Optimizing Beam-Hopping in LEO Constellation Systems with Random Cell Traffic Using Multi-Agent Reinforcement Learning

With the advancement of 6G communication technologies, research on low Earth orbit (LEO) satellite communication has gained significant attention. Among them, beamhopping (BH) technology, explored for its applications in LEO satellites, provides the advantage of dynamically allocating resources based on the traffic demands of ground cells. However, scheduling methods that consider the dynamic nature of ground cell traffic demands, the number of visible satellites, and their elevation angle constraints have yet to be thoroughly investigated. Therefore, this paper proposes a beam-hopping scheduling algorithm based on deep reinforcement learning (DRL), which jointly considers the dynamic traffic distribution, the number of available satellites, and their visibility conditions to optimize resource allocation efficiency in LEO satellite networks.