Advanced Modulation Recognition with CResNet for Mining Complex Information

Automatic modulation recognition is a key technology in the field of cognitive radio, which plays an important role in civil and military fields. In recent years, the increasing number of signal modulation types and the increasing complexity of the communication electromagnetic environment have brought great challenges to automatic modulation identification. In this paper, we introduce a novel automatic modulation recognition (AMR) system, the Complex-convolution Residual Network (CResNet), tailored for accurately identifying modulation modes in challenging low signal-to-noise ratio (SNR) conditions. This model incorporates a complex module at the input layer to capture the key comple features of radio signals, including signal phase and amplitude nuances. Key enhancements include the use of complex Adam optimizer, which bolster feature discernment of the complex radio signal. Our evaluation on the RadioML2016.10a dataset demonstrates robust performance across SNR levels, achieving a notable 0.59 average all-SNR accuracy and over 0.82 accuracy in >0 dB environment. It shows superior performance compared to traditional and contemporary AMR models, particularly in challenging low SNR conditions, highlighting the model's robustness and adaptability in complex electromagnetic environments, providing a solid foundation for future research and applications in advanced complex signal processing and recognition tasks.