【作者】 陈峰；

【导师】 杨德森；

【作者基本信息】 哈尔滨工程大学 ， 水声工程， 2022， 博士

【摘要】 随着水下环境的复杂化、随着水下潜艇的安静化,迫使现有的水下目标探测手段朝着低频段、抗干扰等方面发展,但是传统的标量水听器仅能获取声场中的标量信息,对于恶劣水下环境中低频目标的探测能力略显不足,而矢量水听器作为一种能够同步共点拾取声场中声压、振速具有低频指向性的复合型水听器,它的出现为水下目标探测提供了更加有力的工具。然而由于水下矢量探测手段千差万别,导致不同场景下的探测能力良莠不齐。例如,虽然现有子空间类高分辨方位估计(High-resolution Direction of Arrival,High-resolution DOA)算法经过数十年发展已取得众多优异成果,但是此类算法在工程实施中,尤其是低信噪比、小快拍时子空间类算法存在估计性能急剧下降甚至失效的弊端。这是由于子空间类高分辨算法需将信源个数作为先验知识进行子空间划分,而随着现有水下作战安静化发展对于目标数的获取无疑增加了难度,当信源数获取错误时,将会导致子空间划分错误,进而致使子空间类算法稳健性急剧恶化。为解决此弊端,本文基于矢量水听器提出一类未知信源数的高分辨算法以及低复杂度高分辨DOA估计方法。本文从特定矩阵的特征值排布理论出发,通过构造包含扫描源的特定数据模型,并分析不同参数时扫描源对特征值排布规律的影响,获得不同参数下的特定矩阵准确排布规律,进而实现了信源数未知的高分辨DOA估计。然而,此类信源数未知的高分辨算法需要在每个扫描角度进行特征值分解,快速增长的算法计算量直接影响探测系统的实时性,为此本文通过对协方差矩阵进行分析,基于实虚分离原理提出了一类的低复杂度高分辨DOA估计算法,有效缩减谱搜索范围,减少算法计算量。本文具体研究分为如下几个部分:(1)基于噪声功率不变(Noise Power Invariant,NPI)的DOA估计在研究信号功率与信号特征值的映射关系的基础上,通过参数适当选择,构建了一种未知信源数的高分辨DOA估计算法。算法利用扫描源与参考矩阵构建扫描矩阵并取其最小特征值与参考矩阵最小特征值构建空间谱函数,达到无需信源数的目的。通过后续实验表明:算法受到调节参数以及幂指数的影响,当选取适当调节参数以及幂指数时噪声功率不变方法估计性能优于其他高分辩类算法。(2)基于幂指数的Capon-like(Exponent Capon-like,E-Capon-like)算法然而NPI算法调节参数可选取范围过小,为解决此弊端本文提出了一种E-Caponlike算法。首先利用采样协方差矩阵的幂次方,弱化矩阵中信号分量对于特征值的影响,使得矩阵整体趋近于噪声子空间,从而获得高分辨性质;其次,将此矩阵与扫描源结合,分析不同扫描角度下的特征值排布规律,构建了一种简洁的空间伪谱,实现了高分辨效果。后续仿真实验表明:所提E-Capon-like算法调节参数选取区间有效扩大,此外此算法不仅可以在信源数未知的情况下获得目标方位信息,而且相对于其他高分辨类算法,所提算法拥有更好的估计性能,适用于更加恶劣的水下环境。(3)基于广义Capon(Generalized Capon,G-Capon)的DOA估计算法上述NPI算法以及E-Capon-like算法均是针对白噪声提出,忽略了对色噪声的影响,为此本文提出了一种信源数未知的高分辨G-Capon算法。本文基于白噪声以及色噪声约束下的广义Capon展开研究,分析其特定矩阵下的特征值排布规律,从而实现目标的高分辨估计。为了使得算法性能达到最优化,本文分析了伪峰形成原因、广义检测阈值、调节参数之间的影响因素,理论研究表明:当调节参数选取过大,算法广义检测阈值将会降低,但同时也会导致伪峰的形成,反之当调节参数选取过小,虽不会导致伪峰的形成,但此时算法广义检测阈值将会提高。为此,本文给出了一种调节参数的选取方法,此方法既抑制了伪峰的形成,也降低了算法的最小广义检测阈值,通过后续与其他高分辨类算法的对比,验证了选参方法的正确性以及所提算法的优越性。(4)低复杂度的DOA估计算法虽然上述算法实现了在未知信源数时进行目标DOA估计,但是依旧需要进行整个空间谱搜索,无疑带来了巨大的搜索计算量。为此,本文利用对称虚源的思想,提出了一类基于半谱搜索的DOA估计方法。通过对采样协方差矩阵的内在分析,表明当导向矢量满足奇函数性质时,采样协方差矩阵的实部同时包含目标角度以及目标对称角度。利用此思路,本文提出了半实值化Capon-like(Semi-Real-Value Capon-like,SR-Capon-like)以及(Semi-Real-Value NPI,SR-NPI)算法,此类算法通过半谱搜索并加以判定即可获得完整的目标角度,大幅降低了算法计算量。(5)外场实验验证利用矢量水听器所采集外场数据对所提算法进行验证,分别从阵元数、信号频率进行分析,开展了目标DOA估计、特征值排布以及目标跟踪实验,以表明所提算法的理论正确性以及外场实验的适用性。外场实验结果表明,高分辨类算法不仅可以在目标数未知的情况下准确获得水下目标方位,且此类算法的特征值排布情况基本满足理论推导,说明此类算法能够适用于外场工程;低复杂度DOA估计算法,不仅可以在目标位置处形成谱峰,且可以在其对称位置处形成谱峰,实现了半谱搜索的目的,有效减少了算法的谱搜索计算量,为数据实时处理提供了可能性。更多还原

【Abstract】 With the complexity of the underwater environment and the quietness of underwater submarines,the existing underwater target detection methods are forced to develop in low frequency bands and anti-interference.The vector sensor is a compound sensor that can simultaneously pick up the sound pressure and vibration velocity in the sound field simultaneously and has low-frequency directivity.Its appearance provides a more powerful tool for underwater target detection.Existing subspace highr-resolution algorithms have achieved many excellent results after decades of development,but such algorithms have disadvantages of poor robustness in engineering implementation,especially for low signal-to-noise ratio(SNR)and small snapshots.Subspace algorithms need to use the number of information sources as a priori knowledge to divide the subspace.With the quiet development of existing underwater operations,it will undoubtedly increase the difficulty of obtaining the number of targets.When the number of information sources is obtained incorrectly,it will leads to subspace