weresupported by the majority of FuTURE FORUM members.The opinions contained inthe White Paper does not necessarily represent a unanimous agreement of allFuTURE FORUM members.FuTURE FORUM welcomes all experts and scholars active participation infollow-on working group meetings and workshops.we also highly appreciate yourvaluable contribution to the FuTURE White Paper series.