Reinforcement Learning-Based Joint Resource Allocation for OFDM MU-MIMO Systems

Efficient resource allocation in orthogonal frequency-division multiplexing (OFDM) with multi-user multiple-input multiple-output (MU-MIMO) is crucial for improving link quality in next-generation cellular systems. However, the joint optimization of power levels, resource blocks (RBs), and beamforming vectors is computationally intractable due to its combinatorial complexity. In this paper, we propose a reinforcement learning (RL) framework based on Proximal Policy Optimization (PPO) to maximize the network-wide sum signal-to-interference-plus-noise ratio (SINR). A graph neural network (GNN)-based state encoder captures interference coupling among links and enables efficient policy learning. Simulation results demonstrate that the proposed method significantly outperforms greedy and heuristic baselines in terms of total SINR and computational efficiency.