Deep Learning Based MIMO-NOMA Receiver Research

Non-Orthogonal Multiple Access (NOMA) enables superposition of user signals in power domain. Combining NOMA with Multiple-input Multiple-output (MIMO) technology can improve system spectral efficiency. Successive interference cancellation (SIC) receiver is often used in NOMA. However, most receiver schemes are limited by high complexity and assume perfect channel state information ( CSI) that may not be available in reality. To this end, we propose a novel deep learning based MIMO-NOMA receiver in this paper. We train the deep neural network (DNN) by minimizing the cost function and then feed the testing data into the trained neural network in return for correctly decoded signals. Numerous numerical results show the proposed receiver has strong robustness compared to conventional SIC receiver and also performs well under imperfect CSI.