An Improved Algorithm for Noise Source Localization Based on Equivalent Source Method

Xu, Zhongming; Wang, Qinghua; He, Yansong; Zhang, Zhifei; Li, Shu; Li, Mengran

doi:https://doi.org/10.1155/2016/8302862

Shock and Vibration

On this page

Abstract Introduction Numerical Simulation Conclusions Acknowledgments References Copyright Related Articles

Research Article | Open Access

Volume 2016 | Article ID 8302862 | https://doi.org/10.1155/2016/8302862

An Improved Algorithm for Noise Source Localization Based on Equivalent Source Method

Zhongming Xu,^1,2Qinghua Wang ,^1,2Yansong He,^1,2Zhifei Zhang,^1,2Shu Li,^1,2and Mengran Li^1,2

Academic Editor: Giorgio Dalpiaz

Received10 Sept 2016

Revised02 Nov 2016

Accepted17 Nov 2016

Published15 Dec 2016

Abstract

Near-field acoustical holography (NAH) based on the equivalent source method (ESM) is an efficient method applied for noise source identification. As -norm-based regularization cannot produce a satisfactory solution of the ill-conditioned problem in high frequency, the conventional ESM is restricted to relatively low frequency, and the resolution of conventional ESM in middle to high frequency remains a limitation open to investigation. This article presents an algorithm known as improved functional equivalent source method (IFESM), designed to enhance the resolution of conventional ESM. This method is developed in the framework of wideband acoustical holography (WBH) combining with functional beamforming (FB). Through numerical simulations, it is proved that the proposed method can localize noise with higher resolution compared with WBH and conventional ESM, and the ghosts on noise source map can be suppressed effectively. The validity and the feasibility of the proposed method are manifested by experiments including single-source and coherent-source localization.

1. Introduction

Since near-field acoustic holography (NAH) was first proposed by Williams et al. [1–3] in 1980s, it has been widely studied by the worldwide scholars. The equivalent source method (ESM), which is a kind of NAH reconstruction algorithm for sound sources with arbitrary shape, has some particular advantages such as robust calculation, fast speed, and easy implementation [4, 5]. Its core idea is to substitute the real source with a series of virtual equivalent sources. Then the equivalent source amplitude could be reconstructed with the transfer matrix and acoustic pressure vector, in which the pressure on reconstruction surface can be calculated for noise source identification [6, 7]. However, in the application of the ESM-based NAH, the number of microphones is always less than that of equivalent sources. This leads to a preliminary key problem: ill-posedness, which is often encountered in the inverse problems [8, 9]. Typically, classical -norm-based regularization methods such as Tikhonov regularization [10–12] and truncated singular value decomposition (TSVD) [7, 13, 14] are employed to find the optimal solution of the equivalent source amplitude. As the solution obtained is a least square solution, the ESM with -norm regularization focused on relatively low frequency, and the resolution of ESM-based NAH in middle to high frequency remains a limitation open to investigation.

Recently, sparse regularization based on minimizing a -norm cost function has emerged in the field of noise source identification [15, 16]. The -norm-based regularization has also received much attention, particularly motivated by compressive sensing theory under sparsity condition, where the minimum -norm solution is the sparsest solution [17–19]. A sparse sequence means the number of nonzero components is quite small, compared with its dimension. Under some conditions, it is possible to reconstruct the sparse signals with significantly less measurements (i.e., microphones) than classically required. In ESM-based NAH, the equivalent source amplitude vector can be thought as sparse because most components equal zero. In a sparsity framework, the sparse set of equivalent source amplitude is calculated by solving an objective function containing a penalty on the -norm of equivalent source vector. For example, Chardon et al. [20] used sparsity and compressive sensing principles for near-field acoustic holography to identify vibrating source velocity with fewer microphones than Tikhonov regularization required. Suzuki [16] published a similar method, but he used a monopole/dipole point source model of the same type as ESM, and he enforced sparsity in the source model by use of -norm penalty on the solution vector. Suzuki called his method Generalized Inverse Beamforming, focusing thus on rather long measurement distances. Chardon et al. used readily available MATLAB convex optimization software to solve the inverse problems including -norm minimization, while Suzuki developed his own iterative solver. More recently, Hald [21] proposed wideband acoustical holography method (WBH), in which the gradient iteration algorithm containing a dynamic filter process is employed to find the sparse solution of equivalent source amplitude. Compared with the conventional ESM with -norm-based regularization, WBH uses a dedicated iterative solver that enforces sparsity in the monopole source model. In this way, the reconstruction accuracy of WBH is much better than conventional ESM in middle to high frequency.

