变调制场景下调制信息引导的双分支通信辐射源个体识别方法

王邓曦; 孙佳星; 刘辉; 黄科举; 王万泽; 杨俊安

doi:10.11999/JEIT260349

变调制场景下调制信息引导的双分支通信辐射源个体识别方法

doi: 10.11999/JEIT260349 cstr: 32379.14.JEIT260349

王邓曦¹,
孙佳星¹,
刘辉¹,
黄科举^{1, 2, ,},
王万泽¹,
杨俊安¹

1.
国防科技大学电子对抗学院合肥 230000
2.
安徽省电子制约技术重点实验室合肥 230000

基金项目: 国防科技大学自主科研基金项目

详细信息

作者简介:
王邓曦：男，博士生，研究方向为辐射源个体识别

孙佳星：男，博士，研究方向为小样本学习

刘辉：男，副教授，研究方向为通信对抗、智能信息处理

黄科举：男，讲师，研究方向为智能信号处理

王万泽：男，硕士生，研究方向为辐射源个体识别

杨俊安：男，教授，研究方向为通信对抗、智能信息处理

通讯作者:
黄科举　huangkeju@nudt.edu.cn

中图分类号: TN911.7
计量
- 文章访问数: 161
- HTML全文浏览量: 38
- PDF下载量: 6
- 被引次数: 0
出版历程
- 收稿日期: 2026-03-26
- 修回日期: 2026-09-09
- 录用日期: 2026-09-15
- 网络出版日期: 2026-10-10

A Modulation-Information-guided Dual-Branch Specific Emitter Identification Method Under Variable Modulation Scenarios

1.
College of Electronic Engineering, National University of Defense Technology, Hefei 230000, China
2.
Anhui province key laboratory of electronic restriction, Hefei 230000, China

Funds: Innovation Research Foundation of National University of Defense Technology

摘要

摘要: 通信辐射源调制参数的动态变化，给辐射源个体识别(SEI)性能带来严峻挑战。针对现有基于域适应的变调制SEI方法存在特征鲁棒性不足、过度依赖目标域数据重训练的问题，该文首次将域泛化方法引入该任务，提出一种调制信息引导的双分支训练框架(MIG-DTF)及测试阶段模型适应策略(TMA)。MIG-DTF采用个体识别主分支与调制分类辅助分支协同训练模式。主分支引入余弦聚合损失以强化指纹特征的区分度，辅助分支通过最大化交叉预测熵损失，引导主分支抑制指纹特征中的调制相关信息，从而实现对调制变化鲁棒的指纹特征提取。测试时采用TMA自适应策略单步微调批归一化层参数，以改善失配参数对特征造成的分布偏移，进一步提升识别准确率。基于真实数据集，在变调制样式、变载频、变符号速率3种典型场景下开展对比实验，结果表明，所提方法的识别率显著优于主流域泛化方法，较基线方法在3个场景中分别提升20.39%, 12.43%和14.51%。特征可视化分析与抗噪声实验进一步验证了该方法能够有效提取对调制变化鲁棒的指纹特征，以及良好的抗噪声鲁棒性。此外，得益于TMA单步微调的极低训练开销，及无需目标域训练样本特性，该方法更具实际部署前景。
- 通信辐射源个体识别 /
- 深度学习 /
- 领域泛化 /
- 变调制参数
Abstract: Objective Specific Emitter Identification (SEI) distinguishes individual emitters by exploiting Radio Frequency Fingerprints (RFFs) and has important applications in military reconnaissance and Internet of Things (IoT) authentication. However, conventional SEI methods generally assume fixed modulation parameters, whereas practical communication systems often employ dynamically varying modulation schemes. Existing Domain Adaptation (DA) methods for variable-modulation SEI require target-domain data retraining and still suffer from insufficient robustness of extracted features under unknown modulation variations. To address these problems, this paper introduces Domain Generalization (DG) into variable-modulation SEI and proposes a Modulation-Information-Guided Dual-branch Training Framework (MIG-DTF) with a Testing-stage Model Adaptation strategy (TMA). The proposed framework aims to learn modulation-invariant RFFs from source-domain data and improve SEI performance under unknown modulation parameter variations without large-scale target-domain retraining. Methods The proposed method consists of training and testing stages. During the training stage, MIG-DTF employs an emitter identification main branch and a modulation classification auxiliary branch for collaborative optimization. The main branch utilizes cosine aggregation loss to enhance the intra-class compactness and inter-class discriminability of RFFs. Meanwhile, the auxiliary branch maximizes cross-prediction entropy to suppress modulation-related information contained in the RFFs extracted by the main branch, thereby encouraging the model to learn modulation-invariant fingerprint features. During the testing stage, the proposed TMA strategy performs one-step fine-tuning of the Batch Normalization (BN) layer parameters using test samples under an entropy minimization objective. By adapting BN statistics and affine parameters to the testing distribution, TMA alleviates feature distribution drift caused by modulation mismatch and further improves identification accuracy. Results and Discussions Experiments are conducted on a real-world variable-modulation dataset containing three typical variation scenarios: modulation type, carrier frequency, and symbol rate. The proposed method achieves identification accuracies of 79.31% (Table 1), 87.39% (Table 2), and 88.33% (Table 3) in the three scenarios, respectively, outperforming mainstream DG methods. Ablation experiments demonstrate that both MIG-DTF and TMA independently contribute to performance improvement, and their combination provides further gains (Fig. 4). Feature visualization results show that the proposed method obtains more compact intra-class distributions and clearer inter-class boundaries under different modulation conditions (Fig. 5). Correlation analysis between extracted features and modulation labels further verifies that the proposed method reduces the dependence of RFFs on modulation information while preserving emitter-discriminative characteristics (Table 5). In anti-noise experiments with Signal-to-Noise Ratio (SNR) values from 10 dB to 30 dB, the proposed method consistently outperforms the baseline method and achieves an accuracy improvement of approximately 10% at 10 dB SNR (Fig. 7), demonstrating stronger robustness under noisy conditions. In addition, overhead analysis shows that TMA requires only a small number of parameter updates and eliminates the need for target-domain training data, reducing adaptation costs for practical deployment. Conclusions This paper proposes a robust SEI framework for variable modulation scenarios by combining the training-stage MIG-DTF and testing-stage TMA strategies. MIG-DTF enables the extraction of modulation-invariant RFFs by separating emitter-related fingerprint information from modulation-related information, whereas TMA provides efficient adaptation to unknown testing distributions with low computational overhead. Experimental results on a real-world dataset demonstrate that: (1) the proposed method achieves higher identification accuracy than mainstream DG methods under multiple modulation parameter variations; (2) the extracted features exhibit stronger modulation invariance and emitter discrimination capability; (3) the proposed method maintains better performance under noise interference; and (4) TMA enables efficient adaptation without target-domain retraining. Future work will investigate more complex scenarios, including joint variations of multiple modulation parameters and time-varying channel conditions.
- Specific Emitter Identification(SEI) /
- Deep Learning(DL) /
- Domain Generalization(DG) /
- Variable modulation

