Reconfigurable Intelligent Surface for High-Speed Railway mmWave Communication System: A Deep Reinforcement Learning Approach

This paper presents research on the use of multiple reconfigurable intelligent surfaces (RIS) in high-speed railway (HSR) tunnel communication systems for signal quality improvement and spectral efficiency (SE) maximization. This is done by placing multiple RISs on opposite sides of the tunnel, which then relay the signal to the onboard mobile relay (MR) of the high-speed train (HST), as it passes through the tunnel. To achieve our objective, a joint beamforming at Basestation (BS) and RISs phase shift SE optimization problem is formulated. The formulated problem is non-convex and hard to solve using existing conventional mathematical optimization techniques. We proposed a special class of deep reinforcement learning (DRL) called rank-based batch prioritization replay deep deterministic policy gradient (RBPR-DDPG) to solve the problem. The performance of the proposed solution is then compared with an adapted version of benchmark uniform random batch replay deep deterministic policy gradient (URBR-DDPG). Simulation results show that the placement of multiple RISs along the HSRtunnel line not only leads to signal quality improvement in the tunnel but also the maximization of the signal SE at the HST mobile relay.