Precise Global Navigation Satellite System (GNSS) orbit prediction is critical for real-time positioning applications. Current orbit prediction accuracy for the BeiDou Navigation Satellite System-3 (BDS-3) exhibits a notable disparity compared to GPS and Galileo, with limited advancements from traditional dynamic modeling approaches. This study introduces a novel data-driven methodology, Sample Convolution and Interaction Network with Self-Attention (SCINet-SA), to augment dynamic methods and improve BDS-3 ultra-rapid orbit prediction. SCINet-SA leverages deep learning to model the temporal characteristics of orbit differences between BDS-3 ultra-rapid and final products. By training on historical orbit difference data, SCINet-SA predicts future discrepancies, facilitating the refinement of ultra-rapid orbit estimates. The incorporation of a self-attention mechanism within SCINet-SA enables the model to effectively capture long-range temporal dependencies, thereby enhancing long-term prediction capabilities and mitigating the latency associated with final product availability. Rigorous experimental evaluation demonstrates the superior performance of SCINet-SA in enhancing BDS-3 ultra-rapid orbit prediction accuracy relative to alternative deep learning models. Specifically, SCINet-SA achieved the highest average relative improvement (IMP) in 3D Root Mean Square (RMS) error across 1 d, 7 d, and 15 d prediction horizons, yielding improvements of 21.69%, 18.66%, and 15.42%, respectively. The observed IMP range spanned from 7.78% to 38.91% for 1 d, 4.34% to 35.96% for 7 d, and 1.68% to 31.13% for 15 d predictions, underscoring the efficacy of the proposed methodology in advancing BDS-3 orbit prediction accuracy.
Broadcast ephemeris data are essential for the precision and reliability of the BeiDou Navigation Satellite System (BDS) but are highly susceptible to anomalies caused by various interference factors, such as ionospheric and tropospheric effects, solar radiation pressure, and satellite clock biases. Traditional threshold-based methods and manual review processes are often insufficient for detecting these complex anomalies, especially considering the distinct characteristics of different satellite types. To address these limitations, this study proposes an automated anomaly detection method using the IF-TEA-LSTM model. By transforming broadcast ephemeris data into multivariate time series and integrating anomaly score sequences, the model enhances detection robustness through data integrity assessments and stationarity tests. Evaluation results show that the IF-TEA-LSTM model reduces the RMSE by up to 20.80% for orbital parameters and improves clock deviation prediction accuracy for MEO satellites by 68.37% in short-term forecasts, outperforming baseline models. This method significantly enhances anomaly detection accuracy across GEO, IGSO, and MEO satellite orbits, demonstrating its superiority in long-term data processing and its capacity to improve the reliability of satellite operations within the BDS.
18 November 2025 in IEEE Aerospace and Electronic Systems Magazine
Satellite clock bias (SCB) significantly affects the accuracy of satellite-based positioning, navigation, and timing (PNT) services. While the International GNSS Service (IGS) provides real-time service (RTS) products with precise SCB pre dictions, these products can be unavailable due to interruptions or network issues. To address this, we propose SCITrans, an offline, high-precision SCB prediction model based on deep learning. SCITrans combines a Transformer architecture with inverted data embedding and a sample convolution and interaction (SCI) module. The SCITrans solution leverages the time invariance of downsampled time series to effectively extract temporal features of SCB sequences, while enhancing the capture of long-term dependencies. In experiments involving BDS-3 SCB prediction, the SCITrans model achieved root mean square error (RMSE) values of 0.052 ns, 0.108 ns, 0.143 ns, and 0.166 ns for prediction horizons of 1, 3, 6, and 12 hours, respectively. The total training and prediction time for a single satellite is up to 163 seconds. In BDS-3 kinematic precise point positioning (PPP), SCITrans based solutions converge to a 15 cm level in 14 minutes. A 30-day analysis across eight stations shows a 3D RMS of 8.5-46.3 cm for 9-hour sessions.