DRL-based joint optimization for 3D-oriented multi-IRS communication systems

This paper investigates the achievable rates of multiple intelligent reflecting surface (IRS)-assisted multi-hop communications by exploring the impact of three-dimensional (3D) IRS orientation, represented by elevation and azimuth angles relative to the base station (BS). We first formulate the problem as the joint optimization of the deployment location, 3D orientation, phase shift of IRSs, and power allocation of users, with the goal of maximizing the sum achievable rate. To address this problem, our approach involves the development of a novel algorithm, named deep deterministic policy gradient (DDPG), which leverages deep reinforcement learning (DRL). This algorithm iteratively interacts with the environment, employing a trial-and-error process to improve its performance. The simulation results demonstrate a significant performance improvement achieved by optimizing the IRS orientation compared to other contemporary approaches that do not consider optimizing the IRS deployment orientation.