Sparse Signal Processing for Massive Device Connectivity via Deep Learning

In this paper, we consider the grant-free uplink transmission of Internet of Things (IoT) networks with a multiple-antenna base station (BS) and a large number of IoT devices. To account for the sporadic transmission of IoT devices, we formulate a joint activity detection and channel estimation problem, followed by introducing a group sparse inducing norm to reformulate the problem as the group sparsity estimation problem. Existing methods tackle sparsity estimation problem for real vector signals, and cannot flexibly handle the complex matrix signals. To this end, we extend a learned iterative shrinkage thresholding algorithm for complex group row-sparse matrix signals (LISTA-GS), which maintains the simplicity of ISTA and does not require a prior knowledge of the statistical channel information. In particular, we parameterize the iterative method and consider the ISTA as a recurrent neural network (RNN). Moreover, we develop the RNN architecture and derive the computation complexity for the proposed method. Furthermore, experiments are done on an IoT application example, joint activity detection and channel estimation in grant-free massive connectivity for massive machine-type communications (mMTC). Simulation results show that the proposed method achieves significantly better performance and more robustness comparing with the classic methods for group sparsity estimation.