面向6G的生成对抗网络研究进展综述  

Survey on the research progress of generative adversarial networks for 6G

作　　者：孟婵媛 熊轲[1,2] 高博 张煜[3] 樊平毅 MENG Chanyuan;XIONG Ke;GAO Bo;ZHANG Yu;FAN Pingyi(School of Computer and Information Technology,Beijing Jiaotong University,Beijing 100044,China;Engineering Research Center of Network Management Technology for High Speed Railway of Ministry of Education,Beijing Jiaotong University,Beijing 100044,China;State Grid Energy Research Institute Co.,Ltd.,Beijing 102209,China;Department of Electronic Engineering,Tsinghua University,Beijing 100084,China)

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044 [2]北京交通大学高速铁路网络管理教育部工程研究中心,北京100044 [3]国网能源研究院有限公司,北京102209 [4]清华大学电子工程系,北京100084

出　　处：《物联网学报》2024年第1期1-16,共16页Chinese Journal on Internet of Things

基　　金：国家自然科学基金项目(No.62071033);中央高校基本科研业务费资助项目(No.2022JBGP003)。

摘　　要：人工智能(AI,artificial intelligence)与通信技术的深度融合是6G网络的典型特征。一方面,AI为6G网络发展注入了新动力,能够有效利用网络运行产生的历史数据,使网络具备自维护、自优化的功能,加速了网络智能化进程。另一方面,6G网络丰富的场景和大规模的物联设备入网应用为AI提供了广阔的应用渠道和海量的训练数据,使AI能够更好地训练和部署,充分发挥AI的内在优势,为用户提供更加优质的智能服务。尽管如此,在一些实际应用中,受复杂环境的影响,存在数据样本收集困难、收集成本高和样本普适性不足等问题,难以充分发挥AI的性能优势。为此,学术界和工业界将生成对抗网络(GAN,generative adversarial network)引入无线网络的设计中,利用GAN强大的特征学习和特征表达能力产生大量模拟实际的生成样本,实现无线数据库的扩充,从而有效提升面向无线网络的AI模型的泛化能力。由于其优秀的性能表现,以GAN为代表的生成式模型在无线网络领域受到越来越多的关注,并迅速发展成为6G网络新的研究热点。首先,综述了GAN的原理及其改进衍生模型,对各种衍生模型的框架及优缺点进行了分析归纳;然后,综述了这些模型在无线网络领域的研究及应用现状;最后,面向6G网络的需求展望了GAN在6G网络中的研究趋势,为未来的研究提供了一些有价值的探索。The deep integration of artificial intelligence(AI)and communication technology is the typical feature of the 6G network.On the one hand,AI injects new vitality into the development of the 6G network,which can effectively use the data generated by the historical operation of the network.It enables the network to be self-maintained and self-optimized,and accelerates the process of network intelligence.On the other hand,the rich scenarios and IoT devices of the 6G network provide a large number of application fields and massive data for AI.These can enable the better deployment of AI,fully demonstrate the performance advantages of AI,and provide high-quality services for users.However,in practice,it is difficult to give full play to the performance advantages of AI due to the difficulty of sample collection,high cost of the collection,and lack of universality which caused by the complexity of the environment.Therefore,academia and industry introduce generative adversarial network(GAN)into the design of wireless networks.The powerful feature learning and feature expression ability of GAN can generate a large number of generated samples,which realizes the expansion of the wireless database.The introduction of GAN can effectively improve the generalization ability of AI models for wireless networks.Owing to the excellent performance of GAN,the generative model represented by GAN has attracted increased attention in the field of wireless networks,and rapidly became the new research hotspot of 6G networks.Firstly,the principle of GAN and its different versions of improved derived models were summarized.Then,the framework,advantages and disadvantages of each model were analyzed.Secondly,the research and application status of these models in wireless networks were reviewed.Finally,the research trends of GAN were proposed for the 6G network requirements,which provided some valuable exploration for future research.

关 键 词：生成对抗网络 无线网络 信道估计 物理层安全 无线感知 零和博弈 

分 类 号：TN92[电子电信—通信与信息系统]