A machine learning-based partial ambiguity resolution method for precise positioning in challenging environments

Abstract

Partial ambiguity resolution (PAR) has been widely adopted in real-time kinematic (RTK) and precise point positioning with augmentation from continuously operating reference station (PPP-RTK). However, current PAR methods, either in the position domain or the ambiguity domain, suffer from high false alarm and miss detection, particularly in challenging environments with poor satellite geometry and observations contaminated by non-line-of-sight (NLOS) effects, gross errors, biases, and high observation noise. To address these issues, a PAR method based on machine learning is proposed to significantly improve the correct fix rate and positioning accuracy of PAR in challenging environments. This method combines two support vector machine (SVM) classifiers to effectively identify and exclude ambiguities those are contaminated by bias sources from PAR without relying on satellite geometry. The proposed method is validated with three vehicle-based field tests covering open sky, suburban, and dense urban environments, and the results show significant improvements in terms of correct fix rate and positioning accuracy over the traditional PAR method that only utilizes ambiguity covariances. The fix rates achieved with the proposed method are 93.9%, 81.9%, and 93.1% with the three respective field tests, with no wrong fixes, compared to 72.8%, 20.9%, and 16.0% correct fix rates using the traditional method. The positioning error root mean square (RMS) is 0.020 m, 0.035 m, and 0.056 m in the east, north, and up directions for the first field test, 0.027 m, 0.080 m, and 0.126 m for the second field test, and 0.035 m, 0.042 m, and 0.071 m for the third field test. In contrast, only decimeter- to meter-level accuracy was obtainable with these datasets using the traditional method due to the high wrong fix rate. The proposed method provides a correct and fast time-to-first-fix (TTFF) of 3–5 s, even in challenging environments. Overall, the proposed method offers significant improvements in positioning accuracy and ambiguity fix rate with high reliability, making it a promising solution for PAR in challenging environments.