072 - Navigation Using Machine Learning Applied to Signals of Opportunity

072 - Navigation Using Machine Learning Applied to Signals of Opportunity


There is an existing need to optimise navigation in challenging environments, such as deep urban or indoor, and to offer alternative PNT. For this purpose, the user may complement the GNSS with space-based (e.g. LEO SatCom) or terrestrial-based (e.g. cellular) signals of opportunity (SOOP) to optimise the PNT performance. SOOP could also be exploited similarly in a maritime environment.


Traditionally, a Kalman Filter is used to integrate the sensor data in the navigation kernel. However, the use of Machine Learning may provide numerous benefits when integrating SOOP in specific situations:

  • when the signal characteristics (e.g. modulation, location of transmitter, etc.) are not accurately known;
  • when the statistical distribution of data is non-Gaussian (e.g. in multipath-rich environments).

Objective: To optimize PNT performance in challenging environments and offer alternative PNT. This shall be achieved by integrating GNSS and SOOP, or using SOOP only, applying Machine Learning.

Starting point: Traditional integration techniques (e.g. Kalman Filter), serving as benchmarks.

Description of innovation/Tasks:  The activity will explore innovative use of Machine Learning for positioning and sensor integration at user level using SOOP, whose characteristics are not perfectly known, in challenging environments (e.g. road, maritime). The activity is novel as the previous machine learning activities focused on:

  • EL1-020 Less challenging maritime environments with well-defined sensor characteristics without SOOP, 
  • EL1-053 Exploring machine learning at central processing facility level for integrity products and not at user level, also without SOOP.

Expected output: Optimized navigation in challenging environments, alternative PNT, and a better understanding of the benefits and the limitations of Machine Learning for Navigation