Eigenvalue-Based Spectrum Sensing in Cognitive Radio Networks Using Supervised Learning

We consider supervised machine learning-based detection algorithms for spectrum sensing in cognitive radio (CR) networks. The network comprises of multiple CR nodes, which collect a set of observations over generalized fading channels. These observations are then relayed over a lossless control channel to a fusion center, where they are combined for the classification task. Further, we consider maximum eigenvalue, energy and maximum-minimum eigenvalue of the received sample correlation matrix as a set of features for comparison. For classification, we carry out a performance comparison of support vector machine (SVM), naive Bayes and random forest algorithms, in terms of classification accuracy. Our simulation results show that the maximum eigenvalue feature performs the best, while linear SVM yields the best classification performance.