CN103439689A - System for estimating position of microphone in distributed microphone array - Google Patents

Abstract

The invention discloses a system for estimating the position of a microphone in a distributed microphone array. The system comprises a sound source signal generating module, a loudspeaker, a reference microphone, the microphone to be positioned and a microphone position estimation module. Signals output by the output end of the sound source signal generating module are converted into sound through the loudspeaker to be played in a work place of the distributed microphone array; the sound is received by the reference microphone and the microphone to be positioned; the signals received by the reference microphone and the signals received by the microphone to be positioned are simultaneously transmitted to the microphone position estimation module corresponding to the microphone to be positioned. Three signals different in frequency band are used as sounding signals of three sound sources, the three sound sources are collected and separated in the microphone position estimation process in a one-off mode, and then the position of the microphone can be estimated on line. The system combines the ranging methods based on energy and time delay, rough estimation and accurate estimation of the distance are performed successively, and thus the position can be estimated with less calculation amount and higher accuracy.

Description

Microphone position estimating system in a kind of distributed mike wind array

Technical field

The present invention relates to a kind of location estimation technology of microphone array, particularly microphone position estimating system in a kind of distributed mike wind array.

Background technology

Microphone array is rearranged by a plurality of microphones, each microphone while collection signal also carries out Combined Treatment to signal, not only can effectively suppress noise and reverberation, obtain good tonequality, can also obtain the azimuth information of sound, play an important role so microphone array is listed in the application such as voice enhancing, auditory localization, sound source separation.In recent years, development along with the progress of integrated circuit technique and network service, mobile computing technology, emerged a collection of portable equipment (as notebook computer, smart mobile phone etc.) of usually being furnished with Wi-Fi, bluetooth, audio interface etc., and constantly popularization and application in the routine work life.Utilize these portable equipments, can form ad hoc distributed mike wind array.

Different from regular microphone array, the distributed mike array does not have the information such as predetermined geometry and size, and each microphone may be by processor control alone, independent acquisition and processing data.Because the information such as the geometry of distributed mike wind array, size can't be predicted, need these information and carry out distributed mike wind Array Signal Processing, therefore, the microphone position in how to confirm distributed mike wind array is a technical barrier that must solve.

Chen Di equals in patent that 2011 is CN102123495A at publication number to disclose " a kind of wireless sensor network barycenter location algorithm of proofreading and correct based on RSSI ".The method is by the RSSI ranging technology of proofreading and correct, and the distance of point-to-point between measured node, adopt the triangle centroid algorithm to carry out node locating afterwards.The method calculated amount is little, is easy to realize, has practical value preferably.But for distributed mike wind array, due to voice signal power temporal evolution, and big rise and fall, so adopt signal energy to be found range, precision is often not high enough.

The correction sound source of the Xiao Hua of Chengdu University of Electronic Science and Technology three diverse locations of employing in " the bearing calibration research of microphone array " delivered in 2008 carry out the correction microphone array in microphone position.This scheme successively is positioned over three positions by a sound source, divides collection and the calculating of carrying out signal for three times, finally three result of calculations is merged and obtains microphone position.But the signals collecting of the method operation is more loaded down with trivial details, and efficiency is lower, can only calculated off-line, can't realize the On-line Estimation of microphone position.

Summary of the invention

The problems referred to above that exist for solving prior art, the present invention will design a kind ofly not only to have had degree of precision, but also has less computation complexity, and can realize microphone position estimating system in the distributed mike wind array of On-line Estimation.

To achieve these goals, technical scheme of the present invention is as follows: microphone position estimating system in a kind of distributed mike wind array comprises sound-source signal generation module, loudspeaker, reference microphone, a plurality of microphone to be positioned and a plurality of microphone position estimation module; Described sound-source signal generation module output terminal output signal is converted to sound through loudspeaker and plays in the work-yard of distributed mike wind array; The referenced microphone of this sound and microphone to be positioned receive; The signal that reference microphone and microphone to be positioned receive is passed to the microphone position estimation module corresponding with microphone to be positioned simultaneously; The output terminal of described microphone position estimation module is exported the coordinate of microphone to be positioned; Described microphone to be positioned is corresponding one by one with the microphone position estimation module;

The method of work of described sound-source signal generation module comprises the following steps:

Described sound-source signal generation module comprises white noise generator and bandpass filter, white noise generator produces white noise signal, described white noise signal is by three bandpass filter that passband is different, produce the bandpass signal of three different frequency bands, as the audible signal of three sound sources, input is placed in 3 of distributed mike wind array work-yards loudspeaker respectively; If the unit impulse response of three bandpass filter is respectively h _j(n), j=1,2,3, the white noise signal x (n) respectively white noise generator produced carries out filtering, obtains the bandpass signal s of three different frequency bands _j(n), j=1,2,3,

s _j(n)＝x(n)*h _j(n)，j＝1、2、3

Wherein, the convolution algorithm in symbol " * " representative digit signal process field; N means signal x, unit impulse response h ₁, h ₂, h ₃the sequence number of each sampled value;