This paper focuses on improving the resolution of ESM in middle to high frequency. Firstly, a different filter function and a modification of the step length are applied into WBH to improve its reconstruction accuracy, and the improved WBH is known as iWBH in the following sections. Next, cross spectrum of the equivalent source amplitude vector is decomposed into a specific form. Finally, the high order matrix function [22, 23] is introduced to suppress the ghosts and reduce the beam-width. Through numerical simulation, it is proved that iWBH can reconstruct noise source with better accuracy than WBH, and the proposed method can localize noise sources more accurately in middle to high frequency.

2. Algorithm Description

The ESM-based NAH uses a source model in terms of a set of elementary sources and solves an inverse problem to reconstruct the complex amplitude of all elementary sources. According to the theory of ESM, the arrangements of equivalent source surface, measurement surface, and reconstruction surface are shown in Figure 1.

Assume that there are monopole point sources on the equivalent source surface and microphones set up on the measurement surface. In the framework of ESM-based NAH, the sound field excited by the real sources is considered to be equivalent to the sound field excited by these equivalent sources. Then the relationship between the measured pressure and the amplitude of these equivalent sources can be written in matrix vector notation aswhere (size ) denotes acoustic pressure measured by the microphone array. (size ) represents the complex amplitude of these equivalent sources. (size ) denotes the transfer matrix between acoustic pressure at the microphone positions and equivalent source positions. Each element of the transfer matrix is computed using the monopole radiation formula:with being the distance between the considered source-microphone couple (in m) and the acoustic wave number (in rad/m).

When using ESM to reconstruct the sound field, Tikhonov regularization is one of the efficient methods applied to find the optimal solution of the equivalent source amplitude. The Tikhonov regularization calculates the vector by making a tradeoff between residual norm and the solution norm . It can be described aswhere denotes the -norm and is the regularization parameter. Note that the transfer matrix is underdetermined, because the number of microphones is limited in most acoustic studies. With this additional term , the ill-posed inverse problem becomes well-posed. Generally, the parameter is chosen to be a fraction of the greatest eigenvalue of or determined by L-curve method. For a fixed parameter , Tikhonov regularization always has an explicit analytic solution to formula (3).where is a unit diagonal matrix and represents Hermitian transpose. The main limitation of the ESM with Tikhonov regularization is that the reconstructed sound pressure is much lower than the theoretical value, especially in middle to high frequency.

As mentioned before, the distribution of equivalent source amplitude is known to be sparse in the sense that most components of are close to zero. The limitation encountered in Tikhonov regularization can be alleviated by replacing the -norm in the penalty term by a -norm.where denotes the -norm and denotes noise bound. Formula (5) can be written as a minimizer of a convex unconstrained objective function.

From formulas (5) and (6), one can find that the sparsity is enforced by use of -norm penalty on the solution vector. The sparse solution of the objective function is typically obtained in an iterative manner as there is no analytic solution. Besides, the regularization parameter in formula (6) is determined empirically.

Different from -norm-based regularization, WBH algorithm uses a dedicated iterative solver including a dynamic filter to find the sparse solution of equivalent source amplitude. Firstly, the residual vector is defined related to formula (1).

The residual quadratic function to be minimized is defined as

Denoting by the approximate solution after the th iteration step, the optimal solution of the residual function is computed with the steepest descent method. Besides, the least squares solution is used as the initial value of the iteration. The step length that minimizes in the steepest direction is computed first:Here, is a constant vector related to the initial value , instead of using a relaxation factor in WBH. As the solution obtained with formula (4) is stable, the parameter in formula (10) can modify the step length to avoid big fluctuation. and are search direction and optimal search step length of the th iteration in the steepest decent method, respectively.

