JP2005265466A - Beam forming device based on minimax discipline, and its beam forming method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem wherein a side lobe sometimes becomes large locally hitherto, when a signal from a sound source is received by a sensor such as a microphone and beam forming is performed so as to acquire an output signal having excellent directivity, and it seems that the sound source exists in the nonexistent direction, when performing direction survey of the sound source by scanning in the beam direction. <P>SOLUTION: This beam forming device has a constitution wherein a plurality of microphones 2 are disposed in an optional space, and a sound X from the sound source 1 is received by the plurality of microphones 2, and operation processing is performed so that the side lobe of the output signal Y in its directivity characteristic becomes minimum in a minimax discipline, and updating processing of a filter coefficient of a linear filter 3 is performed as the operation processing result. Thus, the side lobe does not become large locally, and the magnitude of the side lobe can be suppressed small. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a device for beam forming a sensor such as a microphone or an antenna into a directional device, and in particular, a plurality of input sensors are arranged in an arbitrary space, and signals of the plurality of input sensors are processed. It is related with the beam forming apparatus which improves directivity characteristics by doing.

Conventionally, as described in JP-A-2002-48867, input sensors are arranged at equal intervals on a lattice or a circle, and the directivity of a microphone, an antenna, or the like is calculated by delaying and summing the input signals. A beam forming apparatus for improving the above is known.
Further, beam forming is realized by linear filter processing and synthesis processing from input sensor signals arranged on a spherical surface or any space including those arranged at unequal intervals, as described in JP 2000-162308 A The method of doing is already known.

JP 2002-48867 A JP 2000-162308 A

  In the conventional beam forming by the delay and sum calculation processing in the prior art, there is a problem that a sharp directivity cannot be obtained, and in the beam forming performed by the linear filter processing and the synthesis processing, there is a sharp directivity. Although obtained, the filter design method that can reduce the side lobe is based on the least-squares criterion of the side lobe, and the side lobe may be locally increased. When exploring, there is a problem that the output source appears to be in a direction that does not exist.

The gist of the invention according to claim 1 of the present invention is that a plurality of input sensors are arranged in an arbitrary space, and a beam forming apparatus that forms directional characteristics by processing input signals from the plurality of input sensors. An arithmetic device that performs arithmetic processing based on the input signals from the plurality of input sensors so that a side lobe in the directivity characteristic is minimized according to a minimax standard, and the plurality of input sensors based on a calculation result of the arithmetic device The beam forming apparatus includes means for updating a filter coefficient of a filter provided for each filter, and the side lobe is minimized by the minimax standard as much as possible by updating the filter coefficient.
The gist of the invention according to claim 2 of the present invention is that a beam forming apparatus that realizes beam forming by a linear filter is formed by setting an arbitrary filter whose main lobe gain in the target direction is not 0 as an initial value. One or a plurality of processing units for calculating the gain of the side lobes in the non-target direction of the directional characteristics to be low in one or more directions, and updating the filter coefficient of the filter based on the calculation result of the processing unit The present invention resides in a beam forming apparatus including means for performing rotation.
The gist of the invention according to claim 3 of the present invention is a method of forming a directional characteristic by arranging a plurality of input sensors in an arbitrary space and performing arithmetic processing on input signals from the plurality of input sensors. The beam forming method is characterized in that the filter coefficient is updated so that the side lobe in the directivity is minimized according to the minimax standard.
The gist of the invention according to claim 4 of the present invention is that the beam forming is realized by a linear filter, and an arbitrary filter whose main lobe gain in the target direction is not 0 is set as an initial value. The beam forming method is characterized in that the process of updating to a filter in which the sidelobe gain in the non-target direction of the directional characteristic is reduced in one or more directions is performed once or a plurality of times.

  The present invention realizes beam forming having a directional characteristic in which a side lobe is minimized by the minimax standard by arranging a plurality of input sensors in an arbitrary space and processing the signals of the plurality of input sensors. Since the side lobe value does not increase locally, the direction of the output sound source (or radio wave source, light source, etc.) is used when searching the direction of the output source by scanning in the beam direction. It will be possible to correctly explore.

  The present invention realizes beam forming having a directional characteristic in which a side lobe is minimized by the minimax standard by arranging a plurality of input sensors in an arbitrary space and processing the signals of the plurality of input sensors. It is possible to do that.

Hereinafter, the beam forming apparatus and the beam forming method for minimizing the side lobe based on the minimax standard of the present invention will be specifically described with reference to the drawings of the embodiments.
FIGS. 1 to 3 are diagrams for explaining a directivity processing system when microphones are used as a plurality of input sensors according to an embodiment of a beam forming apparatus using a linear filter.

In FIG. 1, 1 is a sound source, 2 is a microphone, 3 is a linear filter having a transfer characteristic G _n (ω), and 4 is an adder.
FIG. 2 is a flowchart for explaining an algorithm for designing a filter that minimizes the side lobe according to the minimax standard of the present invention.

  FIG. 3 is a diagram for explaining the two-dimensional directivity characteristic of the filter having the minimum side lobe according to the minimax standard, and the magnitude of the gain is represented by brightness. (A) is a characteristic in which a beam is formed by a conventional calculation of delay and sum, (b) is a characteristic in which a beam is formed by a linear filter that minimizes a side lobe according to the conventional least square criterion, and (c) is a characteristic of the present invention. It shows the characteristic that the beam is formed by the linear filter that minimizes the side lobe according to the minimax standard. The bright part in the center circle is due to the main lobe, and the brighter side lobe is imaged around it.

