Development of Logistic regression based spectrum sensing algorithm using extreme eigenvalues

Next-generation (NextGen) wireless systems require huge bandwidth for their communication due to the Internet of Things (IoT) and high data rate applications. Cognitive radio (CR) technology provides a solution by finding spectrum holes and opportunistically assigning them to NextGen wireless systems. Among various functions of CR, spectrum sensing is a crucial function that can explore spectrum holes over a specified range of the radio spectrum. Developing an efficient sensing method is a focused area of research. Out of numerous sensing algorithms developed, eigenvalue-based sensing has shown superiority. However, the test statistic designed using the ratio of extreme eigenvalues is not a linear function of signal-to-noise ratio, which may not be an efficient measure for signal detection. On the other hand, machine learning algorithms have not been greatly explored in this area, despite being applied to solving many research problems. Hence, this work focuses on developing a less complex and efficient test statistic using a machine learning algorithm. We have considered extreme eigenvalues of a random covariance matrix as training features to model the logistic regression-based detection technique. The results reveal that the proposed method can detect -7 dB signals and outperforms the eigenvalue-based detection method.