Deep Reinforcement Learning for Hybrid RIS Assisted MIMO Communications

TL;DR

The paper tackles the non-convex, high-dimensional problem of jointly optimizing BS transmit beamforming and HRIS reflection/amplification to maximize spectral efficiency in HRIS-assisted MIMO. It introduces a deep reinforcement learning framework based on Proximal Policy Optimization that learns a CSI-to-configuration mapping offline for fast online inference. The approach yields near-optimal performance (around 95% of the alternating-optimization benchmark) with dramatic reductions in computation time and scalability to large HRISs, even outperforming fully passive RIS benchmarks. This enables real-time, dynamic adaptation in wireless networks with hybrid RIS architectures, reducing latency while maintaining strong SE gains.

Abstract

Hybrid reconfigurable intelligent surfaces (HRIS) enhance wireless systems by combining passive reflection with active signal amplification. However, jointly optimizing the transmit beamforming with the HRIS reflection and amplification coefficients to maximize spectral efficiency (SE) is a non-convex problem, and conventional iterative solutions are computationally intensive. To address this, we propose a deep reinforcement learning (DRL) framework that learns a direct mapping from channel state information to the near-optimal transmit beamforming and HRIS configurations. The DRL model is trained offline, after which it can compute the beamforming and HRIS configurations with low complexity and latency. Simulation results demonstrate that our DRL-based method achieves 95% of the SE obtained by the alternating optimization benchmark, while significantly lowering the computational complexity.

Figures (3)

• Figure 1: HRIS assisted downlink MIMO system.
• Figure 2: The proposed DRL framework.
• Figure 3: Convergence, spectral efficiency, and runtime performance of the proposed DRL scheme compared with benchmarks.