Sparse Blind Deconvolution and Demixing via Block Majorization-Minimization

To support vastly growing intelligent devices, massive connectivity with low-latency communications has become a critical requirement. In this paper, we consider a multiple-input multiple-output network, where a large amount of devices are connected to an access point sporadically. We aim to simultaneously detect the active devices and recover the transmitted signals from the received mixed measurements, without a priori channel information. The problem is mathematically modeled based on the idea of blind deconvolution and demixing for sparse signals. We formulate the optimization problem via nonconvex matrix factorization, and propose an efficient block majorization-minimization algorithm, where the signals and filters are updated with analytical solutions in an alternating way. The proposed algorithm has much lower per-iteration computational complexity compared to state-of-the-art algorithms and hence is more scalable to large-size problems. Numerical results demonstrate that our method is able to recover the sparse signals and filters with higher precision as well as faster convergence in comparison with existing methods.