基于累积量与深度学习融合的水下调制识别模型

doi:10.3969/j.issn.1671-1122.2025.10.007

摘要/Abstract

摘要：

在复杂严苛的水声通信环境中，调制识别技术对于提升水下通信系统的抗截获能力和信息安全具有重要意义。然而，水声信道中的非线性、多径效应与强噪声干扰对自动调制识别任务构成了严峻挑战。为应对这些挑战，文章提出了一种融合小波去噪与高阶累积量特征的深度调制识别模型（CRT）。该模型通过优化残差网络与Transformer编码器结构，分别构建局部与全局时序频域特征模型，并基于水声信号时频分布特性融合高阶累积量特征，在9类典型水下调制方式下实现了93.56%的平均识别准确率，较当前最优模型提升了2.4%。特别是在-10 dB~-2 dB的低信噪比环境下，CRT模型识别准确率提升超过10%，验证了该模型在复杂水声场景下的有效性与实用价值。

关键词: 自动调制识别, 水下声学, 小波变换, 高阶累积量, 深度学习

Abstract:

In complex and demanding underwater acoustic communication environments, modulation recognition technology is crucial for improving the anti-interception capabilities and information security of underwater communication systems. However, nonlinearity, multipath effects, and strong noise interference in underwater acoustic channels pose significant challenges to automatic modulation recognition (AMR). To address these challenges, this paper proposed a deep modulation recognition model (CRT) that integrates wavelet denoising and high-order cumulants. This model optimized the residual network (ResNet) and Transformer encoder (Trans-Encoder) architectures to model local and global temporal features, respectively. Furthermore, it integrated high-order cumulants based on the time-frequency distribution of underwater acoustic signals. This model achieves an average recognition accuracy of 93.56% for nine typical underwater modulation modes, a 2.4% improvement over the current best model. In particular, in low signal-to-noise ratio (SNR) environments of -10 dB to -2 dB, the recognition accuracy improves by over 10%, demonstrating the effectiveness and practical value of the CRT model in complex underwater acoustic scenarios.

Key words: automatic modulation recognition, underwater acoustics, wavelet transform, higher-order cumulants, deep learning

中图分类号:

TP309

李古月, 张子豪, 毛承海, 吕锐. 基于累积量与深度学习融合的水下调制识别模型[J]. 信息网络安全, 2025, 25(10): 1554-1569.

LI Guyue, ZHANG Zihao, MAO Chenghai, LYU Rui. A Cumulant-Deep Learning Fusion Model for Underwater Modulation Recognition[J]. Netinfo Security, 2025, 25(10): 1554-1569.

