A Practical Design Based on Deep Reinforcement Learning for RIS-Assisted mmWave MIMO Systems

Abstract

A revolutionary technology, reconfigurable intelligent surface (RIS), has emerged to enhance the signal transmission quality of wireless communications. This paper a RIS-assisted mmWave multiple-input multiple-output system, where practical finite discrete phase-shift constraints are crucial. Then, we discuss the connection between the channel state information (CSI) and the devices' location information in the mmWave band. To provide a model-free and CSI-free solution, the advanced deep reinforcement learning (DRL) technique is proposed for the RIS-assisted system based on the devices' location information. Moreover, we also apply the deep quantization neural network (DQNN) in the proposed DRL algorithm for the practical finite discrete phase-shift constraint. Finally, simulation results demonstrate the viability and efficacy of our proposed approach.