LEO satellite networks generate predictable RTT spikes every 5–16 seconds due to beam/satellite handovers.
Existing congestion control (BBR/GCC) tends to misinterpret these transitions as congestion, causing large latency oscillations.
This preprint proposes a model-free, dual-loop supervisory controller that uses only RTT signals to anticipate handovers (~1–2s early) and stabilize latency.
It wraps around unmodified BBR v1 and runs in O(1) time (~36 µs per 10 ms cycle).
Single-flow and multi-flow simulations (50k samples per scenario) show reduced p99 latency and improved compliance under Normal/Degraded/Heavy Load LEO conditions.
LEO satellite networks generate predictable RTT spikes every 5–16 seconds due to beam/satellite handovers. Existing congestion control (BBR/GCC) tends to misinterpret these transitions as congestion, causing large latency oscillations.
This preprint proposes a model-free, dual-loop supervisory controller that uses only RTT signals to anticipate handovers (~1–2s early) and stabilize latency. It wraps around unmodified BBR v1 and runs in O(1) time (~36 µs per 10 ms cycle).
Single-flow and multi-flow simulations (50k samples per scenario) show reduced p99 latency and improved compliance under Normal/Degraded/Heavy Load LEO conditions.