Automatic modulation classification is concerned with identifying the modulation present on a radio wave. This can be any type of radar or communication signal. It is employed in fields such as cognitive radio for communications, radar analysis for electronic warfare.
This thesis is dedicated to classifying a variety of modulations used in modern radar. These include unmodulated, various types of frequency modulation, and phase shift keyed waveforms. This task is accomplished through feature extraction and machine learning techniques. The objective is to determine a suitable method applicable for real-time implementation in a complex electronic warfare environment.
Three techniques are proposed: a decision tree combined with Multilayer Perceptron Neural Network, a Multilayer Perceptron Neural Network, and a Convolutional Neural Network. The simulation results show that the decision tree achieves a low classification performance, the Multilayer Perceptron achieves good results in a controlled environment, while the Convolutional Neural Network achieves generalizable results.