Real-time precise point positioning (RT-PPP) depends on accurate satellite clock products, but phase jumps, frequency jumps and complex clock anomalies can destabilize network clock estimation. This study proposes and validates an anomaly-aware real-time Global Navigation Satellite System (GNSS) satellite clock estimation framework that combines atomic clock modeling with anomaly identification, isolation and recovery inside a network solution. The aim is to limit individual clock faults from propagating into the network time reference while preserving real-time operation. Tests with actual GNSS data show effective mitigation of multiple anomaly types and a more stable, resilient timescale for high-accuracy positioning and timing services.
Modern spaceborne atomic clocks are more stable, making model-based clock estimation attractive for real-time precise point positioning. However, conventional clock models impose inter-epoch constraints and cannot respond adequately to abrupt phase or frequency changes. Such anomalies can contaminate correlated parameters, degrade clock estimates and even cause filter divergence. Existing detection methods often work on available clock series rather than within a real-time network solution, where anomaly indicators must be derived before abnormal clocks affect estimation and state constraints. Against these challenges, in-depth research is needed to integrate robust anomaly detection and mitigation into model-based real-time Global Navigation Satellite System (GNSS) clock estimation.
A Wuhan University-led research team, including Feiyu Mao, Xiaopeng Gong, Shengfeng Gu and Yidong Lou, published (DOI: 10.1186/s43020-026-00225-9) the study online on 21 October 2026 in Satellite Navigation . The paper presents an anomaly-aware framework for real-time GNSS satellite clock estimation, combining a four-state atomic clock model with detection, isolation, compensation and reset strategies. It also evaluates confirmation parameters, computational efficiency and real-time performance across the U.S. Global Positioning System (GPS), Russia's Global Navigation Satellite System (GLONASS), BeiDou-3 (BDS-3) and Galileo. The work targets phase jumps, frequency jumps and complex anomalies that can disrupt network time references.
The framework first generates clock prediction residuals from GNSS network preprocessing and an epoch-wise least-squares (LSQ) adjustment. Normalized residuals and normalized frequency corrections are then monitored. A phase jump is confirmed when consecutive outliers fit a constant-offset model; a frequency jump is confirmed when a significant linear trend appears or frequency corrections stay persistently large. Identified anomalies trigger phase and/or frequency compensation, temporary exclusion, additional process noise, and reduced clock-state constraint weights. A reset mechanism handles complex cases that match neither jump model. A robust weighting function keeps full weight below three sigma (3σ), down-weights linearly from 3σ to five sigma (5σ), and assigns zero weight at or above 5σ. Monte Carlo tests selected confirmation windows of seven and nine epochs, with no false alarms in 10,000 nominal trials. Sensitivity analysis showed combined identification accuracy above 99% for high-level anomalies at the selected settings. Case studies captured phase jumps, frequency jumps and severe complex anomalies in satellite and station clocks. The method also prevented uncorrected anomalies from propagating through weighted zero-sum clock-state constraints into other clocks. In real-time tests, average processing took 2.81 s per epoch, below the 5-s update interval.
The authors said the key advance is not simply a new clock model but a practical quality-control layer built around it. They said the framework lets a network recognize when a clock stops behaving like a clock, isolate the faulty signal before it spreads, and bring the clock back into service after compensation. They emphasized that real-time feasibility mattered throughout: the method had to fit within a 5-s update cycle while preserving a stable network time reference. They described the result as a step toward more resilient GNSS timing infrastructure rather than a single-case fix.
The approach is aimed at real-time precise point positioning (RT-PPP), autonomous driving, smart farming, unmanned aircraft and other services that need reliable positioning and timing. By keeping the network timescale stable during clock anomalies, it can help timing users avoid sudden degradation. The researchers also note that users of state-space representation (SSR) corrections should temporarily exclude a satellite when its clock correction is unavailable during an anomaly and reintroduce it after valid corrections return. The framework supports more resilient GNSS timing infrastructure, though operational deployment will require tuning for specific networks and clock populations.