Described microphone position estimation module quantity is identical with the quantity of microphone to be positioned, and the working method method of each microphone position estimation module is identical; If i microphone position estimation module be corresponding to i microphone to be positioned, this module comprises signal validation checking submodule, reference microphone sound-source signal segregant module, reference microphone sub-band ground unrest estimator module, microphone sound-source signal segregant module to be positioned, microphone sub-band ground unrest estimator module to be positioned, apart from rough estimate submodule, the smart estimator module of distance and triangle centroid calculation submodule;

Whether described signal validation checking submodule detects input signal effective, and the signal s that receives of computing reference microphone _r(n) the short-time energy E of every frame _r

$E_r=\underset{n=\mathrm{start}}{\overset{\mathrm{start}+M-1}{\mathrm{\Σ}}}{s}_r^2\left(n\right)$

Wherein, M means frame length, and start means that present frame is at primary reception burst s _r(n) reference position in; Compare E _rwith signal/noise decision threshold E _thrif, E _r<E _thr, judge that present frame is as noise frame, otherwise, judge that present frame is as signal frame;

When judging present frame as noise frame, the estimating background noise comprising ENERGY E _min, upgrade availability deciding threshold value E _thr; This threshold value update method is

if E _min＜E _r

E _min＝E _r

E _thr＝aE _min

end if

Wherein, a is a constant, and the span of a is 1<a<100;

Described reference microphone sound-source signal segregant module utilization three bandpass filter identical with coefficient in the sound-source signal generation module are separated three sound-source signals; The unit impulse response of these three bandpass filter is also h _j(n), j=1,2,3, receive signal s to reference microphone _r(n) carry out respectively filtering, obtain the signal s of three different frequency bands _rj(n), j=1,2,3,

s _rj(n)=s _r(n)*h _j(n)，j=1、2、3

When judging present frame as noise frame, described reference microphone sub-band ground unrest estimator module is estimated respectively the ENERGY E of three sub-frequency bands _rj, j=1,2,3,

$E_{\mathrm{rj}}=\underset{n=\mathrm{start}}{\overset{\mathrm{start}+M-1}{\mathrm{\&Sigma;}}}{s}_{\mathrm{rj}}^2\left(n\right),$ j=1、2、3

By E _rj, j=1,2,3, with a front reference microphone sub-band ground unrest energy estimated result ε _{rj, old}carry out smoothly, level and smooth result is as current ground unrest energy estimated value ε _{rj, curr},

ε _rj，curr＝(1-b)·ε _rj，old+b·E _rj，j＝1、2、3

Wherein, b is smoothing factor, means current ENERGY E _rj, j=1,2,3, proportion when level and smooth, the span of b is 0<b<1;

Described microphone sound-source signal segregant module to be positioned utilization three bandpass filter identical with coefficient in the sound-source signal generation module are separated three sound-source signals; The unit impulse response of these three bandpass filter is also h _j(n), j=1,2,3, receive signal s to microphone to be positioned _i(n) carry out respectively filtering, obtain the signal of three different frequency bands,

s _ij(n)=s _i(n)*h _j(n)，j=1、2、3

When judging present frame as noise frame, described microphone sub-band ground unrest estimator module to be positioned is estimated respectively the ENERGY E of three sub-frequency bands _ij, j=1,2,3,

$E_{\mathrm{ij}}=\underset{n=\mathrm{start}}{\overset{\mathrm{start}+M-1}{\mathrm{\&Sigma;}}}{s}_{\mathrm{ij}}^2\left(n\right),$ j=1、2、3

Again by E _ij, j=1,2,3, with front once microphone sub-band ground unrest energy estimated result ε to be positioned _{ij, old}carry out smoothly, level and smooth result is as current ground unrest energy estimated value ε _{ij, curr},

ε _ij，curr＝(1-b)·ε _ij，old+b·E _ij，j＝1、2、3

Wherein, b is smoothing factor, means current ENERGY E _ij, j=1,2,3, proportion when level and smooth, the span of b is 0<b<1;

Described as follows apart from the rough estimate submodule course of work:

Definition footmark r means reference microphone, and footmark i means i microphone to be positioned, and footmark j means j sound source (i.e. j loudspeaker, j=1,2,3), d _ijmean j sound source and i the Euclidean distance between microphone to be positioned, (x _r, y _r) and (x _j, y _j) mean respectively the coordinate of reference microphone and j sound source, d _rjmean the Euclidean distance between j sound source and reference microphone in the situation that microphone gain is calibrated (be microphone output signal energy identical) under the identical sound intensity, when judging present frame as signal frame, i microphone to be positioned to j sound source apart from rough estimate

for

${\tilde{d}}_{\mathrm{ij}}=d_{\mathrm{rj}}\&CenterDot;\sqrt{\frac{E_{\mathrm{rj}}\left(t\right){-\mathrm{\&epsiv;}}_{\mathrm{rj}}\left(t\right)}{E_{\mathrm{ij}}\left(t\right)-{\mathrm{\&epsiv;}}_{\mathrm{ij}}\left(t\right)}},$ j=1、2、3

