CN101893698B - Noise source test and analysis method and device - Google Patents

Abstract

The invention discloses a noise source test and analysis method, which comprises the following steps of: 1) receiving sound waves radiated by a main noise source to be tested by adopting a microphone; 2) sampling the sound waves radiated by the main noise source to be tested, and saving and displaying the sound waves of the main noise source to be tested after analog-to-digital conversion; 3) processing the sound waves of the main noise source to be tested, which is saved by the step 2) by utilizing wavelet transform and high-order spectrum analysis technology to acquire a signal characteristic value for recognizing the noise source; and 4) estimating frequency response characteristics from the main noise source to be tested to the sensor by utilizing an independent component analysis method to estimate the direction of arrival of the sound source. The wavelet transform and bispectrum analysis-based noise source analysis method can eliminate the influence of non Gaussian noises, cannot cause loss of useful information of signals of the main noise source, and can accurately and effectively extract mode characteristics of the signals of the main noise source.

Description

Noise source test analytical approach and device thereof

Technical field

The present invention relates to a kind of sound source method for testing and analyzing, particularly the noise source test analytical approach based on wavelet transformation and higher-order spectrum analysis and realize the device of the method.

Background technology

Auditory localization technology, can be according to the position of determining target from the information of target, is important technology of location technology.Along with location technology development, the application of auditory localization technology is also increasingly extensive, is particularly applied in the strong industry spot of noise radiation.

Auditory localization, is often used for locating and identifying strong sounding position, provides foundation to noise reduction.Traditional Noise Sources Identification method, as frequency spectrum analysis method, easy analysis is directly perceived, in the noise-source analysis of rotating machinery application comparatively general, but be subject to the impact of the frequency response function of each sensor, can not identify sensitively main noise source.

Microphone is a kind of converting member that acoustic energy is converted to electric energy, and utilizing microphone to position sound source is to adopt microphone array to position the sound source position of the sound sending.Microphone array refers to by multiple microphones and puts according to different locus, jointly forms the device of a certain specific receiving system.At present, microphone array mainly adopts time delay to estimate that (Time-Delay-estimator) technology and time delay summation beam-forming technology carry out auditory localization, and its concrete implementation step is:,

1) first M microphone formed to microphone array according to certain topological structure, pick up voice signal and other noise signal that sound source is sent.

The reception signal F (t) that represents microphone array with mathematical formulae is:

F(t)＝[f ₁(t)f ₂(t)...f _M(t)] ^T (1)

Then the signal, each microphone being received carries out AD conversion

F(n)＝[f ₁(n)f ₂(n)...f _M(n)] ^T (2)

Calculate and the signal of Mei road, choose a frame signal and carry out Short Time Fourier Transform from formula (2) again:

$S\left(\mathrm{\ω}\right)\underset{m=1}{\overset{N}{\mathrm{\Σ}}}F\left(n\right)w(n-m)\exp (-\mathrm{j\ωm})=\left[\begin{array}{c}\underset{m=1}{\overset{n}{\mathrm{\Σ}}}{f}_1\left(n\right)w(n-m)\exp (-\mathrm{j\ωm})\\ \underset{m=1}{\overset{N}{\mathrm{\Σ}}}{f}_M\left(n\right)w(n-m)\exp (-\mathrm{j\ωm})\end{array}\right]---\left(3\right)$

Wherein: w (n) is window function, m is each mobile the counting of window function, and N is frame length.Each the Frequency point S (ω of S (ω) to formula (3) _i):

$S\left({\mathrm{\ω}}_i\right)=\left[\begin{array}{c}{S}_1\left(i\right)\\ \·\\ \·\\ \·\\ S_M\left(i\right)\end{array}\right]---\left(4\right)$

2) step 1 is obtained to each frequency matrix and carried out processing calculating mean space spectrum, its processing procedure is as follows:

1. obtain the spectrum correlation matrix of speech frame

R(i)：R(i)＝E{S(ω _i)S ^H(ω _i)}(5)

