Cross-Channels Identification for Specific Emitters Using Global and Local Semantic Consistency

By analyzing and comparing features extracted from wireless signals, Specific Emitter Identification (SEI) refers to a technology that determines and distinguishes different radio devices. However, in practical applications, wireless signal transmission is affected by varying channel conditions, leading to signal changes such as multipath effects, attenuation, and interference. Furthermore, deep learning is crucial for building automated decision systems in SEI research, but its effectiveness depends on having a certain quantity of labeled training data, a resource that is often challenging and costly to acquire in new environments. This can result in a sharp decline in model performance when applied to new environments. To address these issues, this paper proposes a cross-channel SEI method based on semantic consistency, aiming to tackle the challenge of identification with limited labeled training samples. The method integrates island loss with domain adversarial training to extract domain-invariant features by constraining Global Semantic Consistency (GSC) and Local Semantic Consistency (LSC), thus improving recognition performance under varying channel conditions. Experiments were conducted on a data collection of 16 USRP X310 radios, and the proposed approach achieved recognition accuracies of 81.25 % in the target domain and 89.70% in the source domain with just 1% labeled modeling samples in the target domain, significantly outperforming other methods. The experimental findings show that the proposed approach exhibits excellent recognition capability in both source and target domains, achieving robust and accurate SEI even with limited labeled data.