OATAO - Open Archive Toulouse Archive Ouverte Open Access Week

Modelling of the ionosphere by neural network for equatorial SBAS

Desert, Thibault and Authié, Thierry and Trilles, Sébastien Modelling of the ionosphere by neural network for equatorial SBAS. (2015) In: ION GNSS+ 2015, 14 September 2015 - 18 September 2015 (Tempa, United States).

[img]
Preview
(Document in English)

PDF (Author's version) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
520kB

Abstract

The estimation of the ionosphere delay and associated confidence interval constitutes the major issue to reach APV1 availability performance level for single frequency SBAS above the equatorial area. The ionosphere is a complex physical system which dynamics is particularly disturbed at the Geomagnetic Equator while mid-latitude regions are quieter. Classical methods to compute ionosphere delays, such as those implemented in EGNOS and the WAAS, are specific to a smooth ionosphere behavior and are not really adapted to follow high spatial and temporal gradients, such as those observed in the equatorial area. Thus innovative methods, having flexible and reactivity qualities, shall be defined and adapted to propose efficient equatorial SBAS. Classically in SBAS concept, the knowledge of the ionosphere delay is obtained by a set of lines-of-sight between the network of ground stations and the navigation satellite constellation. Each line of sight intersects the ionosphere layer, assumed infinitely thin, and the dual-frequency combination allows to compute, at first order, the ionosphere delay that affects the GNSS measurements. From this set of heterogeneous information, locally sampled irregularly on the sphere and changing over time, we propose to build an interpolating method to calculate the ionosphere delay on a point of interest by an adaptive mesh, unlike fixed grids usually used. In this paper, we introduce an interpolation method based on the definition of a flexible network that can adapt to the spatial location of the data. It is therefore proposed to create self-organizing maps – as the Kohonen map – defined by a mesh that fits the data. This network ends up sticking to the data by "learning" in real time: the mesh becomes denser and denser in the presence of many measurements and relaxes otherwise. This technique increases the granularity of the ionosphere delay information to compute, in particular to be able to describe the local plasma bubbles or depletions, if they are observable. The adaptive array technology "learning" has been widely studied in the field of modeling neural networks. Their main advantage is to be able to reach an optimal state based on the information they process. The experimentation based on this technique shows a very good behavior in the case of strongly disturbed ionosphere conditions and the preliminary results are promising to bring the expected robustness to deploy SBAS in equatorial area.

Item Type:Conference or Workshop Item (Paper)
HAL Id:hal-01218339
Audience (conference):International conference proceedings
Uncontrolled Keywords:
Institution:Other partners > Thales (FRANCE)
Funders:
DNF
Statistics:download
Deposited By: Sébastien Trilles
Deposited On:21 Oct 2015 06:14

Repository Staff Only: item control page