Joint BS and RIS Element Selection with Large Arrays for EE Maximization

In this work, we investigate the energy effi-ciency (EE) metric in reconfigurable intelligent surface (RIS)-assisted multi-user (MU) massive multiple-input multiple-output (mMIMO) systems. Our analysis involves the optimization of the number of antennas M and reflective elements (REs) N in the base station (BS) and RIS arrays, respectively, along with the optimization of passive beamforming and power allocation. To address this problem, we introduce a novel joint reinforcement-analytical methodology (JRAM) algorithm, utilizing reinforcement learning (RL) for the elements selection in both arrays and subsequently conventional analytical techniques to optimize passive beamforming and power allocation for the selected elements at the BS and RIS. The numerical results underscore the significance of jointly optimizing the number of operative antennas and REs, leading to a remarkable gain of approximately 40.3% compared to optimizing just the operative elements in the BS and passive beamforming optimization.