Active User Detection for Uplink NOMA Communication: Deep Learning Approach

Recently, with the rapid increase in the number of internet of things (IoT) devices, wireless communications based on orthogonal multiple access (OMA) are greatly limited by limited spectrum resources and the data conflicts and loss caused by large-scale IoT device. In order to support massive user devices, non-orthogonal multiple access (NOMA) have been suggested. In this paper, we focus on the problem of active user detection (AUD) of NOMA-based systems in IoT scenarios, and propose a long short-term memory (LSTM) based detection algorithm to solve the problems of low accuracy and high complexity of conventional AUD algorithms. The simulation results demonstrate that the proposed AUD algorithm based on LSTM network is superior to the conventional orthogonal matching pursuit (OMP) algorithm and the sparse Bayesian learning (SBL) algorithm in terms of both the error detection rate and the missed detection rate. Moreover, when the pilot scheme is designed for sparse code multiple access (SCMA), these two rates are further reduced compared with traditional pilot scheme. Furthermore, we present a detailed complexity analysis and comparison between conventional and deep learing (DL) AUD algorithm, which show the DL-based method greatly reduces the computational complexity.