Machine learning algorithms for FCB (fractional cycle bias) estimation in PPP ambiguity resolution

Abstract

This study introduces a novel machine learning–based framework for estimating fractional cycle biases (FCBs) to enhance ambiguity resolution in precise point positioning with ambiguity resolution (PPP-AR). While previous studies have relied on traditional models such as the single difference between satellites (SDBS) technique, our work is the first to modify this model by integrating supervised learning algorithms—specifically support vector machine (SVM) and random forest (RF)—to improve the precision of FCB estimation. The key novelty lies in enabling accurate estimation of even low-magnitude FCB values, which has a direct impact on shortening the convergence time—a known limitation of PPP techniques. Experimental evaluations using real GNSS datasets demonstrate that the SVM-based model significantly outperforms both RF and traditional SDBS approaches in FCB estimation accuracy. These findings establish a new direction for improving PPP-AR performance using data-driven methods, making the approach highly relevant for real-time geodetic and navigation applications where rapid convergence is critical.