Automatic Modulation Recognition Method Basedon Multimodal I/Q-FRFT Fusion

Abstract Automatic modulation recognition (AMR) is a key technology in the domain of cognitive radio communications. Accurately identifying the modulation schemes of signals is crucial for enhancing the robustness and reliability of communication systems. The existing methods usually use deep learning technology to intelligently model the single modality such as in-phase quadrature (I/Q). However, single modality has the disadvantages of low recognition rate and limited feature expression ability. In this paper, we propose an innovative automatic modulation recognition method based on multimodal fusion of I and Fractional Fourier Transform (FRFT). Specifically, we first design the entire multimodal approach as an intermediate fusion mode, that is, performing the unimodal representation first and then multimodal fusion later. In the multimodal fusion stage, we introduce a Multimodal Processing Unit (MPU), which realizes the enhancement and aggregation of different modes by learning the complementary features between and within modalities. Furthermore, we employ contrast learning to categorize signals of different modulation into positive and negative samples to enhance the robustness of the multimodal features. In order to verify the effectiveness of the proposed method, we conduct experiments on two public datasets, RML2016.10a and HisarMod2019.1. The experimental results indicate that the multimodal approach consistently outperforms the unimodal methods significantly, achieving the state-of-the-art.