基于DRL的主动RIS安全无线通信优化方法  

Optimization for active reconfigurable intelligent surface-assisted secure wireless communication based on deep reinforcement learning

作　　者：刘文涛 Manzoor Ahmed 林青 Liu Wentao;Manzoor Ahmed;Lin Qing(College of Computer Science&Technology,Qingdao University,Qingdao Shandong 266071,China)

机构地区：[1]青岛大学计算机科学技术学院,山东青岛266071

出　　处：《计算机应用研究》2023年第9期2808-2814,共7页Application Research of Computers

基　　金：山东省自然科学基金资助项目(ZR2020MF060)。

摘　　要：针对可重构智能表面(reconfigurable intelligent surface,RIS)辅助的安全无线通信系统在保密率优化问题中存在的信道空间连续变化、传统数学优化方法难以逼近最优解等问题,提出一种基于深度强化学习的SEC-DDPG(security deep deterministic policy gradient)算法。通过将RIS通信系统建模为连续变化空间中的马尔可夫决策过程,联合优化传输波束赋形和反射波束赋形达到最大化用户保密率的目的。仿真实验结果显示,在不同的传输功率及反射单元数量下,SEC-DDPG算法在主动和被动RIS系统中得到的最优保密率均比传统的交替优化算法有15%~20%的提升。研究结果表明,主动RIS场景下的安全性要优于被动RIS,与交替优化算法相比,SEC-DDPG算法能显著提高安全无线通信系统的用户保密率且具有鲁棒性,接近系统的最佳保密性能。For reconfigurable intelligent surface(RIS)-assisted secure wireless communication systems in the secrecy rate optimization problem,there are problems such as continuous variation of channel space and difficulty in approximating the optimal solution by traditional mathematical optimization methods.This paper proposed a SEC-DDPG(security deep deterministic policy gradient)algorithm based on deep reinforcement learning.By modeling the RIS communication system as a Markovian decision process in a continuously variable space,this algorithm jointly optimized the transmission beamforming and reflection beamfor-ming to maximize the user secrecy rate.The simulation experimental results show that the SEC-DDPG algorithm obtains the optimal secrecy rate in both active and passive RIS systems with 15%~20%improvement over the conventional alternating optimization algorithm for different transmission power and the number of reflective elements.The study results show that the secu-rity in the active RIS scenario is better than that in the passive RIS,and the SEC-DDPG algorithm can significantly improve the user secrecy rate of the secure wireless communication system with robustness and close to the optimal secrecy performance of the system compared with the alternating optimization algorithm.

关 键 词：主动可重构智能表面 深度强化学习 深度确定性策略梯度 波束赋形 乘法衰落 多输入单输出 物理层安全 

分 类 号：TN926[电子电信—通信与信息系统]