Active User Detection and Channel Estimation for Spatial-Based Random Access in Crowded Massive MIMO Systems via Blind Super-Resolution

This work presents a novel framework for random access (RA) in crowded scenarios of massive multiple-input multiple-output (MIMO) systems. A huge portion of the system resources is dedicated as orthogonal pilots for accurate channel estimation which imposes a huge training overhead. This overhead can be highly mitigated by exploiting intrinsic angular domain sparsity of massive MIMO channels and the sporadic traffic of users, i.e., few number of users are active to send or receive data in each coherence interval. Besides, the continuous-valued angles of arrival (AoA) corresponding to each active user are alongside each other forming a specific cluster. To exploit these features in this work, we propose a blind clustering algorithm based on super-resolution techniques that not only detects the spatial features of the active users but also provides accurate channel estimation. Specifically, an off-grid atomic norm minimization is proposed to obtain the AoAs and then a clustering-based approach is employed to identify which AoAs correspond to which active users. After active user detection, an alternating-based optimization approach is performed to obtain the channels and transmitted data. Simulation results demonstrate the effectiveness of our approach in AoA detection as well as data recovery which indeed provides a high performance spatial-based RA in crowded massive MIMO systems.