• Title: AGORA - Machine Learning for Precise GNSS navigation
  • Funder: EUSPA
  • Partners: Rokubun
  • Duration: 18 months

AGORA is a new Machine Learning (ML) powered GNSS technology aiming at improving the performance of GNSS-only navigation solutions in two key areas: multipath resilience and ionospheric delay estimation. An improvement in any of those two areas would result in better performance navigation, especially in accuracy and convergence time. The objective of the AGORA project is to assess quantitatively the benefit of complementing  GNSS with Machine Learning solutions with respect to GNSS-only solutions. This goal will be assessed in two different hot topics of satellite navigation: multipath effect compensation and ionospheric delay estimation.

GNSS, AI, Machine Learning, Deep Learning, Recurrent Neural Networks, Multipath, Ionosphere

AGORA project will result in a set of ML tools oriented to enhance GNSS solutions performance in those two trending challenges in the navigation community. The resulting ML tools will be seamlessly integrated in the value chain of the GNSS user segment solutions to be the seed of a new generation of ML-powered GNSS receivers and augmentation services. AGORA will generate the following new set of ML-powered GNSS technologies:


An ML-based solution to enhance GNSS-only navigation solutions in the presence of the multipath effect. This will be an ML solution providing location corrections to compensate for the effect of multipath in urban scenarios navigation for a given user location. The solution will be focussed on building ML models for Rokubun’s MEDEA GNSS receiver so ad-hoc data acquisition campaigns will be held in Barcelona city to demonstrate the accuracy improvement using Deep Learning techniques.


An ML-based solution for better estimation of Ionospheric delay. This will be a solution providing an estimation of the ionospheric delay at a given time and location of the user, aiming at over-performing the Klobuchar Ionospheric Model, that is to say, aiming at estimating more than 50% of the ionospheric delay which is normally the portion estimated by Klobuchar.

The concept is illustrated in the following figure, where the ML model to be developed would convert a low-resolution image (i.e. map built with the Klobuchar model) to a high-resolution image, that would have a greater level of detail.

Embedding the models into MEDEA

One of the main objectives of the project is to demonstrate the usage of the models in real scenarios. To do so, the models will be deployed into the MEDEA receiver. The key challenge lies in balancing model performance with its size, given the receiver's limited computational resources compared to a server environment.

AGORA Cloud service

Another objective of AGORA is to simplify the access to ML techniques for those devices with low computational power capabilities where embedding ML algorithms is not feasible. The AGORA Cloud service will act as an interface to the trained models for these kinds of devices simplifying the usage of ML in these receivers.