環境中的未知噪音源識別技術，一直是聲學領域中非常重要的研究課題，近年來聲學陣列已成為一種有效的噪音源分析工具，透過將聲音視覺化並與光學影像結合的聲學檢測技術，能夠以直觀方式呈現出噪音在空間中的分佈情況，有助於了解噪音源發生的切確位置。反捲積去模糊化處理是一種常見於影像的增強處理技術，相關技術也應用於聲學成像以提升噪音分佈圖像的清晰度，本研究針對三種反捲積處理進行討論，「反捲積方法應用聲場成像法」（Deconvolution Approach for the Mapping of Acoustic Sources, DAMAS）、「稀疏反捲積法」（Sparsity Constrained Deconvolution Approach, SC-DAMAS）、「交叉譜矩陣擬合法」（Covariance Matrix Fitting Approach, CMF)。研究中通過數值模擬和實驗分析評估各方法的可行性，在數值模擬中，通過設定不同的訊噪比、快照數（Snapshot）了解不同場景的表現，並以定位誤差、聲場集中度和運算時間三項指標參數進行性能評估。實驗分析透過直徑1公尺的96通道Underbrink麥克風陣列進行量測，將聲場圖與陣列中央的鏡頭拍攝到的影像進行疊合，以此討論聲源定位能力。數值模擬中的場景包含簡單聲場以及複雜聲場，在簡單聲場中DAMAS、SC-DAMAS和CMF的平均定位誤差為DAS的86.4%、67.8%和68.7%，而聲場集中度的均方根差為DAS的12.9%、12.9%和13.3%，在複雜聲場中DAMAS、SC-DAMAS和CMF的平均定位誤差為DAS的85%、83.1%和85%，而聲場集中度的均方根差為DAS的11.9%、13.1%和11.2%，整體運算時間則是DAMAS與SC-DAMAS較為接近，CMF則遠大於另外兩種方法。實驗分析中，透過雙揚聲器的量測，所有反捲積方法皆能實現對成像的去模糊化，除了雙揚聲器外還對實例器材量測進行驗證，反捲積方法同樣能有效將波束合成的成像結果去模糊化，在吹風機量測中對寬頻聲源累加諧頻的成像具有更穩定的結果，實驗整體而言反捲積方法的表現差異不大，但SC-DAMAS和CMF在抑制僞聲源的能力些微優於DAMAS。研究經模擬和實驗驗證了反捲積方法的可行性，確實能有效將聲場成像去模糊化而獲得更精確的聲源位置，綜合模擬和實驗，SC-DAMAS在運算時間以及成像品質下是最佳的選擇，CMF在犧牲運算時間的前提下成像品質是最佳的。

In this study, three different deconvolution methods of acoustic beamforming were reviewed and analyzed. These methods deblur the images from Delay-and-Sum (DAS). Making the images is feasible to recognize the location of sound source. The first method is Deconvolution Approach for the Mapping of Acoustic Sources (DAMAS). The other two methods are Sparsity Constrained Deconvolution Approach (SC-DAMAS) and (Covariance Matrix Fitting Approach, CMF). In addition, the approach of overlapping camera image and acoustic image was introduced. Last by not least, we analyzed the performance and feasibility by simulation and experiment. In simulation three performance indexes were calculated to quantify the degree of improvement. They were the ability of locating sound sources (ALSS), ability of concentrating sound energy (ACSE) and computing time (CT). The results of ALSS in simple sound field simulation (SSFS) showed the error of DAMAS, SC-DAMAS ,and CMF compared to DAS were proportion of 86.4%, 67.8% and 68.7%. And ACSE were 12.9%,12.9% and 13.3%. In complex sound field simulation (CSFS) deconvolution approach ALSS of DAS proportions are 85%、83.1% and 85%. And ACSE were 11.9%, 13.1% and 11.2%. The result of CT showed all deconvolution methods need more time to compute. In comparison, CMF need much more time than DAMAS and SC-DAMAS. In experiment, the test of two speakers showed that deconvolution had the ability of find location of speakers. Besides, we measured hair dryer and pulse oximeter. In acoustic image, the deconvolution method appeared the sound energy appeared in prediction location. In experiment results, the methods didn’t show big difference between others. However, the ability of suppressing false sound source. SC-DAMAS and CMF were better than DAMAS.

[1] J. D. Maynard, E. G. Williams, and Y. Lee, “Nearfield acoustic holography: I. Theory of generalized holography and the development of NAH,” The Journal of the Acoustical Society of America, vol. 78, no. 4, pp. 1395–1413, Oct. 1985.
[2] L. de Santana, “Fundamentals of Acoustic Beamforming,” NATO Educational Notes EN-AVT-287-04, 2017.
[3] T. F. Brooks and W. M. Humphreys, “A deconvolution approach for the mapping of acoustic sources (DAMAS) determined from phased microphone arrays,” Journal of Sound and Vibration, vol. 294, no. 4–5, pp. 856–879, Jul. 2006.
[4] K. Rossmann, “Point Spread-Function, Line Spread-Function, and Modulation Transfer Function,” Radiology, vol. 93, no. 2, pp. 257–272, Aug. 1969.
[5] R. Baker Kearfott, “Preconditioners for the Interval Gauss–Seidel Method,” SIAM Journal on Numerical Analysis, vol. 27, no. 3, pp. 804–822, Jun. 1990.
[6] T. Yardibi, J. Li, P. Stoica, and L. N. Cattafesta, “Sparsity constrained deconvolution approaches for acoustic source mapping,” The Journal of the Acoustical Society of America, vol. 123, no. 5, pp. 2631–2642, May 2008.
[7] O. Lylloff, Efrén Fernández-Grande, F. T. Agerkvist, J. Hald, Elisabet Tiana Roig, and M. S. Andersen, “Improving the efficiency of deconvolution algorithms for sound source localization,” Journal of the Acoustical Society of America, vol. 138, no. 1, pp. 172–180, Jul. 2015.
[8] R. Tibshirani, “Regression Shrinkage and Selection Via the Lasso,” Journal of the Royal Statistical Society: Series B (Methodological), vol. 58, no. 1, pp. 267–288, Jan. 1996.
[9] Z. Prime and C. Doolan, “A comparison of popular beamforming arrays,” Proceedings of Acoustics 2013 Victor Harbor: Science Technology and Amenity, Annual Conference of the Australian Acoustical Society, Nov. 2013.
[10] Hsu, Wei-Jen, “Development of Acoustic Holography for Microphone Array Based on Compressive Sensing and Equivalent Source Method,” master thesis, National Cheng Kung University, Tainan, Taiwan, 2024
[11] Zhang, Y., Tu, X., Abbas, S., Liang, H., Huang, Y., & Ding, X. “ Learning to sound imaging by a model-based interpretable network,” Signal Processing, 215, 2023
[12] M. R. Bai, Jeong-Guon Ih, and J. Benesty, Acoustic Array Systems. John Wiley & Sons, 2013.
[13] Axel Rémi Hamran, “Estimation and correction of deviations between camera and audio images in acoustic imaging,” master's thesis, University of Oslo, Norway, 2020.
[14] M. Grant and S. Boyd, “The CVX Users’ Guide Release 2.2.” Available: https://cvxr.com/cvx/doc/CVX.pdf
[15] L. Stanković, “A measure of some time–frequency distributions concentration,” Signal Processing, vol. 81, no. 3, pp. 621–631, Mar. 2001.