Flexible feedback intelligent routing algorithm based on DQN for large-scale LEO networks

Abstract

The Low Earth Orbit (LEO) satellite constellation has the advantages of low-latency transmission and global coverage and has a wide range of applications in emergency rescue, aviation, and sea voyages. With the highly dynamic topologies, limited onboard computing resources, and large global routing computation overheads of the large-scale LEO constellation, centralized routing algorithms always encounter difficulties in such LEO satellite networks. Therefore, this paper proposes a flexible feedback intelligent routing algorithm based on deep Q-network (DQN) for large-scale LEO networks. By dynamically designing the reward function of DQN and adding a congestion alleviation scheme to the Kshortest path method in the training process, the delay and packet loss performances are improved. The delay performance is improved by 8%-12% over the Shortest Path First (SPF) algorithm, which is closer to the Shortest Delay First (SDF) algorithm with a minimum difference of only 6%. At the same time, the proposed algorithm has better packet loss performances than SPF and SDF algorithms to avoid congestion. When the congestion occurs, the algorithm can choose other paths.