Deep Learning-Based Joint Activity Detection and Channel Estimation for Massive Access: When More Antennas Meet Fewer Pilots

We consider the problem of joint activity detection and channel estimation (JADCE) for massive grant-free random access with sporadic traffic devices. Due to the deployment of a large-scale antennas array and a massive number of Internet-of-Things (IoT) devices in the sixth-generation (6G) wireless networks, conventional JADCE approaches usually incur exceedingly high computational complexity and require long pilot sequences. To address these challenges, this paper develops a deep learning-based JADCE framework which consists of a dimension reduction module, a deep learning network module, an active device detection module, and a channel estimation module. In particular, a deep learning network is developed for the recovery of device state matrix based on a designed denoiser, which can adaptively adjust the density parameters characterizing the device state matrix and effectively adapt to general complex channel settings with a finite number of training data. Both theoretical analysis and simulation results confirm that the proposed deep learning framework is computationally-efficient for achieving excellent performance in practical scenarios.