Share
Export Citation
DRaGon: Robust Emitter Geolocation via Deep Learning with NTP-Synchronized RTL-SDR Receivers
Hamid Z.
IEEE Asia Pacific Conference on Geoscience Electronics and Remote Sensing Technology Agers
Abstract
Accurate radio frequency (RF) geolocation is vital for spectrum management, public safety, and defense. The Time Difference of Arrival (TDoA) method is effective but requires sub-nanosecond synchronization, traditionally achievable only with costly GPS-disciplined hardware. In contrast, the Network Time Protocol (NTP) over public networks have millisecond-level jitter, making it not suitable for conventional TDoA. This paper introduces DRaGon (Distributed Radio Geolocation), a deep learning framework that bypasses explicit time-difference estimation by learning an end-to-end mapping from raw in-phase and quadrature (I/Q) samples to geographic coordinates. Using three low-cost RTL-SDR receivers synchronized solely by NTP over cellular networks, we trained and evaluated DRaGon on a real-world dataset of more than $\mathbf{1 9 8, 0 0 0 ~ F M}$ broadcast samples collected in Semarang, Indonesia. Results show that DRaGon achieves a mean absolute localization error of 426 meters, a 63.24% improvement over sophisticated RSSI-based trilateration baselines. These findings challenge the assumption that precise hardware synchronization is indispensable for TDoA and demonstrate that synchronization errors and multipath effects can be treated as part of a learnable $\mathbf{R F}$ fingerprint, enabling affordable and scalable RF geolocation systems.