Sum Product Network to Indicate Proximate Objects in Space (SNIPS)

ABSTRACT: Space system operators require the ability to detect and track potential threats to their platforms. One way to detect and track objects in the proximity of a satellite or other space vehicle without dedicated sensors and subsystems is to perform additional processing on existing system sensor data streams to identify perturbations caused by a nearby object. Of the sensors commonly deployed on space vehicles, the Global Positioning System (GPS) is the most suitable for this purpose. The GPS signal stream will be perturbed by any nearby radio-reflective object in the form of multipath noise. We demonstrate the use of this multipath noise to perform a trilateration which exploits both time and frequency perturbations in the data to detect and track radio-reflective objects up to several thousand meters from the vehicle. To implement this algorithm we describe a radiation-hardened Sum Product Network to Indicate Proximate Objects in Space (SNIPS) which infers the existence of potential threats to space platforms using GPS multipath noise and validates the detection via anomaly detection on concomitant effects such as perturbation of communication signals. A verifiable sum product network localizes nearby objects and uses belief propagation to determine the track and threat indications of the objects. BENEFIT: SNIPS benefits a range of spacecraft and platforms by enabling immediate and autonomous reaction to threats from the local area. Government transition opportunities include satellite programs such as the Wideband Global Satellite Communications (SATCOM) program, Defense Satellite Communications System (DSCS), and TACSAT. Potential commercial licensing includes placement with major satellite operators. SNIPS will enhance the existing AgentWorks product suite with algorithms and aspects resulting from this research.