The codes for paper "Spectrum Sensing for Cognitive Radio based on Feature Extraction and Deep Learning", in which a method that uses feature extraction and deep learning to do spectrum sensing for cognitive radio is introduced.

The codes in MATLAB are used to simulate QPSK/8PSK signal sequenses that pass through frequency-selective Rayleigh fading channel and AWGN channel.Then the signal feature matrixes with different SNR are generated. Use the two types of signal feature matrix as dataset to train CNN model. The codes can be modified to generate other types of signal matrixes. Run 'features.m' to generate data for training.

The codes in Python are used to load the dataset generated by MATLAB and train CNN networks by PyTorch. We set up both linear network and convolutional network(CNN), and it can be evaluated that the performance of CNN is better. Use Train_CovNet to train a model based on the dataset generated and use Test_CovNet to test the probability of detection and probability of false alarm.