HTML全文

图 1 本文方法的总体框架图

下载: 全尺寸图片幻灯片

图 2 测试阶段TMA方法示意图

下载: 全尺寸图片幻灯片

图 3 Net_F与Net_M网络结构图

下载: 全尺寸图片幻灯片

图 4 3种变参数场景下的消融实验

下载: 全尺寸图片幻灯片

图 5 Baseline与本文方法的特征可视化对比

下载: 全尺寸图片幻灯片

图 6 指纹特征与调制特征进行调制分类的训练准确率曲线

下载: 全尺寸图片幻灯片

图 7 识别准确率随信噪比变化曲线

下载: 全尺寸图片幻灯片

表 1 变调制样式场景下不同方法的识别准确率对比(%)

算法	BP, QP→8P	BP, 8P→QP	QP, 8P→BP	QP, 8P→QAM	BP, QP→QAM	均值
Baseline	70.65±4.83	85.04±4.15	62.76±4.85	34.66±0.45	41.49±7.36	58.92
CORAL	91.00±0.24	86.53±3.10	65.56±2.57	33.57±3.84	48.27±6.20	64.99
V-REx	80.29±1.77	86.56±1.86	66.00±1.36	38.13±0.27	48.33±2.00	63.86
RIEI	89.62±3.64	87.31±2.65	65.31±0.98	37.38±0.36	39.84±0.31	63.89
CDDG	90.45±4.44	89.31±0.16	66.62±0.80	32.66±2.89	49.27±8.23	65.66
本文方法	84.89±2.72	89.11±0.44	81.74±2.67	66.34±0.31	74.49±2.67	79.31

下载: 导出CSV

表 2 变载频场景下不同方法的识别准确率对比(%)

算法	F1, F2→F3	F1, F3→F2	F2, F3→F1	均值
Baseline	80.45±3.03	66.93±2.98	77.49±2.37	74.96
CORAL	86.87±2.89	63.56±2.73	75.69±2.18	75.37
V-REx	88.69±1.52	73.51±2.33	80.38±1.08	80.86
RIEI	82.64±0.93	68.98±2.04	77.73±1.46	76.45
CDDG	86.40±1.42	66.24±1.51	76.45±1.30	76.36
本文方法	89.26±0.13	88.49±0.14	84.43±0.27	87.39

