Research Article
Distributed Routing Algorithm for LEO Satellite Network Based on Deep Reinforcement Learning
@INPROCEEDINGS{10.1007/978-3-031-37139-4_15, author={Yudie Chen and Lan Wang and Houze Liang and Dong Lv and Weizhi Wu and Xiang Chen and Terngyin Hsu}, title={Distributed Routing Algorithm for LEO Satellite Network Based on Deep Reinforcement Learning}, proceedings={IoT as a Service. 8th EAI International Conference, IoTaaS 2022, Virtual Event, November 17-18, 2022, Proceedings}, proceedings_a={IOTAAS}, year={2023}, month={7}, keywords={low earth orbit satellite deep Q networks routing algorithm}, doi={10.1007/978-3-031-37139-4_15} }- Yudie Chen
Lan Wang
Houze Liang
Dong Lv
Weizhi Wu
Xiang Chen
Terngyin Hsu
Year: 2023
Distributed Routing Algorithm for LEO Satellite Network Based on Deep Reinforcement Learning
IOTAAS
Springer
DOI: 10.1007/978-3-031-37139-4_15
Abstract
Low earth orbit (LEO) satellites have the advantages of low transmission delay and wide coverage, which are widely used in current and future satellite communication systems. However, the large-scale and unevenly distributed traffic will easily cause severe inter-satellite-link (ISL) congestion. Therefore, the inter-satellite routing design of the LEO satellite networks is of vital importance. Traditional shortest path-based routing algorithms in LEO networks might cause overlapped loops that aggravate network delay. In this paper, by introducing the deep Q-network (DQN) into routing design, a distributed routing algorithm is proposed to reduce the delay and relieve congestion. Particularly, a reward function, considering routing loop as restriction to identify the best pathways, is designed to prevent the agent from getting stuck in a loop owing to duplicate paths. Meanwhile, instead of simply selecting a fixed shortest path, the proposed DQN model selects the next hop with a certain random probability, which helps to avoid overlapped paths and efficiently relieve congestion. A hybrid OPNET-STK simulation platform is built up for a typical LEO constellation with more than 200 satellites. Simulation results reveal that, compared with the traditional algorithms, the proposed approach can effectively relieve the ISL congestion and brings higher throughput for the LEO networks.
