104 - AI for Anomaly Detection in Multi-Sensor PNT

There is currently considerable research effort being devoted to the development of multi-sensor PNT approaches which promise greater resilience and potentially greater accuracy than GNSS alone. New sensors are also being developed which have less well understood characteristics and vulnerabilities. These include quantum inertial sensors and quantum gravimeters and magnetometers. For such complex multi-sensor systems with less well understood sensors, it may not be possible to fully understand the error sources and failure modes a priori.

Traditional methods for sensor fusion rely on pre-defined error models. However, with novel sensors like quantum devices, these models might not exist or be inaccurate. While machine learning excels at learning complex, non-linear relationships from data, it can identify error patterns and dependencies between sensors that traditional methods might miss.

This project will seek to develop machine learning approaches to characterising sensors and detecting anomalies that will impact the contributions of these sensors to the position and time solution. For systems which include the use of external signals such as GNSS and signals of opportunity, the machine learning algorithms can also be used to identify spoofing or changes in the signals structure or orbits from satellites of opportunity. As this is not the first attempt to apply Machine Learning to complex systems, the highest novelty possible in the development will be sought (typically considering the newest quantum sensors models).

The objective of this activity is to develop machine learning algorithms for anomaly detection and characterization in complex multi-sensor PNT systems with novel sensors for robust and adaptable positioning and timing.

The tasks to be performed shall include:

Sensor Characterization: Machine learning algorithms will be developed to analyse sensor data and identify patterns that correlate with sensor health and performance. These algorithms will be trained on data from various sensors (including novel ones) under different operating conditions

Multi-Sensor Fusion with Anomaly Detection: The activity will design algorithms that combine data from multiple sensors (including GNSS, Signals of Opportunity (SoOP), and novel sensors) for PNT. These algorithms will integrate anomaly detection models to isolate and attribute suspicious readings to specific sensors or external signal sources (e.g., GNSS spoofing)

Machine Learning for Signal Integrity Monitoring: Machine learning models will be developed to analyse the characteristics of GNSS and SoOP signals for integrity monitoring. These models will be trained to identify deviations that might indicate spoofing attempts or changes in signal structure.

The main outputs of the activity will consist of:
 

• A suite of machine learning models for:
  • Characterizing sensor behaviour and identifying anomalies
  • Fusing multi-sensor data for PNT with real-time anomaly detection
  • Monitoring the integrity of GNSS and SoOP signals for spoofing attempts
• A deployable PNT system with anomaly detection capabilities, enabling robust and adaptable positioning and timing
• An adaptable framework that continuously improves anomaly detection through machine learning and integration of new data and threats.