【作者】 董姝敏；

【导师】 梁国龙；

【作者基本信息】 哈尔滨工程大学 ， 信号与信息处理， 2012， 博士

【摘要】 水声被动信号处理中，往往由于实际水声环境复杂多变，干扰和噪声又难以确定，由水听器（或水听器阵）直接观测到的信号，大多是多源目标信号经多途传输后在接收点叠加并与干扰背景噪声（海洋环境噪声）混合的一种复杂信号，严重影响着后期的目标定位、检测、识别等方面的性能和精度。盲信号处理技术可以无需源信号先验知识而仅仅利用接收信号和某些统计假设条件获取期望信息，比较适合应用于水声被动信号处理，盲信号分离技术作为其中一个重要分支内容，它的目标是获得多个源信号的最佳估计，在需要准确获得水下目标的真实源信号时具有重要应用需求。论文主要研究内容包括三方面：欠定情况下源数目估计方法；单波束信号内多源信号的盲分离方法及含噪波束信号的净化方法；适合单矢量传感器的盲信号分离方法。实现盲信号有效分离的前提条件是源数目的准确估计，所以论文首先研究了源数目的估计问题。众多的源数目估计方法中，盖尔圆（GDE）准则法既适用于高斯白噪声背景，又适用于色噪声背景，具有重要的应用价值。但传统GDE准则法存在一定的弊端：低信噪比或者小快拍数条件下的检测性能不尽如人意，而且需要人为确定调整因子。论文在传统GDE准则基础上提出了一种改进盖尔圆准则（MGDE）方法，该方法主要利用盖尔圆圆心信息压缩盖尔圆半径，使噪声盖尔圆进一步远离信号盖尔圆，从而更有利于低信噪比或者小快拍数条件下源数目的准确估计；同时，针对GDE准则设计中存在人为确定调整因子的问题，论文设计了一种由盖尔圆信息自动确定调整因子的方法，GDE准则使用该调整因子，不但检测性能得到改善，而且避免了人为因素的影响。欠定条件下源数目估计问题是盲信号分离问题中的难点之一，论文选用相空间重构法拓展观测信号个数，将欠定条件转换为超定条件。原相空间重构法只能增加与阵元数目相同的重构信号，针对这一局限，给出了可重构多个信号的改进相空间重构方法，使用该方法，并结合适当的源数目估计方法可以从单阵元信号中正确估计大于两个源的数目。许多水声信号处理方法和理论都建立在高斯噪声模型基础上，多数情况下这种假设是合理有效的，但研究表明：水声环境噪声、舰船辐射噪声等常常具有α稳定分布（α-Stable Distribution）的特性：其概率密度函数具有比较厚重的拖尾，表现在时间波形上就是具有显著的脉冲特性，这种情况下基于高斯噪声模型的处理算法性能退化，甚至不能得到正确结果。针对这一情况，论文初步探讨了服从α稳定分布的随机向量的源数目估计和DOA估计问题。一方面提出使用协共变矩阵代替协方差矩阵，从而实现超定源数目估计，并仿真验证了源数目估计性能；另一方面：对二维单矢量水听器的振速向量采用经验函数法和投影法两种谱测度估计方法进行源数目估计和DOA估计，仿真表明：前者只能实现单个声源的DOA估计，后者可以实现多源的数目估计和DOA估计。论文使用盲信号分离的手段进一步对波束信号进行分析处理，目的是解决两方面问题：一是分离波束内来自同一方位、或者相互靠得很近的多个源信号；二是进一步净化含有噪声的波束信号，因为仅仅依靠波束形成无法实现这样的目的。论文首先从增加波束域信号维数的角度，主要运用相空间重构法对波束域信号升维；然后在最大化负熵的基础上，推导了最小化边缘熵准则，给出了基于概率密度估计的独立边缘熵计算方法，以该最小化边缘熵为分离准则，采用正定盲分离技术进一步分离或净化单波束信号。研究表明：将盲信号分离技术用作波束形成的后置信号处理方法是合理有效的，它既可以辅助波束形成法分离波束内来自同一方位、或相互靠得很近的多个源信号，也可以对含有噪声的单源波束信号进行净化处理。另外，从盲信号分离的角度考虑，将其与波束形成法结合，可以减小盲分离算法的阶数，降低盲分离算法的复杂度。使用盲信号分离技术从单矢量水听器接收信号中准确获知多个目标的参数信息，是近几年兴起的研究课题。为此，论文根据矢量水听器观测信号的特点，给出了两种适合单矢量水听器的盲信号分离方法——基于混合矩阵的盲分离方法和基于二阶统计量的解析方程法，这两种方法都可以实现非相干源信号分离和DOA估计，仿真分析了算法的性能和各种影响因素。更多还原

