Global Navigation Satellite System (GNSS) ephemerides contain important information about GNSS satellites. They provide essential data for Single Point Positioning (SPP) and Real-time Kinematic (RTK) users to compute satellite positions and clock errors at specific times for positioning. The International GNSS Service (IGS) collects ephemerides from global GNSS stations and combine them to generate daily broadcast ephemeris files for post-processing users. However, these daily ephemeris files are occasionally contaminated by anomalies, impairing the integrity of GNSS systems and severely degrading positioning performance.
Global Navigation Satellite System (GNSS) ephemerides contain important information about GNSS satellites. They provide essential data for Single Point Positioning (SPP) and Real-time Kinematic (RTK) users to compute satellite positions and clock errors at specific times for positioning. The International GNSS Service (IGS) collects ephemerides from global GNSS stations and combine them to generate daily broadcast ephemeris files for post-processing users. However, these daily ephemeris files are occasionally contaminated by anomalies, impairing the integrity of GNSS systems and severely degrading positioning performance.
In this study, daily GNSS ephemerides over several years will be downloaded from the IGS. To identify abnormal ephemeris records, precise satellite orbit and clock products on the same day will be used to evaluate the error of each record. Ephemerides with significant evaluation error will be labeled as abnormal. The Keplerian elements, clock polynomials and other ephemeris parameters can be used to form input features. The student will design a machine learning (ML)-based classification model to automatically identify abnormal GNSS ephemerides. This study will also investigate and compare the classification performance of different ML models, e.g., random forest and support vector machine (SVM). Upon successful completion of this study, the ML-based classifier could be utilized to generate clean daily ephemeris files operationally, benefiting the GNSS community.
In this study, daily GNSS ephemerides over several years will be downloaded from the IGS. To identify abnormal ephemeris records, precise satellite orbit and clock products on the same day will be used to evaluate the error of each record. Ephemerides with significant evaluation error will be labeled as abnormal. The Keplerian elements, clock polynomials and other ephemeris parameters can be used to form input features. The student will design a machine learning (ML)-based classification model to automatically identify abnormal GNSS ephemerides. This study will also investigate and compare the classification performance of different ML models, e.g., random forest and support vector machine (SVM). Upon successful completion of this study, the ML-based classifier could be utilized to generate clean daily ephemeris files operationally, benefiting the GNSS community.
Yuanxin Pan (yxpan@ethz.ch)
Marcel Iten (miten@ethz.ch)
Prof. Benedikt Soja (soja@ethz.ch)
Yuanxin Pan (yxpan@ethz.ch) Marcel Iten (miten@ethz.ch) Prof. Benedikt Soja (soja@ethz.ch)