基于麦克风平面阵列的运动噪声源定位及算法研究

发布时间：2019-01-20 08:52

【摘要】：噪声源识别是进行噪声抑制和降噪的前提条件，由于高速列车工作状态和速度的不同而导致高铁运动噪声源频率、强弱等特性发生变化，给外部噪声源的空间指向性建模、测试分析带来很大的困难。经典“延时-累加”波束形成方法在识别运动声源中存在不足，无法克服运动噪声源存在的多普勒效应。本课题对运动噪声源定位的理论算法、阵列设计、建模仿真进行了研究，具有重要的实用参考价值和广阔应用前景。本文详细分析和总结国内外研究经验，以经典波束形成和运动声源辐射理论为基础，以运动噪声源识别技术为主，采用理论研究与仿真分析相结合的方法，对运动噪声源识别和声场重建技术进行研究。论文的主要研究内容包括以下几个部分：（1）运动声源辐射问题。在经典“延时-累加”波束形成对静止声源识别的基础上提出该方法在运动声源识别中存在的不足，阐述了运动声源辐射的相关问题，给出了运动声场辐射数学描述。（2）运动声源多普勒效应处理。分析运动列车阵列信号基本理论，建立运动声源点与测量点几何模型。针对常规运动声源识别方法对幅度和频率进行简单校正，波数域仍然存在多普勒效应和简化模型存在定位误差的情况，本论文提出基于非简化模型的声源测量方法，通过微积分方法建立消除多普勒效应的运动声源数学模型，该模型为一种无损精度数学模型，并由此推导出基于非简化模型的运动声源波束形成声场重建公式。为最大程度提高系统的抗干扰能力，增强系统声源指向的准确性，采用去自相关项方法来降低降噪声污染。（3）麦克风阵列性能研究。详细分析噪声信号频率、阵列间距、阵元数目、阵列尺寸等阵列几何参数对阵列分辨率等性能的重要影响。以十字阵、矩形阵、六角阵为对象，在阵列方向性、圆对称性、角度分辨率等性能方面进行定性和定量对比仿真与分析。（4）波束形成算法性能研究。在方位谱估计的理论基础上，给出了常规波束形成器、MVDR（Minimum Variance Distortionless Response）波束形成器、MUSIC(Multiple Signal Classification)波束形成器在两声源方位间隔Δθ=5°和Δθ=10°的扫描方位谱。通过信噪比分别为-5dB和5dB时分辨概率和方位间隔的关系以及输入信噪比SNR的值在-30~30dB间变化时SINR与输入SNR的关系对三种波束形成器稳定性、阵列增益等性能进行了对比研究并给出相关结论，为后续声场重建仿真选择算法提供依据。（5）基于波束形成算法噪声源识别声场仿真。基于麦克风阵列在噪声源识别实际应用中阵列布局、阵列选型等方面的相关结论和波束形成算法性能对比研究，，本文以MATLAB为工具，结合常规波束形成算法、MVDR波束形成算法分别对单频声源、多频声源、偶极子、分布源等声源进行基于非简化模型和简化模型的运动噪声源识别声场重建对比仿真。
[Abstract]:Noise source identification is a prerequisite for noise suppression and noise reduction. Because of the different operating state and speed of high-speed train, the frequency, intensity and other characteristics of high-speed train moving noise source are changed, and the spatial directivity of external noise source is modeled. Test analysis presents great difficulties. The classical "delay-cumulation" beamforming method can not overcome the Doppler effect of moving noise sources because of its shortcomings in identifying moving sound sources. In this paper, the theoretical algorithm, array design, modeling and simulation of moving noise source localization are studied, which has important practical reference value and broad application prospect. This paper analyzes and summarizes the domestic and foreign research experiences in detail, based on the classical beamforming and the theory of moving sound source radiation, and mainly based on the recognition technology of moving noise source, and adopts the method of combining theoretical research with simulation analysis. The identification of moving noise sources and the technique of sound field reconstruction are studied. The main contents of this paper are as follows: (1) the radiation of moving sound source. On the basis of the classical "time-delay cumulative" beamforming method for the recognition of static sound sources, the shortcomings of this method in the recognition of moving sound sources are proposed. The related problems of moving sound source radiation are expounded, and the mathematical description of moving sound field radiation is given. (2) Doppler effect processing of moving sound source. The basic theory of moving train array signal is analyzed and the geometric model of moving sound source point and measuring point is established. In view of the simple correction of amplitude and frequency by conventional moving sound source recognition methods, the Doppler effect in wavenumber domain and the localization error in simplified model are still existed. In this paper, a method of sound source measurement based on unsimplified model is proposed in this paper. The mathematical model of moving sound source for eliminating Doppler effect is established by means of calculus method. The model is a mathematical model with lossless accuracy and the formula of acoustic field reconstruction of moving sound source beamforming based on non-simplified model is derived. In order to improve the anti-interference ability of the system and enhance the accuracy of the sound source pointing, the method of de-autocorrelation is adopted to reduce the noise pollution. (3) performance of microphone array. The effects of array geometry parameters, such as noise signal frequency, array spacing, array number and array size, on array resolution are analyzed in detail. Taking cross array, rectangular array and hexagonal array as objects, qualitative and quantitative simulation and analysis of array directivity, circular symmetry and angle resolution are carried out. (4) performance of beamforming algorithm. Based on the theory of azimuth spectrum estimation, the scanning azimuth spectrum of conventional beamformer, MVDR (Minimum Variance Distortionless Response) beamformer, MUSIC (Multiple Signal Classification) beamformer between two sound source azimuth 螖胃 = 5 掳and 螖胃 = 10 掳is given. The stability of the three beamforming devices is obtained by the relationship between the resolution probability and the azimuth interval when the SNR is-5dB and 5dB, and the relation between SINR and the input SNR when the input SNR changes between-30~30dB and the signal to noise ratio (SNR). The performance of array gain is compared and the relevant conclusions are given, which provide the basis for the selection algorithm of subsequent acoustic field reconstruction simulation. (5) Acoustic field simulation based on beamforming algorithm. Based on the relative conclusions of array layout and array selection in the practical application of microphone array in noise source recognition and the comparative study of beamforming algorithm performance, this paper uses MATLAB as a tool and combines with conventional beamforming algorithm. MVDR beamforming algorithm is used to reconstruct the acoustic field of moving noise source recognition based on non-simplified model and simplified model, respectively, for single frequency sound source, multi-frequency sound source, dipole, distributed source and so on.
【学位授予单位】：中国计量学院
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U270.16;TB53

【参考文献】