Deep Learning Based Active User Detection for Uplink Grant-Free Access

Abstract

In massive machine-type communication (mMTC) systems, a large number of devices transmit small packets sporadically. Grant-free (GF) nonorthogonal multiple access turns to be a competitive candidate, since it avoids the granting access and reduces the signaling overheads. Exploiting the active users' inherent sparsity nature, we formulate the active user detection (AUD) problem as a single measurement vector (SMV) problem, and prove that our SMV model could support more active users than conventional multiple measurement vector (MMV) model. Based on the iterative soft thresholding (IST) algorithm, we propose a learning IST network (LISTnet), which is easy to be trained and performs better than conventional methods when the users' active rate is high. Besides, we add connections between the layers of LISTnet and develop a residual LISTnet (ResLISTnet), which can adaptively adjust the number of layers to reduce the computational complexity. Numerical simulation results show the superiority of our methods.