To suppress the ghosts, all components in below a certain threshold are set to zero. Different from the filter process in WBH, a new filter function including natural logarithm is used here, in which the filter function is smoother than that in WBH. Although there is no proof of how formula (11) performs better, the new filter function can improve reconstruction accuracy of WBH in the simulations, particularly for noise sources at middle to high frequency. The threshold value in each step can be expressed asHere, is updated during the iteration as shown in formula (13), and is the amplitude of the largest element in . Accordingly, the approximate solution after ()th iteration step can be written aswhere the range of is determined by initial value , maximum value and the constant . In formula (13), for , it could lead the filter process to be invalid. In simulations, the parameter in formula (13) is found to have a significant influence on the iteration result. When comparing iWBH with WBH, the value of , , and in WBH is set to be the same as in [21]. For iWBH, the following values have been found to work very well:

Suppose that final solution of is calculated after the th iteration step.

As shown in Figure 1, is the transfer matrix between equivalent source surface and reconstruction surface. The sound pressure on the reconstruction surface can be calculated by

Due to the fact that iterative algorithm can converge to the optimal solution, the resolution of WBH would reach a certain level when increasing iterative steps. To further improve the resolution, the output of iWBH is modified with a coefficient matrix obtained with FB.

Firstly, the cross spectrum of is defined as follows:where is nonnegative and Hermitian matrix. The high order matrix function with the order is introduced, and it is applied to by computing the spectral decomposition of and applying the function to the eigenvalues:where is a unitary matrix of of dimension . is the diagonal eigenvalue matrix. The output of FB at th grid can be expressed as follows:with being the -norm and the th row of the transfer matrix . Note that the components in matrix are not the sound intensity on the reconstruction surface but the coefficients related to the cross spectral matrix of reconstruction sound intensity.

In order to explain how the FB algorithm focuses more on main lobe and suppress side lobe, a simple model is constructed as shown in Figure 2. In the model, there is only a single sound source and only a corresponding nonzero equivalent source. Under this condition, we define as the single-source amplitude and as propagation vector between the sound source and equivalent source. represents the propagation vector from the only equivalent source to side lobe, and denotes the propagation vector to the main lobe.

The relative level of the side lobe can be expressed by ratio between the amplitude of and .where , , and are assumed to be unit vectors. is the included angle of and , and is the included angle between and . As the distribution of equivalent source amplitude and the sound intensity on reconstruction surface are almost consistent in terms of position, is approximately parallel to . Then, in formula (20) can be simplified as a function that depends on and . For a fixed , the reconstruction sound intensity in side lobe direction would decay faster than that in the main lobe direction as . Finally, the output of FB would focus more on the main lobe with the increase of . However, the output of FB is just the coefficient matrix corresponding to the amplitude of sound pressure matrix produced by iWBH. The final output of IFESM is generated by modifying the iWBH output with .

In formula (21), every component in is modified with to suppress the side lobe, and the amplitude remains accurate due to the good reconstruction accuracy of iWBH. In other words, IFESM combines the advantages of iWBH and FB into one algorithm to produce better resolution than iWBH and conventional ESM.

3. Numerical Simulation

In this section, noise source maps obtained with the methods previously presented are compared. The aim is to highlight the abilities in terms of the reconstruction accuracy of iWBH and resolution of IFESM. The noise sources, measurement array, and equivalent source surface are arranged as shown in Figure 1. The simulation objects are pulsating ball sources with radius of 0.01 m, vibrating velocity is 2.5 × 10⁻² m/s, and the sound velocity is 340 m/s. In all simulations, the distance between measurement surface and origin of coordinates is 0.3 m, and the distance between reconstruction surface and origin of coordinates is 0.08 m. The equivalent source surface is located at 0.001 m from sound sources. Noise sources, measurement surface, reconstruction surface, and equivalent surface are in the same coordinates. The aperture of measurement surface is 0.1 m × 0.1 m including 8 × 8 measurement points. The mesh on reconstruction surface has 41 × 41 grids with 0.02 m spacing, and the mesh on the equivalent source surface is the same as reconstruction surface.

