Efficient User Clustering and Reinforcement Learning Based Power Allocation for NOMA Systems

Non-orthogonal multiple access (NOMA) is a popular multiplexing technique for the 5th generation (5G) network due to its spectral efficiency, high reliability, and massive connectivity support. However, some technical challenges in NOMA needs to be addressed properly. Usually, NOMA exploits the channel gains of multiple users to serve them from the same radio resource block at different power levels using a complex power allocation policy. In this paper, we propose a reinforcement learning-based power allocation algorithm with a simple and efficient user clustering technique. We use a Q-learning algorithm among all reinforcement learning techniques, which can easily obtain an optimal strategy to allocate power efficiently to maximize the sum data-rate of the NOMA system. In addition, we propose a channel gain-based user clustering technique that also contributes to the maximization of sum data-rate. To verify the performance of the proposed scheme, we conduct extensive simulations under various environments. We can confirm that the proposed Q-learning algorithm with user clustering achieves the maximum sum data-rate compared to other scenarios.