When mmWave High-Speed Railway Networks Meet Reconfigurable Intelligent Surface: A Deep Reinforcement Learning Method

Abstract

Reconfigurable intelligent surface (RIS) is an efficient and promising way to enhance millimeter-wave (mmWave) high-speed railway (HSR) network resilience. However, due to the harsh nature of HSR environment and the unpredictable hardware damage of RIS, it is challenging to perceive accurate and complete channel state information (CSI). To tackle the challenge, this letter proposes a novel deep reinforcement learning framework to design the base station (BS) transmission beamforming and RIS phase shifts for the RIS aided mmWave HSR networks, which combines long short-term memory (LSTM) and deep deterministic policy gradient (DDPG), termed as LSTM-DDPG. The simulation results demonstrate that the proposed LSTM-DDPG scheme outperforms the benchmark schemes in terms of spectral efficiency with comparatively low execution time, which makes real-time decision-making truly viable in the dynamic HSR network.