Power Allocation Optimization in Cr-Noma Using Reinforcement Learning

This paper addresses the power allocation challenge in cognitive radio non-orthogonal multiple access (CR-NOMA) systems through a reinforcement learning (RL) approach, specifically Q -learning. The model-free Q -learning algorithm explores state and action spaces to maximize cumulative rewards through iterative strategy optimization. The power allocation problem is formulated as a Markov decision process (MDP), where the state reflects the current signal-to-interference-plusnoise ratio (SINR) of secondary users (CU). A novel framework is proposed to enhance system throughput and energy efficiency while managing interference with the primary user (PU). Simulation results demonstrate that Q-learning outperforms traditional methods regarding throughput, energy efficiency, and adaptability to dynamic network conditions. This approach significantly improves CR-NOMA systems by addressing resource allocation and interference control questions.