Machine Learning based Characterization of GPS Satellite Oscillator Anomaly
Yunxiang Liu, and Y. T. Jade Morton, Smead Aerospace Engineering Sciences Department, University of Colorado Boulder
Location: Beacon A
Date/Time: Thursday, Jan. 27, 2:12 p.m.
This paper applies a machine learning-based satellite oscillator anomaly detection method to high rate GNSS receiver carrier phase data collected by the University of Colorado Boulder Satellite Navigation and Sensing Lab network. The result is used to characterize the GPS satellite oscillator anomaly from 2013 to 2018. A total of 5328 days of data from 20 stations distributed all over the world are processed and 224361 satellite oscillator anomaly events are detected. On average, 42 events are detected per station per day. For those detected anomalies that are verified by multiple stations, the histogram of maximum L1 phase deviation, distribution of event duration, and distributions of inter-event times, as well as the magnitude of anomalies distributions are investigated. The results show that most of the detected anomalies are micro anomalies and nearly all verified anomalies are observed from block IIF satellites with a Rubidium clock.