基于进化技术的近场声源定位研究

发布时间：2018-03-30 14:19

  本文选题：波达方向　切入点：差分演化　出处：《中国科学技术大学》2017年博士论文

【摘要】：在存在多个窄带源的环境中,远场和近场信源定位存在诸多挑战。为分析信源定位的性能,常见的挑战有:阵列输出的快拍,未知参数的联合估计,未知参数的匹配,计算复杂度,对噪声的鲁棒性,信源与阵列之间距离的估计,阵列扰动和多维波达方向(DOA)。这里提到的一些挑战可以在系统建模中得到解决,例如阵列的几何结构,阵元的间隔和待测的信源数量,其他的挑战则可以在算法的层面上进行处理。选择近场窄带信源定位是受到室内通信和信源定位,超声波成像,电子监控,射频识别(RFID)通信,水下信源定位和地震勘探等领域不断增长的应用需求的启发。本论文中,主要贡献之一是在近场信源定位中仅使用阵列输出的单快拍数据进行参数估计,这使其可以在实时应用中使用。此外,我们通过利用演化计算的优势,实现无需匹配的距离和DOA联合估计。文中使用均匀线阵(ULA)和L-型阵列等阵列构型,是因其具有成本效益,计算简便和易于使用等性质。本文的主要贡献简要总结如下:1.均匀线阵(ULA)近场窄带信源定位(1DDOAs和距离)的统计分析和建模。当窄带信源存在于阵列的菲涅耳区(近场)时,信源定位问题变得更加复杂。由于信源入射的波前变成球面,同时需要信源的距离和到达角信息才能实现准确定位。这将使联合估计的未知参数数目翻倍,并使计算过程复杂化。现有的近场窄带信源定位模型需要大量的阵列输出快拍。此外,大多数现有模型不能联合估计未知参数,而是需要逐个进行估计。这也使得这些方法在需要实时性的应用中无法使用。为了克服大量快拍的需求,提出了一种称为差分演化的演化技术,采用均方误差作为适应度评估函数。所提算法十分高效,能够在基于均匀线阵的近场窄带信源定位中使用单快拍进行未知参数的联合估计。所提算法的统计分析是通过大量的蒙特卡罗模拟进行的。仿真结果表明,所提出的方法更接近于Cramer-Rao界,且随着信噪比的增加逐步趋近。此外,结果表明,当信源远离阵列时,根据远场中来波距离无限远的理论,所提算法表现会受到影响。且当信源的数量大于所使用的阵元数时,所提方法失效,因为这是已经变成一个欠定问题。2.存在阵元位置扰动的情况下,均匀线阵的增强建模和近场窄带信源定位性能的统计分析。通常,在阵列信号处理中,传感器位置被认为是已知的。在实际情况下,外部因素和制造精度限制会导致被称为阵列扰动的传感器位置误差。使用具有传感器位置扰动的阵列会降低参数估计的性能和精度。一些现有的模型通过阵列预校准精确地获得传感器位置,然后进行信源定位。为了避免阵列预校准的需求,针对存在随机阵元位置误差的均匀线阵的近场窄带信源定位进行建模。在这种情况下,未知参数的数量是信源数量的两倍多。为了简化,将该过程分为三个步骤。首先,在假设不存在传感器位置误差的情况下,联合估计距离和DO As。由于实际中是存在传感器位置误差的,这些参数估计结果并不准确。第二步,使用上面估计出的距离和DOAs作为校准源来估计阵元位置扰动。然后,考虑到第二步中估计出的阵元位置不确定性,对近场窄带信源的DOAs和距离进行更新。由于适应度函数为均方误差的差分演化算法所具有的能力,有效性,易用性和单快拍可收敛到最优解的特性,将其作为全局优化器。大量的蒙特卡罗模拟及其统计分析展示了所提方法的有效性。所提方法的实验结果与其他方法进行以及Cramer-Rao界进行了比较。结果表明,即使在传感器扰动的情况下,所提出的方法也优于其他方法,并且趋近于Cramer-Rao界。3.在不使用任何配对匹配条件下,将均匀线性阵列的统计分析和建模扩展到L型阵列,用于联合估计近场窄带源的距离和二维DOA。对近场窄带信源距离和二维DOAs无需匹配的联合估计中,将均匀线阵拓展到L-型阵列的统计分析和建模历经数十年,二维波达方向估计现已颇受重视。现有的二维阵列几何构型包括圆形阵列,平面阵列,球形阵列等。随着DOA估计的维数增加,估计过程的计算复杂度持续受阵列几何构型的影响,DOAs配对匹配变得至关重要,甚至会导致配对不准确和角度估计性能差。用于二维DOA估计的现有模型需要二维搜索和非线性优化的匹配算法。为克服匹配的困难,并联合估计近场窄带源的范围和二维DOA(即俯仰角和方位角),提出了一种由两个ULA首尾垂直相接而成L-型平面阵列。与矩形阵列和圆形阵列相比,L型阵列在覆盖区域和使用上具有优势,因为它只需要较少的阵元。使用L型阵列的另一个优点是能通过将阵列解耦为两个ULA来独立估计DOA从而得到二维DOA。由于适应度函数为均方误差的差分演化算法不需要谱峰搜索,额外的角度匹配过程和具有实时优势的单快拍收敛特性,将其由于优化估计过程。理论分析和实验结果表明,即使在未知参数是信源数三倍的情况下,所提算法在俯仰角,方位角和距离的估计中仍优于其他算法,并趋近于Cramer-Rao界。
[Abstract]:In the presence of multiple narrowband sources in the far field and near field source localization has many challenges. To analyze the performance of source location, the common challenges: the array output snapshots, joint estimation of unknown parameters, unknown parameters, computational complexity, robustness to noise, estimation of the distance between the source with the array, array perturbations and multidimensional direction of arrival (DOA). Some of the challenges mentioned here can be solved in the system modeling, the geometric structure of array for example, the number of array element spacing and measured the source, other challenges can be processed in the algorithm level. Selection of near-field narrowband source localization by indoor communication and source localization, ultrasonic imaging, electronic monitoring, radio frequency identification (RFID) communication, inspired applications field of underwater source localization and seismic exploration growing. In this thesis, the main contributions is