Compressed Sensing-Based CSI Feedback by Leveraging Perception Information

To significantly reduce channel state information (CSI) feedback overhead in massive multipleinput and multiple-output (mMIMO) systems, compressed sensing (CS)-based CSI feedback methods are widely studied. However, these CS methods typically need accurate channel sparsity, high computational complexity for reconstruction processing, hindering their practical applications. To tackle these issues, a perception-assisted CS-based CSI feedback method is proposed in this paper. By using the active perception, the support-set of downlink CSI is extracted from the echo signals, thereby tackling the issue of unknown channel sparsity. With the perceived support-set, the proposed method reconstructs downlink CSI by solving a least square (LS) problem. Since the iterative processing in classic CS-based reconstruction is avoided, the proposed method significantly reduces computational complexity. In addition, this method utilizes the perceived support-set to suppress receiver noise, significantly enhancing the reconstruction accuracy. Simulation results show that, compared to classical CS-based reconstruction methods, the proposed method notably reduces the computational complexity, effectively improves the recovery accuracy of downlink CSI, and demonstrates the robustness to parameter variations.