The accuracy of estimating for improving follow-up distance essence, by current distance rough estimate result

j=1,2,3, with previous level and smooth di as a result _{j, old}carry out smoothly, obtaining the level and smooth d as a result of current distance rough estimate _{ij, curr}

$d_{\mathrm{ij},\mathrm{curr}}=(1-\mathrm{\&beta;})\&CenterDot;d_{\mathrm{ij},\mathrm{old}}+\mathrm{\&beta;}\&CenterDot;{\tilde{d}}_{\mathrm{ij}},$ j=1、2、3

Wherein, β is smoothing factor, means current distance rough estimate result

proportion in smoothing process, the span of β is 0<β<1;

The smart estimator module routine of described distance is as follows:

Definition α _rjand α _ijbe respectively sound and propagate into the decay factor of reference microphone and i microphone to be positioned, τ from sound source j _rjand τ _ijbe respectively sound and propagate into the time delay of reference microphone and i microphone to be positioned, v from sound source j _rand v (n) _i(n) be respectively the ground unrest at reference microphone and i microphone to be positioned place, when judging present frame as signal frame, the individual microphone to be positioned of reference microphone and i receives comes from same sound source s _isignal be respectively

$\begin{array}{c}{s}_{\mathrm{rj}}\left(n\right)={\mathrm{\&alpha;}}_{\mathrm{rj}}{s}_j(n-{\mathrm{\&tau;}}_{\mathrm{rj}})+v_r\left(n\right)\\ s_{\mathrm{ij}}\left(n\right)={\mathrm{\&alpha;}}_{\mathrm{ij}}{s}_j(n-{\mathrm{\&tau;}}_{\mathrm{ij}})+v_i\left(n\right)\end{array},$ j=1、2、3

Usually, ground unrest is the stationary noise that average is zero, and uncorrelated mutually with signal, different microphones place ground unrest is also uncorrelated mutually, s _rjand s (n) _ij(n) cross correlation function R _rij(τ) be expressed as

j=1、2、3

The distance essence is estimated to adopt the distance-finding method based on time delay, with the time delay of better simply cross-correlation method estimated signal; The τ possible to each calculates R _rij(τ), search maximum value position, obtain signal and arrive the delay time T of i microphone to be positioned with respect to reference microphone _rij,

${\mathrm{\&tau;}}_{\mathrm{rij}}={\mathrm{\&tau;}}_{\mathrm{rj}}-{\mathrm{\&tau;}}_{\mathrm{ij}}=\arg \underset{\mathrm{\&tau;}}{\max }{R}_{\mathrm{rij}}\left(\mathrm{\&tau;}\right),$ j=1、2、3

Because distance rough estimate module has estimated rough apart from d between each sound source of microphone to be positioned _{ij, curr}, near now only need thick distance, in certain limit, search for microphone meticulous distance; The definition velocity of sound is c, roughly apart from d _{ij, curr}corresponding thick time delay is t _{j, curr}=d _{ij, curr}/ c, at d _{ij, curr}near Δ d _{ij, curr}in scope, the meticulous distance of search, be converted to corresponding temporal information Δ t by range information _{j, curr}=Δ d _{ij, curr}/ c, the hunting zone of meticulous time delay is [t _{j, curr}-Δ t _{j, curr}, t _{j, curr}+ Δ t _{j, curr}];

Signal arrival time difference is multiplied by velocity of sound c, obtains the range difference that signal arrives two microphones

d _rij＝(τ _ij-τ _rj)·c＝-τ _rij·c，j＝1、2、3

According to reference microphone to j sound source apart from d _rj, and j sound source is to the range difference d of reference microphone and i microphone to be positioned _rij, obtain i microphone to be positioned to j sound source apart from d _ij

d _ij＝d _rj+d _rij＝d _rj-τ _rij·c (12)

The course of work of described triangle centroid calculation submodule is as follows:

Definition (x _s1, y _s1), (x _s2, y _s2) and (x _s3, y _s3) be respectively three sound source coordinates, d _i1, d _i2, d _i3be respectively the distance of microphone to three a to be positioned sound source.When judging present frame as signal frame, obtaining meticulous apart from d to j sound source of above-mentioned i microphone to be positioned _ijafterwards, according to geometric relationship, by solving, take sound source position as the center of circle, apart from the intersection point that is three circles of radius, determine i microphone to be positioned position,

$\left\{\begin{array}{c}{(x-x_{s1})}^2+{(y-y_{s1})}^2={\mathrm{<figure-callout\; id="1"\; label="d\; i"\; filenames="HDA0000369914230000032.png,HDA0000369914230000033.png"\; state="\{\{state\}\}">d</figure-callout>}}_i^1\\ {(x-x_{s2})}^2+{(y-y_{s2})}^2={\mathrm{<figure-callout\; id="2"\; label="d\; i"\; filenames="HDA0000369914230000031.png,HDA0000369914230000032.png"\; state="\{\{state\}\}">d</figure-callout>}}_i^2\\ {(x-x_{s3})}^2+{(y-y_{s3})}^2={\mathrm{<figure-callout\; id="3"\; label="d\; i"\; filenames="HDA0000369914230000011.png,HDA0000369914230000032.png"\; state="\{\{state\}\}">d</figure-callout>}}_i^3\end{array}\right.---\left(13\right)$

