面向联邦学习的免授权随机接入研究  

Research on Grant-free Random Access for Federated Learning

作　　者：洪波 张一萌 赵仕杰 王昱晓 马国玉 HONG Bo;ZHANG Yimeng;ZHAO Shijie;WANG Yuxiao;MA Guoyu(School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学电子信息工程学院,北京100044

出　　处：《移动通信》2024年第10期136-143,共8页Mobile Communications

基　　金：北京市自然科学基金-昌平创新联合基金资助项目“面向电力场景大规模物联网的多址接入技术研究”(L234083);北京交通大学大学生创新创业训练计划资助项目“一种面向联邦学习的大规模多址接入方案”;国家自然科学基金委青年项目“面向高速铁路大规模物联网的多址方法研究”(62101024);北京交通大学大学生创新创业训练计划资助项目“一种面向联邦学习的大规模多址接入方案”(202510004154)。

摘　　要：联邦学习作为一种可以保障用户数据隐私的分布式学习范式,在无线边缘场景中体现出了突出的应用潜力。在联邦学习中,用户终端需要与服务器进行接入已传输其更新的模型参数,因而随机接入是联邦学习中无线传输的一个重要环节。相较于传统的随机接入方式,免授权随机接入由于节省了大量的控制信令开销使其成为6G大规模物联网实现海量低功耗低时延连接的关键。因此有必要研究免授权随机接入与联邦学习的融合。现有研究面向海量终端提出了资源跳跃多址技术与联邦学习的结合方案。然而,目前的结合方案仅考虑了用户数量固定的情况,免授权随机接入导致的用户数量随机对结合方案的影响仍有待研究。因此提出将免授权随机接入融合进新型多址与联邦学习的结合方案中,以探究其所产生的影响。结果表明,考虑用户数量随机不仅能够提高联邦学习模型的训练速度,而且可以提高系统的稳定性。As a distributed learning paradigm that can ensure user data privacy,federated learning has shown outstanding application potential in wireless edge scenarios.In federated learning,user terminals need to access the server to transmit their updated model parameters,so random access is an important link in the wireless transmission of federated learning.Compared with traditional random access methods,grant-free random access has become key to achieving massive low-power low-latency connectivity in 6G massive Internet of Things(IoT)due to its significant reduction in control signaling overhead.Therefore,it is necessary to study the integration of grant-free random access and federated learning.Existing research has proposed a combination of resource hopping multiple access technology and federated learning for massive terminals.However,the current combination scheme only considers the case where the number of users is fixed.The impact of the random number of users caused by grant-free random access on the combination scheme still needs to be studied.Therefore,this paper proposes that the grant-free random access is integrated into a novel combination scheme of multiple access and federated learning to explore its effects.The results show that considering the random number of users can improve both the training speed of the federated learning model and the system stability.

关 键 词：联邦学习 大规模物联网 免授权随机接入 

分 类 号：TN929.5[电子电信—通信与信息系统]