Return to Session A5 Next Abstract

Session A5: Integrity and Assurance

A Modular Machine Learning Software Application for GPS Integrity Monitoring
Wilbur Myrick, Brian Sexton, W. Todd Faulkner, Daniel T. Goff and Stanley Radzevicius, ENSCO, Inc.; and Joseph Stevanak, US Army CCDC C5ISR
Location: Ballroom E
Date/Time: Wednesday, Aug. 25, 4:05 p.m.

Accurate real-time integrity assessment of positioning, navigation, and timing (PNT) sensor data is critical to ensuring mission success when operating in navigation warfare (NAVWAR) environments. Leveraging multiple sensors is also important for resilient operation and maintaining of PNT situational awareness in challenging threat environments. ENSCO presents a scalable Machine Learning (ML) software application for GPS integrity monitoring that meets these goals by leveraging a modular ML architecture that accepts a variety of sensor inputs and data formats, operates on a general-purpose processor (GPP), and supports causal real-time operation. The ML architecture presented is designed to provide PNT integrity and assurance to the warfighter.
ENSCO designed and demonstrated a prototype Machine Learning Spoof Detection Software Application (ML-SDSA) that implements this concept for a modular ML GPS integrity monitoring architecture. The ML-SDSA was developed using a multi-input machine learning data architecture that supports mixed sensor data types and optimally blends multiple sensor data ML inference branches to provide a single GPS integrity score. This sensor-combining approach supports integration of diverse PNT data and provides multi-input scalability to counter constantly evolving NAVWAR threats to PNT systems. The ML-SDSA also incorporates a physical model data pre-filter approach that enables well-conditioned ML models to be trained on sparse datasets collected from GPS receivers operating in realistic NAVWAR environments.
The ML-SDSA was trained with a combination of simulated GPS data, lab-based collections, and open-air NAVWAR environment test event data collections with two uBlox GPS receivers and an oven-controlled crystal oscillator (OCXO). This system operated at the 2020 Positioning, Navigation and Timing Assessment Exercise (PNTAX) to evaluate the scalable ML software architecture and integrity monitoring approach. Empirical results from the exercise will be presented to illustrate the performance potential of this architecture and approach to provide a real-time ML-based GPS integrity indicator that supports Army operations in NAVWAR environments.



Return to Session A5 Next Abstract