Extending Galileo Broadcast Orbits Beyond Valid Ephemeris Using Physics-Constrained Neural Forecasting

Satya S. Vemuri and Jari Nurmi

Peer Reviewed

Abstract:	Global Navigation Satellite Systems (GNSS) rely on broadcast ephemerides for real-time positioning, but their validity is limited, typically 2 hours for GPS and up to 4 hours for Galileo, beyond which orbit prediction errors grow rapidly. This research presents a novel, deployable method to extend the usable validity of broadcast ephemerides using physics-constrained neural forecasting. Residuals are modeled in the Radial-Intrack-Crosstrack (RIC) frame and predicted using deep learning architectures such as BiLSTM and N-HiTS with direct multi-horizon outputs. Forecasts are softly constrained by a two-body orbital model with J2 perturbation, enhancing physical consistency without overriding data-driven learning. A two-pass training strategy explicitly addresses both valid and stale ephemeris regimes, enabling reliable orbit corrections up to 6–24 hours beyond the nominal Time of Ephemeris. Results for the Galileo constellation show that the BiLSTM-based forecaster suppresses broadcast orbit drift by approximately 98–99.7%, while N-HiTS provides moderate improvement with horizon-specific trade-offs. This approach offers a practical solution for real-time receivers and bridging gaps between precise orbit products, without inheriting the degradation of stale broadcast data. The proposed PyTorch-based pipeline demonstrates efficient orbit prediction capabilities, requiring only 2 hours of training data to forecast satellite trajectories up to 24 hours ahead. Leveraging GPU acceleration (NVIDIA A6000), full-constellation training and prediction are completed in under 5 minutes. This enables real-time SP3 ingestion and correction, supporting low-latency GNSS integrity monitoring.
Published in:	Proceedings of the 2026 International Technical Meeting of The Institute of Navigation January 26 - 29, 2026 Hyatt Regency Orange County Anaheim, California
Pages:	726 - 740
Cite this article:	Vemuri, Satya S., Nurmi, Jari, "Extending Galileo Broadcast Orbits Beyond Valid Ephemeris Using Physics-Constrained Neural Forecasting," Proceedings of the 2026 International Technical Meeting of The Institute of Navigation, Anaheim, California, January 2026, pp. 726-740. https://doi.org/10.33012/2026.20520
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In