Convolutional neural network based approach for estimating ionospheric delay from GNSS observables

Abstract

With our increasing reliance on GNSS-based services for Position, Navigation, and Time (PNT), the end users require higher level corrections, for example on ionospheric delay, for more accurate positioning and navigation applications. The accuracy of the PNT services can be improved by applying correction parameters, or by utilising post-processing. In this paper we introduce a convolutional neural network based solution for estimating the ionospheric delay directly from the GNSS observables measured by the Finnish national reference station network FinnRef. Our model is able to reproduce the general shape of the ionosphere, compared against a global ionospheric map, but the model is overestimating the ionospheric delay derived from the global map. A machine learning model is computationally too heavy to be run at receiver level, however, the ionospheric delay estimates could be broadcast by the monitoring station network to increase situational awareness or as correction parameters for more precise positioning services.