2. above-mentioned correlation matrix is carried out to Eigenvalues Decomposition, obtain eigenwert and characteristic of correspondence vector;

$R\left(i\right)=U_i\mathrm{diag}({\mathrm{\λ}}_1,...,{\mathrm{\λ}}_M)V_i^H---\left(6\right)$

3. be decomposed into noise subspace S and signal subspace according to the signal space of the large wisp proper vector composition of eigenwert

G：V _i＝[S _iMG _i](7)

4. obtain the position vector a of each searching position according to formula (8) _i(r, θ):

$a_i(r,\mathrm{\θ})=\left[\begin{array}{ccc}\frac{1}{r_1}{e}^{-j{\mathrm{\ω}}_i{\mathrm{\τ}}_1}& \frac{1}{r_m}{e}^{-j{\mathrm{\ω}}_i{\mathrm{\τ}}_m}& \frac{1}{r_M}{e}^{-j{\mathrm{\ω}}_i{\mathrm{\τ}}_M}\end{array}\right]---\left(8\right)$

Wherein: r _mit is the distance from search point (r, θ) to m microphone; τ _mthe travel-time of sound source from search point (r, θ) to m microphone; ω _irepresent i Frequency point.

Definition b _i(r, θ) is:

$b_i(r,\mathrm{\θ})=\frac{a_i(r,\mathrm{\θ})}{\left|\right|a_i(r,\mathrm{\θ})\left|\right|}---\left(9\right)$

The two-dimensional space spectrum matrix of obtaining i Frequency point is:

$p_i(\mathrm{\θ},r,{\mathrm{\ω}}_i)=\frac{1}{\left|\right|b_i^H(r,\mathrm{\θ})G_i\left|\right|}---\left(10\right)$

After processing by 4 steps above, select a frequency range and obtain the mean space spectrum in this frequency range:

$p_i(r,{\mathrm{\ω}}_i)=\frac{1}{k}\underset{i={\mathrm{\ω}}_L}{\overset{{\mathrm{\ω}}_H}{\mathrm{\Σ}}}{P}_i(\mathrm{\θ},r,{\mathrm{\ω}}_i)---\left(11\right)$

Wherein: ω _l, ω _hbe respectively lower bound and the upper bound frequency of this frequency range, K=ω _h-ω _l+ 1.

3) P (r, θ) obtaining in step 2 is carried out to two-dimentional spectrum peak search, thereby find the position of echo signal, at two-dimensional space spectrum matrix P (r, θ), find out peak value, the coordinate that peak value is corresponding is exactly sound source position estimated value: (r, θ).

But the major defect of the method is location out of true, and all Frequency points are all asked to spatial spectrum, operand is large, and real-time is low, realizes cost high, thereby can not effectively analyze and extract noise source, has larger limitation in auditory localization.

Summary of the invention

The key point of auditory localization technology is how to identify tested sound source and estimate the orientation of sound wave from ground unrest, and wherein in the urgent need to address is the resolution characteristic problem of basic matrix.In order to address the above problem, the object of the present invention is to provide a kind of noise source test analytical approach and device thereof, make it have higher signal resolution ability.

The technical scheme that the present invention solves its technical matters employing is:

1, a noise source test analytical approach, comprises the following steps:

1) adopt microphone to receive the sound wave of tested main noise source radiation;

2) sound wave of tested main noise source radiation is sampled; After analog-to-digital conversion, the acoustical signal of tested main noise source is preserved, shown;

3) utilize the acoustical signal of the tested main noise source that wavelet transformation and higher-order spectrum analytical technology preserve step 2 to process, obtain signal characteristic value, for identifying noise source;

4) utilize independent component analysis method to estimate the frequency response characteristic from tested main noise source to sensor, and then realize the estimation of sound source direction of arrival.

Further, the microphone of described step 1 is microphone array, and microphone number is greater than 1.

