Lightweight Multi-scale Convolution Neural Networks for CSI feedback in Massive MIMO

In the frequency division duplex (FDD) massive multiple-input multiple-output (MIMO) systems, channel state information (CSI) needs to be fed back to the base station (BS) by the resource-limited user equipment (UE). Although deep learning (DL)-based methods have been widely applied to improve performance of CSI feedback in FDD massive MIMO systems, there is still potential for improvement in feedback accuracy and computational complexity. In this paper, we propose a lightweight channel state information feedback scheme, aiming to improve feedback performance and reduce complexity of model. This scheme utilizes lightweight depthwise separable convolution to process channel state information for reducing information loss during compression, and employs a multi-scale feature and residual learning techniques to recover and reconstruct CSI. Simulation results demonstrate that proposed feedback scheme exhibits better feedback accuracy compared to other lightweight DL-based networks.