Global Navigation Satellite Systems (GNSS) are vital to contemporary infrastructure but remain heavily susceptible to Radio Frequency Interference (RFI), particularly deliberate jamming. Crowdsourcing interference data from consumer smartphones is a scalable detection solution; however, it poses challenges regarding hardware heterogeneity, signal propagation in urban areas, and user privacy. This thesis develops a privacy-preserving framework for crowdsourced jammer localization with a two-stage pipeline trained via Federated Learning (FL). The first stage addresses the absence of calibrated power measurement tools for smartphones by introducing a physics-informed signal-fusion architecture that combines raw Automatic Gain Control (AGC) and Carrier-to-Noise Density (C/N₀) observables via a regime-level adaptive gating mechanism, reconstructing a physically consistent jammer Received Signal Strength Indicator (RSSI) without special hardware.