Described step 3 comprises the steps:

1) voice signal is carried out to wavelet transformation, the details component that is different scale by audio-signal resolution and approximate component, by at different levels points of quantitative analyses, determine the scale component that main noise source and ground unrest are concentrated, and by the scale component filtering that contains ground unrest, only retain the scale component at main noise source place, the burst of estimating as two spectrums;

2), by the burst after wavelet transformation, the two spectrums of the property entered are estimated, calculate the bispectrum feature of main noise source.

The concrete steps of described step 4 are:

1) utilize microphone array to receive tested main noise source signal vector x (t);

2) while utilization-time-frequency matrix of frequency analysis technique construction acoustical signal;

3) utilize Independent Component Analysis Technology, frequency response characteristic---the separation matrix by time-frequency Matrix Estimation from tested main noise source to sensor;

4) estimate that from separation matrix sound source ripple reaches the value of azimuth angle theta.

Further, use formula

${\mathrm{WT}}_x(a,b)=\frac{1}{\sqrt{a}}\&Integral;x\left(t\right)\mathrm{\&Psi;}\left[\frac{t-b}{a}\right]=<x\left(t\right),{\mathrm{\&Psi;}}_{a,b}\left(t\right)>---\left(1\right)$

Carry out wavelet transformation, wherein a is scale factor, and b is the time shift factor.Ψ _{a, b}(t) be that female small echo Ψ (t) is through displacement and the flexible cluster function producing.

Use formula

$C_{3X}({\mathrm{\&omega;}}_1,{\mathrm{\&omega;}}_2)=\underset{{\mathrm{\&tau;}}_1=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}\underset{{\mathrm{\&tau;}}_2=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}{c}_{3X}({\mathrm{\&tau;}}_1,{\mathrm{\&tau;}}_2)\exp \{-j\left({\mathrm{\&omega;}}_1{\mathrm{\&tau;}}_1\text{+}{\mathrm{\&omega;}}_2{\mathrm{\&tau;}}_2\right)\}---\left(2\right)$

Calculate the bispectrum feature of acoustical signal.

Realizing this noise source test analytical equipment comprises: microphone, signal conditioner, capture card, computing machine, data acquisition software and data analysis; After the sound wave of the tested main noise source that described microphone receives is amplified by signal condition joint, transfer to computing machine through capture card, carry out analog to digital conversion and show in real time, store acoustical signal by data acquisition software, acoustical signal after above-mentioned digitizing, realizes the identification of noise source by data analysis software computing.

Beneficial effect of the present invention:

Common statistical signal processing has three basic assumptions: linear, Gauss and stationarity, and in location, actual noise source, signal process using mainly with non-linear, non-Gauss and non-stationary signal as analyzing and the object of processing.Double-spectrum analysis is to process the powerful of gaussian signal, and it characterizes random signal from order Probability Structure more, can suppress in theory Gaussian noise completely, in higher-order spectrum, the exponent number of two spectrums is minimum, and disposal route is the simplest, also comprises all characteristics of higher-order spectrum simultaneously.But in the time that tested main noise source is submerged in non-Gaussian Background noise, it is unable to do what one wishes that double-spectrum analysis just seems.And the advantage of wavelet transformation is non-stationary signal, there is obvious time domain localization, can effectively strengthen the prompting message being hidden in vibration signal.

A kind of noise source test analytical approach provided by the invention, it is the noise-source analysis method based on small echo variation and double-spectrum analysis, can both eliminate like this impact of non-Gauss's noise like part, can not cause damage to the useful information of main noise source signal again, can extract comparatively accurately and efficiently the pattern feature of main noise source signal.

Accompanying drawing explanation

Fig. 1 is a kind of noise source test analytical equipment structural representation

Specific embodiment mode

Below in conjunction with drawings and the specific embodiments, principle of the present invention and particular content are elaborated.

