A Residual Neural Network for Modulation Recognition of 24 kinds of Signals

With the development of wireless communication technology and the updates of communication equipment, the modulation of signal becomes more complex, and modulation recognition is becoming more and more difficult. Traditional signal modulation recognition methods rely on human experience, its feature extraction process is complex, and the empirical threshold is difficult to find. The recognition method combined with manual feature extraction and deep neural network can achieve better recognition accuracy, but it is still limited by the process of feature extraction. Compared with the above, automatic modulation recognition method based on deep learning is more efficient in complicated open environment. In this paper, a residual neural network for automatic modulation recognition was designed, and the experiment had achieved remarkable results. When SNR is 10dB, we got an accuracy of 95.3% faced to 24 kinds of signals, and when SNR is 12dB, we got an accuracy of 96.3%. Compared with existing models, this model reduces the network parameters, greatly shortens the training time, and lower the hardware requirements. This model shows a good result on the recognition of high-level modulation signal. When SNR is 10dB, the recognition accuracy of 128APSK, 128QAM and 256QAM is 97%, 88% and 88%.