Hybrid Model-Data-Driven User-Activity Detection Network for Massive Random Access

Massive Machine-Type Communications (mMTC) features a massive number of low-cost user equipments (UEs) with sparse activity. Tailor-made for these features, grant-free random access (GF-RA) serves as an efficient access solution for mMTC. In GF-RA systems, the covariance-based maximum likelihood detection (CB-MLD) algorithm is extensively employed to achieve the user-equipment activity detection (UAD). However, the limited receiving antennas and UAD decision made upon the hard threshold may undermine the UAD accuracy of the CB-MLD algorithm. To address this problem, we propose a hybrid-driven deep neural network (DNN) for UAD termed as the hybrid-driven user-equipment activity detection network (HyD-UADNet), which is composed of a model-driven coordinate descent network (MD-CDNet) and a data-driven soft thresholding network (DD-STNet). Specifically, the MD-CDNet is designed to modify the step size for the coordinate descent operation in each CB-MLD iteration and alleviate the impact of the limited antennas. Following the MD-CDNet, the DD-STNet is constructed as an adaptive soft thresholding function to replace the hard threshold for the UAD decision. Simulation results are provided to demonstrate the effectiveness of the hybrid-driven method and the performance of the HyD-UADNet.