in the past A single snapshot data array output using only the field source localization in parameter estimation, which can be used in real-time applications. In addition, we use evolutionary computing, without matching distance and DOA joint estimation. Using the uniform linear array (ULA) and L- type array array configuration. Because of its cost-effective, simple calculation and easy to use. The main contributions of this paper are briefly summarized as follows: 1. uniform linear array (ULA) near-field narrowband source localization (1DDOAs and distance) of the statistical analysis and modeling. When the Fresnel zone exists in the array of narrowband sources (near field), the source localization problem become more complex. Because the source of incident wavefront into a spherical surface, and the source of the distance and angle of arrival information to achieve accurate positioning. This will double the number of joint estimation of unknown parameters, and the calculation process of the existing complex. Near-field narrowband source localization model needs a large number of snapshots of array output. In addition, the joint estimation of unknown parameters in most existing models can not be, but need one by one estimate. It also makes these methods cannot be used in applications requiring real-time demand. In order to overcome a lot of snapshots, this paper proposes a model called differential evolution evolution technology, using mean square error as the fitness evaluation function. The proposed algorithm is very efficient in estimation of near-field narrowband source localization using uniform linear array in a single snapshot of the unknown parameters. The proposed algorithm based on the statistical analysis is simulated by a Monte Carlo. The simulation results show that the proposed method the more close to Cramer-Rao, and with the increase of the SNR gradually approaching. In addition, results show that when the source is far away from the array, according to the far field wave to infinite distance theory, The proposed algorithm performance will be affected. And when the number of array source quantity is greater than the use of the proposed method, the failure, because it has become a underdetermined problem of.2. array element position disturbance, statistical analysis of enhanced modeling of uniform linear array and near-field narrowband source localization performance. Usually, in array signal processing, sensor position is assumed to be known. In the actual situation, the external factors and the accuracy of manufacturing restrictions may lead to be called sensor position error array perturbations. With the sensor array position disturbance will reduce the parameter estimation accuracy and performance. Some of the existing model to obtain accurately the position sensor through the array of pre calibration, then the source location. In order to avoid pre array calibration needs, modeling the near-field narrowband source localization random uniform linear array element position error matrix. 鍦ㄨ繖绉嶆儏鍐典笅,鏈�鐭ュ弬鏁扮殑鏁伴噺鏄�淇℃簮鏁伴噺鐨勪袱鍊嶅��.涓轰簡绠，

本文编号：1686287