下载: 导出CSV

表 3 变符号速率场景下不同方法的识别准确率对比(%)

算法	R1, R2→R3	R1, R3→R2	R2, R3→R1	均值
Baseline	55.33±0.99	94.51±4.11	71.62±4.67	73.82
CORAL	56.33±1.87	98.98±0.38	69.07±1.20	74.79
V-REx	61.29±2.76	95.53±2.08	69.82±4.42	75.55
RIEI	56.50±1.06	98.53±1.16	82.60±3.91	79.21
CDDG	57.78±0.22	99.13±0.11	79.86±6.71	78.92
本文方法	66.58±0.71	99.80±0.04	98.62±0.53	88.33

下载: 导出CSV

表 4 TMA与DA的开销对比

阶段	更新参数量	单轮耗时(s)
MIG-DTF训练	1.12 M	5.36
测试	0	0.75
TMA模型微调	1.60 K	0.79
DA模型重训练	561.27 K	2.15

下载: 导出CSV

表 5 特征与调制标签、个体标签的第一典型相关系数$ \rho $值

方法	实验组	特征与调制标签	特征与个体标签
Baseline	BP, 8P→QP	0.9636	0.9975
Baseline	R1, R3→R2	0.9823	0.9905
本文方法	BP, 8P→QP	0.3833	0.9926
本文方法	R1, R3→R2	0.4185	0.9870

下载: 导出CSV

参考文献(26)

