Abstract
In this thesis, we discuss two different approaches to modulation classifiers: we first propose a hybrid method for automatic modulation classification that lies in the intersection between likelihood-based and feature-based classifiers. Specifically, the proposed method relies on statistical moments along with a maximum likelihood engine. We show that the proposed method offers a good trade-off between classification accuracy and complexity relative to the Maximum Likelihood (ML) classifier. Furthermore, our classifier outperforms state-of-the-art machine learning classifiers, such as genetic programming-based K-nearest neighbor (GP-KNN) classifiers, the linear support vector machine classifier (LSVM) and the fold-based Kolmogorov-Smirnov (FB-KS) algorithm. In the second part of thesis, we propose a distribution-based modulation classifier using neural networks. We show that our proposed classifier outperform state-of-the-art classifiers, even when the pool of possible candidate modulations are unknown to the receiver.