摘要
在复杂无线环境下,针对空中智能反射面(Aerial Reconfigurable Intelligent Surface,ARIS)辅助毫米波通信系统的能效优化问题,考虑加入对ARIS的朝向设计,利用深度强化学习方法对ARIS轨迹、朝向及其相移进行联合优化。在建立的系统和信道模型基础上,将ARIS辅助通信任务建模为马尔可夫决策过程,对其状态和动作空间及奖励函数进行设计,分别应用竞争双深度Q网络(Dueling Double Deep Q Network,D3QN)和近端策略优化(Proximal Policy Optimization,PPO)两种算法优化系统能效。仿真结果显示,优化ARIS朝向有效提高了系统能效,且D3QN相较于PPO算法得到的单位时间能效高出约830 b/J/s,表明基于D3QN的能效优化算法性能更优。
For the energy efficiency optimization problem of aerial reconfigurable intelligent surface(ARIS)-assisted millimeter-wave communication systems in complex wireless environments,the incorporation of ARIS orientation design is considered,and a deep reinforcement learning method is utilized for the joint optimization of the ARIS trajectory,orientations and their phase shifts.On the basis of the system and channel models established,the ARIS-assisted communication task is modeled as a Markov decision process,the state and action spaces and reward functions are designed,and the two algorithms of Dueling Double Deep Q Network(D3QN)and Proximal Policy Optimization(PPO)are applied respectively to optimize the energy efficiency of the system.Simulation results show that optimizing the ARIS direction effectively improves the system energy efficiency,and the energy efficiency per unit time obtained by D3QN is about 830 b/J/s higher than that obtained by the PPO algorithm,which indicates the better performance of the energy efficiency optimization algorithm based on D3QN.
作者
席云光
李素月
XI Yunguang;LI Suyue(School of Electronic and Information Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China)
出处
《电讯技术》
北大核心
2024年第12期1971-1980,共10页
Telecommunication Engineering
基金
山西省基础研究计划面上项目(20210302123205)
山西省回国留学人员科研资助项目(2022-162)。
关键词
毫米波通信
智能反射面
无人机
深度强化学习
能量效率
mmWave communications
reconfigurable intelligent surface
unmanned aerial vehicle
deep reinforcement learning
energy efficiency