Novel Robust Labeling Methods Applied to GNSS Signal Classification Based on Deviation and Differential

Global navigation satellite system (GNSS) signals are highly susceptible to surrounding conditions in complex urban environments, resulting in multipath (MP) effects and non-line-of-sight (NLOS) reception. These phenomena significantly compromise positioning precision in GNSS-based applications. Consequently, the detection and suppression of MP and NLOS have become the focus of research. This study delves into the issue of signal type labeling in the emerging research area of utilizing supervised learning for GNSS signal classification. Two novel robust labeling methods are proposed, namely receiver clock error non-excluded pseudorange error deviation (NEPED) and differential receiver clock error non-excluded pseudorange error (DNEPE). In the experimental validation process, we conducted tests on the accuracy and robustness of labels based on dynamic GNSS data collected by the Google team in the San Francisco Bay Area, and comparative analyses were performed with three other methods. The experimental results indicate that both NEPED and DNEPE exhibit superior accuracy and robustness. In particular, DNEPE consistently demonstrates much better performance across all experiments in comparison to the alternatives and is a more recommended method.