NTN-CSInet: Joint design of CSI feedback and prediction in NTN scenario

In the non-terrestrial networks (NTN) scenario, long transmission delays induce channel aging and result in outdated channel state information (CSI) feedback. Meanwhile, codebook-based CSI feedback schemes face the challenge of codebook complexity and feedback overhead in massive MIMO systems. To address these issues, this paper proposes a joint-design deep learning(DL)-based model for CSI feedback and prediction in NTN scenario, called NTN-CSInet. An encoder is implemented on the user equipment (UE) side to compress the CSI, while the decoder and predictor are implemented on the base station (BS) side to both recover and predict the future CSI, considering the time dependence of the channel. A ConvLSTM network is employed to construct a predictor that processes the spatiotemporal features in the time series data and predicts the subband feature matrix of future slots. The effectiveness of this scheme in mitigating the adverse effects of satellite communication channel aging is verified by simulation experiments, with its performance compared with that of classical DL-based feedback schemes and traditional codebook schemes. The results show that the NTN-CSInet can improve the accuracy and timeliness of CSI in NTN systems, thereby to improve the system performance.