Machine Learning-Based Multi-User Semi-Blind Signal Detection in Non-Orthogonal Multiple Access

With the rapid development of fifth generation (5G) and sixth generation (6G) wireless communication technologies, non-orthogonal multiple access (NOMA) systems have become crucial to improve spectral efficiency and support more user connections. However, multi-user signal detection faces challenges due to complex channel environments and signal interference, and efficient and accurate detection methods need to be developed. In this paper, we propose a novel machine learning (ML) algorithm for semi-blind signal detection in NOMA uplink systems. Uniquely, the algorithm does not require separate channel estimation and can directly and accurately recover user signals received at the base station. Specifically, a convolutional neural network long term memory (CNN-LSTM) framework is used to analyse the synthetic signals received at the base station and estimate the channel state information. The estimation results are then fed into a deep neural network (DNN) based on joint maximum likelihood (JML) detection. By integrating with the synthetic signals, the transmitted signals of each user are detected separately. We evaluate the detection performance of both users under different modulation schemes. Simulation results show that using the proposed machine learning algorithm, our approach achieves significantly lower BER than traditional serial interference cancellation methods and the best JML detection performance.