Specific emitter identification (SEI) is extracting the features of the received radio signals and determining the emitter individuals that generate the signals. Although deep learning-based methods have been effectively applied for SEI, their performance declines dramatically with the smaller number of labeled training samples and in the presence of significant noise. To address this issue, we propose an improved Bootstrap Your Own Late (BYOL) self-supervised learning scheme to fully exploit the unlabeled samples, which comprises the pretext task adopting contrastive learning conception and the downstream task. We designed three optimized data augmentation methods for communication signals in the former task to serve the contrastive concept. We built two neural networks, online and target networks, which interact and learn from each other. The proposed scheme demonstrates the generality of handling the small and sufficient sample cases across a wide range from 10 to 400, being labeled in each group. The experiment also shows promising accuracy and robustness where the recognition results increase at 3-8% from 3 to 7 signal-to-noise ratio (SNR). Our scheme can accurately identify the individual emitter in a complicated electromagnetic environment.

Specific emitter identification (SEI) refers to distinguishing emitters using individual features extracted from wireless signals. The current SEI methods have proven to be accurate in tackling large labeled data sets at a high signal-to-noise ratio (SNR). However, their performance declines dramatically in the presence of small samples and a significant noise environment. To address this issue, we propose a complex self-supervised learning scheme to fully exploit the unlabeled samples, comprised of a pretext task adopting the contrastive learning concept and a downstream task. In the former task, we design an optimized data augmentation method based on communication signals to serve the contrastive conception. Then, we embed a complex-valued network in the learning to improve the robustness to noise. The proposed scheme demonstrates the generality of handling the small and sufficient samples cases across a wide range from 10 to 400 being labeled in each group. The experiment also shows a promising accuracy and robustness where the recognition results increase at 10–16% from 10–15 SNR.

Radio Frequency Fingerprint Identification (RFFI) is a promising technology for Internet of Things (IoT) device authentication, which uses a device’s unique hardware imperfections patterns as the “fingerprint”, providing a highly secure, device-specific form of authentication for IoT devices. Due to the influence of inherent hardware imperfections of receivers, the performance of RFFI is markedly reduced when the receiver is changed. In engineering practice, when the receiver changes, the data on the new receiver needs to be relabeled and retrained, resulting in significant resource consumption. To address this limitation, a few studies utilized Domain Adaptation (DA) to avoid relabeling data on the new receiver. However, existing DA methods focused exclusively on aligning the domain distribution at the feature level, resulting in performance limitations and neglecting the utilization of unlabeled target domain data. To address these shortcomings, we propose a two-stage Domain Adaptation method for cross-receiver RFFI. The proposed method sequentially performs data-level domain alignment based on Tensor Alignment Invariant Subspace Learning and feature-level domain alignment based on Subdomain Adaptation to achieve a comprehensive domain alignment. In addition, we introduce a pseudo-labeling method based on Confident Learning to improve the utilization of unlabeled data. Experimental results on the publicly available WiSig dataset show that the proposed method achieves at least 30% performance improvement in cross-receiver scenarios, and outperforms other existing Domain Adaptation methods.

Radio frequency fingerprint identification (RFFI) is critical for securing Internet of Things (IoT) devices and wireless communication systems. However, existing deep learning approaches often suffer a sharp degradation in accuracy when labeled data is limited. To address this issue, this letter introduces a novel RFFI method, SRP-CBL, which combines signal recurrence plots and convolutional broad learning. It converts RF time series into recurrence plots and applies convolution operations for feature extraction within the broad learning framework. By leveraging sparse connectivity and weight sharing, the model reduces complexity and improves generalization in low-label regimes. Experiments on a public dataset demonstrate that SRP-CBL consistently outperforms state-of-the-art methods in accuracy under limited training data. The dataset can be downloaded from https://cores.ee.ucla.edu/downloads/datasets/wisig/.

Specific emitter identification (SEI) is essential for enhancing the security of Internet of Things (IoT) devices and wireless communication networks. Incremental learning, capable of adapting to real-time data streams with new categories, performs effectively when abundant labeled samples are available. However, the identification accuracy of current incremental learning-based SEI methods decreases in the presence of limited labeled samples. To address this issue, we integrate self-supervised learning (SSL) into incremental learning, structured into three stages: pre-training task, pretext task, and downstream task. The proposed method utilizes contrastive learning in the pretext task to effectively utilize unlabeled samples, thus reducing the necessary number of labeled samples and improving recognition accuracy even with small samples. Experimental results on real radio stations and ADS-B datasets demonstrate that the proposed method achieves an average accuracy of 80.50% on the ADS-B dataset and 92.97% on the radio stations dataset with 100 labeled new category training samples. The code can be downloaded from https://github.com/Simple-up/ISSL.

Specific emitter identification (SEI) is a critical technology for wireless device security authentication. However, in practical applications, time-varying channel characteristics often obscure the radio frequency fingerprint features of emitters, significantly degrading recognition performance. Additionally, the high cost of sample labeling limits the effectiveness of incremental learning methods that rely on labeled data. To address this challenge, we propose a cross-domain semi-supervised IL approach for SEI (CDSIL-SEI). This method leverages domain adaptation to effectively utilize unlabeled samples, narrowing the feature gap between the source and target domains, and enabling incremental learning with limited labeled samples. Experimental results on the WiSig dataset show that CDSIL-SEI improves target-domain accuracy by about 15% over domain adaptation methods and achieves more than 10% higher source-domain accuracy than incremental learning methods, effectively alleviating catastrophic forgetting.