Satellite Beamforming based on Federated Learning

As the landscape of sixth-generation (6G) wireless networks advances, low-Earth-orbit (LEO) satellite communication emerges as a promising solution for offering comprehensive global communication services, though its potential is challenged by severe large-scale path loss that significantly impairs the received channel capacity available to terrestrial users. To address this limitation, this paper investigates a federated learning aided distributed beamforming network for LEO satellite communications, namely the FederSat network. In order to enhance the average achievable rate of the LEO satellite networks, we propose a novel code-book-based distributed beamforming strategy in the FederSat networks. Through numerical analysis, we demonstrate that the proposed FederSat networks substantially outperform in average achievable rate. Additionally, more LEO satellites are encouraged for further enhancing the received signal power, which indicates that a mega-constellation LEO satellite network is preferable.