Deep Learning-Based Successive Interference Cancellation for Non-Orthogonal Multiple Access in 5G Wireless Networks

Abstract

This research proposes a hybrid deep learning-based Successive Interference Cancellation (SIC) framework using Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) to improve interference management in 5G wireless communication systems, particularly in Non-Orthogonal Multiple Access (NOMA) networks. The proposed CNN-LSTM architecture combines the ability of CNN to extract spatial features and local signal patterns with the capability of LSTM to capture temporal dependencies in wireless signals. Unlike conventional SIC methods, which basically rely on perfect Channel State Information (CSI) and fixed decoding orders, the proposed framework predicts the optimal interference cancellation order and estimates residual interference. This helps minimize error propagation and improve signal detection reliability under varying channel conditions. The proposed framework is evaluated through simulations involving different signal-to-noise ratios (SNRs) and user densities. The results demonstrate improved interference cancellation accuracy, reduced Bit Error Rate (BER), and enhanced overall system performance compared with conventional SIC and standalone deep learning approaches.