Multiple Input Multiple Output (MIMO) systems have been gaining significant attention from the research community due to their potential to improve data rates. However, a suitable scheduling mechanism is required to efficiently distribute available spectrum resources and enhance system capacity. This paper investigates the user selection problem in Multi-User MIMO (MU-MIMO) environment using the multi-agent Reinforcement learning (RL) methodology. Adopting multiple antennas' spatial degrees of freedom, devices can serve to transmit simultaneously in every time slot. We aim to develop an optimal scheduling policy by optimally selecting a group of users to be scheduled for transmission, given the channel condition and resource blocks at the beginning of each time slot. We first formulate the MU-MIMO scheduling problem as a single-state Markov Decision Process (MDP). We achieve the optimal policy by solving the formulated MDP problem using RL. We use aggregated sum-rate of the group of users selected for transmission, and a 20% higher sum-rate performance over the conventional methods is reported.
Keywords: MIMO; MU-MIMO; channel capacity; fairness; next-generation networks; reinforcement learning; sumrate; user scheduling.