Joint Deep Reinforcement Learning and Unfolding: Beam Selection and Precoding for mmWave Multiuser MIMO With Lens Arrays

The millimeter wave (mmWave) multiuser multiple-input multiple-output (MU-MIMO) systems with discrete lens arrays (DLA) have received great attention due to their simple hardware implementation and excellent performance. In this work, we investigate the joint design of beam selection digital precoding matrices for mmWave MU-MIMO DLA maximize sum-rate subject transmit power constraint constraints matrix structure. investigated non-convex problem variables coupled is challenging solve an efficient framework neural network (NN) proposed tackle it. Specifically, consists a deep reinforcement learning (DRL)-based NN deep-unfolding NN, which are employed optimize matrices, respectively. As DRL-based formulate as Markov decision process double Q-network algorithm developed base station considered be agent, where state, action, reward function carefully designed. Regarding matrix, develop iterative weighted minimum mean-square error induced unfolds into layer-wise structure introduced trainable parameters. Simulation results verify that jointly trained remarkably outperforms existing algorithms reduced complexity stronger robustness.