Investigation and application of Machine Learning algorithm for GNSS spoofing detection

Nowadays, Global Navigation Satellite System (GNSS) is used in many domains from civil to military applications typically for navigation and precise positioning. The growth of these systems occurs alongside the growth of Spoofing attacks. Spoofing is a malicious attempt to disrupt a user’s GNSS measurements, making their estimated position unreliable. The aim of this dissertation is the development of Spoofing Detection algorithms implementing several Machine Learning (ML) techniques. Artificial Intelligence (AI) with all its related sub-fields is become pervasive in our life and even GNSS systems are not immune to the impetuous development of these cutting edge technologies. Even though only a specific use case will be presented in this work, the AI and ML techniques could help the receiver to increase its spoofing/jamming detection capabilities and errors mitigation, such as Multipath/NLOS signals and Ionopsheric/Tropospheric delays. The main goal of this Thesis is to create a set of learning models that implements several Supervised and Unsupervised learning techniques, in order to increase the robustness of the GNSS receiver against malicious attacks. Finally, these models are compared to understand which one brings the best response to Spoofing attacks. Particular attention is given to the description of the features that populate the dataset and the innovative techniques that allow the creation of these models. Ultimately, the two macro areas presented in this Thesis, GNSS and ML, will inevitably merge, combining two technologies apparently incompatible. In fact, this dissertation demonstrates the advantages of the implementation of ML algorithms inside GNSS receivers. All the considerations and implementations done during this dissertation are the starting point for a future deployment process of these techniques on a real GNSS-type device.