A kind of main noise source method for testing and analyzing, comprises the steps

1) adopt microphone to receive the sound wave of tested main noise source radiation.

The present embodiment adopts the MPA416 wish polarization free field measuring microphone of Beijing popularity, and this sensitivity of microphone is high, can be connected with data collecting card, and strong signal output and lower background noise are provided, and is mainly used in array and acoustic measurement.Microphone number is selected 4.

2) sound wave of tested main noise source radiation is sampled; After analog-to-digital conversion, the acoustical signal of tested main noise source is preserved, shown.

By the acoustic signals of the tested main noise source radiation collecting, amplify through signal conditioner, by data collecting card, convert the simulating signal collecting to digital signal, transfer to the processing such as data acquisition software in computing machine stores, demonstration.This example is selected the MC104 signal conditioner of popularity, selects the PCI1714UL high-performance data capture card of NI company as the intermediate equipment that connects signal conditioner and computing machine.The data acquisition software of the present embodiment is the data acquisition software based on the design of LabView Software Development Platform.

3) utilize the acoustical signal of the tested main noise source that wavelet transformation and higher-order spectrum analytical technology preserve step 2 to process, obtain signal characteristic value, for identifying noise source.

This enforcement, by the noise-source analysis system of Matlab software development, can be analyzed the noise signal collecting by this software.Concrete steps:

First, the voice signal collecting is carried out to wavelet transformation, the details component that is different scale by audio-signal resolution and approximate component, by at different levels points of quantitative analyses, determine the scale component that main noise source and ground unrest are concentrated, and by the scale component filtering that contains ground unrest, only retain the scale component at main noise source place, the burst of estimating as two spectrums.

Wherein wavelet transformation is defined as:

For given signal x (t) ∈ L ²(R), the wavelet transformation of x (t) is defined as:

${\mathrm{WT}}_x(a,b)=\frac{1}{\sqrt{a}}\&Integral;x\left(t\right)\mathrm{\&Psi;}\left[\frac{t-b}{a}\right]=<x\left(t\right),{\mathrm{\&Psi;}}_{a,b}\left(t\right)>---\left(1\right)$

${\mathrm{\&Psi;}}_{a,b}\left(t\right)=\frac{1}{\sqrt{a}}\mathrm{\&Psi;}\left[\frac{t-b}{a}\right]---\left(2\right)$

In formula: a > 0 is scale factor, b is the time shift factor.Ψ _{a, b}(t) be female small echo Ψ (t) through displacement and the flexible cluster function producing, be referred to as wavelet basis.Define thus knownly, wavelet transformation is in fact original signal and the related operation of wavelet function after flexible bunch.By adjusting yardstick, the small echo that can obtain having different time-frequency width, to mate the diverse location of original signal, reaches the localization analysis of signal.

A, b parameter be corresponding to the frequency in time-frequency analysis and time variable, and the time-frequency characteristic according to unlike signal always in wavelet transformation, carries out adaptive selection by algorithm.For example, in noise source signal analysis, in the time that analysis yardstick a is more than or equal to 4, what acquired results was substantially corresponding is the ground unrest interference of high frequency.Therefore the analysis yardstick that, we choose actual noise source signal wavelet transformation is conventionally 1: 3.

Then,, by the burst after wavelet transformation, the two spectrums of the property entered are estimated, calculate the bispectrum feature of main noise source.Wherein orientating as of double-spectrum analysis:

The defined formula of two spectrums is as follows:

$C_{3X}({\mathrm{\&omega;}}_1,{\mathrm{\&omega;}}_2)=\underset{{\mathrm{\&tau;}}_1=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}\underset{{\mathrm{\&tau;}}_2=-\&infin;}{\overset{\&infin;}{\mathrm{\&Sigma;}}}{c}_{3X}({\mathrm{\&tau;}}_1,{\mathrm{\&tau;}}_2)\exp \{-j\left({\mathrm{\&omega;}}_1{\mathrm{\&tau;}}_1\text{+}{\mathrm{\&omega;}}_2{\mathrm{\&tau;}}_2\right)\}---\left(3\right)$

By the symmetry of three rank semi-invariants, can obtain symmetry and the symmetric domains thereof of two spectrums.

