GPS Orbital Prediction Using Artificial Neural Networks

Hamad Yousif and Ahmed El-Rabbany

Abstract:	Precise real-time and near real-time GPS orbit is required for a number of applications, including real-time Precise Point Positioning (PPP), long range RTK and weather forecast. At present, users may take advantage of the predicted part of the IGS ultra-rapid orbit for real-time applications. Unfortunately, however, the precision of the predicted part of the ultra-rapid orbit is limited to about 10 cm for the 24-hour predicted part, which may not be sufficient for the above applications. In this paper an improved 6-hour predicted orbit is developed in three steps. First, an initial predicted orbit is generated by extrapolating a concatenated group of previous precise ephemeris for 5 days. GPS observations from 35 globally distributed tracking stations, corresponding to the 24-hour period preceding the predicted part, are then utilized within the Bernese software to make further enhancement to the predicted orbit. Finally, the predicted orbit is refined by implementing a modular, three-layer feed-forward backpropagation neural network. A comparison is made between our predicted orbit and the IGS ultra-rapid orbit to verify the efficiency of the newly developed neural network-based model. It is shown that the newly developed neural network-based model improved the orbit prediction by up to 47%, 22%, and 37% for three randomly-selected satellites from Blocks IIA, IIR and IIR-M, respectively.
Published in:	Proceedings of the 2008 National Technical Meeting of The Institute of Navigation January 28 - 30, 2008 The Catamaran Resort Hotel San Diego, CA
Pages:	773 - 780
Cite this article:	Yousif, Hamad, El-Rabbany, Ahmed, "GPS Orbital Prediction Using Artificial Neural Networks," Proceedings of the 2008 National Technical Meeting of The Institute of Navigation, San Diego, CA, January 2008, pp. 773-780.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In