Modulation Recognition Method of Cyclic Stationary Signal Based on Graph Theory

Aiming at the existing shortcomings of recognition algorithms based on signal cyclic spectra, such as incomplete feature extraction and poor robustness against noise, a signal recognition method based on a deep residual network and noise reduction cyclic spectrum is discussed in this paper. The Wiener filtering is used to preprocess the two-dimensional cyclic spectrum of the signal. Then the preprocessed signal is input into the modified 50-residual layer network for feature extraction and identification of the modulation type of the signal. Experimental results report that the proposed method can effectively identify eight different modulation types of cyclostationary signals, with the recognition accuracy reaching 96% when SNR is greater than −2dB. Furthermore, the recognition performance of the proposed method is better than some existing deep learning recognition algorithms, such as AlexNet, GoogleNet, etc.