基于深度强化学习的无蜂窝系统无线接入点选择算法  

An Access Point Selection Algorithm for Cell-free SystemsBased on Deep Reinforcement Learning

作　　者：赵婉楠 宋晓阳 赵迎新[1] 吴虹[1] 刘之洋[1] ZHAO Wannan;SONG Xiaoyang;ZHAO Yingxin;WU Hong;LIU Zhiyang(College of Electronic Information and Optical Engineering,Nankai University,Tianjin 300350,China)

机构地区：[1]南开大学电子信息与光学工程学院,天津300350

出　　处：《电讯技术》2024年第6期821-829,共9页Telecommunication Engineering

基　　金：国家自然科学基金资助项目(61871239)。

摘　　要：面向以用户为中心的无蜂窝分布式多输入多输出(Multiple-Input Multiple-Output, MIMO)架构,研究利用不完备信道状态信息(Channel State Information, CSI)下实现无线接入点(Access Point, AP)与用户(User Equipment, UE)之间的选择,提出基于深度强化学习(Deep Reinforcement Learning, DRL)的高效分配算法,通过使用不完备CSI快速生成以用户为中心的AP集合,减少了对前馈链路容量的占用。仿真结果表明,与其他传统选择算法相比,所提出的DRL接入点选择算法可以获得至少22.48%的总遍历频谱效率增益;与深度Q网络(Deep-Q-Network, DQN)算法相比,可以获得约14.17%的总频谱效率增益。The selection problem between wireless access point(AP)and user equipment(UE)in a user-centric cell-free distributed multiple-input multiple-output(MIMO)system is investigated when only partial channel state information(CSI)is available.Based on deep reinforcement learning(DRL),an efficient AP selection algorithm is proposed,which uses partial CSI to rapidly generate a user-centric set of APs to reduce the occupancy of the fronthaul link.Simulation results demonstrate that the proposed DRL-based AP selection algorithm can achieve sum ergodic spectrum efficiency gain of at least 22.48%compared with other traditional selection algorithms.Additionally,compared with the Deep-Q-Network(DQN)algorithm,the DRL-based AP selection algorithm can achieve sum ergodic spectrum efficiency gain of about 14.17%.

关 键 词：MIMO 以用户为中心的无蜂窝网络 接入点选择 深度强化学习 频谱效率增益 

分 类 号：TN929.5[电子电信—通信与信息系统]