A Few-Shot Modulation Recognition Method Based on Multi-Modal Feature Fusion
Yanping Zha, Hongjun Wang, Zhexian Shen, Jiangzhou Wang
Abstract
Few-Shot learning (FSL) is a new deep learning method, which has been successfully applied in many fields, but has not been well explored in the field of communication. To adapt to the application scenario of signal modulation recognition when only a small number of samples of new types of signals can be collected, a novel FSL algorithm based on a new multi-channel feature fusion network named FFFNet is proposed in this paper. Specifically, the time sequence signals firstly are preprocessed into time-frequency and constellation diagrams which are enhanced by the reciprocal attenuation model. Then, a multi-channel feature extraction network is designed to generate prototypes by extracting and integrating the multi-modal features. The modulation type of the sample to be tested can be obtained by measuring the distance between it with the prototypes. Simulation results prove the proposed algorithm can effectively identify the modulation type of new signal categories just with a demand of few samples due to its strong generalization and robustness ability.