C _3X(ω ₁，ω ₂)＝C _3X(ω ₂，ω ₁)＝C _3X(-ω ₂，ω ₁)＝C _3X(-ω ₁-ω ₂，ω ₂)

＝C _3X(ω ₁，-ω ₁-ω ₂)＝C _3X(-ω ₁-ω ₂，ω ₁)＝C _3X(ω ₂，-ω ₁-ω ₂) (4)

For real stochastic process, two spectrums have 12 symmetrical regions.Therefore, from above formula, as long as known main trigonum ω ₂>=0, ω ₁>=ω ₂, ω ₁+ ω ₂two spectrums in≤π, just can describe all two spectrums completely.The flow process that two spectrums are calculated is as follows:

1. N the data of acoustical signal x (t) are divided into K section, every section containing M sample, i.e. N=KM;

2. remove the average of every segment data;

3. calculate the estimated value of three rank semi-invariants of each data segment;

4. getting the mean value of the three rank semi-invariants of all sections estimates as three rank semi-invariants of whole observation data group;

5. three rank semi-invariants are estimated to carry out two-dimension fourier transform, the two spectrums that obtain acoustical signal x (t) are estimated.Finally, estimated the eigenwert of the voice signal that just can obtain collecting by the two spectrums on above-mentioned wavelet transformed domain.

4) utilize independent component analysis method to estimate the frequency response characteristic from tested main noise source to sensor, and then realize the estimation of sound source direction of arrival, its step is as follows:

1. utilize hyperchannel microphone array or measurement mechanism to receive original sound-source signal vector x (t);

2. while utilization-frequency analysis technology (as short time discrete Fourier transform) builds time-frequency matrix of acoustical signal;

3. utilize independent component analysis method, frequency response characteristic---the separation matrix by time-frequency Matrix Estimation from tested main noise source to sensor;

4. estimate that from separation matrix sound source ripple reaches the value of azimuth angle theta.

The ultimate principle that orientation is estimated is:

Suppose the source signal s that has P to mix _pand Q sensor observation signal (t) wherein, h _qp(k) represent the impulse response from source p to sensor q.Allow d _qrepresent the position (linear sensor arrangement) of sensor q, θ _pexpression source s _porientation (becoming 90 to spend with the orthogonal of sensor array).

By L point short time DFT, time-domain signal x _q(t) x _q(t) convert frequency domain time series signal X to _q(f), wherein f=0, f _s/ L ..., f _s(L-1)/L, m is frame index.Therefore, mixture model can be by frequency represent.

Although signal mixes under the condition of echoing,, can be similar to and obtain an impulse response h _qp(k) frequency response H _pq(f) be wherein c is velocity of propagation.If θ _pas a variable θ, can obtain a guiding vector:

$a(f,\mathrm{\&theta;})={[e^{j2\mathrm{\&pi;}{\mathrm{fc}}^{-1}{d}_1\cos \mathrm{\&theta;}},...,e^{j2\mathrm{\&pi;}{\mathrm{fc}}^{-1}{d}_Q\cos \mathrm{\&theta;}}]}^T---\left(5\right)$

Thereby the observed quantity that sensor obtains can be expressed as wherein, X (f) is a Q dimensional vector X (f)=[X ₁(f) ..., X _q(f)] ^t.

