Assessment of spectrum sensing using support vector machine combined with principal component analysis

Abstract

Cognitive radio (CR) is an up-and-coming technology to rectify the problem of under-utilisation of the allocated spectrum and meet the increasing demand for free spectrum. Spectrum sensing empowers the CR to adjust to its surroundings by locating free spectrum. Although spectrum sensing using a support vector machine (SVM) is already found in literature, an SVM combined with principal component analysis (PCA) and varying the kernel scale is yet to be investigated. In this paper, we perform spectrum sensing using an SVM and evaluate the performances of various kernel functions used in the SVM as well as how the performances of the learning algorithm change as we apply PCA and vary the kernel scales. We then compare the training time of the SVM kernels. Finally, we calculate the contributions of power, variance, skewness, and kurtosis of the received signal towards the decision-making process of the learning algorithm.