Classification of Pulse Repetition Interval Modulations Using Neural Networks

Abstract

Repetition Intervals (PRI)-the distances between consecutive times of arrival of radar pulses-are important characteristics that help identify the emitting source. The recognition of various PRI modulation types under the assumption of missing and spurious pulses is a classical yet challenging problem. We introduce in this paper a novel learning-based method for the classification of 7 popular PRI modulations. In this classifier, a set of 6 features, extracted from the preprocessed PRI sequences, are fed into a simple feed-forward neural network. The proposed scheme, while computationally fast, outperforms existing methods by a significant margin on a variety of PRI parameters and under different levels of pulse miss-detections and false alarms.