SOM-GA-SVM Detection Based Spectrum Sensing in Cognitive Radio

Spectrum sensing is a key function for the second users (SUs) to determine availability of a channel in the primary user's (PUs) spectrum in cognitive radio(CR). In order to achieve that, much research of energy detection has been studied, but they play poor performance in low signal-to-noise (SNR) environment. In this paper, we proposed Support Vector Machines (SVM) based on Genetic Algorithms (GA), which is a novel method of classifying in real time. GA has the ability to find out the optimal nuclear function parameter and penalty parameter of SVM. Due to the massive training data increase the computational complexity and make lower performance of SVM. So we re-sample the massive data by using Self Organizing Map (SOM) to obtain representative compressed data. So we set up the SOM-GA-SVM classification model for detecting and it has idle recognized ability in low SNR compared with SVM model and energy detection. Simulation results show that the proposed sensing method can obtain excellent performance.