Research on interference recognition technique based on adaptive multimodal convolutional denoising network

Interference recognition is a crucial component of communication anti-jamming systems. To address the challenge of interference recognition in complex adversarial environments, this paper proposes a novel deep learning model based on multi-scale convolutions: the Adaptive Multimodal Convolutional Denoising Network (AMCDN). This model incorporates Variational Modal Decomposition (VMD) combined with a channel attention mechanism to suppress misleading information within the interference signals. Furthermore, leveraging frequency domain features on top of the temporal convolutional network (TCN) enhances the model's robustness and improves the accuracy of interference recognition. Experimental results on our simulated dataset demonstrate that AMCDN achieves a 60.01% improvement in F1-score compared to the TCN baseline.