[1]	刘高辉, 席宏恩. 改进变分模态分解与多特征的通信辐射源个体识别方法[J]. 电子与信息学报, 2024, 46(10): 4044–4052. doi: 10.11999/JEIT231348. LIU Gaohui and XI Hongen. Individual identification method for communication emitters based on improved variational modal decomposition and multiple features[J]. Journal of Electronics & Information Technology, 2024, 46(10): 4044–4052. doi: 10.11999/JEIT231348.
[2]	DANEV B, HEYDT-BENJAMIN T S, and ČAPKUN S. Physical-layer identification of RFID devices[C]. The 18th Conference on USENIX Security Symposium, Montreal, Canada, 2009: 199–214.
[3]	ZHANG Jingwen, WANG Fanggang, DOBRE O A, et al. Specific emitter identification via Hilbert-Huang transform in single-hop and relaying scenarios[J]. IEEE Transactions on Information Forensics and Security, 2016, 11(6): 1192–1205. doi: 10.1109/TIFS.2016.2520908.
[4]	SATIJA U, TRIVEDI N, BISWAL G, et al. Specific emitter identification based on variational mode decomposition and spectral features in single hop and relaying scenarios[J]. IEEE Transactions on Information Forensics and Security, 2019, 14(3): 581–591. doi: 10.1109/TIFS.2018.2855665.
[5]	FU Hua, SUN Yanfeng, PENG Linning, et al. Channel-resilient RF fingerprint identification based on nonlinear features with memory effect[J]. IEEE Communications Letters, 2024, 28(4): 798–802. doi: 10.1109/LCOMM.2024.3364494.
[6]	肖易寒, 王博煜, 于祥祯, 等. 基于双路射频指纹卷积神经网络与特征融合的雷达辐射源个体识别[J]. 电子与信息学报, 2024, 46(8): 3238–3245. doi: 10.11999/JEIT231236. XIAO Yihan, WANG Boyu, YU Xiangzhen, et al. Radar emitter identification based on dual radio frequency fingerprint convolutional neural network and feature fusion[J]. Journal of Electronics & Information Technology, 2024, 46(8): 3238–3245. doi: 10.11999/JEIT231236.
[7]	GIUSTINIANO D, SCHALCH M, LIECHTI M, et al. Interference suppression in bandwidth hopping spread spectrum communications[C]. The 11th ACM Conference on Security & Privacy in Wireless and Mobile Networks, Stockholm, Sweden, 2018: 134–143. doi: 10.1145/3212480.3212484.
[8]	HUANG Keju, YANG Junan, LIU Hui, et al. Deep adversarial neural network for specific emitter identification under varying frequency[J]. Bulletin of the Polish Academy of Sciences-Technical Sciences, 2021, 69(2): e136737. doi: 10.24425/bpasts.2021.136737.
[9]	ZHANG Xinliang, LI Tianyun, GONG Pei, et al. Variable-modulation specific emitter identification with domain adaptation[J]. IEEE Transactions on Information Forensics and Security, 2023, 18: 380–395. doi: 10.1109/TIFS.2022.3223794.
[10]	YIN Lijun, FU Xue, SHI Shengnan, et al. Few-shot domain adaption-based specific emitter identification under varying modulation[C]. 2023 IEEE 23rd International Conference on Communication Technology (ICCT), Wuxi, China, 2023: 1439–1443. doi: 10.1109/ICCT59356.2023.10419733.
[11]	HE Yazhe and ZHANG Qin. Cross-modulation specific emitter identification based on source-free domain adaptation[C]. 2025 10th International Conference on Computer and Communication System (ICCCS), Chengdu, China, 2025: 458–463. doi: 10.1109/ICCCS65393.2025.11069648.
[12]	LI Ying, WANG Xingwei, ZENG Rongfei, et al. Federated domain generalization: A survey[J]. Proceedings of the IEEE, 2025, 113(4): 370–410. doi: 10.1109/JPROC.2025.3596173.
[13]	王雪安. 基于循环谱的调制方式盲识别与频偏估计[D]. [硕士论文], 北京邮电大学, 2017. WANG Xuean. Modulation blind recognition and frequency offset estimation based on cyclic spectrum[D]. [Master dissertation], Beijing University of Posts and Telecommunications, 2017.
[14]	DI ROSA G and RICHTER A. Low complexity blind carrier phase recovery for probabilistically shaped QAM[J]. IEEE Photonics Technology Letters, 2020, 32(17): 1109–1112. doi: 10.1109/LPT.2020.3012970.
[15]	LIM D. A modified Gardner detector for symbol timing recovery of M-PSK signals[J]. IEEE Transactions on Communications, 2004, 52(10): 1643–1647. doi: 10.1109/TCOMM.2004.836450.
[16]	MAO Anqi, MOHRI M, and ZHONG Yutao. Cross-entropy loss functions: Theoretical analysis and applications[C]. 40th International Conference on Machine Learning, Honolulu, USA, 2023: 23803–23828.
[17]	MAHAJAN D, TOPLE S, and SHARMA A. Domain generalization using causal matching[C]. The 38th International Conference on Machine Learning, 2021: 7313–7324.
[18]	WU Zhize, DU Changjiang, WANG Xinqi, et al. Domain Adaptation via Feature Disentanglement for cross-domain image classification[J]. Applied Soft Computing, 2025, 172: 112868. doi: 10.1016/j.asoc.2025.112868.
[19]	SU Kejia, LIU Kai, WAN Bo, et al. Multisource adversarial feature disentanglement method for cross-subject gesture recognition using sEMG signals[J]. IEEE Transactions on Instrumentation and Measurement, 2025, 74: 4508912. doi: 10.1109/TIM.2025.3557823.
[20]	WANG Dequan, SHELHAMER E, LIU Shaoteng, et al. Tent: Fully test-time adaptation by entropy minimization[C]. 9th International Conference on Learning Representations, 2021.
[21]	SUN Baochen and SAENKO K. Deep CORAL: Correlation alignment for deep domain adaptation[C]. Computer Vision – ECCV 2016 Workshops, Amsterdam, The Netherlands, 2016: 443–450.
[22]	KRUEGER D, CABALLERO E, JACOBSEN J H, et al. Out-of-distribution generalization via risk extrapolation[C]. Proceedings of the 38th International Conference on Machine Learning, 2021: 5815–5826.
[23]	ZHANG Ying, LI Qiang, LIU Hongli, et al. Domain generalization for cross-receiver radio frequency fingerprint identification[J]. IEEE Internet of Things Journal, 2025, 12(5): 5207–5218. doi: 10.1109/JIOT.2024.3486068.
[24]	JIA Linshan, CHOW T W C, and YUAN Yixuan. Causal Disentanglement Domain Generalization for time-series signal fault diagnosis[J]. Neural Networks, 2024, 172: 106099. doi: 10.1016/j.neunet.2024.106099.
[25]	LU Weipeng and YAN Xuefeng. Visual high dimensional industrial process monitoring based on deep discriminant features and t-SNE[J]. Multidimensional Systems and Signal Processing, 2021, 32(2): 767–789. doi: 10.1007/s11045-020-00758-5.
[26]	HARDOON D R, SZEDMAK S, and SHAWE-TAYLOR J. Canonical correlation analysis: An overview with application to learning methods[J]. Neural Computation, 2004, 16(12): 2639–2664. doi: 10.1162/0899766042321814.