Communication Signal Automatic Modulation Classification Based on Feature Fusion

Communication signal automatic modulation classification technology plays a crucial role in non-cooperative communication, and improving the efficiency and accuracy of automatic modulation classification technology has profound significance. Previous researchers mostly relied on using either deep features obtained from machine learning or artificial features based on prior knowledge for classification. These methods not only have limitations in robustness and generalization but also struggle to effectively classify high-order modulation types. This paper proposes a communication signal modulation classification method based on feature fusion, which can be divided into four steps. First of all, the method performs deep feature extraction based on ResNet18 network and artificial feature extraction based on a prior knowledge for IQ signals, respectively. Secondly, the above features are sequentially spliced to obtain fused features. After that, the fused features are reduced in dimensionality through principal component analysis and then classified by support vector machine to achieve the final results. Simulation data experiments and real-world data experiments demonstrate that the proposed method exhibits high accuracy in classifying signals of fifteen modulation types, outperforming CNN networks and decision trees under the same conditions. Moreover, this method shows good classification performance for high-order modulation types. Additionally, the method demonstrates adaptability to signal-to-noise ratio, confirming its universality and robustness.