Only have in the ideal case, three circles m that just can intersect at a point, coordinate is (x _m, y _m); And in the ordinary course of things, due to the existence of the error of calculation, this system of equations, without solution, is divided into two kinds of situations without solution: the first is that three circles can not meet at a bit, but surrenders mutually a zone; The second is wherein to have at least two circles there is no intersection point;

For the first, without the solution situation, select at a distance of three nearest intersection point p ₁(x _p1, y _p1), p ₂(x _p2, y _p2) and p ₃(x _p3, y _p3) form triangle, ask for this triangle barycenter m, coordinate is (x _m, y _m), as sound source position,

$(x_m,y_m)=\frac{1}{3}\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}(x_{\mathrm{ni}},y_{\mathrm{ni}})---\left(14\right)$

For the second, without the solution situation, consider when estimating microphone position, each microphone to be positioned and correction sound source position are all fixed, and two round intersection points of this calculating by former frame serve as this frame intersection point.

The optimum value of a of the present invention is 4; The optimum value of b is 0.05; The optimum value of β is 0.05.

Compared with prior art, the present invention has following beneficial effect:

1, the present invention selects the audible signal of the voice signal of three different frequency bands as three sound sources, can disposable collection in the microphone position estimation procedure and isolate three sound sources, needn't calculated off-line, and can realize the On-line Estimation of microphone position.

2, the present invention is combined with the distance-finding method based on energy and time delay, successively carries out thick, the essence of distance and estimates, has guaranteed that range finding has less calculated amount and higher precision.

3, the present invention adopts the position that the triangle centroid algorithm obtains each microphone in array, to various, without the solution situation, corresponding processing is all arranged, and guarantees continuation and the stability of positioning result.

The accompanying drawing explanation

14, the total accompanying drawing of the present invention, wherein:

Fig. 1 is distributed mike wind array microphone position estimating system functional block diagram.

Fig. 2 is sound-source signal generation module functional block diagram.

Fig. 3 is the functional block diagram of i microphone position estimation module.

Fig. 4 is the three round schematic diagram that intersect at a point.

Fig. 5 is that three circles intersect at an area schematic.

Fig. 6 is the non-intersect schematic diagram of three circles.

Distributed microphone array and sound source distribution plan when Fig. 7 is emulation testing.

Fig. 8 is the amplitude-frequency response schematic diagram of three bandpass filter.

Fig. 9 is the horizontal ordinate estimated result that coordinate is (8,5) m place microphone.

Figure 10 is the ordinate estimated result that coordinate is (8,5) m place microphone.

Figure 11 is the coordinate evaluated error that coordinate is (8,5) m place microphone.

Figure 12 is the horizontal ordinate estimated result that coordinate is (9,2) m place microphone.

Figure 13 is the ordinate estimated result that coordinate is (9,2) m place microphone.

Figure 14 is the coordinate evaluated error that coordinate is (9,2) m place microphone.

In figure: 1, sound-source signal generation module, 2, loudspeaker, reference microphone, 3, the microphone position estimation module, 4, microphone to be positioned.

Embodiment

Below in conjunction with accompanying drawing, the present invention is described further.

As shown in Figure 1, microphone position estimating system in a kind of distributed mike wind array, comprise sound-source signal generation module 1, loudspeaker, reference microphone 2, a plurality of microphone 4 to be positioned and a plurality of microphone position estimation module 3; Described sound-source signal generation module 1 output terminal output signal is converted to sound through loudspeaker and plays in the work-yard of distributed mike wind array; The referenced microphone 2 of this sound and microphone to be positioned 4 receive; The signal that reference microphone 2 and microphone to be positioned 4 receive is passed to the microphone position estimation module 3 corresponding with microphone 4 to be positioned simultaneously; The output terminal of described microphone position estimation module 3 is exported the coordinate of microphone 4 to be positioned; Described microphone to be positioned 4 is corresponding one by one with microphone position estimation module 3;

As shown in Figure 2, described sound-source signal generation module 1 comprises white noise generator and bandpass filter, white noise generator produces white noise signal, described white noise signal is by three bandpass filter that passband is different, produce the bandpass signal of three different frequency bands, as the audible signal of three sound sources, input is placed in 3 of distributed mike wind array work-yards loudspeaker respectively.