division errors,which in turn leads to a sharp deterioration in the robustness of subspace algorithms.In order to solve this drawback,this paper proposes a class of robust highr-resolution algorithms based on vector sensors that do not require the number of sources.Starting from the theory of eigenvalue arrangement of a specific matrix,this thesis constructs a specific data model that includes scanning sources,and analyzes the influence of scanning sources on the eigenvalue arrangement rules when different parameters are used to obtain the exact arrangement rules of specific matrices under different parameters.Furthermore,the highrresolution DOA estimation with unknown source number is realized.However,such highrresolution algorithms with unknown number of sources need to perform eigenvalue decomposition at each scanning angle,and the fast-growing algorithm calculations directly affect the real-time performance of the detection system.For this reason,the paper analyzes the covariance matrix and proposes a kind of low-complexity DOA estimation algorithm based on the principle of real-virtual separation,which effectively reduces the spectrum search range and reduces the calculation amount of the algorithm.The specific research is divided into the following parts:(1)DOA estimation based on noise power invariant(NPI)Based on the research of signal power and signal eigenvalues,an efficient highr-resolution DOA estimation algorithm without the number of sources is constructed through proper selection of parameters.This algorithm uses the scan source and the reference matrix to construct the scan matrix,and takes the smallest eigenvalue and the smallest eigenvalue of the reference matrix to construct the spatial pseudo-spectrum,so as to achieve the goal of not requiring the number of information sources.Follow-up experiments show that the algorithm is affected by the adjustment parameters and the power exponent.When the appropriate adjustment parameters and the power exponent are selected,the noise power estimation performance is better than other high-resolution algorithms.(2)Capon-like(Exponent Capon-like,E-Capon-like)algorithm based on power exponentHowever,the selection range of the adjustment parameters of the NPI algorithm is too small.In order to solve this problem,the paper proposes an E-Capon-like algorithm.First,the power of the sampling covariance matrix is used to weaken the influence of the signal components in the matrix on the eigenvalues,making the matrix as a whole approach the noise subspace,thereby obtaining high-resolution properties;secondly,combining this matrix with the scanning source to analyze the differences.The arrangement of eigenvalues under the scanning angle constructs a simple spatial pseudo-spectrum and achieves a high-resolution effect.Subsequent simulation experiments show that this algorithm can not only obtain target orientation information when the number of sources is unknown,avoiding the acquisition of the target number,but also compared with other high-resolution algorithms,the proposed algorithm has better estimation performance and is suitable for more severe underwater environment.