Work

Constraint-Aware Temporal Graph Policy for Beam Hopping Scheduling in LEO Satellite Communications

Year: 2026
Type: conference-paper
Abstract: Beam hopping (BH) is a practical mechanism for low Earth orbit (LEO) satellite payloads to match limited active beams with uneven and time-varying traffic demand. Rule-based scheduling is simple, but it reacts slowly to burst queues, beam-edge channel variation, and resource constraints. This paper presents a constraint-aware temporal graph policy trained with proximal policy optimization (CTG-PPO) for LEO BH scheduling. The method represents service cells as graph nodes, encodes recent demand and link-state trajectories with a causal temporal module, aggregates spatial coupling through graph attention, and projects the selected beam pattern onto active-beam, separation, bandwidth, power, and switching constraints before reward evaluation. A simulation-based verification is built for a normalized Ka-band LEO downlink scenario with 24 candidate cells, six active beams per slot, stochastic hotspot traffic, moderate rain attenuation, and residual Doppler loss after frequency compensation. Across 1800 scheduling episodes, CTG-PPO reaches 95.0% demand satisfaction, 40.8 ms average queueing delay, 2.4% packet drop rate, and a 0.951 Jain fairness index. Stress tests under asymmetric hotspot traffic, training-convergence traces, and full-episode power-utilization statistics are also reported. Compared with an LSTM policy baseline, CTG-PPO improves demand satisfaction by 1.9 percentage points and reduces packet drop rate by 0.8 percentage points. The results indicate that combining temporal learning with explicit resource projection is useful for simulation-based BH scheduling.
Language: English
Open Access status: hybrid
OpenAlex ID: W7204274385
FWCI:
Cites: 18
Cited by:
Related to: 10