Identification of pulse radar signals using machine learning algorithms

This thesis aims to use machine learning methods for radar identification in radar warning receivers. A comprehensive threat library was created using pulsed signals of known radars and pulse descriptor words were created with the signal frequency, signal width, signal repetition frequency and beamwidth parameters obtained after parameter extraction. To optimize the pre-programming phase of the radar warning receivers, machine learning algorithms were evaluated, and their performances were compared with each other and with conventional algorithms. Five machine learning algorithms were identified as suitable for the problem and models were created with these algorithms. The preliminary tests on K-Nearest Neighbors, Decision Tree, Ensemble Methods, Support Vector Machines and Artificial Neural Networks algorithms achieved learning accuracy levels above the targeted 95%. In combat environment simulations with the prepared threat library, test accuracy of over 98% was achieved. The strengths of the machine learning algorithms obtained from the completed simulation results in terms of speed and accuracy compared to traditional algorithms are evaluated.