基于张量分解的声矢量传感器阵列信号处理方法研究

发布时间：2018-05-07 19:24

本文选题：张量 + 声矢量传感器　；参考：《吉林大学》2017年硕士论文

【摘要】：声矢量传感器是在传统声压传感器基础上发展而成的,它由声压传感器以及质点振速传感器复合而成,是一种新型的水声测量设备。声矢量传感器的结构组成决定了声矢量传感器既可以测量声压,同时还可以测量声场中质点振速的三维正交分量。由于接收的信息具有多维性,所以声矢量传感器在声呐、雷达、导航定位等众多民用与军事领域均具有广泛应用。因此,有关声矢量传感器阵列信号处理方法也受到各界学者的广泛关注。在对声矢量传感器阵列接收信号进行空间信号到达方向(Direction of Arrival,DOA)估计时,传统处理方式是将声矢量传感器接收到的声压及振速信息排列成为长矢量矩阵形式,该形式并没有充分利用声矢量传感器蕴含的多维结构信息,因此,本文引进了高维数组张量分析模型,建立声矢量传感器阵列信号张量模型,并与传统的DOA估计算法结合,提出了基于张量分解的声矢量传感器阵列信号DOA估计方法。本文的研究工作如下:针对声矢量传感器阵列包含的多维结构信息,提出了一种声矢量传感器阵列张量模型,并对该模型进行推导。针对空间信号到达方向估计问题,提出了声矢量传感器阵列直接张量分解的多重信号分类(Multiple Signal Classification method,MUSIC)算法以及协方差张量分解MUSIC算法。首先,对测量张量或者测量张量的协方差张量进行高阶奇异值分解,估计张量信号子空间和张量噪声子空间,并证明这两种张量高阶奇异值分解估计的张量信号子空间具有等效性,由于信号处理时高阶奇异值分解比传统矩阵奇异值分解对噪声抑制能力更强,所以得到的信号子空间更加准确。结合传统MUSIC算法给出基于张量子空间的谱估计公式,对其进行谱峰搜索,估计声源信号的入射角。实验证明基于张量分解的MUSIC算法比传统长矢量矩阵MUSIC算法的参数估计性能更好,说明张量能够更好地利用多维结构数据信息。针对MUSIC算法需要进行谱峰搜索,具有计算量大的缺点,选取具有相同结构的阵元偶,采用旋转不变技术(Estimating Signal Parameters via Rotational Invariance Techniques,ESPRIT),推导出该阵元偶的旋转不变因子,构造张量r维旋转不变方程,提出了基于声矢量传感器阵列的张量分解ESPRIT算法,仿真实验证明该算法比张量MUSIC算法具有更快的计算速度,比传统声矢量ESPRIT算法具有更高的参数估计性能。针对张量高阶奇异值分解估计子空间不是最优的问题,提出了一种基于张量最佳秩逼近的声矢量传感器阵列MUSIC算法。首先,求解测量张量的最佳秩逼近张量,得到张量信号子空间,并结合传统MUSIC算法对声源的入射角进行估计。实验证明,基于张量最佳秩逼近声矢量传感器阵列参数估计算法比基于高阶奇异值分解算法的估计精度更高,参数估计性能更好。将矩阵实值ESPRIT算法扩展到张量情况,提出基于张量最佳秩逼近的声矢量传感器阵列酉ESPRIT(Unitary-ESPRIT)算法。首先,对测量张量进行双向平滑处理,然后对得到的复张量进行变换得到实数张量,利用张量最佳秩逼近算法,得到实数张量的信号子空间,构造选择矩阵和旋转因子,对入射角进行估计。实验证明,与传统Unitary-ESPRIT算法相比,该算法有更高的参数估计性能。
[Abstract]:Acoustic vector sensor is developed on the basis of the traditional sound pressure sensor. It is composed of sound pressure sensor and particle vibration speed sensor. It is a new type of underwater acoustic measurement equipment. The structure of acoustic vector sensor determines that acoustic vector sensor can not only measure sound pressure, but also measure the velocity of the vibration of the particle in the sound field by three Since the received information is multidimensional, the acoustic vector sensor is widely used in many civil and military fields, such as sonar, radar, navigation and positioning. Therefore, the signal processing method of acoustic vector sensor array is also widely used by scholars from all walks of life. When the Direction of Arrival (DOA) is estimated, the traditional way is to arrange the sound pressure and speed information received by the acoustic vector sensor into a long vector matrix form. This form does not make full use of the multi-dimensional structure information contained in the acoustic vector sensor. Therefore, the high dimension array tensor analysis model is introduced in this paper. A tensor model of acoustic vector sensor array signal is established and combined with the traditional DOA estimation algorithm, a DOA estimation method for acoustic vector sensor array signal based on tensor decomposition is proposed. The research work of this paper is as follows: a tensor model of acoustic vector sensor array is proposed in view of the multi-dimensional structure information contained in the acoustic vector sensor array. In order to estimate the direction of arrival of space signals, the multiple signal classification (Multiple Signal Classification method, MUSIC) algorithm and covariance tensor decomposition MUSIC algorithm for the direct tensor decomposition of the acoustic vector sensor array are proposed. First, the covariance tensor of measurement tensor or measurement tensor is carried out. The high order singular value decomposition (SVD) is used to estimate the tensor signal subspace and tensor noise subspace. It is proved that the tensor signal subspace of the two tensor singular value decomposition estimation is equivalent. Because the high order singular value decomposition is more powerful than the traditional singular value decomposition, the signal subspace is better. Combined with the traditional MUSIC algorithm, the spectral estimation formula based on the quantum space is given, the spectrum peak is searched and the incidence angle of the source signal is estimated. The experiment shows that the MUSIC algorithm based on tensor decomposition is better than the traditional long vector matrix MUSIC algorithm. It shows that the tensor can make better use of the multi-dimensional structure data information. In view of the need for the MUSIC algorithm to search for peak spectrum, with the disadvantage of large computational complexity, the array element with the same structure is selected and the rotation invariant (Estimating Signal Parameters via Rotational Invariance Techniques, ESPRIT) is used to derive the rotation invariant factor of the array element pair, and a tensor R dimension rotation invariant equation is constructed, and a base is proposed. In the tensor decomposition ESPRIT algorithm of acoustic vector sensor array, the simulation experiment shows that the algorithm has faster computation speed than the tensor MUSIC algorithm and has a higher parameter estimation performance than the traditional sound vector ESPRIT algorithm. MUSIC algorithm for the rank approximation of acoustic vector sensor array. First, the best rank approximation tensor of the measurement tensor is solved. The tensor signal subspace is obtained, and the incidence angle of the sound source is estimated with the traditional MUSIC algorithm. The experiment shows that the estimation algorithm based on the tensor best rank approximation for the array parameter estimation of the acoustic vector sensor is based on the higher order singular value decomposition The estimation accuracy of the algorithm is higher and the performance of the parameter estimation is better. The ESPRIT algorithm is extended to the tensor condition and the unitary ESPRIT (Unitary-ESPRIT) algorithm based on the best rank approximation of tensor is proposed. First, the measurement tensor is processed by two-way smoothing, and the real tensor is obtained by changing the obtained complex tensor. By using the tensor best rank approximation algorithm, the signal subspace of real number tensor is obtained, the selection matrix and rotation factor are constructed and the incidence angle is estimated. The experiment shows that the algorithm has higher parameter estimation performance compared with the traditional Unitary-ESPRIT algorithm.

【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN911.7;TP212

【相似文献】