Attention Aided Channel Prediction Scheme For Satellite-Terrestrial Networks

Low Earth orbit (LEO) communication satellites are becoming an essential component of the integrated spaceterrestrial network for sixth-generation (6G) networks nowadays. However, due to the time-varying communication environment and long round-trip time (RTT), the downlink channel state information (CSI), which is always obtained by satellites from user equipments feedback, is severely outdated. Furthermore, as the satellite moves, changes in elevation angle result in received CSI with varying correlation characteristics and levels of outdatedness. Motivated by these issues, this paper proposes a deep learning (DL)-based prediction scheme named ATNet, which is designed to make channel prediction in frequency division duplexing (FDD) satellite Communication systems. Specifically, the predictor integrate convolutional neural network (CNN) and attention mechanism with the foundation of long short-term with memory (LSTM) network. This scheme considers the relationship between elevation angle and channel correlation, as well as the degree of CSI outdatedness. Simulation results demonstrate that our ATNet scheme has achieved a $7 \%-17 \%$ improvement in prediction accuracy compared to the baseline schemes.