Feature Based Automatic Modulation Classification

Abstract

Automatic modulation classification (AMC) is a key step for blind signal demodulation. AMC is widely used in military as well as commercial applications. Without prior information of the received data, AMC becomes a difficult task. This thesis follows Feature based classification approach of AMC using Multi-Class Support Vector Machine (SVM) for the classification of digital modulation schemes in the presence of Additive White Gaussian Noise (AWGN). Eight types of digital modulation schemes are classified using four features extracted from the blind modulated signal. In this thesis different types of features are discussed which is used in modulation classification. Spectral feature, higher order statistics features and cyclostationary features are presented. Implementation of classification is done in two steps. Firstly, inter modulation schemes (ASK, PSK and QAM) are identified. Next, the order (M) of modulation scheme is identified. This thesis presents comparison results of SVM and Decision Tree (DT) classifiers at different SNR values.