Modulation Recognition using Wavelet-assisted Convolutional Neural Network

Automatic modulation recognition is considered as a very important part in cognitive radio (CR). The artificial neural network is the research hot spots of pattern recognition. In recent years, researchers have found that convolutional neural network (CNN) can be a powerful tool to accomplish this task. In this paper, we present a CNN-based modulation recognition method which creates the spectrogram images of different complex signals using Short-Time Fourier Transform (STFT) so that the complex modulation recognition problem is converted to an image recognition problem. In addition, in order to improve the recognition accuracy under low SNR region, wavelet denoising method is applied to the test set before feeding to the CNN. The test results show that the recognition accuracy under lower signal-to-noise ratio (SNR) is improved after wavelet denoising and the method has good generalization ability.