Signal classification with an SVM-FFT approach for feature extraction in cognitive radio

The estimation of the spectrum usage from the point of view of number of users and modulation types is addressed in this paper. The techniques used here are based on Support Vector Machines (SVM). SVMs are machine learning strategies which use a robust cost function alternative to the widely used Least Squares function and that apply a regularization which provides control of the complexity of the resulting estimators. As a result, estimators are robust against interferences and nongaussian noise and present excellent generalization properties where the number of data available for the estimation is small. The structure presented here has a feature extraction part that, instead of using an FFT approach, uses the SVM criterion for spectrum estimation, feature extraction and modulation classification.