水下航行器低频线谱噪声源的结构化稀疏匹配场超分辨定位方法研究

It is a necessary procedure to perform the high-resolution localization of low-frequency tone noise sources for enhancing the acoustic stealth property of underwater vehicles. Exploiting the sparse distribution pattern of noise sources in space, sparse matched field method could achieve a high-resolution localization. However, the complicated acoustic propagation structure and non-Gaussian noises in shallow water zones, together with incomplete prior knowledge of noise sources, seriously restrict its excellent performance on low-frequency tone noise source localization..Structured sparsity representation and regularization is utilized in this research to comprehensively exploit the structured prior knowledge of shallow water environment and low-frequency tone noise source. This structured prior knowledge could effectively reduce the array aperture size and the requirement on high-precise replicas when performing the high-resolution localization of noise sources. Firstly, an incoherence basis is developed for the physical intrinsic structure of shallow water, a data-driven mutual incoherence basis design strategy is also proposed for the underlying structure of non-Gaussian noises, and a multiple sparse collaboration regularization is proposed for the distribution patterns of low-frequency tone noise sources in space-time-frequency domain. Integrating all structured prior knowledge above into the super-resolution model of sparse matched field method, a structured sparsity matched field framework is then established, moreover, utilizing the location gain from multiple temporary stationary data block, the super-resolution localization of low-frequency tone noise sources could be achieved robustly. Experiments on the ocean will be performed. The proposed algorithmic framework will be evaluated and improved through applying it to detecting weak tone noise source under strong sources, discriminating two sources with similar frequency structure and alleviating non-Gaussian noises.

低频线谱噪声源的高分辨定位是提高“安静型”水下航行器声隐身性能的基本前提。稀疏匹配场技术通过探索噪声源的空域稀疏分布结构实现高分辨定位，但是浅海环境的复杂声场结构、背景噪声的非高斯结构、以及线谱噪声源的不完备先验知识严重地限制了其在低频线谱噪声源定位中的应用。. 本项目拟采用结构化稀疏先验建模思路，全面深入地探索浅海测试环境和低频线谱噪声源的结构化先验知识来弥补高分辨定位时阵列孔径和拷贝场精度的不足。首先研究浅海声场结构的不耦合稀疏基优化、背景噪声非高斯结构的低互相关稀疏基学习和线谱噪声源空时频结构的联合稀疏正则建模；然后，基于建立的三类结构化先验知识，集成稀疏匹配场技术的高分辨模型，提出结构化稀疏匹配场技术框架，融合移动多快拍数据的定位增益，实现低频线谱噪源的稳健超分辨定位。开展海上实验研究，评估并完善技术框架在弱噪声源检测、相干噪声源鉴别、非高斯噪声抑制等方面的超分辨性能。

低频线谱噪声源的高分辨定位是提高“安静型”水下航行器声隐身性能的基本前提。稀疏匹配场技术通过探索声源的空域稀疏分布结构实现高分辨定位，但是海洋环境的复杂声场结构、背景噪声的非高斯结构、声信道的非线性调制以及线谱噪声源的不完备先验知识严重限制了其在低频线谱噪声源定位中的应用。本项目采用结构化稀疏先验建模策略，并结合深度网络的非线性描述，开展了结构化稀疏匹配场高分辨定位方法研究。.（1）针对稀疏匹配场不耦合稀疏基构造问题，建立了参数化加性结构的不耦合基表征方法，设计了分层概率先验分布和共轭先验分布约束下的自适应稀疏正则函数，开发了多快拍观测数据的快速贝叶斯后验推断算法，实现了不耦合表征基下声源方位的高分辨估计。（2）针对水下声信道和非高斯噪声干扰的非线性调制，提出了非线性低互相关基端到端学习的深度建模策略，设计了基于声传播模型的仿真数据集和水下声辐射数据的线谱定位数据集，提出了非线性深度网络定位学习算法，实现了强噪声干扰下的高精度声源定位。（3）针对凸正则的稀疏描述能力不足问题，开发了参数化稀疏正则函数，提出了耦合阵列流形矩阵的最优凸稀疏定位模型，开发了收敛的Nesterov加速的块梯度下降上升优化算法，实现了高精度、高稳健的声源高分辨定位。（4）利用声传播模型、水下声源数据和国际标准水下数据集，建立了水下目标线谱定位数据库，开展了算法的综合分析与工程验证，形成了一套完备、稳健的结构化稀疏匹配场定位技术，实现了水下目标的准确定位。.基于本项目相关的研究，发表相关学术论文7篇，其中中科院二区以上SCI论文5篇，《西安交通大学学报》论文1篇，会议EI论文1篇。在本项目的基础上获批陕西省自然科学基础研究计划1项。