SPECTRUMNET: Cooperative Spectrum Monitoring Using Deep Neural Networks

Spectrum monitoring is one of the significant tasks required during spectrum sharing process in cognitive radio networks (CRNs). Although widely used to monitor usage allocated resources, this work focuses on detecting a primary user (PU) presence secondary (SU) signals. For signal classification, existing methods, including cooperative, noncooperative, and neural network-based models, are frequently used, but they still inconsistent because lack sensitivity accuracy. A deep network model for intelligent wireless identification perform proposed efficient sensing at low SNR (signal noise ratio) preserve hyperspectral image features. hybrid learning called SPECTRUMNET (spectrum using network) presented. It can quickly accurately from spectrogram images by utilizing cyclostationary features convolutional (CNN). The class imbalance issue solved uniformly spreading samples throughout classes oversampling method known as SMOTE (Synthetic Minority Oversampling Technique). achieves classification accuracy 94.46% −15 dB, which an improvement over CNN models with minor trainable parameters.