(3)DOA estimation algorithm based on generalized Capon(G-Capon)The above-mentioned NPI algorithm and E-Capon-like algorithm are both proposed for white noise,ignoring the impact on color noise.For this reason,the paper proposes a highresolution G-Capon algorithm with unknown source number.Firstly,the research is carried out based on the generalized Capon model under white noise and color noise,and the eigenvalue arrangement rule under the specific matrix is analyzed,so as to realize the robust highresolution estimation of the target.In order to optimize the performance of the algorithm,the article analyzes the reasons for the formation of false peaks,the detection threshold,and the influencing factors among the adjustment parameters.The theory shows that when the adjustment parameters are selected too large,the algorithm detection threshold will be reduced,but at the same time it will also lead to false peaks.Peak formation;On the contrary,when the adjustment parameter is selected too small,although it will not lead to the formation of false peaks,the detection threshold of the algorithm will increase at this time.To this end,the article presents a method for selecting adjustment parameters.This method not only suppresses the formation of false peaks,but also reduces the minimum detection threshold of the algorithm.Through subsequent comparison with other high-resolution algorithms,the method of selecting parameters is explained.The correctness of and the superiority of the proposed algorithm.(4)Low-complexity DOA estimation algorithmAlthough the above algorithm realizes the target DOA estimation without the number of sources,it still needs to search the entire spatial spectrum,which undoubtedly brings heavy search calculations.For this reason,the article uses the idea of symmetric virtual source to propose a kind of half-spectrum search low-complexity DOA estimation method.Through the internal analysis of the sampling covariance matrix,it is shown that when the steering vector satisfies the odd function property,the real part of the sampling covariance matrix contains both the target angle and the target symmetric angle.Using this idea,the article proposes SR-Caponlike and SR-NPI algorithms,which can obtain a complete target angle through half-spectrum search and judgment,which greatly reduces the computational complexity of the algorithm.(5)Field experiment verificationThe sea trial data collected by the acoustic vector sensors is used to verify the proposed algorithm.The verification experiments are carried out from the number of array elements and signal frequency.The target DOA estimation,eigenvalue arrangement and target tracking experiments are carried out to show the proposed algorithm’s performance.The correctness of the theory and the applicability of field experiments.Experimental results in the field show that robust high-resolution algorithms can not only accurately obtain the azimuth of underwater targets when the number of targets is unknown,but also that the eigenvalue arrangement of such algorithms basically meets the theoretical derivation,indicating that this type of algorithm can be applied to the field Engineering;the low-complexity DOA estimation algorithm can not only form a spectrum peak at the target position,but also form a spectrum peak at its symmetrical position.This achieves the purpose of half-spectrum search,effectively reducing the calculation of the algorithm’s spectrum search,and it is the data Real-time processing offers possibilities.更多还原