A Data-Driven Approach for GPS Interference Source Detection and Positioning via Weighted Least Squares

Abstract

The safety of the aviation radio frequency spectrum is a fundamental prerequisite for civil aviation operations. Recently, frequent GPS interference events across multiple regions have caused flight diversions and false ground proximity warnings, posing serious threats to flight safety. Traditional ground-based monitoring methods face significant challenges in rapidly locating interference sources over large airspaces. To address this, we propose a data-driven, semi-cooperative spatial anomaly detection and coarse positioning framework. Instead of relying on specialized monitoring hardware, our approach mines spatial proximity relationships and extracts interference distances directly from continuously logged GPS signal quality indicators of affected aircraft. A key data mining innovation in this study is the introduction of a weighted least squares (WLS) estimator that utilizes signal quality metrics (HFOM/HIL) as reliability weights to mitigate the impact of noisy spatial data. Unlike prior equal-weight approaches that treat all measurements identically, our data-driven weighting strategy dynamically prioritizes high-fidelity signals. The nonlinear spatial distance equation is linearized via differential processing, yielding a computationally efficient closed-form WLS solution well-suited for processing large-scale flight data. The Cramér-Rao lower bound (CRLB) is derived to characterize the theoretical positioning accuracy limit. Extensive Monte Carlo simulations demonstrate that the proposed method achieves a root mean square error (RMSE) of approximately 109 meters under typical operational scenarios (n = 10, σr = 10 m), outperforming the standard equal-weight baseline by 10.5% in RMSE. An ablation study further validates that this weighting strategy contributes an average accuracy improvement of 6%–12% across varying noise levels and aircraft counts. The proposed method offers an effective, automated, and low-cost spatial data mining solution for large-area GPS interference localization.