A q-learning approach for enhanced routing in dynamic LEO satellite networks

Abstract

As global communication demand rises, Low Earth Orbit (LEO) satellite systems offer high-speed data transmission and extensive coverage options but face routing challenges due to dynamic topologies. This paper introduces a Q-Learning-based routing approach that converts dynamic networks into virtually static topologies at different snapshot intervals. Simulation results on a 66-satellite Starlink constellation demonstrate that Q-Learning outperforms Dijkstra’s algorithm, achieving faster convergence and reduced latency. These findings highlight the potential for Q-Learning in enhancing efficient, cost-effective satellite communications.