Artificial neural network performance evaluation for a hybrid power domain orthogonal/non-orthogonal multiple access (OMA/NOMA) system

Abstract

Next-generation wireless technologies face considerable challenges in terms of providing the required latency and connectivity for new heterogeneous mobile networks. Driven by these problems, this study focuses on increasing user connectivity together with system throughput. For doing so, we propose and evaluate a hybrid machine learning-driven orthogonal/non-orthogonal multiple access (OMA/NOMA) system. In this work, we use an artificial neural network (ANN) to assign an OMA or NOMA access method to each user equipment (UE). As part of this research we also evaluate the accuracy and training time of the three most relevant learning algorithms of ANN (L-M, BFGS, and OSS). The main objective is to increase the sum-rate of the mobile network in the introduced beamforming and mmWave channel environment. Simulation results show up to a $20%$ sum-rate average performance increase of the system using the ANN management in contrast to a random non-ANN managed system. The Leveberg-Marquard (L-M) training algorithm is the best overall algorithm for this proposed application as presents the highest accuracy of around $77%$ despite 37 minutes of training and lower accuracy of $73%$ with approximately 28 seconds of training time.