Joint Channel Estimation and Signal Detection in MIMO-NOMA Wireless Systems Using Deep Learning

This paper presents a the performance of deep learning (DL) in joint channel estimation and multi-user detection in non-orthogonal multiple access (NOMA) based multiple-input multiple-output (MIMO) wireless systems. Generally, in NOMA systems, the sequential decoding of multiple users is performed by the conventional successive interference cancellation (SIC) method at the receiver. Due to the effects of error propagation, with SIC, the accuracy of detection is dependent on the accuracy of detection of preceding users. DL-based MIMO-NOMA receiver is intended to detect messages for several users in a single step, without explicit estimation of the channel state information. The DL-based receiver for the MIMO-NOMA system can be implemented by a deep neural network (DNN) that conducts the estimation of channel parameters and the detection of signals jointly. The DNN is trained offline and is then used to directly detect the transmitted data during the online testing phase. Results indicate that the DL-based systems outperform the systems with traditional channel estimation methods and are more tolerant to variation in the number of pilot symbols and length of cyclic prefixes (CP). It is validated that the DL-based scheme is able to mitigate the SIC-based detector's error propagation effects. It is also observed that the DL method can perform better than the least square (LS) and minimum mean square error (MMSE) detector even in severe inter-symbol interference (ISI).