图/表 13

图1

图2

图3

图4

图5

表1

图6

图7

图8

图9

图10

表2

表3

参考文献 36

[1]	SONG Aijun, STOJANOVIC M, CHITRE M. Editorial Underwater Acoustic Communications: Where We Stand and What is Next?[J]. IEEE Journal of Oceanic Engineering, 2019, 44(1): 1-6. doi: 10.1109/JOE.2018.2883872
[2]	JIANG Shengming. On Securing Underwater Acoustic Networks: A Survey[J]. IEEE Communications Surveys & Tutorials, 2018, 21(1): 729-752.
[3]	JIANG Shengming. On Reliable Data Transfer in Underwater Acoustic Networks: A Survey from Networking Perspective[J]. IEEE Communications Surveys & Tutorials, 2018, 20(2): 1036-1055.
[4]	HUANG Jianchun, DIAMANT R. Adaptive Modulation for Long-Range Underwater Acoustic Communication[J]. IEEE Transactions on Wireless Communications, 2020, 19(10): 6844-6857. doi: 10.1109/TWC.2020.3006230 URL
[5]	ALI M F, JAYAKODY D N K, CHURSIN Y A, et al. Recent Advances and Future Directions on Underwater Wireless Communications[J]. Archives of Computational Methods in Engineering, 2020, 27(5): 1379-1412. doi: 10.1007/s11831-019-09354-8
[6]	GHOREYSHI S M, SHAHRABI A, BOUTALEB T. Void-Handling Techniques for Routing Protocols in Underwater Sensor Networks: Survey and Challenges[J]. IEEE Communications Surveys & Tutorials, 2017, 19(2): 800-827.
[7]	ZHAO Caidan, CHEN Jingqian, WU Zhiqiang. Automatic Modulation Recognition Algorithm Based on Multi-Channel Joint Learning[J]. Netinfo Security, 2023, 23(4): 20-29.
	赵彩丹, 陈璟乾, 吴志强. 基于多通道联合学习的自动调制识别网络[J]. 信息网络安全, 2023, 23(4):20-29.
[8]	XU J L, SU Wei, ZHOU Mengchu, Likelihood-Ratio Approaches to Automatic Modulation Classification[J]. IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), 2010, 41(4): 455-469. doi: 10.1109/TSMCC.2010.2076347 URL
[9]	LI Yuxing, TANG Bingzhao, JIAO Shangbin. SO-Slope Entropy Coupled with SVMD: A Novel Adaptive Feature Extraction Method for Ship-Radiated Noise[EB/OL]. (2023-05-01)[2025-06-30]. https://doi.org/10.1016/j.oceaneng.2023.114677.
[10]	MENG Qingxin, YANG Shie. A Wave Structure Based Method for Recognition of Marine Acoustic Target Signals[EB/OL]. (2015-04-01)[2025-06-30]. https://doi.org/10.1121/1.4920186.
[11]	ZHU Daimei, MATHEWS V J, DETIENNE D H. A Likelihood-Based Algorithm for Blind Identification of QAM and PSK Signals[J]. IEEE Transactions on Wireless Communications, 2018, 17(5): 3417-3430. doi: 10.1109/TWC.2018.2811802 URL
[12]	HONG Feng, LIU Chengwei, GUO Lijuan, et al. Underwater Acoustic Target Recognition with Resnet18 on Shipsear Dataset[C]// IEEE. 2021 IEEE 4th International Conference on Electronics Technology (ICET). New York: IEEE, 2021: 1240-1244.
[13]	JIANG Zenghui, ZENG Weijun, CHEN Pu, et al. Review of Adversarial Samples for Modulation Recognition[J]. Netinfo Security, 2023, 23(6): 74-90.
	蒋曾辉, 曾维军, 陈璞, 等. 面向调制识别的对抗样本研究综述[J]. 信息网络安全, 2023, 23(6):74-90.
[14]	WANG Yan, ZHANG Hao, SANG Zhanliang, et al. Modulation Classification of Underwater Communication with Deep Learning Network[EB/OL]. (2019-04-01)[2025-06-30]. https://pubmed.ncbi.nlm.nih.gov/31065254/.
[15]	HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep Residual Learning for Image Recognition[C]// IEEE. The IEEE Conference on Computer Vision and Pattern Recognition. New York: IEEE, 2016: 770-778.
[16]	TIAN Xiao, CHEN Chao. Modulation Pattern Recognition Based on Resnet50 Neural Network[C]// IEEE. 2019 IEEE 2nd International Conference on Information Communication and Signal Processing (ICICSP). New York: IEEE, 2019: 34-38.
[17]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is All You Need[EB/OL]. (2017-09-02)[2025-06-30]. https://papers.nips.cc/paper/2017/file/3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf.
[18]	CAI Jingjing, GAN Fengming, CAO Xianghai, et al. Signal Modulation Classification Based on the Transformer Network[J]. IEEE Transactions on Cognitive Communications and Networking, 2022, 8(3): 1348-1357. doi: 10.1109/TCCN.2022.3176640 URL
[19]	LI Bing, CUI Wei, WANG Wei, et al. Two-Stream Convolution Augmented Transformer for Human Activity Recognition[C]// AAAI. The AAAI Conference on Artificial Intelligence. Menlo Park: AAAI, 2021: 286-293.
[20]	ZHANG Rui, YIN Zhendong, WU Zhilu, et al. A Novel Automatic Modulation Classification Method Using Attention Mechanism and Hybrid Parallel Neural Network[EB/OL]. (2021-01-02)[2025-06-30]. https://doi.org/10.3390/app11031327.
