Deep Learning Based Sparse Signal Processing for Massive Wireless Access

In this tutorial, we introduce recent research of data-driven methods for compressive sensing-based grant free massive access, where deep learning is used to solve a sparse recovery problem to complete jointly activity detection, channel estimation, and data recovery. We first introduce different sparse problem formulations in communication systems, and then present direct and indirect data-driven solutions, respectively. Then, we discuss several variations of neural networks for sparse signal processing. Especially, we show how to construct special neural networks with domain knowledge in wireless networks. Besides, we present limitations in classical data-driven methods in this context, and how to deal with them. Finally, we give future challenges and potential directions for using data-driven methods for massive access.