Exploratory Feature Analysis for Deep Learning based Spectrum Sensing

In the present work, a feature based deep learning spectrum sensing model is developed wherein the features are extracted from the received radio signals captured using GNU Radio. The efficacy of proposed model does not depend on the a priori knowledge of the signal structure. The data set used to develop the model supports multiple modulation techniques, reasonable range of signal to noise ratio with embedded noise and interference. The dependency of the features selected and extracted are investigated using deep learning (DL) based classi- fier which learns the hidden class distribution of the signal based on the labelled data. A deep convolutional neural net (CNN) is designed to attain acceptable classification accuracy. In addition, the proposed deep CNN model is tested against various other state of the art machine learning (ML) models and conventional model to prove its efficacy. From the results, it is evident that the proposed model outperforms the other models in most of the cases.