麥克風陣列影像系統應用於近場聲源辨識暨演算法比較

Abstract

利用麥克風陣列實現噪音源辨識的技巧可分為兩大類：一類為近場聲學全像術，而另一類為波束合成法。此篇論文比較了由近場聲學全像術及波束合成法衍生出的數個演算法應用於近場上的效果。在近場聲學全像術衍生出的演算法中，此論文再訪傳統的傅立葉近場聲學全像術及我們先前發表的近場等效聲源成像法。進一步介紹在頻率域執行近場等效聲源成像法的技巧及應用，並且討論模擬及實驗的結果。近場等效聲源成像法利用多通道反算濾波器可重建數個聲學變數，包含聲壓、粒子速度與主動聲強。而當可供使用的麥克風數量不足時，可運用虛擬麥克風技巧裡的內插及外插方法分別增加解析度及減低邊緣效應。在到達方向估測及波束合成法衍生出的演算法中，將討論及比較延遲相加法、時間反轉法、最小變異無失真響應法與多重信號分類法。且由於應用於近場，上述演算法的入射波皆假設為球面波。經由數值模擬及實驗結果比較及討論上述所有的演算法。

Noise source identification (NSI) techniques using microphone arrays can be divided into two categories: Nearfield Acoustical Holography (NAH) and Beamforming. This thesis compares several algorithms of the methods derived from NAH and the methods derived from Beamforming in the nearfield. In the methods derived from NAH, this paper revisits traditional Fourier NAH and Nearfield Equivalent Source Imaging (NESI) proposed previously by the authors. The techniques and applications of the NESI that implemented in frequency domain termed as FDNESI are introduced and discussed by simulations and experiments. In FDNESI, acoustical variables including sound pressure, particle velocity, and active intensity are reconstructed by using multichannel inverse filters. A virtual microphone approach is employed to improve imaging resolution and minimize edge effects by using interpolation and extrapolation when sparse microphones are available. In the methods derived from Beamforming and direction of arrival (DOA) estimation: the delay and sum algorithm (DAS), the time reversal (TR) algorithm, the minimum variance distortionless response (MVDR) algorithm, and the multiple signal classification algorithm (MUSIC) are discussed and compared. As application in nearfield, the algorithms above are based on the assumption that the incoming waves are spherical waves. All the algorithms were compared and discussed by several numerical simulations and experiments in the nearfield.