In order to quantify the reconstruction accuracy of IWBH, the error is defined aswhere and are, respectively, the theoretical pressure and reconstruction sound pressure at the th measurement point.

3.1. Simulation One

The simulation one aims to analyze the reconstruction performance of iWBH for different type of sources. Figure 3 shows the reconstruction error of conventional ESM, WBH, and iWBH versus frequency. A single source is located at () m, and random noise is added to the complex measurement data at a level of 15 dB below the average sound pressure across the microphones. The reconstruction error is the average value of 20 simulations. Generally, the reconstruction accuracy of WBH and iWBH is better than conventional ESM. As conventional ESM focuses on relatively low frequency and the regularization parameter of Tikhonov method does not fit well in high frequency, the reconstruction error of conventional ESM increases rapidly versus frequency. Actually, differences between WBH and iWBH are the different filter function and modification on step length. The result confirms that iWBH performs better, especially in the frequencies higher than 600 Hz.

Figure 4 presents sound pressure on reconstruction surface for two coherent noise sources at 800 Hz. The two sources with equal strength are located at () m and () m, and random noise is also set to be 15 dB. As shown in Figure 4, WBH and iWBH have identified the peak of the sources which represent the source position and strength. It can be seen from Figure 4(d), in which a comparison along the middle row (the 21th row) is shown, that iWBH can identify the amplitude of reconstruction pressure more accurately than WBH. According to formula (22), the reconstruction error based on WBH and iWBH is −18.4 dB and −23.5 dB, respectively. Figure 5 shows the comparison of reconstruction sound pressures for two coherent noise sources at 1800 Hz, from which the same conclusion can be made that iWBH can reconstruct noise sources with better reconstruction accuracy.

(a)

(b)

(c)

(d)

(a)

(b)

(c)

(d)

3.2. Simulation Two

To verify the better resolution of IFESM, localization results obtained with iWBH, WBH, and IFESM are compared in this section. As an important parameter, represents the resolution of these three algorithms. It describes the ability of the algorithm to distinguish one source from another close to it and is given asHere, is the “diameter” of −15 dB down output of the algorithm and is the side length of the equivalent source surface.

Figure 6 shows the contour diagram of sound pressure on reconstruction surface. According to formula (23), res of different methods is shown in Table 1. The results show that IFESM can produce better resolution, and IFESM can further reduce the beam-width as the order increases from 1 to 5.

(a)

(b)

(c)

(d)

Noise source maps of coherent noise sources located at () m and () m are presented in Figure 7. The WBH noise source map is shown in Figure 7(a). Two main spots are apparent at the given positions, but strong side lobes prevent the actual detection of the source position. Figure 7(b) shows the iWBH noise source map. The noise source map obtained with iWBH exhibits two main spots at the known positions, with side lobes partially removed. The IFESM noise source map further reduces the level of side lobes and exhibits two main spots at the given positions with smaller beam-width. Figure 7(d) is IFESM noise source map with the order set to 6, and the resolution is further improved compared with Figure 7(c) in which the order is set to 2.

(a)

(b)

(c)

(d)

Figure 8 shows the noise source maps of IFESM versus order . Figures 8(a)–8(f) confirm that the resolution can be further improved as the order increases appropriately. However, there would be only a single source shown on the sound pressure distribution diagram (see Figure 8(e)) if order increases to a certain extent. From the results, the order is found to have a significant influence on the localization results. However, there is unknown method to determine the optimal value of order . In simulations and experiments, the order is hand-tuned for the best performance.

(a)

(b)

(c)

(d)

(e)

(f)

4. Experimental Validation

In order to verify the validity and feasibility of the application of IFESM, experimental examples are designed corresponding to the simulation in MATLAB as in Figure 9. The Brüel & Kjær 18-channel sector wheel array is employed to measure the sound pressure. Sampling frequency is 12.56 kHz and sampling time is 5 s. The frequency domain data is obtained by time domain pressure using fast Fourier transform. The stationary-random noise is excited from D&S type 139-5 loudspeakers fixed in front of the microphone array. Computing size is 0.8 m × 0.8 m with 2 cm spacing, and the dynamic range of output display is set to be 30 dB.

