Joint User Activity, Delay, and Data Detection for Heterogeneous Asynchronous Massive Random Access

Massive random access is a key challenge in massive machine-type communications (mMTC), which aims to support connectivity for a large number of devices. Most existing studies assume strict user synchronization; however, this assumption is often unrealistic in practical scenarios without base station coordination. To address the issue of random access in asynchronous scenarios, this paper proposes two innovative algorithms. First, we propose a detection algorithm based on generalized approximate message passing with the sum product algorithm (GAMP-SPA). By incorporating user delay characteristics into a factor graph model and performing iterative message updates, the algorithm progressively converges to the optimal posterior probability distribution, thereby enabling joint detection of user activity, delay, and transmitted data. Second, we propose AMP-Net, a model-driven approximate message passing network that integrates deep neural architectures with message passing principles to reduce reliance on prior knowledge of both channel state information (CSI) and user activity patterns. Simulation results demonstrate that the two proposed algorithms achieve superior performance in asynchronous scenarios, significantly outperforming traditional compressed sensing algorithms in non-coherent joint detection tasks.