Real-time uncertainty quantification using correlated noise models for GNSS positioning
A. D. Martin, Andrew Walden Ross Soundy, Bradley J. Panckhurst, C. P. Brown, Dániel Schumayer, T. C. A. Molteno, Matthew F. Parry · 2017
We develop Bayesian algorithms to perform realtime positioning and uncertainty quantification on Global Positioning System (GPS) data. We test the algorithms on GPS data from several global locations and score their predictions using the log-score. The best algorithm is a Kalman filter that assumes an Ornstein-Uhlenbeck (OU) noise model. The OU model accounts for the observed autocorrelated process-noise in the latitude, longitude and altitude measurements. It outscores both a Kalman filter that assumes an independent and identically distributed (iid) Gaussian noise model (white noise), and a basic method that uses the raw positions with a constant uncertainty estimate. Inference with the OU noise model enables reliable real-time position inference with improved uncertainty quantification, and is particularly suitable for sensor-fusion applications.