In this paper, a split of super-resolution (SR) to reconstruct the channel state information (CSI) through transfer learning for a multiple-input-multiple-output (MIMO) system is proposed. The detailed structure of the CSI is degraded once the compression process. Compared with the existing system, a split of SR into two disjoint sub-blocks through transfer learning as well as a small number of filters are used in each layer to improve the CSI detailed structures in the reconstruction process. Simulation results demonstrate that the proposed system significantly enhances the quality of the CSI after reconstruction against the existing system in terms of normalized mean square error (NMSE), cosine similarity $\rho$, and training and validation loss stages, which are essentials for a MIMO system.
1. Rejected but resubmission allowed (IEEE WCL)
2. Rejected (IEEE WCL)
3. ICT Express (Preparation)