Edinburgh Napier University

  In this work, the performance of support vector machine (SVM)-based classifier, applied for spectrum sensing in cognitive radio (CR) networks, is analyzed. A single observation given input to classifier is composed of three statistical features extracted from the primary user (PU) sensing signal and residual energy in percent of the secondary user (SU). The trained classifier predicts PU’s presence based on the input signal. The SU starts transmission if PU is predicted absent, otherwise continues sensing other frequency bands. The hypothesis that PU is absent, is further classified in multi classes. The secondary user varies the transmission power based on the output class. This technique increases the quality of service (QoS) due to low interference from SU to PU even if failed to detect. The cross validation technique increases the generalization of classifier. The performance of classifier is examined in terms of accuracy results. The signal-to-noise (SNR) ratio from PU to SU is varied to investigate effect on classifier’s performance. Furthermore, the receiver operating characteristics (ROC) is presented for more evaluation. The parameter ‘area under curve (AUC)’ is given for comparison. The simulation results show the efficiency of proposed features with SVM-based classifier for spectrum sensing in CR applications.