Jammer Source Localization with Federated Learning

Global Navigation Satellite System (GNSS) signals are increasingly vulnerable to jamming, disrupting critical applications like autonomous navigation and aviation. Traditional jammer localization relies on centralized data processing, raising privacy concerns. This work proposes a federated learning (FL) framework for privacy-preserving jammer localization using crowdsourced received signal strength (RSS) measurements. We explore three models: a neural network (NN) for initial localization, a path-loss model (PL), and an augmented physics-based model (APBM) combining both PL and NN models. Evaluations in open-sky, suburban and urban environments show that PL and APBM outperform a non-FL baseline in open-sky and suburban settings, while urban scenarios remain challenging due to multipath and shadowing. In addition, we analyze the impact of client distribution, observation density, and measurement noise on localization accuracy