False Data Injection (FDI) Modeling and Detection in Global Positioning Systems (GPS) for UAS Environments

Global Positioning System (GPS) is critical for precise navigation of Unmanned Aerial Systems (UAS) trajectory planning. GPS signals may experience sudden or intermittent interference due to multi-path fading, spoofing, jamming, or difficult terrain environments leading to “dropout” phenomena. The paper models one such dropout scenario by injecting false data into original GPS timestamps representing interference. We consider 'dropouts' as insertion of false GPS data (FDI). The end goal is to investigate the performance of FDI detection rates using several clustering algorithms: 1) Density-Based Spatial Clustering, 2) Ordering Points to Identify the Clustering Structure, 3) Gaussian mixture models, and 4) Hierarchical Clustering. Multiple grouping indices such as Silhouette Coefficient (SC) and David Bouldin's (DB) Index were used to evaluate normal vs. abnormal points grouping. The preliminary results are mixed indicating that the OPTICS clustering algorithm outperforms for scenario 3, while scenarios 1 and 2 are challenging for any clustering methods to capture the injected points. Also, SC is a good metric over DB index to evaluate the cluster groups. This paper also investigated non-clustering approach such as Isolation forest, which seems to be a better candidate for such FDI detection.