Random forest aided single frequency doppler based GNSS cycle slip detection

Cycle slip detection is a prerequisite for high-precision Global Navigation Satellite System (GNSS) positioning. However, for single-frequency receivers operating at low sampling rates (1 Hz), Doppler-based detection methods are susceptible to failure due to the indistinguishability between cycle slips and slowly varying errors such as atmospheric drift and clock bias. This limitation is particularly pronounced in low-cost consumer devices, where detection ambiguities can severely degrade positioning integrity. To address this, this study proposes a machine learning-aided framework utilizing a Random Forest classifier to assess the reliability of Doppler-based cycle slip detection. The model utilizes a dual-epoch feature set, including Doppler-derived Phase Change Variation (DCV), carrier-to-noise density ( C / N 0 ), and elevation angle, trained and validated on data from consumer-level devices including u-blox ZED-F9P receiver, Google Pixel 4, Google Pixel 7 Pro and Google Pixel 8a. Results demonstrate that while the Doppler method exhibited high failure rates on challenge data, the proposed model achieved strong generalization with a 96.48% recall rate for unreliable epochs. By prioritizing the elimination of false negatives, the proposed method effectively filters out gross errors, serving as a robust integrity augmentation tool for low-cost, single-frequency GNSS applications. • Cycle slip detection is a prerequisite for high-precision Global Navigation Satellite System (GNSS) positioning. • However, for single-frequency receivers operating at low sampling rates (1 Hz), Doppler-based detection methods are susceptible to failure due to the indistinguishability between cycle slips and slowly varying errors such as atmospheric drift and clock bias. • This limitation is particularly pronounced in low-cost consumer devices, where detection ambiguities can severely degrade positioning integrity. • To address this, this study proposes a machine learning-aided framework utilizing a Random Forest classifier to assess the reliability of Doppler-based cycle slip detection. • The model utilizes a dual-epoch feature set, including Doppler-derived Phase Change Variation (DCV), carrier-to-noise density (C/N0), and elevation angle, trained and validated on data from consumer-level devices including u-blox ZED-F9P receiver, Google Pixel 4, Google Pixel 7 Pro and Google Pixel 8a. • Results demonstrate that while the Doppler method exhibited high failure rates on challenge data, the proposed model achieved strong generalization with a 96.48% recall rate for unreliable epochs. • By prioritizing the elimination of false negatives, the proposed method eXectively filters out gross errors, serving as a robust integrity augmentation tool for low-cost, single-frequency GNSS applications.