The deep reinforcement learning (DRL)-based deep deterministic policy gradient (DDPG) framework is proposed to solve the joint optimization of the IRS phase-shift matrix and the precoding matrix in an IRS-assisted multi-stream multi-user MIMO communication.

The combination of multiple-input multiple-output(MIMO) communications and intelligent reflecting surfaces(IRSs) is foreseen as a key enabler of beyond 5G (B5G) and 6Gsystems. In this work, we develop an innovative deep reinforcement learning (DRL)-based approach to the joint optimization of the MIMO precoders and the IRS phase-shift matrices that is proved to be efficient in high dimensional systems. The proposed approach is termed deep deterministic policy gradient (DDPG)and maximizes the sum rate of an IRS-assisted multi-stream(MS) multi-user MIMO (MU-MIMO) system by learning the best matrix configuration through online trial-and-error interactions. The proposed approach is formulated in terms of continuous state and action spaces, and a sum-rate-based reward function. The computational complexity is reduced by using artificial neural networks (ANNs) for function approximations and it is shown that the proposed solution scales better than other state-of-the-art methods, while reaching a competitive performance.