If the separation matrix being obtained by ICA method can extract sound-source signal, and guarantee W (f) H (f)=I, thereby by H (f)=W ^-1(f) estimate the frequency response of commingled system.Here need to consider the yardstick of ICA decomposition and the uncertainty of signal source sequence.Therefore, with the real frequency response of commingled system should compared with, each row of H (f) have yardstick and signal source sequence arbitrarily.Each element H of matrix H (f) _pq(f) there is amplitude arbitrarily.Because the approximate value of commingled system do not meet this condition above, so, rebuild at initial point and there is decay A _qp(real-valued) and phase-modulation the model of commingled system:

By calculating two the element Hs corresponding with same source p _qpand H (f) _{q ' p}(f) ratio between, can ignore the uncertainty of yardstick:

$H_{\mathrm{qp}}/H_{q^{\&prime;}p}=A_{\mathrm{qp}}/A_{q^{\&prime;}p}{e}^{j2\mathrm{\&pi;}{\mathrm{fc}}^{-1}(d_q-d_{q^{\&prime;}})\cos {\mathrm{\&theta;}}_p}---\left(7\right)$

Can obtain computer azimuth angle θ by formula above _pa formula

${\mathrm{\&theta;}}_p={\cos }^{-1}\frac{\mathrm{angle}(H_{\mathrm{qp}}/H_{q^{\&prime;}p})}{2\mathrm{\&pi;}{\mathrm{fc}}^{-1}(d_q-d_{q^{\&prime;}})}---\left(8\right)$

If to discussed cos ^-1absolute value be greater than 1, θ _pbecome plural number, and can not get azimuth information.In this case, formula (7) can calculate q and q ' with another.From formula, this method can obtain any amount of source signal.

Since the uncertainty of sequence, θ _pmay be not and s _pcorresponding, but corresponding with another one source signal.But, by calculating all p=1 ..., the θ of P _p, can obtain the orientation of institute's active signal.By formula (8), what orientation can be similar to estimates.

For the situation of two sound sources, can prove θ _pcan be obtained by formula (7), same, the minimum value of zero direction in directional diagram also can be obtained by (7).In directional diagram, | B _r(f, θ) |=w _r(f) a (f, θ), wherein w _r(f) be that the r of separation matrix frequency response is capable.When | B _r(f, θ) | for hour, what θ was corresponding is zero direction.Work as omission during with f, minimize expression formula and be

$J\left(\mathrm{\&theta;}\right)=B_r\left(\mathrm{\&theta;}\right)\&CenterDot;B_r{\left(\mathrm{\&theta;}\right)}^{*}(W_{r1}{e}^{j{\mathrm{\&alpha;}}_1\cos \mathrm{\&theta;}}+W_{r2}{e}^{j{\mathrm{\&alpha;}}_2\cos \mathrm{\&theta;}})\&CenterDot;(W_{r1}^{*}{e}^{-j{\mathrm{\&alpha;}}_1\cos \mathrm{\&theta;}}+W_{r2}^{*}{e}^{-j{\mathrm{\&alpha;}}_2\cos \mathrm{\&theta;}})---\left(9\right)$

As α=α ₂-α ₁time, first order derivative and second derivative are respectively

$\frac{\mathrm{dJ}}{\mathrm{d\&theta;}}=-\mathrm{\&alpha;}\sin \mathrm{\&theta;}\&CenterDot;2\mathrm{im}\left(W_{r1}{W}_{r2}^{*}{e}^{-\mathrm{j\&alpha;}\cos \mathrm{\&theta;}}\right),$

wherein, re and im represent real part and imaginary part.If be zero, for positive number, the value of J (θ) is minimum.

$\mathrm{angle}(-W_{r1}{W}_{r2}^{*})=\mathrm{\&alpha;}\cos {\mathrm{\&theta;}}_r^{\mathrm{null}}\&DoubleLeftRightArrow;{\mathrm{\&theta;}}_r^{\mathrm{null}}={\cos }^{-1}\frac{\mathrm{angle}(-W_{r1}/W_{r2})}{2\mathrm{\&pi;}{\mathrm{fc}}^{-1}(d_2-d_1)}---\left(10\right)$

Therefore, zero direction be by r capable form, and to pass through with obtain.Thereby can see θ ₁with of equal value:

${\mathrm{\&theta;}}_1={\cos }^{-1}\frac{\mathrm{angle}(H_{21}/H_{11})}{2\mathrm{\&pi;}{\mathrm{fc}}^{-1}(d_2-d_1)}={\cos }^{-1}\frac{\mathrm{angle}(-W_{21}/W_{22})}{2\mathrm{\&pi;}{\mathrm{fc}}^{-1}(d_2-d_1)}={\mathrm{\&theta;}}_2^{\mathrm{null}}---\left(11\right)$

Can obtain all position angles according to formula (8) and (11).

