ADS-B Replay Attack Detection using Transformers

Automatic Dependent Surveillance Broadcast (ADSB) is one of the most prominent surveillance protocols for Air Traffic Control (ATC) that has gained great momentum in recent years. This has been motivated by several advantages it brings to the table, compared to traditional surveillance protocols, such as higher location accuracy, lower deployment and maintenance costs, and better financial viability. This led to the deployment of ADS-B in several parts of the world including Europe, China and USA. However, despite these numerous benefits, there have been reports of serious security flaws that could be exploited by a variety of cyber attacks, which may hinder the wider adoption of this technology. There were several techniques to deal with this important problem including secure location verification, cryptography, and Artificial Intelligence (AI). This latter has an edge over the two others as no modification to the original protocol and no special requirements, such as clock synchronization or number of nodes, are needed. In this context, novel AI-based classifiers for threat detection are constantly emerging and one of the most promising ones is Transformer Neural Networks known as Transformers. In this paper, we propose a novel approach for detecting replay attacks targeting ADS-B using a model based on transformers. We validate our findings with a performance evaluation of our model in comparison with other deep learning classifiers. Our results show that our model outperformed all other classifiers with regards to the performance metrics we used which indicate a promising success of our approach.