Supervised ADS-B Anomaly Detection Using a False Data Generator
Ralph Karam, Michel Salomon, Raphaël Couturier · 2022
The ADS-B is an air traffic monitoring technology based on broadcasting messages to transfer information in between aircraft as well as between aircraft and ground stations. It was created to increase the surveillance coverage as well as reduce the cost of operation relative to traditional radars. However the messages used to communicate under the ADS-B protocol are not encrypted and thus are prone to false data injection attacks which can, for example, modify the values of the messages' components. In this paper, a supervised deep learning strategy is designed to detect attacks that modify components of ADS-B messages such as altitude, ground speed, trajectory, latitude and longitude. A false data generator based on a domain specific language was used to attack ADS-B data and obtain a dataset containing both normal and anomalous data for supervised learning. The detection performance of two types of attacks were evaluated: gradual attacks and waypoints attacks which diverge aircraft trajectories to pass through specific waypoints. The experimental results show that the proposed supervised deep learning strategy is able to recall on average 99% of anomalies in ADS-B messages, mainly property modification attacks.