Iterative SICNet for Non-Orthogonal Multiple Access

In this paper, we study signal detection in a multi-user non-orthogonal multiple-access (NOMA) network using deep learning. Deep learning is implemented to replace the iterative process of successive interference cancellation (SIC) for receivers at the downlink transmission. We propose the iterative SICNet (ISICNet), which is based on a deep neural network (DNN), to replace the SIC algorithm at each user receiver. Numerical results show that the proposed ISICNet is less prone to channel state information (CSI) uncertainty compared to the classical SIC algorithm and outperforms existing methods.