Signal type detection in CRN: A hierarchical modulation classification framework using SVM and Decision tree approaches

To ensure effective dynamic spectrum access in cognitive radio proper signal classification in fading channels and low signal to noise ratio environment is essential. Due to lack of information about the modulation scheme used it cannot distinguish between signal present is that of the primary or any other secondary user communication. In this paper a multiclass modulation classification hierarchical framework is proposed which exploit cyclostationary features. The hierarchical framework uses two known binary classifiers SVM and Decision tree at each level which is based on one-against-all approach. The proposed system assumes no a priori knowledge about signal properties like frequency, phase or symbol rate. The performance of the classifiers is compared based upon the accuracy, training time and classification time needed for each stage of the framework. It is demonstrated through simulation for four modulation types with the optimal features required for classifying each scheme in the proposed hierarchical framework. Compared with two existing methods the proposed framework is found to be effective for modulation detection of signals.