Communication Signal Recognition Based on Time-Frequency and Bispectral Multi-Feature Fusion

In the complex electromagnetic environment, the existing single-feature-domain methods for non-cooperative communication signal recognition have limited representation ability in diverse signal scenarios, leading to insufficient recognition accuracy. To address this issue, this paper proposes a deep learning method based on time-frequency bispectrum multifeature fusion. The proposed method innovatively constructs a joint feature extraction framework that combines time-frequency analysis and bispectrum analysis. Time-frequency analysis is employed to capture the dynamic characteristics of signals, while bispectrum analysis is used to extract higher-order statistical information. This combination enables complementary feature representation. Furthermore, a deep model integrating the ResNet18 network and SE attention mechanism is designed. The residual structure in ResNet18 is utilized to mine spatial deep features, and the channel attention mechanism adaptively optimizes the weights of key features. This enhances the robustness of the model under low signal-to-noise ratio (SNR) conditions. Experimental results demonstrate that the proposed method achieves better recognition accuracy than traditional single-feature-domain methods in low SNR scenarios.