Communication Specific Emitter Identification Based on Deep Convolutional Feature Matching

Communication specific emitter identification (CSEI) exploits the unintentional modulation features embedded in an emitter’s transmitted signals, known as radio frequency fingerprints (RFFs), to distinguish individual emitters. CSEI has been widely applied in both military and civilian domains for reliable emitter identification. However, traditional approaches often depend on prior information and handcrafted features, which restrict their identification performance. To overcome these limitations, this paper proposes a novel deep convolutional feature matching framework for communication specific emitter identification. The method leverages data augmentation and preprocessing strategies to enrich the training dataset, thereby enhancing model identification accuracy and robustness. A multi-dimensional convolution architecture is designed, integrating two-dimensional, one-dimensional, and dilated convolutions for hierarchical feature extraction. Moreover, a joint loss function combining cross-entropy and matrix correlation loss is introduced for effective model optimization. Experimental evaluations demonstrate that, compared to state-of-the-art methods, the proposed approach achieves superior identification accuracy and robustness in complex electromagnetic environments.