Title: Deep learning-based detector for downlink IM-NOMA systems

Authors: Issa Chihaoui; Mohamed Lassaad Ammari

Addresses: Innovation of Communicating and Cooperative Mobile Laboratory, University of Carthage, Tunis, Tunisia ' Innovation of Communicating and Cooperative Mobile Laboratory, University of Carthage, Tunis, Tunisia

Abstract: This work proposes a deep learning (DL)-based detector for downlink index modulation non-orthogonal multiple access (IM-NOMA) systems, which we call DL-IM-NOMA. The proposed detector involves a multilayer fully connected deep neural network (DNN) block. The received signal is preprocessed using zero-forcing (ZF) and channel knowledge at the receiver before being fed into the DNN block. Architecture complexity and bit error rate (BER) performance of DL-IM-NOMA have been presented for several configurations, and compared to these of the suboptimal log-likelihood ratio (LLR) detector. Simulation results show that DL-IM-NOMA outperforms LLR detector for high signal-to-noise ratios (SNRs) with lower complexity. It has been shown that DL-IM-NOMA avoids BER floor occurred at strong users using LLR detector. In contrast, for low SNRs, DL-IM-NOMA causes an insignificant performance loss, especially for systems using large constellation size.

Keywords: deep learning; NOMA; subcarrier index modulation; OFDM.

DOI: 10.1504/IJAHUC.2024.140440

International Journal of Ad Hoc and Ubiquitous Computing, 2024 Vol.46 No.4, pp.209 - 215

Received: 05 Jan 2024
Accepted: 23 Apr 2024
Published online: 09 Aug 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article