A New Method Based on Local Integral Bispectra and SVM for Radio Transmitter Individual Identification

Abstract

To resolve the difficult problem of identifying radio transmitters with the same model, a new method using support vector machine with mixtures of kernels is present for classification of individual transmitters. In this method, the selected local integral bispectra and parameters significant for classification of the received signal form the new identification feature vector. To optimize the classifier, different parameters of kernel function are discussed. The performance of classifier which based on mixtures of kernels is compared with which based on conventional kernel functions. The result of experiments on FM individual transmitters shows that this method is able to achieve better classification rate than conventional kernels even in low SNR.