Wavelet transformation and higher-order spectrum analysis are mainly used in analysis and the identification of noise source, estimate the frequency response characteristic from source to sensor with independent component analysis method, and then realize the estimation of sound source direction of arrival.They have formed the noise source test analytic system based on wavelet transformation and higher-order spectrum analysis jointly in conjunction with data acquisition equipment.

As shown in Figure 1, realize the noise source test analytical equipment of this algorithm, comprise microphone (1), signal conditioner (2), capture card (3), computing machine (4), data acquisition software (5) and data analysis software (6); The sound wave that is stung main noise source of described microphone (1) reception, after being amplified by signal conditioner (2), transfer to computing machine (4) through capture card (3), carry out analog to digital conversion and show in real time, store acoustical signal by carrying out data acquisition software (5), realize the test of tested main noise source by data analysis software (6) computing, comprise, orientation is estimated and identification of sound source.

Above embodiment only, for explanation technological thought of the present invention, can not limit protection scope of the present invention with this, every technological thought proposing according to the present invention, and any change of doing on technical scheme basis, within all falling into protection domain of the present invention.

Claims (2)

1. a noise source test analytical approach, is characterized in that, comprises the following steps:

1) adopt microphone to receive the sound wave of tested main noise source radiation, described microphone is microphone array;

2) sound wave of tested main noise source radiation is sampled; After analog-to-digital conversion, the acoustical signal of tested main noise source is preserved, shown;

3) utilize the acoustical signal of the tested main noise source that wavelet transformation and higher-order spectrum analytical technology preserve step 2 to process, obtain signal characteristic value, for identifying noise source;

4) utilize independent component analysis method to estimate the frequency response characteristic from tested main noise source to sensor, and then realize the estimation of sound source direction of arrival;

Described step 3) comprise the steps:

1) voice signal is carried out to wavelet transformation, the details component that is different scale by audio-signal resolution and approximate component, by at different levels points of quantitative analyses, determine the scale component that main noise source and ground unrest are concentrated, and by the scale component filtering that contains ground unrest, only retain the scale component at main noise source place, the burst of estimating as two spectrums;

2) by the burst after wavelet transformation, carry out two spectrums according to the scale component at main noise source place and estimate, calculate the bispectrum feature of main noise source; Use formula

$W{T}_x(a,b)=\frac{1}{\sqrt{a}}\&Integral;x\left(t\right)\mathrm{\&psi;}\left[\frac{t-b}{a}\right]=<x\left(t\right),{\mathrm{\&psi;}}_{a,b}\left(t\right)>$

Carry out wavelet transformation, wherein a is scale factor, and b is the time shift factor, ψ _{a, b}(t) be that female small echo ψ (t) is through displacement and the flexible cluster function producing; Use formula calculate the bispectrum feature of acoustical signal;

Described step 4) concrete steps be:

1) utilize microphone array to receive tested main noise source signal vector x (t);

2) while utilization-time-frequency matrix of frequency analysis technique construction acoustical signal;

3) utilize Independent Component Analysis Technology, frequency response characteristic---the separation matrix by time-frequency Matrix Estimation from tested main noise source to sensor;

4) estimate that from separation matrix sound source ripple reaches the value of azimuth angle theta.

2. a kind of noise source test analytical approach according to claim 1, is characterized in that, the microphone number of described step 1 is greater than 1.