基于深度强化学习的多功能超表面辅助无蜂窝网络资源分配  

Deep Reinforcement Learning-Based Resource Allocation for Multi-Functional RIS-Assisted Cell-Free Networks

作　　者：王雯 倪万里 魏昊[4] 张铖 黄永明[1,2] WANG Wen;NI Wanli;WEI Hao;ZHANG Cheng;HUANG Yongming(Purple Mountain Laboratories,Nanjing 211111,China;School of Information Science and Engineering,Southeast University,Nanjing 210096,China;Department of Electronic Engineering,Tsinghua University,Beijing 100084,China;School of Artiicial Inteigence,Beijing University of Posts and Telecommunications,Bejing 100876,China)

机构地区：[1]紫金山实验室,江苏南京211111 [2]东南大学信息科学与工程学院,江苏南京210096 [3]清华大学电子工程系,北京100084 [4]北京邮电大学人工智能学院,北京100876

出　　处：《移动通信》2025年第4期20-27,共8页Mobile Communications

基　　金：国家资助博士后研究人员计划(GZB20240386);中国博士后科学基金资助项目(2024M761669);北京邮电大学博士研究生创新基金项目(CX2023110);国家自然科学基金面上项目(62271140);江苏省优秀青年科学基金项目(BK20240174)。

摘　　要：近年来兴起的无蜂窝网络架构,消除了由时频资源划分的小区边界,提高了频谱利用率、系统覆盖能力和用户公平性,被认为是第六代移动通信的潜在关键技术。然而,这一架构的实施通常需要大规模的接入点(AP,Access Point)部署,这将带来较高的系统成本和功耗。此外,无蜂窝网络的可达性能往往受限于多变的、难以控制的信号传播环境。为此,将多功能超表面(MF-RIS,Multi-Functional Reconfigurable Intelligent Surface)引入无蜂窝网络。通过同时反射、折射和放大入射信号,MF-RIS能以一种高效的方式重构无蜂窝网络的无线传播环境。为了最大化最小的用户端可达速率,构建AP的发射波束赋形,MF-RIS的模式选择和系数优化的联合调度问题。为了求解该混合整数非线性规划问题,提出一种基于深度强化学习的双智能体算法,其中一个智能体确定MF-RIS单元的工作模式,另一个智能体根据模式选择策略设计AP的波束形成和MF-RIS系数。仿真结果验证了所提算法的有效性。In recent years,the emerging cell-free network architecture has eliminated cell boundaries defined by time-frequency resources,thereby enhancing spectrum utilization,system coverage,and user fainess.Itis regarded as a potential key technology for the sixthgeneration mobile communications.However,implementing such an architecture typically requires the large-scale deployment of access points(APs),resulting in high system costs and power consumption.Moreover,the achievable performance of cell-free networks is often constrained by dynamic and uncontrollable signal propagation environments.To address these issues,this paper introduces the multi-functional reconfigurable intelligent surface(MF-RIS)into cel-free networks.By simultaneously reflecting,refracting,and amplifying incident signals,MF-RIS can reconfigure the wireless propagation environment of cell-free networks in an effective manner.To maximize the minimum achievable rate at the user side,this paper formulates a joint scheduling problem involving AP beamforming,and MF-RIS mode selection and coefficient optimization.To solve this mixed-integer nonlinear programming problem,this paper proposes a dual-agent algorithm based on deep reinforcement learning,where one agent determines the operating mode of MF-RIS elements,and the other designs the beamforming at APs and MF-RIS coeficients based on the selected modes.Simulation results demonstrate the effectiveness of the proposed algorithm.

关 键 词：多功能超表面 无蜂窝网络 深度强化学习 资源管理 

分 类 号：TN929.5[电子电信—通信与信息系统]