(a)

(b)

(c)

(d)

(e)

The WBH source map is shown in Figure 10(a). A large main spot is apparent at the position of the loudspeaker, but the beam-width is too large to detect the actual source position accurately. Figure 10(b) shows the iWBH noise source map. Compared with Figure 10(a), Figure 10(b) is more regular and the beam-width is reduced. The IFESM noise source map (Figures 10(c) and 10(d)) exhibits a main spot at source position with smaller beam-width. The results shown in Figure 10 indicate that IFESM offers better resolution compared with WBH and iWBH when applied to single source.

(a)

(b)

(c)

(d)

Figure 11 shows the noise source maps of WBH, iWBH, and IFESM for two coherent sources at. In WBH noise source map (Figure 11(a)), the strong side lobes prevent the actual identification of the source position. Figure 11(b) presents the iWBH source map, in which the side lobe is partially removed. With IFESM, the best result is obtained, in which the noise source map shows two spots at the position of loudspeakers. In general, IFESM is somewhat more accurate than WBH and iWBH for localizing the noise sources.

(a)

(b)

(c)

(d)

5. Conclusions

To improve the resolution of conventional ESM in middle to high frequency, IFESM is proposed by combining WBH with FB. Firstly, the underdetermined model of equivalent source method for noise source identification is conducted. Next, the theoretical background of conventional ESM and the proposed method are presented. For the known noise sources, numerical simulation results demonstrate the better reconstruction accuracy of iWBH in middle to high frequency compared with WBH and conventional ESM. Moreover, experimental results also indicate that IFESM can localize the single/coherent noise sources accurately and beam-width can be greatly reduced with reasonable increase of order . It is worth highlighting that the IFESM is useful and reliable to identify noise sources at middle to high frequency.

Competing Interests

The authors declare that they have no competing interests.

Acknowledgments

The authors gratefully acknowledge the Foundation and Advanced Research Project of Chongqing (no. CSTC2015jcyjBX0075) and graduate student research and innovation project of Chongqing (no. CYB16028).

