Enhancing Network Performance in LEO Satellite Networks through DRL-Based Routing Optimization

Routing in Low Earth Orbit (LEO) satellite networks is complex due to the constantly changing network topology, unpredictable channel conditions, and the need for high QoS. This paper puts forth an innovative routing algorithm that leverages deep reinforcement learning to tackle the challenge. The algorithm considers the available network resources, traffic demand, and QoS requirements to make dynamic and efficient routing decisions. We represent the network topology as a graph and use a deep Q-learning network (DQN) to learn the optimal routing policy. The DQN is trained offline using a dataset of simulated network scenarios and then deployed online to make routing decisions based on real-time network conditions. The algorithm employs a reward function that considers network performance metrics such as delay, packet loss, and throughput to guide the learning process. To assess the effectiveness of the suggested algorithm, we carried out simulations utilizing the ns-3 network simulator. The outcomes reveal that the proposed algorithm surpasses present routing algorithms concerning network performance metrics and QoS requirements. Additionally, the algorithm exhibits robustness to network conditions and traffic demand changes.