As shown in Figure 3, described microphone position estimation module 3 quantity are identical with the quantity of microphone 4 to be positioned, and the working method method of each microphone position estimation module 3 is identical; Described microphone position estimation module 3 comprises signal validation checking submodule, reference microphone 2 sound-source signal segregant modules, reference microphone 2 sub-band ground unrest estimator modules, microphone to be positioned 4 sound-source signal segregant modules, microphone to be positioned 4 sub-band ground unrest estimator modules, apart from rough estimate submodule, the smart estimator module of distance and triangle centroid calculation submodule;

As shown in Fig. 4-6, the course of work of described triangle centroid calculation submodule is as follows:

Definition (x _s1, y _s1), (x _s2, y _s2) and (x _s3, y _s3) be respectively three sound source coordinates, d _i1, d _i2, d _i3be respectively the distance of i 4 to three sound sources of microphone to be positioned.When judging present frame as signal frame, obtaining meticulous apart from d to j sound source of above-mentioned i microphone 4 to be positioned _ijafterwards, j=1,2,3, according to geometric relationship, take sound source position as the center of circle, apart from the intersection point that is three circles of radius by solving, and determines i microphone to be positioned position,

$\left\{\begin{array}{c}{(x-x_{s1})}^2+{(y-y_{s1})}^2={\mathrm{<figure-callout\; id="1"\; label="d\; i"\; filenames="HDA0000369914230000032.png,HDA0000369914230000033.png"\; state="\{\{state\}\}">d</figure-callout>}}_i^1\\ {(x-x_{s2})}^2{+(y-y_{s2})}^2={\mathrm{<figure-callout\; id="2"\; label="d\; i"\; filenames="HDA0000369914230000031.png,HDA0000369914230000032.png"\; state="\{\{state\}\}">d</figure-callout>}}_i^2\\ {(x-x_{s3})}^2+{(y-y_{s3})}^2={\mathrm{<figure-callout\; id="3"\; label="d\; i"\; filenames="HDA0000369914230000011.png,HDA0000369914230000032.png"\; state="\{\{state\}\}">d</figure-callout>}}_i^3\end{array}\right.$

Only have in the ideal case, three circles m that just can intersect at a point, coordinate is (x _m, y _m) (as shown in Figure 4); And in the ordinary course of things, due to the existence of the error of calculation, this system of equations, without solution, is divided into two kinds of situations without solution: the first is that three circles can not meet at a bit, but surrenders mutually a zone (as shown in Figure 5); The second is wherein to have at least two circles there is no intersection point (as shown in Figure 6);

For the first, without the solution situation, select at a distance of three nearest intersection point p ₁(x _p1, y _p1), p ₂(x _p2, y _p2) and p ₃(x _p3, y _p3) form triangle, ask for this triangle barycenter m, coordinate is (x _m, y _m), as sound source position,

$(x_m,y_m)=\frac{1}{3}\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}(x_{\mathrm{ni}},y_{\mathrm{ni}})---\left(14\right)$

For the second, without the solution situation, consider when estimating microphone position, each microphone 4 to be positioned and correction sound source position are all fixed, and two round intersection points of this calculating by former frame serve as this frame intersection point.

The optimum value of a of the present invention is 4; The optimum value of b is 0.05; The optimum value of β is 0.05.

For verifying advantage of the present invention, carried out following emulation testing.Suppose the zone of a 10m * 10m, the coordinate on four summits is respectively (0,0) m, (0,10) m, (10,0) m and (10,10) m.The coordinate of three calibration sound sources is respectively (0,0) m, (0,1) m and (1,0) m, and the reference microphone coordinate is (1,1) m.The microphone that 8 positions to be determined are arranged in zone, and the gain of each microphone is calibrated.Calibration sound source, reference microphone, microphone position to be positioned distribute as shown in Figure 7.

If three bandpass filter are 8 rank Butterworth wave filters, cut-off frequecy of passband is respectively [450Hz, 650Hz], [950Hz, 1150Hz] and [1450Hz, 1650Hz], and coefficient is respectively:

(1) [450Hz, 650Hz] bandpass filter:

B1=1.0×10 ^-4×[0.021505689961486,0,-0.086022759845944,0,0.129034139768916,0,-0.086022759845944,0,0.021505689961486];

A1=[1.0,-7.619560406446450,25.571208572772093,-49.362204407327965,59.944291732842998,-46.892314668407494,23.076325245729549,-6.532148379124553,0.814405997727282];

(2) [950Hz, 1150Hz] bandpass filter:

B2=1.0×10 ^-4×[0.021505687637543,0,-0.086022750550173,0,0.129034125825259,0,-0.08602275055017,0,0.021505687637543];

A2=[1.0,-7.147003817816104,22.953870141090817,-43.181776132672987,51.995025280573671,-41.021175158245249,20.714374288994392,-6.127031864547846,0.814405997727283];

(3) [1450Hz, 1650Hz] bandpass filter:

B3=1.0×10 ^-4×[0.021505687407555,0,-0.086022749630222,0,0.129034124445332,0,-0.08602274963022,0,0.021505687407555];

A3=[1.0,-6.399791880553000,19.157164010994535,-34.615721568256149,41.140865143886366,-32.883785334287396,17.288132742285850,-5.486456951495592,0.814405997727280].

