Variable-Channel Specific Emitter Identification Method Using Semi-Supervised Domain Adaptation | IEEE Conference Publication | IEEE Xplore

Variable-Channel Specific Emitter Identification Method Using Semi-Supervised Domain Adaptation


Abstract:

Specific emitter identification (SEI) is a technique that identifies radio equipment by analyzing and comparing the characteristics of radio frequency signals. Deep learn...Show More

Abstract:

Specific emitter identification (SEI) is a technique that identifies radio equipment by analyzing and comparing the characteristics of radio frequency signals. Deep learning (DL), which fully exploits the hidden features of data and automatically makes classification decisions, has shown good performance in SEI. However, the success of DL is based on the training of abundant labeled samples, but collecting and annotating such data is challenging. In addition, the received data contains not only device-relevant features, but also device-irrelevant features, models trained under fixed channels often lack channel robustness. Considering these problems, we propose a semantic consistency-powered semi-supervised domain adaptation (SSDA) method for variable-channel SEI (VC-SEI). Specifically, we innovatively start from the perspective of semantic consistency, introduce domain adversarial training to constrain global semantic consistency (GSC) to extract channel-irrelevant features, and design two loss functions to constrain local semantic consistency (LSC) to extract category-relevant features, so as to jointly realize domain adaptation. The proposed SSDA-based VC-SEI method is evaluated on a dataset from 16 USRP X310 Radios, our method obtains 84.20% identification accuracy on the target domain and 92.00% identification accuracy on the source domain when only 1% of training samples in the target domain are labeled, which is much higher than other methods. The results demonstrate that the proposed SSDA-based VC-SEI method achieve better variable-channel emitter identification performance than other methods. Our code can be downloaded from https://github.com/frownean/SSDA.
Date of Conference: 20-22 October 2023
Date Added to IEEE Xplore: 12 February 2024
ISBN Information:

ISSN Information:

Conference Location: Wuxi, China

References

References is not available for this document.