MSCA-COFUND-CLEAR-Doc-PhD Position#CD22-43: GNSS noise prediction based on deep networks with the supervision of 3D city models

With the emergence of autonomous vehicles, performance requirements for localisation solutions are strongly growing. Indeed, the localisation information needs to be provided with accuracy, but also availability, and more and more with integrity, i.e. with a certain guarantee on the level of performance provided by the system, in particular when dealing with safety of the users.

GNSS positioning systems are widely used because of their (cheap) cost, their provision of an absolute position, wherever and whenever the user needs. According to the market report released by EUSPA (EU Agency for the Space Programme) in 2019, the revenue of GNSS downstream market is reaching €150 billion and out of 1.7 billion GNSS shipped units in 2019, more than 40% will be Galileo. In principle, the GNSS receivers rely on the measurement of time of arrival of several satellite signals simultaneously, in order to estimate the pseudo-range, i.e. distance between each satellite and the user’s receiver, for position estimation.

In most of the usages today, the user is an urban user. The city, unfortunately, generates an unexpected noise for GNSS. Indeed, buildings, trees and interferences, called local effects, degrade the measurements, thus the positioning performance. This is one of the major bottlenecks to extend the GNSS application in urban areas.

This noise is composed of multipath, masking effects, or interferences created by other systems in the surroundings. It has multiple causes and dependencies: the close environment of the user, antenna choice and installation, time (as GNSS satellites are moving around the earth), user dynamics… It can also differ from one receiver to another one due to various solutions implemented in the receiver, usually used as a black box by the end user. The noise error model has a non-zero mean, large variance, can be non-symmetric with heavy tails.

Although noise and in particular multipath noise concentrates huge efforts from the scientific community, the urban noise remains very complex to model. So many parameters are involved that it is difficult to model it considering physics or mathematics only. Looking at the challenges of GNSS multipath and at the potential of Machine Learning algorithms, it becomes straightforward to investigate the benefits of the artificial intelligence in the GNSS domain.

First Machine Learning related publications focus on multipath and NLOS mitigation [Hsu 2017][Suzuki 2017]; some others on context detection [Gao 2020] (is the user in a city, in a rural environment…?), spoofing detection [Semanjski 2020]. When signal processing can be accessed, some proposals apply machine learning to NLOS Multipath Detection [Suzuki 2020] or Jammer Classification [Ferre 2019] but this is not available from COTS receivers.

In this study, we intend to focus on the use of Machine learning to model the noise observed in urban environments [Lee 2022]. These models are of primary use for a better localization as previous studies shown that accurate noise modelling can enhance GNSS accuracy [Marais 2010], as well as multi-sensor fusion [Zhang 2018][Chen 2020]. They can be used in enhanced integrity monitoring concepts bounding the errors [No 2021].

The use of these technics raises several issues among which the choice of the relevant features that will characterize the noise the better; the availability of data and 3D building model of large reference database for learning processes; the authentication of the labelled multipath data; the scalability of the datasets and models obtained considering that a city on Hong Kong strongly differ from one in France in terms of masking effects…

The candidate will be part of the LEOST lab from the university Gustave Eiffel and will be welcomed inside the Intelligent Positioning and Navigation Laboratory from the Hong Kong Polytechnic University for the mobility stay. Both teams have a strong scientific background on GNSS, experimentation capacities, as well as the experience of collaborative international projects with academics as well as with industry. He/She will benefit from the past experience of both teams on GNSS local effects detection and mitigation studies, as well as their international network to build his/her research work and prepare the future.

References

Chen P.Y., Chen H., Tsai M.H., Kuo H.K., Tsai Y.M., Chiou T.Y., Jau P.H. “Performance of Machine Learning Models in Determining the GNSS Position Usage for a Loosely Coupled GNSS/IMU System,” ION GNSS+ 2020, virtually, September 21-25, 2020.

Ferre, R. M., Fuente, A. D. La, & Lohan, E. S. (2019). Jammer classification in GNSS bands via machine learning algorithms. Sensors (Switzerland), 19(22). https://doi.org/10.3390/s19224841

Gao H, Groves PD. (2020) Improving environment detection by behavior association for context-adaptive navigation. NAVIGATION, 67:43–60. https://doi.org/10.1002/navi.349

Guohao Zhang, Li-Ta Hsu, Intelligent GNSS/INS integrated navigation system for a commercial UAV flight control system, Aerospace Science and Technology, Volume 80, 2018, Pages 368-380. https://doi.org/10.1016/j.ast.2018.07.026.

Lee Y., Park B., (2022) Nonlinear Regression-based GNSS Multipath Map Construction and Its Dynamic Application in Deep Urban Area, ION GNSS+, 21-23 sept. Denver, USA.

Marais J., Viandier N., Rabaoui A., Duflos E., GNSS multipath bias models for accurate positioning in urban environments, ITST 2010, 9-11 nov. 2010, Kyoto, Japon.

Hsu L.T. “GNSS Multipath Detection Using a Machine Learning Approach,” IEEE ITSC 2017, Yokohama, Japan.

No H., Milner C., Machine Learning Based Overbound Modeling of Multipath Error for Safety Critical Urban Environment, ION GNSS+ 2021, virtual, September 2021.

Semanjski, S., Semanjski, I., De Wilde, W., & Muls, A. (2020). Use of supervised machine learning for GNSS signal spoofing detection with validation on real-world meaconing and spoofing data—Part I. Sensors, 20(4), 1171.

Suzuki T., Nakano, Y., Amano, Y., NLOS Multipath Detection by Using Machine Learning in Urban Environments, ION GNSS+ 2017, Portland, Oregon, pp. 3958-3967.

Zhang, G., Xu, P., Xu, H., & Hsu, L. T. (2021). Prediction on the urban GNSS measurement uncertainty based on deep learning networks with long short-term memory. IEEE Sensors Journal, 21(18), 20563-20577.