Examination of the Non-Orthogonal Multiple Access System Using Long Short Memory Based Deep Neural Network

Abstract

This paper investigates deep learning (DL) non-orthogonal multiple access (NOMA) receivers based on long short-term memory (LSTM) under Rayleigh fading channel circumstances. The performance comparison between the DL NOMA detector and the traditional NOMA method is established, and results have shown that the DL-based NOMA detector performance is far better in comparison with conventional NOMA detectors. Simulation curves are compared with the performance of the DL detector in terms of minimum mean square estimate (MMSE) and least square error (LSE) estimate, taking all realistic circumstances, except the cyclic prefix (CP), and clipping distortion into account. The simulation curves demonstrate that the performance of the DL-based detector is exceptionally good when it equals 1 when the noise signal ratio (SNR) is more than 15 dB, assuming that the DL method is more resilient to clipping distortion.