[21]	YU Yang, CAO Xu, ZHANG Xiaomin. Underwater Target Classification Using Deep Neural Network[C]// IEEE. 2018 Oceans-Mts/IEEE Kobe Techno-Oceans (Oto). New York: IEEE, 2018: 1-5.
[22]	STOJANOVIC M, PREISIG J. Underwater Acoustic Communication Channels: Propagation Models and Statistical Characterization[J]. IEEE Communications Magazine, 2009, 47(1): 84-89.
[23]	CAO Xu, ZHANG Xiaomin, YU Yang, et al. Deep Learning-Based Recognition of Underwater Target[C]// IEEE. 2016 IEEE International Conference on Digital Signal Processing (DSP). New York: IEEE, 2016: 89-93.
[24]	PENG Shengliang, SUN Shujun, YAO Yudong. A Survey of Modulation Classification Using Deep Learning: Signal Representation and Data Preprocessing[J]. IEEE Transactions on Neural Networks and Learning Systems, 2021, 33(12): 7020-7038. doi: 10.1109/TNNLS.2021.3085433 URL
[25]	CHANG Shuo, HUANG Sai, ZHANG Ruiyun, et al. Multitask-Learning-Based Deep Neural Network for Automatic Modulation Classification[J]. IEEE Internet of Things Journal, 2021, 9(3): 2192-2206. doi: 10.1109/JIOT.2021.3091523 URL
[26]	PENG Shengliang, JIANG Hanyu, WANG Huaxia, et al. Modulation Classification Based on Signal Constellation Diagrams and Deep Learning[J]. IEEE Transactions on Neural Networks and Learning Systems, 2018, 30(3): 718-727. doi: 10.1109/TNNLS.2018.2850703 URL
[27]	ZHANG Zufan, LUO Hao, WANG Chun, et al. Automatic Modulation Classification Using CNN-LSTM Based Dual-Stream Structure[J]. IEEE Transactions on Vehicular Technology, 2020, 69(11): 13521-13531. doi: 10.1109/TVT.2020.3030018
[28]	KE Ziqi, VIKALO H. Real-Time Radio Technology and Modulation Classification via an LSTM Auto-Encoder[J]. IEEE Transactions on Wireless Communications, 2021, 21(1): 370-382. doi: 10.1109/TWC.2021.3095855 URL
[29]	GHASEMZADEH P, HEMPEL M, SHARIF H. GS-QRNN: A High-Efficiency Automatic Modulation Classifier for Cognitive Radio IoT[J]. IEEE Internet of Things Journal, 2022, 9(12): 9467-9477. doi: 10.1109/JIOT.2022.3141032 URL
[30]	XU Jialang, LUO Chunbo, PARR G, et al. A Spatiotemporal Multi-Channel Learning Framework for Automatic Modulation Recognition[J]. IEEE Wireless Communications Letters, 2020, 9(10): 1629-1632. doi: 10.1109/LWC.2020.2999453
[31]	YUE Zhou, WEI Kong, QING Xu. A Novel Modeling and Recognition Method for Underwater Sound Based on HMT in Wavelet Domain[C]// Springer. Australasian Joint Conference on Artificial Intelligence. Heidelberg: Springer, 2004: 332-343.
[32]	SAFI S, FRIKEL M, ZEROUAL A, et al. Higher Order Cumulants for Identification and Equalization of Multicarrier Spreading Spectrum Systems[J]. Journal of Telecommunications and Information Technology, 2011 (2): 74-84.
[33]	ZHANG Ruoyun, CHANG Shuo, WEI Zhiqing, et al. Modulation Classification of Active Attacks in Internet of Things: Lightweight MCBLDN with Spatial Transformer Network[J]. IEEE Internet of Things Journal, 2022, 9(19): 19132-19146. doi: 10.1109/JIOT.2022.3163892 URL
[34]	LIANG Zhi, TAO Mingliang, XIE Jian, et al. A Radio Signal Recognition Approach Based on Complex-Valued CNN and Self-Attention Mechanism[J]. IEEE Transactions on Cognitive Communications and Networking, 2022, 8(3): 1358-1373. doi: 10.1109/TCCN.2022.3179450 URL
[35]	ZHANG Weilong, YANG Xinghai, LENG Changli, et al. Modulation Recognition of Underwater Acoustic Signals Using Deep Hybrid Neural Networks[J]. IEEE Transactions on Wireless Communications, 2022, 21(8): 5977-5988. doi: 10.1109/TWC.2022.3144608 URL
[36]	LI Juan, JIA Qingning, CUI Xuerong, et al. Automatic Modulation Recognition of Underwater Acoustic Signals Using a Two-Stream Transformer[J]. IEEE Internet of Things Journal, 2024, 11(10): 18839-18851. doi: 10.1109/JIOT.2024.3367852 URL

参数	参数值
实验环境	峡湾
实验时间	6个月
水深	10 m
多普勒频率覆盖范围	7.8 kHz
发射与接收距离	750 m

模型	准确率	模型参数量	训练时间/h
CRT	93.56%	503000	0.66
CNN^[19]	60.51%	2803000	0.59
LSTM^[27]	66.67%	201000	0.26
MCLDNN^[33]	69.00%	406000	2.70
CCNN&Atten^[34]	76.23%	481000	1.61
DHN^[35]	87.06%	537000	1.23
TSTR^[36]	91.12%	346000	1.54

模块	参数量	密集区域	优化后的参数量	优化后的计算量占比（FLOPs）
ResNet主干网络	544000	深层卷积层	384000	68.2%
Transformer编码器	376000	自注意力投影矩阵	94000	28.5%
分类头（FC层）	13000	全连接层	13000	3.3%