Joint Resource Allocation in LEO Satellite System: A Hypergraph Neural Network Enhanced Reinforcement Learning Approach

In the context of the Large-scale satellite network, beam interference poses a significant obstacle to the transmission efficacy of Low Earth Orbit (LEO) satellite downlinks. Given the relatively swift motion of LEO satellites relative to ground-based users, this interference manifests pronounced time-varying traits, thereby presenting a formidable challenge to contemporary transmission resource allocation methodologies. To address this problem, we propose the Hypergraph Neural Network (HGNN) enhanced Reinforcement Learning Resource Allocation (HRL) algorithm tailored for alleviating downlinks interference. The core objective of this algorithm is to resolve a transmission-rate maximization problem, flexibly managing time-varying interference across multiple downlink beams. Specifically, considering the dynamic nature of satellite beam coverage, we construct signal interference models of users at different times based on hypergraphs to accurately describe their time-varying relationships. We use HGNN to extract features from users' interference relationships, and then use reinforcement learning to learn optimal resource allocation strategies based on the extracted features. The simulation results show the effectiveness of the proposed HRL algorithm in optimizing the transmission rate.