Abstract
Classifiers based on machine learning are usually trained to distinguish between several known classes. For an electronic intelligence application, however, it is of great importance to recognize if an intercepted signal belongs to an unknown radar emitter. In the machine learning literature, this task is called open-set recognition. This article investigates six approaches in several configurations to recognize unknown emitters. It is based on a hierarchical emission model that understands emissions as a language with an inherent hierarchical structure. We consider two general approaches, which are the “memoryless” Markov chain and the Long Short-Term Memory recurrent neural network, which is especially designed to “remember” the past. The performance is demonstrated with two evaluation metrics in ten scenarios that contain different combinations of known and unknown emitters. An evaluation with corrupted data provides an estimate on the methods’ accuracies under challenging conditions. The results show that unknown emitters that do not use known waveforms are reliably recognized even with corrupted data, while unknown emitters that are more similar to known ones are harder to detect.
Highlights
T HE goal of electronic intelligence (ELINT) is to collect information about radar systems by intercepting and analysing their signals
This paper investigates open-set recognition to detect unknown radar emitters in an electronic intelligence (ELINT) context
The presented methods are based on a hierarchical emission model, which interprets the radar emissions as a language that consists of letters, syllables, words, commands, and functions
Summary
T HE goal of electronic intelligence (ELINT) is to collect information about radar systems by intercepting and analysing their signals. The major differences between open-set recognition, classification, and anomaly detection are the goal, the available training data, and the output that the classifier provides in each task. This paper investigates six methods to recognise unknown radar emitters, which are either based on a “memoryless” Markov chain (MC) or a Long Short-Term Memory (LSTM) recurrent neural network [14] that possesses a “memory”. These approaches build on previous work of the authors on modelling and predicting emissions [15], [16], as well as the identification of the radar emitter type [2], [16]
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have