DNN-Based Decoder for Four-Dimensional Modulation Superposition NOMA

Abstract

As a critical technology for the fifth-generation (5G) mobile communication system, non-orthogonal multiple access (NOMA) has drawn substantial attention due to its high spectrum efficiency with successive interference cancellation (SIC). However, SIC requires relatively high computation complexity since it needs to decode the interferer’s information first. In this paper, we propose a DNN-based decoder for four-dimensional modulation superposition NOMA. Spherical code is utilized as the four-dimensional modulation method to increase the Euclidean distance between constellations. We design the DNN-based decoder and analyze the effect of different training set on the detection performance. The performance of the DNN-based decoder is compared with the traditional maximum likelihood (ML) decoder. The simulation results show that, the DNN-based decoder can work well with a low complexity.