An enhanced signal detection method for a set of multiple access sonar detection waveforms
-
摘要:
为了提高浅海水声环境多址声呐探测性能, 提出了一种多址声呐探测波形及其增强检测方法。建立了浅海回波信道模型, 生成了浅海多址声呐回波数据; 将基于生成对抗网络(GAN)结构的信号增强网络与基于卷积–全连接网络结构的分类网络相结合, 引入融合梯度(FG)训练方法, 设计了WGAN-FG信号增强检测器; 基于WGAN-FG信号增强检测器和传统卷积神经网络、循环神经网络、生成对抗网络及副本相关检测器, 对浅海多址声呐回波检测性能进行了仿真分析。结果表明, 基于深度学习的神经网络检测器相比传统的副本相关检测器具有更好的多径、多普勒和互干扰抑制能力, 同时具备目标速度识别能力; 而在神经网络检测器中, WGAN-FG信号增强检测器在强干扰或强畸变条件下表现出更优的检测性能和目标速度判别能力。
Abstract:To improve the detection performance of multiple access sonar in shallow water acoustic channels, a set of multiple access sonar detection waveforms and their enhanced detection method are proposed. The shallow water target echo channel is modeled and the multiple access sonar data are generated. The WGAN-FG enhanced signal detector is designed, which consists of a generate adversarial network (GAN) signal enhancer and a fully connected convolutional neural network (CNN) classifier, using the fusion gradient (FG) training method. The detection performances of shallow water multiple access sonar echoes are analyzed by simulation methods using the WGAN-FG enhanced signal detector and the traditional detectors such as the CNN, recurrent neural network (RNN), GAN and replica correlation (RC). Simulation results show that the deep learning based neural network detectors have better multi-path, Doppler and mutual interference suppression capabilities than the RC detector. The neural network detectors also have the ability to measure the target speed. And the WGAN-FG enhanced signal detector has better detection performance and target speed identification ability than the other neural network detectors under strong interference or distortion conditions.
-
-
图 5 4组不同编码多重正交DFCW-OFD-LFM波形时频图 (a)
pn ,cn ,qn 分别为(1,1,−1,−1,1,1,1), (1,2,7,5,4,6,3), (1,−1,1,1,1,−1,1); (b)pn ,cn ,qn 分别为(−1,−1,−1,1,−1,−1,1), (1,3,6,2,7,4,5), (−1,−1,1,1,1,−1,−1); (c)pn ,cn ,qn 分别为(1,−1,1,−1,1,−1,1), (1,4,2,6,7,3,5), (1,1,−1,1,1,−1,1); (d)pn ,cn ,qn 分别为(1,1,−1,1,1,−1,1), (2,5,1,7,6,3,4), (1,−1,1,−1,1,−1,1)表 1 发射波形参数设置
采样率fs(kHz) 40 信号总脉宽T(ms) 100 信号起始频率fl(kHz) 10 子信号带宽Bs(kHz) 1.64 信号终止频率fh(kHz) 15 子信号频率间隔Δf(kHz) 0.56 信号总带宽B(kHz) 5 子信号脉宽Tp(ms) 14.29 表 2 虚警率统计结果
模型 WGAN -FG WGAN CNN-FCN GRU-FCN RC 虚警率 0.31% 0.32% 0.32% 0.30% 0.31% 表 3 WGAN-FG, WGAN, CNN-FCN, GRU-FCN的Macro-F1与Micro-F1
WGAN-FG WGAN CNN-FCN GRU-FCN Macro-F1 0.8291 0.7898 0.7142 0.7413 Micro-F1 0.8294 0.7902 0.7163 0.7415 -
[1] 许彦伟, 薛勐, 刘明刚, 等. 多无人水下航行器协同探测声呐宽带波形设计与性能分析. 电子与信息学报, 2023; 45(10): 3796−3804 DOI: 10.11999/JEIT221265 [2] Kroh P K, Simon R, Rupitsch S J. Classification of sonar targets in air: A neural network approach. Sensors, 2019; 19(5): 1176 DOI: 10.3390/s19051176
[3] Ahmed F, Xiang X, Jiang C, et al. Survey on traditional and AI based estimation techniques for hydrodynamic coefficients of autonomous underwater vehicle. Ocean Eng., 2023; 268: 113300 DOI: 10.1016/j.oceaneng.2022.113300
[4] 周天, 司吉坤, 杜伟东, 等. 采用GAF-D3Net深度学习网络的水下目标有源识别方法. 声学学报, 2023; 48(5): 950−958 DOI: 10.12395/0371-0025.2022016 [5] 李琛, 黄兆琼, 徐及, 等. 使用深度学习的多通道水下目标识别. 声学学报, 2020; 45(4): 506−514 DOI: 10.15949/j.cnki.0371-0025.2020.04.007 [6] Nie J, Xiao Y, Huang L, et al. Time-frequency analysis and target recognition of HRRP based on CN-LSGAN, STFT, and CNN. Complexity, 2021: 1−10 DOI: 10.1155/2021/6664530
[7] Sigillito V G, Wing S P, Hutton L V, et al. Classification of radar returns from the ionosphere using neural networks. Johns Hopkins APL Technical Digest, 1989; 10(3): 262−266
[8] Wan C, Si W, Deng Z. Research on modulation recognition method of multi-component radar signals based on deep convolution neural network. IET Radar Sonar Navig., 2023; 17(9): 1313−1326 DOI: 10.1049/rsn2.12421
[9] Wang L, Tang J, Liao Q. A study on radar target detection based on deep neural networks. IEEE Sens. Lett., 2019; 3(3): 1−4 DOI: 10.1109/LSENS.2019.2896072
[10] Lee K G, Oh S J. Detection of frequency-hopping signals with deep learning. IEEE Commun. Lett., 2020; 24(5): 1042−1046 DOI: 10.1109/LCOMM.2020.2971216
[11] Goodfellow I, Pouget-Abadie J, Mirza M, et al. Generative adversarial networks. Commun. ACM, 2020; 63(11): 139−144 DOI: 10.1145/3422622
[12] Radford A, Metz L, Chintala S. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint: 1511.06434, 2015
[13] Yin Y, Qiu J, Sun S, et al. Sonar pulse detection and recognition based on deep denoising. Proceedings of the 6th International Conference on Electronic Information Technology and Computer Engineering, 2022: 384−389
[14] Gemici M, Akata Z, Welling M. Primal-dual Wasserstein Gan. arXiv preprint: 1805.09575, 2018
[15] Gröchenig K. Foundations of time-frequency analysis. Birkhäuser Boston, MA: Springer, 2001: 37−57
[16] Donoho D L, Stark P B. Uncertainty principles and signal recovery. SIAM J. Appl. Math., 1989; 49(3): 906−931 DOI: 10.1137/0149053
[17] Heil M M, Hammoodi A T, Rasool J M. A comparative analysis of Bohman windowing performance for UFMC with higher QAM modulation. 2023 Al-Sadiq International Conference on Communication and Information Technology, IEEE, Al-Muthana, Iraq, 2023: 13−17
[18] Chung J, Gulcehre C, Cho K H, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling. arXiv preprint: 1412.3555, 2014
[19] Ioffe S, Szegedy C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. International Conference on Machine Learning, PMLR, 2015: 448−456