An Spatial Feature Compensation Network for Massive MIMO CSI Feedback

Massive MIMO is a promising technology for enhancing link capacity and improving energy efficiency. In FDD massive MIMO systems, user equipments (UEs) are required to transmit downlink channel state information (CSI) to base stations (BSs). However, processing high-dimensional CSI matrices incurs unacceptable feedback overhead. In this paper, we introduce a novel neural network spatial feature compensation network (SFCNet), to address CSI feedback issues. SFCNet integrates a local spatial feature compensation (LSFC) module with a global compensation mechanism to enhance the quality of CSI feedback. This capacity allows the module to focus on key information from both dimensions, which is essential for enhancing the quality of CSI feedback. Furthermore, the global compensation mechanism aims to minimize the loss of lowfrequency information and boost the network's resilience against interference, thereby enhancing the reconstruction effectiveness. Experimental validations confirm that the proposed SFCNet significantly outperforms existing methods in CSI reconstruction accuracy.