Resource Allocation in Uplink NOMA Systems: A Hybrid-Decision-Based Multi-Agent Deep Reinforcement Learning Approach

This correspondence investigates a joint power control, channel selection, and user dynamic access scheme to maximize the sum rate of non-orthogonal multiple access (NOMA) systems. The highly dynamic and uncertain environment hinders the collection of accurate instantaneous channel state information (CSI) at the base station for centralized resource management. Therefore, we propose a hybrid-decision-based multi-agent actor-critic (HD-MAAC) approach to optimize the sum rate of the system. The proposed scheme improves the standard actor-critic (AC) algorithm to obtain both discrete and continuous actions. Simulation results verify that the proposed scheme achieves higher sum rate than that of existing popular schemes under different settings.