【Abstract】 In passive underwater acoustics signal processing, observed directly signals from thehydrophone (or hydrophone array) are a complex mixture signal which is a superposition ofmulti-sources via multipath propagation and interference noise or ocean backgroundnoise,because the actual acoustic environment is complex and changeable, interference andnoise are also difficult to be determined. Those affect seriously the performance and accuracyof target location, the detection and identification. Blind signal processing technology issuitable for application in passive underwater acoustic signal processing, because it can onlymake use of received signals and some statistical assumptions to gain expected information,without prior knowledge of source signals. The goal of blind separation as one of theimportant branches is to get the best estimation of multiple source signals, so blind separationtechnology has great application value in the many fields with the need to get the accurateunderwater source signals. The paper includes three main research contents.These areestimation method about the number of source under under-determined condition, blindmulti-source separation method from single beam signal or purification method in the light ofnoisy beam signal, and blind signal separation method suitable for single vector sensor.A prerequisite realizing blind signal separation is the accurate estimation of sourcenumbers, so the paper studies estimation problem of source number. In many estimationmethods, Gerschgorin Criterion method has important application value, because it can applyto both the Gauss white noise background and colored noise background. But the traditionalGDE criterion method has some disadvantages. Detection performance is not just as onewishes under low SNR condition or a small number of snapshots, and adjustment factor needto determine by people. According to this, a modified method of Gerschgorin radii wasintroduced based on the original Gerschgorin Criterion method. That modified method(MGDE) lessened independently Gerschgorin radii using center information of Gerschgorincircle in order to compression speed of noise Gerschgorin radii faster than that of signalGerschgorin radii, so the noise Gerschgorin disks could be kept as remote from the signalGerschgorin disks as possible. Therefore, the source number could be easily determined underlow SNR condition or a small number of snapshots. Meanwhile, in view of the existingproblems which adjustment factor need to determined by people, this paper proposes amethod, in which can determine automatically the adjustment factor by Gerschgorin disksinformation. And GDE criterion not only improved the detection performance but also avoiding the influence of artificial factors through use of proposed adjustment factor.Source number estimation problem under under-determined condition is one of thedifficulties in blind signal separation problem.According to this, we convert under-determinedcondition into overdetermined conditions by phase space reconstruction method to expand theobserved signal dimensions.For limiting of phase space reconstruction method can onlyreconstruct a signal, we will present a modified phase space reconstruction method which canreconstruct multiple times signals by modifying the original method.We can estimationcorrectly the number of sources more than2from single array signal combining with theappropriate source number estimation method.A lot of methods and theory of underwater acoustic signal processing are establishedbased on the Gauss noise model.In most cases, this assumption is reasonable and effective.But research show: acoustic environment noise and warship radiated noise often haveα-stable distribution characteristics. Probability density function has a relatively thick tail,that is In time waveform they have significant pulse characteristics.So algorithmsperformance degrade based on the Gauss noise models, even it can’t get the right result. Inview of this situation, the paper initially explores source number estimation and DOAestimation of random vector obeying to α-stable distribution. On the one hand sourcenumber is estimated using diagonal covariance matrix instead of the covariance matrix, andthe simulation results verify the source number estimation performance.On the other hand,wemake use of two spectral measure(empirical function method and the projection method) toestimate the number of sources and DOA. The former can only achieve single sound sourceDOA estimation. The latter can realize DOA estimation of multi-sources.This paper make use of blind signal separation means to analyze and process furtherbeam signal. The purposes are to solve the two aspects problems because rely solely onbeamforming cannot achieve these objectives. One is that more source signals can be separatefrom single beam signal, when they come from the same direction, or they are closer eachother. The other is the beam signal can be purified further containing noise. From the aspectof increasing signal dimensions in this paper, phase space reconstruction method is used toincrease signal dimensions. Then the determined blind separation technology further can beadopted to separate or purified single beam signal.Research show: it is reasonable andeffective that blind signal separation technology be used as post signal processing method ofbeamforming. It can not only separate more source signals from single beam signal when theycome from the same direction or they are closer each other, but also clear beam signalpolluted by noises. In addition, from the point of view of blind signal separation, it can reduce orders of blind separation algorithm, and reduce the complexity of the algorithms.It is a new hot research topic in recent years that parameter informations of multipletargets can be obtained accurately from signal of a single vector hydrophone using blindseparation technology. Therefore, two blind separation methods are propose based on thevector hydrophone signal characteristics. They are blind separation method based on hybridmatrix and blind separation method based on two order statistics analytic equation method.They are suitable for noncoherent source.And simulations analyzed performance and affectionfactors of separation algorithm.更多还原