Deep Inception-based Siamese Network for Active User Detection in Grant-free NOMA System

Abstract

Recent years have seen a rapid growth and development in the field of wireless communication networks. Specifically, the grant-free access and non-orthogonal multiple access (NOMA) in connection with deep learning algorithms. Which facilitate massive machine-type communication devices and improve performance in terms of active user detection (AUD). The detection procedure in the grant-free NOMA systems is difficult due to the signal being received is superimposed. Existing studies focused on deep learning methods to increase the detection performance. However, the models show limitations over the computational complexity. Integration of LSTM and GRUs can only handle temporal modeling not the spatial correlations. The aim of this paper is to add inception modules with Siamese network. The proposed S-net goes wider instead of deeper which reduces computational complexity and increase detection performance Furthermore, parameter sharing characteristics of S-Net helps in generalizing the performance for large sparse matrices with varying SNR values. The comparative analysis show that the proposed S-Net outperforms existing state-of-the-art methods in an effective manner.