References

E. G. Williams, J. D. Maynard, and E. Skudrzyk, “Sound source reconstruction using a microphone array,” Journal of Sound and Vibration, vol. 68, no. 1, pp. 340–344, 1980.
View at: Google Scholar
J. D. Maynard, E. G. Williams, and Y. Lee, “Nearfield acoustic holography: I. theory of generalized holography and the development of NAH,” Journal of Sound and Vibration, vol. 78, no. 4, pp. 1395–1412, 1984.
View at: Google Scholar
W. A. Veronesi and J. D. Maynard, “Nearfield acoustic holography: II. Holographic reconstruction algorithms and computer 514 implementation,” Journal of Sound and Vibration, vol. 81, no. 5, pp. 1307–1321, 1987.
View at: Google Scholar
G. H. Koopmann, L. Song, and J. Fahnline, “A method for computing acoustic fields based on the principle of wave superposition,” Journal of Sound and Vibration, vol. 86, no. 5, pp. 2433–2438, 1989.
View at: Google Scholar
L. Song, G. H. Koopmann, and J. B. Fahnline, “Numerical errors associated with the method of superposition for computing acoustic fields,” Journal of the Acoustical Society of America, vol. 89, no. 6, pp. 2625–2633, 1991.
View at: Publisher Site | Google Scholar
A. Sarkissian, “Extension of measurement surface in near-field acoustic holography,” Journal of the Acoustical Society of America, vol. 115, no. 4, pp. 1593–1596, 2004.
View at: Publisher Site | Google Scholar
A. Sarkissian, “Method of superposition applied to patch near-field acoustic holography,” Journal of the Acoustical Society of America, vol. 118, no. 2, pp. 671–678, 2005.
View at: Publisher Site | Google Scholar
P. A. Nelson and S. H. Yoon, “Estimation of acoustic source strength by inverse methods—part I, conditioning of the inverse problem,” Journal of Sound and Vibration, vol. 233, no. 4, pp. 643–668, 2000.
View at: Google Scholar
S. H. Yoon and P. A. Nelson, “Estimation of acoustic source strength by inverse methods: part II, experimental investigation of methods for choosing regularization parameters,” Journal of Sound and Vibration, vol. 233, no. 4, pp. 665–701, 2000.
View at: Google Scholar
A. N. Tikhonov, “Solution of incorrectly formulated problems and the regularization method,” Soviet Mathematics Doklady, vol. 4, no. 4, pp. 1035–1038, 1963.
View at: Google Scholar
A. Schuhmacher, J. Hald, K. B. Rasmussen, and P. C. Hansen, “Sound source reconstruction using inverse boundary element calculations,” Journal of the Acoustical Society of America, vol. 113, no. 1, pp. 114–127, 2003.
View at: Publisher Site | Google Scholar
Q. Leclère, “Acoustic imaging using under-determined inverse approaches: frequency limitations and optimal regularization,” Journal of Sound and Vibration, vol. 321, no. 3–5, pp. 605–619, 2009.
View at: Publisher Site | Google Scholar
T. Delillo, V. Isakov, N. Valdivia, and L. J. Wang, “The detection of the source of acoustical noise in two dimensions,” SIAM Journal on Applied Mathematics, vol. 61, no. 6, pp. 2104–2121, 2001.
View at: Publisher Site | Google Scholar
M. B. S. Magalhães and R. A. Tenenbaum, “Sound sources reconstruction techniques: a review of their evolution and new trends,” Acta Acustica United with Acustica, vol. 90, no. 2, pp. 199–220, 2004.
View at: Google Scholar
T. Suzuki, “Generalized inverse beam-forming algorithm resolving coherent/inherent distributed and multipole sources,” in Proceedings of the 14th AIAA/CEAS Aeroacoustics Conference (29th AIAA Aeroacoustics Conference), AIAA-2008-2954, Vancouver, Canada, May 2008.
View at: Publisher Site | Google Scholar
T. Suzuki, “ $L_{1}$ generalized inverse beam-forming algorithm resolving coherent/incoherent, distributed and multipole sources,” Journal of Sound and Vibration, vol. 330, no. 24, pp. 5835–5851, 2011.
View at: Publisher Site | Google Scholar
E. J. Candes and M. B. Wakin, “An introduction to compressive sampling,” IEEE Signal Processing, vol. 25, no. 2, pp. 21–30, 2008.
View at: Publisher Site | Google Scholar
G. F. Edelmann and C. F. Gaumond, “Beamforming using compressive sensing,” Journal of the Acoustical Society of America, vol. 130, no. 4, pp. EL232–EL237, 2011.
View at: Publisher Site | Google Scholar
S. Zhong, Q. Wei, and X. Huang, “Compressive sensing beamforming based on covariance for acoustic imaging with noisy measurements,” Journal of the Acoustical Society of America, vol. 134, no. 5, pp. EL445–EL451, 2013.
View at: Publisher Site | Google Scholar
G. Chardon, L. Daudet, A. Peillot, F. Ollivier, N. Bertin, and R. Gribonval, “Near-field acoustic holography using sparse regularization and compressive sampling principles,” Journal of the Acoustical Society of America, vol. 132, no. 3, pp. 1521–1534, 2012.
View at: Publisher Site | Google Scholar
J. Hald, “Fast wideband acoustical holography,” The Journal of the Acoustical Society of America, vol. 139, no. 4, pp. 1508–1517, 2016.
View at: Publisher Site | Google Scholar
R. P. Dougherty, “Functional beamforming,” in Proceedings of the 5th Berlin Beamforming Conference, BeBeC-2014-01, Berlin, Germany, February 2014.
View at: Google Scholar
R. P. Dougherty, “Functional beamforming for aero-acoustic source distributions,” in Proceedings of the 20th AIAA/CEAS Aero-Acoustics Conference, AIAA-2014-3066, pp. 2014–3066, Atlanta, Ga, USA, June.
View at: Google Scholar

Copyright

Copyright © 2016 Zhongming Xu et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

PDF Download Citation

Download other formats

Order printed copies

Views

807

Downloads

911

Citations