RadArnomaly: Protecting Radar Systems from Data Manipulation Attacks.

Area of Science:

• Cybersecurity
• Radar Systems Engineering
• Machine Learning

Background:

• Radar systems are critical for tracking various objects, with data integrity essential for real-time decision-making.
• Existing cybersecurity research has not addressed anomaly detection specifically within radar data streams.
• The heterogeneous nature of radar data (numerical and categorical features) presents unique challenges for anomaly detection.

Purpose of the Study:

• To develop an unsupervised deep learning method for detecting anomalies in heterogeneous radar data streams.
• To address the need for enhanced reliability and availability of information from radar systems.
• To identify and mitigate malicious manipulations and message-dropping attempts in radar data.

Main Methods:

• An unsupervised deep learning approach was employed to analyze radar data streams.
• A novel technique was developed to learn correlations between numerical and categorical features using embeddings.
• A timing-interval anomaly detection mechanism was integrated to detect message-dropping attacks.

Main Results:

• The method achieved high detection accuracy for data-stream manipulation attacks (88% average detection rate, 1.59% false alarm rate).
• The technique demonstrated high accuracy in detecting message-dropping attacks (92% average detection rate, 2.2% false alarm rate).
• Experiments on real radar system data validated the effectiveness of the proposed anomaly detection method.

Conclusions:

• The proposed unsupervised deep learning method is effective for anomaly detection in radar systems.
• The technique successfully identifies both data manipulation and message-dropping attacks.
• This research contributes to improving the security and reliability of critical radar infrastructure.