Improvements to Steady State Spoof Detection with Experimental Validation using a Dual Polarization Antenna

Abstract

Recent work has demonstrated that signal-geometry-based approaches are a powerful mean for a GNSS receiver to detect spoofing. These techniques, leveraging the spatial diversity in the received signals, usually rely on all or at least most signals being spoofed for successful detection. This paper presents several complimentary improvements to these techniques to allow for robust spoof detection even in the presence of light multipath and a mixture of spoofed and genuine satellites. The improved spoof detection can accomplish this while satisfying a constraint on the false alert probability. For this, the paper makes three main contributions: 1) Using the example of a Dual Polarization Antenna (DPA) as a case study, it presents a characterization of measurement quality based on data and physics-derived features. 2) It introduces a hypothesis test iteration algorithm that breaks an m-ary hypothesis test of an unknown number of spoofed satellites from an unknown number of independent signal sources down into a sequence of binary hypothesis tests that identifies the largest subset of satellites triggering an alarm. 3) It derives a likelihood-ratio based decision threshold independent of prior probabilities similar to a Neyman-Pearson test. It is applicable to any spoof defense employing hypotheses tests resulting in conditional probabilities. The theoretical work is supported by Monte Carlo simulations and tested on data collected by a DPA during a government sponsored live spoofing event. Despite the low quality of azimuth-only spatial measurements, the presence of multipath and the spoofing of only a subset of GPS satellites, more than half of the attacks are detected within a few measurement epochs. No false alarms are raised by a large margin. Compared to the same data processed without individually characterizing the measurement quality and using binary “all satellites nominal” vs. “all satellites spoofed” hypotheses, a 2-3 fold improvement in detection is achieved using the derived decision threshold.