Efficient Multiple-Input–Multiple-Output Channel State Information Feedback: A Semantic-Knowledge-Base- Driven Approach

Massive multiple-input–multiple-output (MIMO) systems encounter substantial challenges in relation to channel state information (CSI) feedback; this is particularly the case in frequency division duplex systems, where the lack of channel reciprocity necessitates high-dimensional CSI transmission, resulting in substantial feedback overheads. Most existing deep learning methods have improved compression efficiency but still suffer from high feedback overheads due to their transmission of entire compressed vectors. To address this, we propose SKBNet, an innovative semantic knowledge base (SKB)-driven framework for efficient MIMO CSI feedback. By sharing the SKB between the transmitter and receiver, SKBNet transmits only index values, significantly reducing feedback overheads while achieving efficient CSI compression and accurate reconstruction. The simulation results demonstrate that SKBNet outperforms many existing methods in normalized mean square error, cosine similarity, and feedback bit-count at various compression ratios. This framework offers a promising solution for low-overhead, high-precision CSI feedback for future semantic communication networks.