Federated MADRL for Resource Optimization in Beam-Hopping LEO Satellite Networks with Integrated Sensing-Communication-Computing

This paper investigates a beam-hopping low Earth orbit satellite network that integrates sensing, communication, and computing to support post-disaster monitoring and timely rescue decision-making. To enable low-latency and efficient operation under constrained satellite resources, we propose a joint resource allocation framework that simultaneously optimizes beam illumination, subcarrier assignment, and satellite power allocation, with the objective of minimizing the average task completion delay of disaster-affected ground cells. The proposed scheme adopts an enhanced federated learning-driven distributed multi-agent proximal policy optimization (MAPPO) algorithm. Simulation results demonstrate that the proposed algorithm incurs only about 20% of the network parameter communication overhead of MAPPO while achieving lower average delay compared with other PPO-based algorithms.