Modulation Waveform Recognition Method Based on Feature Fusion

Signal modulation waveform recognition is an important technical means to promote sharing spectrum resources and ensure the reliable operation of spatial information systems. However, the modulation waveform recognition methods based on single-feature extraction have difficulties in feature selection. To improve recognition accuracy and efficiency, in this article, we propose a modulation waveform recognition method based on feature fusion. First, the method constructs the feature vectors of instantaneous characteristics, higher-order cumulants, and constellation diagrams of modulated signals. Then a convolutional neural network (CNN) is used to extract the higher-level features of original data and its instantaneous characteristics, higher-order cumulants, and constellation diagrams. Moreover, we propose to add a channel attention mechanism to make the model selectively focus on the features which are more useful for the recognition task, so as to reduce the influence of redundant information. Simulation results show that the maximum recognition accuracy of the proposed method can reach 91%, which is nearly 16% higher than the traditional machine learning recognition methods, and the number of parameters decreased by nearly 77%.