Deep Reinforcement Learning-Driven Resource Allocation for Enhanced Performance in Multiple Access Networks

Resource allocation has always been one of the most critical aspects of cellular communications. With the advent of new generations of telecommunications, the exponential increase in the number of users and the demand for higher bandwidth and improved user experience have significantly heightened the importance of resource allocation techniques. Non-Orthogonal Multiple Access (NOMA) is considered one of the most efficient user access methods. It remains a focus of interest due to its relatively straightforward implementation compared to newer access methods. In this paper, we address the optimization problem using one of the most promising optimization techniques, reinforcement learning (RL), which outperforms traditional methods regarding computational efficiency and accuracy. Specifically, we employ the dueling architecture and demonstrate its efficacy in calculating the users’ data rate. In the simulation section, we also compute the spectrum efficiency (SE) for users at various power levels and showcase its performance. We have compared the performance of our proposed model with the exhaustive search method, and the results validate its effectiveness.