The frequency response curve of three wave filters as shown in Figure 8.

Near the meticulous distance of search in the 0.1m scope distance rough estimate, approximate 340 meter per seconds that are taken as of velocity of sound c.Sample frequency is 16KHz, and it is 0.2 second (being that frame length M is 3200 points) that data are processed frame length.

Table 1 provides each microphone position estimated result, from table 1, when, snr of received signal near apart from sound source at microphone is larger, only less based on energy range finding positioning error, but signal to noise ratio (S/N ratio) reduces, and positioning error increases.And two kinds of distance-finding methods are combined, by the time delay distance-finding method, energy range finding result is revised, signal to noise ratio (S/N ratio) size no matter, positioning error is all smaller.

Table 1 microphone position estimated result

Fig. 9-Figure 14 be take coordinate and is that the microphone of (8,5) m and (9,2) m is that example provides each frame estimated result in position fixing process.From Fig. 9-Figure 14, what dotted line meaned only has fluctuation all the time based on the energy positioning result, be convergent tendency and solid line means based on energy and the every frame alignment result of time delay associated methods, and speed of convergence is very fast, therefore adopt associated methods can realize that rapid, stable microphone position estimates.

Aspect calculated amount, if only adopt the distance-finding method based on time delay, directly carry out cross-correlation calculation,, under 16KHz sampling rate condition, while calculating 0.2 second, the cross correlation function of long data need to carry out about 1.024 * 10 ⁷inferior multiplying and 1.023 * 10 ⁷the sub-addition computing; And the energy that first calculates two paths of signals carries out apart from rough estimate, then carry out cross-correlation calculation at a small amount of some place (under this simulated conditions, being equivalent to 11 sampled points in the 0.1m scope), only need carry out about 4.16 * 10 ⁴inferior multiplying and 4.159 * 10 ⁴the sub-addition computing.Calculated amount only is down to based on approximately 1/240 of time delay distance-finding method.

Claims (2)

1. microphone position estimating system in a distributed mike wind array, is characterized in that: comprise sound-source signal generation module (1), loudspeaker, reference microphone (2), a plurality of microphone to be positioned (4) and a plurality of microphone position estimation module (3); Described sound-source signal generation module (1) output terminal output signal is converted to sound through loudspeaker and plays in the work-yard of distributed mike wind array; The referenced microphone of this sound (2) and microphone to be positioned (4) receive; The signal that reference microphone (2) and microphone to be positioned (4) receive is passed to the microphone position estimation module (3) corresponding with microphone to be positioned (4) simultaneously; The output terminal of described microphone position estimation module (3) is exported the coordinate of microphone to be positioned (4); Described microphone to be positioned (4) is corresponding one by one with microphone position estimation module (3);

The method of work of described sound-source signal generation module (1) comprises the following steps:

Described sound-source signal generation module (1) comprises white noise generator and bandpass filter, white noise generator produces white noise signal, described white noise signal is by three bandpass filter that passband is different, produce the bandpass signal of three different frequency bands, as the audible signal of three sound sources, input is placed in 3 of distributed mike wind array work-yards loudspeaker respectively; If the unit impulse response of three bandpass filter is respectively h _j(n), j=1,2,3, the white noise signal x (n) respectively white noise generator produced carries out filtering, obtains the bandpass signal s of three different frequency bands _j(n), j=1,2,3,

s _j(n)＝x(n)*h _j(n)，j=1、2、3

Wherein, the convolution algorithm in symbol " * " representative digit signal process field; N means signal x, unit impulse response h ₁, h ₂, h ₃the sequence number of each sampled value;

Described microphone position estimation module (3) quantity is identical with the quantity of microphone to be positioned (4), and the working method method of each microphone position estimation module (3) is identical; I microphone position estimation module (3) be corresponding to i microphone to be positioned (4), and this module comprises signal validation checking submodule, reference microphone (2) sound-source signal segregant module, reference microphone (2) sub-band ground unrest estimator module, microphone to be positioned (4) sound-source signal segregant module, microphone to be positioned (4) sub-band ground unrest estimator module, apart from rough estimate submodule, the smart estimator module of distance and triangle centroid calculation submodule;

Whether described signal validation checking submodule detects input signal effective, and the signal s that receives of computing reference microphone (2) _r(n) the short-time energy E of every frame _r

$E_r=\underset{n=\mathrm{start}}{\overset{\mathrm{start}+M-1}{\mathrm{\&Sigma;}}}{s}_r^2\left(n\right)$

Wherein, M means frame length, and start means that present frame is at primary reception burst s _r(n) reference position in; Compare E _rwith signal/noise decision threshold E _thrif, E _r<E _thr, judge that present frame is as noise frame, otherwise, judge that present frame is as signal frame;

When judging present frame as noise frame, the estimating background noise comprising ENERGY E _min, upgrade availability deciding threshold value E _thr; This threshold value update method is

if E _min＜E _r

E _min＝E _r

E _thr＝aE _min

end if

Wherein, a is a constant, and the span of a is 1<a<100;

