Multi-Domain Networks for Wireless Interference Recognition

In the field of military communications, electromagnetic interference has posed a serious threat to wireless communication systems, and wireless interference recognition (WIR) is considered as one of the most indispensable steps to defend against adversarial attacks for anti-interference communication. Following the success in advancing many disciplines, deep learning (DL) brings revolutionary changes to WIR. However, most conventional DL-based methods only use single domain information of interference signals, such as time domain or frequency domain, resulting in low recognition performance under low interference-to-noise ratio (INR). To this end, we propose to exploit multi-domain information of interference signals to take full advantage of the complementarity between domains for WIR. We propose multi-domain networks (MDN) that consist of convolutional layers and transformers to simultaneously strengthen locality and establish long-range dependencies for extracting the features of each domain information. Additionally, to fuse the extracted features from multiple transformation domains together, we propose novel fusion mechanisms, which can be seamlessly incorporated into the MDN. The experimental results demonstrate that the proposed methods notably boost recognition performance as compared to conventional methods for WIR.