Parameter selection for SVM in automatic modulation classification of analog and digital signals

Abstract

Cognitive radio is a revolutionary technology that aims to solve the spectrum-underutilization problem, through spectrum sensing, which is a technique focused on detecting spectrum holes. Automatic modulation classification plays an important role in this scenario, as it can provide information about primary users, with the goal of aiding in spectrum sensing tasks. In the present work, an implementation methodology for a multiclass classification system, using support vector machines (SVM) for recognizing seven types of modulation (AM, FM, BPSK, QPSK, 16QAM, 64QAM and GMSK), is described, where test signals are generated in a more realistic way than usually found in the related literature. In the classification stage, the parameter selection for SVM and classifier validation steps are performed with grid search and k-fold cross-validation techniques. Finally, one-against-one and one-against-all multiclass approaches are compared. The overall correct classification percentage, with one-against-one, was approximately 94%, which is very good, considering that SNR levels range from 0 to 30 dB.