Semi-Supervised Techniques for Detecting Previously Unseen Radar Behaviors

The rapid advance in multifunction radars (MFRs) whose behavior model can be promptly reprogrammed complicates the task of electrical warfare (EW) systems. It is crucial that an EW system be able to detect the change in a radar's behavior when it happens. In this article, two contextual anomaly detection techniques based on Hidden Markov Models (HMM) and Long Short-Term Memories (LSTM) are applied to detect an MFR's behavior change. Both of them are trained based on known radar modes and indicate the presence of an anomaly radar mode when the signal sequence emitted by the radar does not match the EWsystem's prediction based on known modes. This topic is important as the EWsystems relying on libraries of knowns radar signals would be in a dangerous situation when encountering an unknown radar behavior in practice and knowing the existence of such a radar mode can increase the system's survivability. The results demonstrate the great potential of these techniques when applied in EW applications. However, the HMM holds the advantage over the LSTM if the MFR changes its mode more frequently.

Automated summary

This article applies two contextual anomaly detection techniques based on Hidden Markov Models (HMM) and Long Short-Term Memories (LSTM) to detect changes in the behavior of multifunction radars (MFRs). The techniques are trained on known radar modes and can identify an anomaly radar mode when the emitted signal sequence does not match the prediction based on known modes. This is important for enhancing the survivability of electrical warfare (EW) systems in the face of unknown radar behaviors.