A Synergistic Cross-Domain Network Integrating Time-Frequency-Spatial Features for Modulation Recognition

Automatic Modulation Recognition (AMR) analyzes the modulation schemes of wireless communication signals to achieve automatic classification and decoding, with widespread applications in dynamic spectrum management, radio interference monitoring, and military communications. However, existing AMR models typically rely only on raw In-phase and Quadrature (I/Q) data or a single modality, and the extracted feature information is insufficient to fully describe the signal characteristics. This letter proposes a synergistic cross-domain network (SyCoNet) that integrates time-domain, frequency-domain, and spatial features. By using depthwise separable gated complex-valued convolutions, the model learns the intrinsic dependencies in signal data, while incorporating channel and spatial attention mechanisms to fuse multiple features. Despite its lower complexity, the model achieves high recognition accuracy, outperforming current state-of-the-art models on multiple public datasets.