A GAT Based Robust Beamforming Method in Satellite-Terrestrial Integrated Network

The integration of satellite and terrestrial networks has ushered in a new era of communication capabilities. In satellite-terrestrial integrated network (STIN), full-frequency reuse introduces substantial inter-satellite interference and inter-beam interference, leading to a reduction in system capacity. Considering the extensive coverage area and high mobility of low Earth orbit (LEO) satellites, the acquisition of satellite-ground channel state information (CSI) is inherently imprecise, significantly affecting the performance gains of beamforming techniques. We propose a Graph Attention Network(GAT)-based approach to reduce system interference by optimizing beamforming. In detail, we represent the network topology as a graph, where nodes represent desired links and edges signify interference links. This transforms the NP-hard problem into a graph optimization task. In GAT, it can uncover the concealed influences of inter-satellite and inter-beam interference. This entails learning high-dimensional feature information from networks with similar structures while disregarding state variations in low-dimensional spaces. This approach enhances system robustness. Through simulation results, our proposed beamforming method demonstrates effective interference suppression with low computational complexity, leading to increased system throughput. Moreover, the method exhibits robustness in the presence of corrupted input data.