Adaptive Traffic-Aware Routing for Dynamic LEO Satellite Networks

Abstract

Low Earth orbit (LEO) satellite networks require adaptive routing because satellite mobility, inter-satellite link availability, and traffic congestion change over time. This paper proposes a double deep Q-Network-based traffic-aware routing method for dynamic LEO satellite networks. The routing agent selects the next-hop satellite by considering neighboring satellite traffic loads, link quality, and destination information. The routing problem is formulated as a Markov decision process, where the reward penalizes hop usage, congestion, and low-quality links while encouraging successful packet delivery. The proposed formulation provides a learning-based routing framework that can avoid congested satellites using local traffic and link-state observations.