Threshold-Gated Conflict-Aware Arc Selection for Satellite Task Scheduling: Evidence from Starlink Mega-Constellation Simulations

Abstract

Greedy schedulers are widely used for low-Earth-orbit satellite task scheduling for their computational efficiency and determinism. However, they select candidate arcs by primary criteria alone, so near-equivalent alternatives are separated arbitrarily, potentially reducing scheduling flexibility for later tasks under contention. We present FAMAS-G, a lightweight conflict-aware extension of greedy scheduling rather than a new optimization framework. Its threshold-gated conflict-aware arc selection activates only when primary scores cannot clearly distinguish among candidates, applying a bounded conflict adjustment that favors lower-conflict arcs without displacing clearly superior ones. It relies only on pre-computed arc-level conflict degrees, adding no message passing, backtracking, or iterative search. We evaluate FAMAS-G against Greedy-Central, Base-CNP, and the original FAMAS in Starlink mega-constellation simulations at 100, 300, and 500 tasks per 6 h period across four TLE epochs. Hierarchical bootstrap analysis shows statistically supported improvements over Greedy-Central at all scales (ΔWFR=+0.0111, +0.0051, and +0.0024), with the advantage attenuating as contention rises. Ablation isolates the conflict-aware tiebreak as the primary positive contributor; urgent priority shows no measurable effect, whereas the risk tiebreak shows a small but significant negative effect. Lightweight conflict awareness thus improves greedy scheduling under contention while retaining single-pass, sub-second execution. Significance at the largest scale is sensitive to epoch inclusion; results are specific to the evaluated Starlink scenarios and require further validation.