Automatic Modulation Recognition via PointPN Recognition Network
Tao Chen, Boyi Yang, Lei Yu · 2025
In this paper, a method for automatic modulation recognition of modulated signals via point cloud recognition networks is proposed. The method aims to bypass existing recognition methods based on image classification and develop a novel deep learning recognition method that is different from mainstream methods. The proposed method maps the modulated signals to the point cloud space based on time-frequency analysis to complete the conversion of time-domain signals to spatial data. Next, the converted modulation signal is downsampled twice to reduce the amount of point and meet the basic requirements of fixed points in the point cloud recognition network. Then, the spatial geometric information of different modulated signals is extracted and mapped into implicit features using the point cloud recognition network. Finally, a linear classifier is used to aggregate discriminative features and output the category information. Simulation results show that the method is effective for both communication signals and radar modulated signals. The proposed method provides a novel recognition approach for modulation recognition problems.