The trajectory of vehicle-mediated experiments at Google Earth. Panel A presents the entire bird's-eye view of the experiment, where panels B and C are snapshots of vehicles starting and driving under the elevated railway, respectively. (Credit: Satellite navigation)
High-precision GNSS applications such as real-time displacement monitoring and vehicle navigation rely heavily on solving carrier phase ambiguity. However, traditional methods such as R-ratio and W-ratio testing often use empirical thresholds, which can lead to unreliable results due to bias and environmental variation.
These limitations hinder the efficiency of accurate point positioning of ambiguity resolution (PPP-AR), particularly in dynamic or challenging conditions. Based on these challenges, more robust and adaptive techniques need to be developed for the verification of ambiguity.
Released on June 9, 2025 (doi: 10.1186/s43020-025-00167-8) Satellite navigationresearchers at the Royal Observatory in Belgium and the state major laboratory of China's precision geodetic law have announced a Support Vector Machine (SVM)-based method for GNSS Ambiguity verification.
This study utilizes machine learning to combine multiple diagnostic metrics to achieve higher accuracy and reliability than traditional approaches. This model is trained on a wide range of datasets and verified through real experiments, demonstrating the possibility of transforming high-precision positions.
The key innovations in this study are integrated into the SVM model into the integration of seven diagnostic indicators including R-Ratio, adoption, and ambiguity dimensions. This approach addresses the limitations of traditional methods that rely on a single threshold and cannot explain complex dependencies between variables.
The SVM model achieved a success rate of 92% on ambiguity verification, surpassing 82% of R-Ratio tests in Kinematic scenarios. In particular, the model reduced the convergence time prediction error to just 1.0 minutes compared to the 5.0 minutes of the traditional method.
The highlights of the study are as follows:
- Improved reliability. The ability to adaptively weight multiple metrics ensures a more consistent resolution of ambiguity.
- Real-world verification. Vehicle-mediated experiments demonstrated a success rate of 92%, demonstrating the practicality of the method in a dynamic environment.
- Scalability. This framework adapts to both single and composite GNSS systems and expands applicability.
Despite its advances, the study acknowledges an error rate of 5% of unresolved ambiguity, pointing to future research directions.
“Our SVM model represents a paradigm shift in the verification of ambiguity,” emphasized Jianghui Geng, co-author of the study. “By leveraging machine learning, we not only improved accuracy, but also provided scalable solutions for a wide range of GNSS applications, from autonomous vehicles to geodetic monitoring.”
SVM-based methods have great promise for industries that require ultra-fast positioning, including autonomous navigation, aerospace and infrastructure monitoring. The ability to reduce convergence times and increase reliability can revolutionize real-time GNSS applications, particularly in urban or occlusion environments where signal disruption is common.
Future iterations of models that incorporate additional data layers can further bridge the gap between theoretical accuracy and actual performance, and set new standards for GNSS technology.
