Deep reinforcement learning for IRS-aided multiuser MIMO systems

Abstract

Intelligent Reflecting Surfaces (IRS) have emerged as a transformative technology for sixth-generation (6G) wireless networks, enabling dynamic manipulation of the wireless propagation environment through programmable reflectors. When integrated into multiuser multiple-input multiple-output (MU-MIMO) systems, IRS can significantly enhance spectral efficiency. However, the joint optimization of transmit beamforming at the base station (BS) and phase shift control at the IRS presents a complex, high-dimensional, and non-convex problem. To tackle this problem, we propose a deep reinforcement learning (DRL)-based framework that jointly optimizes BS beamforming and IRS phase shifts in an end-to-end manner. Unlike conventional alternating optimization approaches that iteratively update each component separately, our method employs the deep deterministic policy gradient (DDPG) algorithm to simultaneously optimize both parameters while dynamically adapting to channel variations. The use of deep neural networks enables efficient learning of temporal dependencies, ensuring stable convergence and improved performance. Simulation results demonstrate that the proposed DRL-based framework significantly outperforms traditional optimization techniques in terms of sum-rate performance. Furthermore, the trained DRL agent continuously refines its decision-making policy through interaction with the wireless environment, highlighting the potential of DRL in efficiently optimizing IRS-aided MU-MIMO systems. These findings pave the way for intelligent, adaptive wireless networks in future 6G communications.