Learning Spot Beam Power Control in Multi-LEO Satellite Networks

This paper proposes a deep learning approach for optimizing spot beam power control policies in multi low Earth orbit (LEO) satellite networks. Individual LEO satellites determine their power allocation variables in a decentralized manner for serving multiple ground users efficiently. To this end, we propose a graph neural network approach which characterizes the multi-LEO communication systems into a computational graph. Individual spot beams correspond to vertices of the graph, whereas edges connecting them are associated with satellite-to-ground links. The proposed method can be universally applied to any network size and save the coordination overhead among LEOs. Numerical results verify the effectiveness of the proposed scheme.