Antijamming Routing-Based Graph Neural Network for Satellite Constellation Defense

This paper investigates the routing problem for mega low earth orbit (LEO) satellite constellations in adversarial environments. Due to the predictable orbital motion of LEO satellites, they are highly susceptible to malicious jamming from other constellations. We analyze the end-to-end delay performance under the threat of jamming attacks from adversarial medium earth orbit (MEO) constellation. To cope with the dynamic changes of satellite networks, we employ the graph neural network to process the satellite network topology graphs and propose an intelligent anti-jamming routing scheme based on deep reinforcement learning. Simulation results demonstrate that the proposed strategy outperforms baseline strategies in terms of delay and the ability to avoid the jammed satellites during routing.