On the Performance of Digital Modulation Classification for Cooperative Multiple Relays Network System without Direct Channel

Abstract

Automatic modulation classification (AMC) is defined as the automatic identification of modulation format of the sensed signal. We consider in this work a features-based digital modulation identification approach for wireless cooperative multi-relay networks using amplify-and-forward (AF) protocol and without direct channel between source and destination nodes. We compare the performances of three models among the most commonly used classifiers, namely decision tree (TREE), K-nearest neighbors (KNN), and support-vector machine (SVM). Comparison criteria are probability of identification, recall, precision and F-Scores. Our purpose is to discriminate between the modulation format and order of different M-ary shift keying modulations using higher order cumulants (HOCs) of the received signal. To select the most suitable classifier, we firstly consider a single AF relay cooperative network system with direct link (source to destination nodes). Then, the best classifier is applied to the cooperative multiple relays network alternative ignoring the direct link. Obtained results, within the considered scenario, show that the SVM classifier is more efficient and achieves high probability of correct identification in acceptable snr level. This performance improves with the increase of the number of relays.