Spatial‐Aware Reinforcement Learning for Mitigating Inline Interference in Satellite Constellations
International Journal of Satellite Communications and Networking · 22.07.2026
DOI: 10.1002/sat.70077 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Satellite Communications and Networking, online erschienen am 22.07.2026. Verfasst von Huiliang Liu und 3 weiteren Autor:innen. Die Arbeit nennt 26 Literaturangaben. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
With the unfolding vision of sixth‐generation (6G) space‐air‐ground integrated networks (SAGIN), the rapid deployment of low Earth orbit (LEO) mega‐constellations has introduced severe spectrum coexistence challenges. Inline co‐frequency interference, occurring when satellites from different systems align geometrically with a ground station, significantly degrades link quality. Conventional beam management strategies, such as highest elevation (HE) and longest access time (LAT), primarily focus on single‐satellite visibility and struggle to adapt to the highly dynamic interference topologies of multi‐constellation environments. In this paper, we propose a state‐action‐reward‐state‐action (SARSA) reinforcement learning framework for proactive beam management, featuring a novel projected hexagonal state‐space discretization method. By mapping the continuous hemispherical observation space into a compact projected hexagonal grid, the proposed mechanism transforms complex angular topologies into discrete states, effectively reducing dimensionality while preserving spatial neighborhood relationships. A reward function is meticulously designed to balance the trade‐off between interference mitigation and handover signaling costs. Simulation results based on realistic constellation parameters, where the victim system employs a Walker‐Delta configuration suited for mid‐inclinations, and the interferer system utilizes a Walker‐Star configuration typical for near‐polar orbits, demonstrate that the proposed framework significantly outperforms baseline HE and LAT strategies. Specifically, at a 7° grid resolution, the severe interference probability (isolation angle < 5°) is reduced to 0.0278%, a substantial improvement over the traditional HE (1.1181%) and LAT (0.7222%) methods. Furthermore, the framework exhibits robust generalization across different geographic latitudes and maintains a stable daily handover frequency. This study provides an efficient and scalable solution for interference‐aware resource management in the era of LEO satellite constellation coexistence.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Liu, H., Sun, Q., Zhang, Z., & Chu, Y. (2026). Spatial‐Aware Reinforcement Learning for Mitigating Inline Interference in Satellite Constellations. International Journal of Satellite Communications and Networking. https://doi.org/10.1002/sat.70077
Vancouver
Liu H, Sun Q, Zhang Z, Chu Y. Spatial‐Aware Reinforcement Learning for Mitigating Inline Interference in Satellite Constellations. International Journal of Satellite Communications and Networking. 2026. doi:10.1002/sat.70077
BibTeX
@article{Liu2026,
author = {Liu, Huiliang and Sun, Qian and Zhang, Zhe and Chu, Yao},
title = {Spatial‐Aware Reinforcement Learning for Mitigating Inline Interference in Satellite Constellations},
journal = {International Journal of Satellite Communications and Networking},
year = {2026},
doi = {10.1002/sat.70077}
}
EUVIMED