Described reference microphone (2) sound-source signal segregant module utilizes three bandpass filter identical with coefficient in sound-source signal generation module (1) that three sound-source signals are separated; The unit impulse response of these three bandpass filter is also h _j(n), j=1,2,3, receive signal s to reference microphone (2) _r(n) carry out respectively filtering, obtain the signal s of three different frequency bands _rj(n), j=1,2,3,

s _rj(n)=s _r(n)*h _j(n)，j=1、2、3

When judging present frame as noise frame, described reference microphone (2) sub-band ground unrest estimator module is estimated respectively the ENERGY E of three sub-frequency bands _rj, j=1,2,3,

$E_{\mathrm{rj}}=\underset{n=\mathrm{start}}{\overset{\mathrm{start}+M-1}{\mathrm{\&Sigma;}}}{s}_{\mathrm{rj}}^2\left(n\right),$ j=1、2、3

By E _rj, j=1,2,3, with a front reference microphone (2) sub-band ground unrest energy estimated result ε _{rj, old}carry out smoothly, level and smooth result is as current ground unrest energy estimated value ε _{rj, curr},

ε _rj，curr＝(1-b)·ε _rj，old+b·E _rj，j＝1、2、3

Wherein, b is smoothing factor, means current ENERGY E _rj, j=1,2,3, proportion when level and smooth, the span of b is 0<b<1;

Described microphone to be positioned (4) sound-source signal segregant module utilizes three bandpass filter identical with coefficient in sound-source signal generation module (1) that three sound-source signals are separated; The unit impulse response of these three bandpass filter is also h _j(n), j=1,2,3, receive signal s to microphone to be positioned (4) _i(n) carry out respectively filtering, obtain the signal of three different frequency bands,

s _ij(n)=s _i(n)*h _j(n)，j=1、2、3

When judging present frame as noise frame, described microphone to be positioned (4) sub-band ground unrest estimator module is estimated respectively the ENERGY E of three sub-frequency bands _ij, j=1,2,3,

$E_{\mathrm{ij}}=\underset{n=\mathrm{start}}{\overset{\mathrm{start}+M-1}{\mathrm{\&Sigma;}}}{s}_{\mathrm{ij}}^2\left(n\right),$ j=1、2、3

Again by E _ij, j=1,2,3, with front once microphone to be positioned (4) sub-band ground unrest energy estimated result ε _{ij, old}carry out smoothly, level and smooth result is as current ground unrest energy estimated value ε _{ij, curr},

εi _j，curr＝(1-b)·ε _ij，old+b·E _ij，j＝1、2、3

Wherein, b is smoothing factor, means current ENERGY E _ij, j=1,2,3, proportion when level and smooth, the span of b is 0<b<1;

Described as follows apart from the rough estimate submodule course of work:

Definition footmark r means reference microphone (2), and footmark i means i microphone to be positioned (4), and footmark j means j sound source (i.e. j loudspeaker, j=1,2,3), d _ijmean j sound source and i the Euclidean distance between microphone to be positioned (4), (x _r, y _r) and (x _j, y _j) mean respectively the coordinate of reference microphone (2) and j sound source, d _rjmean the Euclidean distance between j sound source and reference microphone (2)

in the situation that microphone gain is calibrated (be microphone output signal energy identical) under the identical sound intensity, when judging present frame as signal frame, i current microphone to be positioned (4) to j sound source apart from rough estimate

for

${\tilde{d}}_{\mathrm{ij}}=d_{\mathrm{rj}}\&CenterDot;\sqrt{\frac{E_{\mathrm{rj}}\left(t\right){-\mathrm{\&epsiv;}}_{\mathrm{rj}}\left(t\right)}{E_{\mathrm{ij}}\left(t\right)-{\mathrm{\&epsiv;}}_{\mathrm{ij}}\left(t\right)}},$ j=1、2、3

The accuracy of estimating for improving follow-up distance essence, by current distance rough estimate result j=1,2,3, with previous level and smooth d as a result _{ij, old}carry out smoothly, obtaining the level and smooth d as a result of current distance rough estimate _{ij, curr}

$d_{\mathrm{ij},\mathrm{curr}}=(1-\mathrm{\&beta;})\&CenterDot;d_{\mathrm{ij},\mathrm{old}}+\mathrm{\&beta;}\&CenterDot;{\tilde{d}}_{\mathrm{ij}},$ j=1、2、3

Wherein, β is smoothing factor, means current distance rough estimate result proportion in smoothing process, the span of β is 0<β<1;

The smart estimator module routine of described distance is as follows:

Definition α _rjand α _ijbe respectively sound and propagate into the decay factor of reference microphone (2) and i microphone to be positioned (4), τ from j sound source _rjand τ _ijbe respectively sound and propagate into the time delay of reference microphone (2) and i microphone to be positioned (4), v from sound source j _rand v (n) _i(n) be respectively the ground unrest that reference microphone (2) and i microphone to be positioned (4) located, when judging present frame as signal frame, reference microphone (2) and the individual microphone to be positioned (4) of i receive comes from same sound source s _isignal be respectively

