Please ensure that JavaScript is enabled in your browser to view this page.
Researchers from the Royal Observatory of Belgium and the State Key Laboratory of Precision Geodesy at the Chinese Academy of Sciences (CAS) have developed a new algorithm that, they say, can deliver a boost to the precision achievable from existing Global Navigation Satellite Systems (GNSS).
"Our SVM [Support Vector Machine] model represents a paradigm shift in ambiguity validation," claims co-author Jianghui Geng of the team's work, which does not require any change to the hardware of satellites nor their planetside receivers. "By harnessing machine learning, we've not only improved accuracy but also provided a scalable solution for diverse GNSS applications, from autonomous vehicles to geodetic monitoring."
The team's algorithm is designed for so-called "ambiguity resolution," the process of resolving uncertainties in carrier-phase signals in order to increase the precision of a GPS or other GNSS fix. Compared to existing approaches, the team says its Support Vector Machine delivers improvements to both accuracy and reliability when delivering Precise Point Positioning Ambiguity Resolution (PPP-AR) — necessary for high-precision tasks like autonomous vehicle navigation.
The trick: integrating seven different diagnostic metrics into a single model. With a machine learning model trained on real-world GPS data and tested on more, the team showed a boost from an 82 percent success rate for kinematic-scenario (temporary loss or degradation of GNSS signal) vehicle navigation to 92 percent — with convergence time prediction errors reduced from five minutes to just one.
"While the machine learning-based model improves the ambiguity validation success rate," the researchers admit, "especially for the solutions in the converging period, we acknowledge that about five percent of wrongly resolved ambiguities cannot be identified by the current model. Future research will focus on addressing these false cases by incorporating variance–covariance information into the model."
The team's work has been published in the journal Satellite Navigation under open-access terms; the SVM model itself is available from the paper's authors upon request.
Hackster.io, an Avnet Community © 2025

source