Politecnico di Torino (logo)

SINS/GNSS Tighty Coupled Integration based on a Radial Basis Function Neural Network

Giovanni Prestigiacomo

SINS/GNSS Tighty Coupled Integration based on a Radial Basis Function Neural Network.

Rel. Fabio Dovis, Falin Wu. Politecnico di Torino, Corso di laurea magistrale in Communications And Computer Networks Engineering (Ingegneria Telematica E Delle Comunicazioni), 2020

PDF (Tesi_di_laurea) - Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives.

Download (2MB) | Preview

The following thesis dissertation has been carried out at the Beijing's University of Aeronautics and Astronautics (Beihang University), with the aid of the SNARS Research Group and of professors Falin Wu and Fabio Dovis. The aim of this thesis project is to investigate techniques capable to improve the performances of a tightly coupled SINS/GNSS Integrated System with the assistance of a Radial Basis Function Neural Network. Indeed, precise and accurate navigation is nowadays required also for civil applications. The combination of a INS (Inertial Navigation System) and of a GNSS (Global Navigation Satellite System) is already able to improve the performances of such an integrated navigation system. Unfortunately GNSS outages are still not predictable in their length and their occurrence because they strongly depend on the satellites visibility. As a consequence, there is the strong necessity to understand how to compensate the effects of those events, during which all the responsibility is left on the INS. A stand-alone INS is indeed known for being reliable, since it works without the aiding of a network, but not accurate on the long run, because the approximated results drift with time even if the errors committed by the equipped instrumentation is negligible. Tightly coupled SINS/GNSS systems are able to integrate the solutions of both the systems combined together by working on a pseudo-range level. In this context, Neural Networks are employed to substitute in the most accurate way the GNSS during its outages, providing pseudo-range approximations which are consequently merged with the INS solutions time by time. The aim of the approach investigated in the following thesis is therefore to avoid the navigation solution to drift over time during a GNSS outage. This result is accordingly obtained by substituting the GNSS performances with a RBFNN (Radial Basis Function Neural Network), also known with the acronym RBNN, which is known to be suitable for real-time applications, to have a simple topology and also to require not a lot of efforts to be trained. In order to simplify the simulation algorithms carried out on Matlab, the model of the new SINS/GNSS Integrated System has been built separately with two different block schemes. The first generalized scheme represents the system as long as the GNSS is working properly. This scheme is similar to the classical tightly coupled implementation without the Neural Network aiding. The pseudo-ranges retrieved by the GNSS are combined with the ones approximated from the INS solution and then their difference is used to feed the Kalman Filter and also the RBNN, only if the system is built with an online training approach. The output of the Kalman Filter represents the error committed by the INS and consequently it is subtracted from its navigation solution in order to retrieve the corrected PVT solution.

Relators: Fabio Dovis, Falin Wu
Academic year: 2019/20
Publication type: Electronic
Number of Pages: 47
Corso di laurea: Corso di laurea magistrale in Communications And Computer Networks Engineering (Ingegneria Telematica E Delle Comunicazioni)
Classe di laurea: New organization > Master science > LM-27 - TELECOMMUNICATIONS ENGINEERING
Ente in cotutela: Beihang University (CINA)
Aziende collaboratrici: UNSPECIFIED
URI: http://webthesis.biblio.polito.it/id/eprint/14407
Modify record (reserved for operators) Modify record (reserved for operators)