Deep Learning-aided Successive Interference Cancellation for MIMO-NOMA

This paper introduces a novel deep learning (DL) based successive interference cancellation (SIC) scheme for an uplink multiple-input multiple-output non-orthogonal multiple access (MIMO-NOMA) system. The proposed scheme is aimed at mitigating the problems of error propagation and high computational complexity encountered with traditional SIC schemes. A separate deep neural network (DNN) is used to directly decode each user's signal at every SIC step. In particular, the DNN simulates the following operations: channel estimation, signal detection and canceling of decoded signals from the received combined signal. Results from simulation show superior performance and the effectiveness of the proposed approach. It outperforms traditional SIC schemes including the DL based approaches in terms of the bit error rate while maintaining low computational complexity.