$\begin{array}{c}{s}_{\mathrm{rj}}\left(n\right)={\mathrm{\&alpha;}}_{\mathrm{rj}}{s}_j(n-{\mathrm{\&tau;}}_{\mathrm{rj}})+v_r\left(n\right)\\ s_{\mathrm{ij}}\left(n\right)={\mathrm{\&alpha;}}_{\mathrm{ij}}{s}_j(n-{\mathrm{\&tau;}}_{\mathrm{ij}})+v_i\left(n\right)\end{array},$ j=1、2、3

Usually, ground unrest is the stationary noise that average is zero, and uncorrelated mutually with signal, different microphones place ground unrest is also uncorrelated mutually, s _rjand s (n) _ij(n) cross correlation function R _rij(τ) be expressed as

j=1、2、3

The distance essence is estimated to adopt the distance-finding method based on time delay, with the time delay of better simply cross-correlation method estimated signal; The τ possible to each calculates R _rij(τ), search maximum value position, obtain signal and arrive the delay time T of i microphone to be positioned (4) with respect to reference microphone (2) _rij,

${\mathrm{\&tau;}}_{\mathrm{rij}}={\mathrm{\&tau;}}_{\mathrm{rj}}-{\mathrm{\&tau;}}_{\mathrm{ij}}=\arg \underset{\mathrm{\&tau;}}{\max }{R}_{\mathrm{rij}}\left(\mathrm{\&tau;}\right),$ j=1、2、3

Because distance rough estimate module has estimated rough apart from d between each sound source of microphone to be positioned (4) _{ij, curr}, near now only need thick distance, in certain limit, search for microphone meticulous distance; The definition velocity of sound is c, roughly apart from d _{ij, curr}corresponding thick time delay is t _{j, curr}=d _{ij, curr}/ c, at d _{ij, curr}near Δ d _{ij, curr}in scope, the meticulous distance of search, be converted to corresponding temporal information Δ t by range information _{j, curr}=Δ d _{ij, curr}/ c, the hunting zone of meticulous time delay is [t _{j, curr}-Δ t _{j, curr}, t _{j, curr}+ Δ t _{j, curr}];

Signal arrival time difference is multiplied by velocity of sound c, obtains the range difference that signal arrives two microphones

d _rij＝(τ _ij-τ _rj)·c＝-τ _rij·c，j＝1、2、3

According to reference microphone (2) to j sound source apart from d _rj, and j sound source to reference microphone (2) the range difference d with i microphone to be positioned (4) _rij, obtain meticulous apart from d to j sound source of microphone to be positioned (4) _ij

d _ij＝d _rj+d _rij＝d _rj-τ _rij·c

The course of work of described triangle centroid calculation submodule is as follows:

Definition (x _s1, y _s1), (x _s2, y _s2) and (x _s3, y _s3) be respectively three sound source coordinates, d _i1, d _i2, d _i3be respectively the meticulous distance of i microphone to be positioned (4) to three sound sources; When judging present frame as signal frame, obtaining meticulous apart from d to j sound source of above-mentioned i microphone to be positioned (4) _ijafterwards, j=1,2,3, according to geometric relationship, take sound source position as the center of circle, apart from the intersection point that is three circles of radius by solving, and determines i microphone to be positioned position,

$\left\{\begin{array}{c}{(x-x_{s1})}^2+{(y-y_{s1})}^2=d_{i1}^2\\ {(x-x_{s2})}^2+{(y-y_{s2})}^2=d_{i2}^2\\ {(x-x_{s3})}^2+{(y-y_{s3})}^2=d_{i3}^2\end{array}\right.$

Only have in the ideal case, three circles m that just can intersect at a point, coordinate is (x _m, y _m); And in the ordinary course of things, due to the existence of the error of calculation, this system of equations, without solution, is divided into two kinds of situations without solution: the first is that three circles can not meet at a bit, but surrenders mutually a zone; The second is wherein to have at least two circles there is no intersection point;

For the first, without the solution situation, select at a distance of three nearest intersection point p ₁(x _p1, y _p1), p ₂(x _p2, y _p2) and p ₃(x _p3, y _p3) form triangle, ask for this triangle barycenter m, coordinate is (x _m, y _m), as sound source position,

$(x_m,y_m)=\frac{1}{3}\underset{i=1}{\overset{3}{\mathrm{\&Sigma;}}}(x_{\mathrm{ni}},y_{\mathrm{ni}})$

For the second, without the solution situation, consider when estimating microphone position, each microphone to be positioned (4) and correction sound source position are all fixed, and two round intersection points of this calculating by former frame serve as this frame intersection point.

2. microphone position estimating system in a kind of distributed mike wind array according to claim 1, it is characterized in that: the optimum value of described a is 4; The optimum value of b is 0.05; The optimum value of β is 0.05.

Images