Page 159 - 《应用声学》2025年第1期

第 44 卷 第 1 期 Vol. 44, No. 1
2025 年 1 月 Journal of Applied Acoustics January, 2025

⋄ 研究论文 ⋄

信号分离在深海定位中的应用

袁 博 1,2 钱 鹏 3 赵 猛 1,2 杨馥锦 1,2 鹿力成 1†

(1 中国科学院声学研究所 中国科学院水声环境特性重点实验室 北京 100190)
(2 中国科学院大学 北京 100049)
(3 中山大学海洋工程与技术学院 珠海 519082)
摘要：声波在深海中远距离传播时海水吸收、扩展导致传播损失大，接收到的声波能量非常小，同时受到航船
风浪等强噪声干扰，声波信号的信噪比非常低。在低信噪比的情况下，信号增强、信号降噪等数据处理方法的
效果均降低，对水下目标定位、检测和识别造成很大影响。该文针对水下目标低信噪比定位问题，应用全卷积
时域网络，基于信号幅度和相位的解耦，提出了一种快速信噪分离方法。该方法利用了端到端时域分离的深度
学习框架，通过线性编码器编码信号，编码之后的信号波形可以通过一组加权函数分离出信号和噪声，最后再
使用线性编码器将分离后的信号反转到时域进行目标定位。通过数据仿真验证了该方法的可行性，并对海上
实验数据进行处理，取得较好结果。
关键词：全卷积时域网络；信噪分离；被动定位系统；深海定位
中图法分类号: P751 文献标识码: A 文章编号: 1000-310X(2025)01-0155-07
DOI: 10.11684/j.issn.1000-310X.2025.01.016

Application of signal separation in deep-sea positioning

3
YUAN Bo 1,2 , QIAN Peng , ZHAO Meng 1,2 , YANG Fujin 1,2 and LU Licheng 1
(1 Key Laboratory of Underwater Acoustic Environment, Institute of Acoustics,
Chinese Academy of Sciences, Beijing 100190, China)
(2 University of Chinese Academy of Sciences, Beijing 100049, China)
(2 School of Ocean Engineeringa and Technology, Sun Yat-sen University, Zhuhai 519082, China)

Abstract: The propagation of sound waves in deep sea experiences significant losses due to absorption and
spreading, resulting in very small received sound energy. Additionally, strong noise interference from ships and
waves leads to a low signal-to-noise ratio (SNR) in the sound wave signals. In conditions of low SNR, the
effectiveness of signal enhancement and noise reduction methods diminishes, which greatly impacts underwater
target localization, detection, and recognition. This paper addresses the issue of low SNR localization for
underwater targets by applying a fully convolutional time-domain network and proposing a rapid signal-noise
separation method based on the decoupling of signal amplitude and phase. The method utilizes an end-to-end
time-domain separation deep learning framework, where a linear encoder encodes the signal. The encoded
signal waveform can be separated into signal and noise using a set of weighted functions. Finally, the separated
signal is inverted back to the time domain using the linear encoder for target localization. The feasibility of
this method is validated through data simulation, and it is applied to process experimental data from the sea,
yielding favorable results.
Keywords: Fully convolutional time-domain network; Signal-to-noise separation; Passive localization system;
Deep-sea positioning
2023-09-05 收稿; 2023-10-23 定稿
作者简介: 袁博 (1997– ), 男, 河南新乡人, 硕士研究生, 研究方向: 电子信息。
† 通信作者 E-mail: lulicheng@mail.ioa.ac.cn