AI Based Power Allocation for NOMA

Abstract

Novel methods using artificial intelligence for downlink power allocation problem in non-orthogonal multiple access networks are presented. The proposed machine learning and deep learning based methods achieved performance close to the optimum in terms of sum capacity with significantly lower computational costs. The numerical results also demonstrated up to 120 times a boost in computation time as compared to the conventional exhaustive search approach. Furthermore, the training and testing accuracy of the deep learning model reached 0.92 and 0.93 with the loss value dropping up to 0.002.