Intelligent Implicit CSI Feedback Approaches in Wideband Wireless Communications Systems

Deep learning (DL) significantly improves various wireless communications tasks, including the implicit channel state information (CSI) feedback, which considers CSI feedback and precoding collaboratively. In this paper, we design two DL-based implicit CSI feedback approaches for specific scenarios in wideband wireless communications systems. We apply the idea of implicit feedback in the limited feedback situation for the multi-user scenario with intra-cell interference, which is a valuable scenario among implicit feedback methods for wideband communications systems. Then, we design a robust module to fuse the codewords of users for better performance. Next, to realize efficient and flexible adjustment between performance and computational complexity, an adaptive implicit CSI feedback approach is designed for the zero feedback situation for the single-user (SU) scenario based on the scalable neural network (NN). Finally, we analyze the relationship between spectral efficiency (SE) and squared error, showing the rationality of using SE as the loss function. The numerical experiments verify that the fusion module in the designed method can improve the performance of different NN structures. Besides, in the SU scenario, the designed method provides a flexible way to adaptively adjust the size of the NN to meet specific requirements, which reduces memory cost by approximately half compared with using the fixed NNs.