A PSO-based weighting method to enhance machine learning techniques for cooperative spectrum sensing in CR networks

Abstract

Cognitive radio (CR) is a recent technology to tackle the problem of radio spectrum scarcity. Successful spectrum sensing is fundamental in performance of CR networks; hence, a PSO-based weighting method is proposed in order to improve the functionality of machine learning techniques which are used with the aim of detecting the activity of secondary users in cooperative cognitive radio (CCR) networks. Regarding classification methods, three supervised classifiers which are supported vector machines (SVM), K-nearest neighbors (K-NN) and naive Bayes are used for pattern classification. Since our goal is spectrum sensing in CCR networks, the vector of energy levels in radio which is considered as a feature vector is fed into the classifier to determine the availability of the channel. The classifier labels each feature vector as two classes: the channel available class or the channel unavailable In our proposed method, first, the three mentioned classifiers go through a training phase. Next, for new feature vectors, a label is assigned to the feature vector by each classifier and the final decision about the availability of the is made by a weighted voting method based on the PSO algorithm in an online fashion. The performance of our technique is measured in terms of the classification error. Also, the comparative results show twofold merit over previous methods since it not only reduces the error rate but also decreases the error of the available class.