Spatio-Temporal Focal Modulation Network for Massive MIMO CSI Feedback

Channel state information (CSI) compression and recovery are vital for the communication system to maximize the throughput. In this paper, a CSI feedback method based on spatio-temporal focal modulation (STFM) is proposed. By using a focal modulation mechanism that incorporates the interaction between the spatial and temporal features, the proposed STFM jointly exploits the cross-domain correlation. With multiple successive CSI snapshots, STFM can achieve excellent compression and reconstruction performance. Besides, a residual vector quantization (RVQ) with a shared and learnable codebook is introduced. The RVQ includes multiple stages to quantize the output of the encoder recursively and the corresponding quantization error can be reduced. Simulation results show that the STFM with RVQ outperforms the existing CSI feedback compression and quantization methods in different configurations.