That is, as shown in FIG. 1, the characteristic of the sound source 1 at the distance r, the azimuth angle θ, and the elevation angle φ is X (ω), and the sound source 1 to the microphone 2 (N microphones from Mic.1 to Mic.N are arranged. _{Hr, θ, φ, 1} (ω) to H _{r, θ, φ, n} (ω) (where n = 1, 2, ..., N), linear filter 3 is G ₁ (ω) to G _n (ω), the ratio between the characteristic X (ω) of the sound source 1 and the output signal Y (ω) that is added by the adder 4 and output. The directivity characteristic D (r, θ, φ) including the distance direction of this processing system is

It becomes.
Further, the discretized direction in the target range is represented by a subscript m (m = 1, 2,..., M). For this discretized transfer system, a directional characteristic vector d, a transfer function matrix H, Consider a filter vector g (T is a transpose matrix),

given that,

It can be expressed as a simple expression.
Here, the linear filter 3 having the transfer characteristic G _n (ω) is specifically constituted by a digital filter or an analog filter. When the linear filter 3 is configured by a digital filter, the filter coefficient is obtained by inverse discrete Fourier transform of the transfer characteristic G _n (ω). Further, when the linear filter 3 is configured by an analog circuit, each fractional polynomial is converted into an analog by LCR if it is approximated by a fractional polynomial of jω and then decomposed into a fractional polynomial of second order or less using partial fraction expansion or the like. It can be realized by a filter, and a transfer characteristic G _n (ω) can be realized as a combination thereof.

Next, the filter g k having a gain (product of the transfer function matrix H and the filter vector g) of 1 in the target direction _k and a peak in the vicinity of the target direction coincides with the target direction, , Φ each component has an extreme value, drh _k , dθh _k , dφh _k are respectively _expressed as vectors having differential values of H _{r, θ, φ n} (ω) at r, θ, φ as elements.

It can be expressed as a solution of
This solution is generally innumerable with N> 4, and uses a matrix Z composed of independent vectors spanning a zero space of solutions g to _k satisfying an arbitrary expression [3] and an arbitrary vector y. The

G ~ _k are, for example, using the minimum norm solution and + as a pseudo inverse matrix,

Z is the singular value decomposition of H _k

Is obtained as a column vector of V corresponding to singular value 0.
As a result, by adjusting the arbitrary vector y, it is possible to adjust the gain in the non-target direction (that is, the gain of the side lobe) while maintaining the gain and peak in the target direction. U, transfer function matrix U in non-target direction

If

It becomes.
The already proposed filter that minimizes the sum of squares of the sidelobe size (filter based on the least-squares criterion) can be designed using y ~ obtained by the least-squares solution of Equation = 0 in [9] And y ~

By substituting y˜ into the equation [5], a filter that minimizes the sum of squares of the side lobe sizes can be obtained.

A filter that minimizes the maximum value of the side lobe (a filter based on the minimax criterion) obtains y that minimizes the maximum absolute value u _max of each element of the gain u in the non-target direction in the side lobe direction region. However, since u _max is a non-linear function with respect to y, an optimal solution cannot be obtained by linear inverse operation. In the present invention, iterative processing is performed according to the algorithm shown in FIG. In order to achieve this, u _max is successively reduced.

  In FIG. 2, since there are usually a plurality of side lobes, first, pay attention to the direction of the side lobe with the maximum gain among the plurality of side lobes, and use this as the direction of interest mi so that the gain in the direction of interest mi decreases. The solution is updated (one round of steps S1 to S6 in FIG. 2).

Steps S1 to S6 will be described. First, in step S1, a direction m in which the side lobe is maximum is searched. In step S2, the direction m searched in step S1 is added to the direction of interest mi. In step S3, y is calculated so that the gains in the direction of interest mi are all zero. In step S4, it is checked whether or not all the gains in the direction of interest mi have been reduced. If all the gains in the direction of interest mi have decreased, the process proceeds to step S5, where g is calculated by replacing W of g = g _old + WZy (W: 0 to 1), and W _opt that minimizes the maximum value of the side lobe. Search for. In step S6, the solution is updated with g = g _old + W _opt Zy.

W _opt in step S6, when completely 0 the gain of interest direction mi, to gain other direction m is increased, the gain of the side lobes of the other direction m is decreasing directions mi Side This is a parameter that is adjusted to update the solution to a point that does not become larger than the lobe gain.

Thereafter, steps S1 to S6 in FIG. 2 are repeated.
In this way, when the gain of the side lobe in the direction of interest mi and the gain of the next largest side lobe are substantially the same, the direction m of the next largest side lobe is added to the direction of interest mi and the focus includes a plurality of directions. Similarly, the solution is repeatedly updated so that the gain in the direction of interest mi decreases in the direction mi. In this way, side lobes in the direction of interest mi are reduced and other directions m are sequentially added to the direction of interest, so that the gain of the side lobe as a whole can be reduced.

  In step S4, it is checked whether all the gains of mi have been reduced by updating the solution. If not, it is checked whether all the gains in the direction of interest mi are the same (step S7). If all the gains in the direction of interest mi are the same, the side lobe gain is minimized, and the process ends. In step S7, when all the gains in the direction of interest mi are not the same, it is considered that the direction of interest mi was not appropriate, so the direction of interest mi is reviewed (step S8).

  Next, the process of FIG. 2 will be described more specifically. First, the side lobe peak is set to 0, and as a result, the side lobe becomes higher at other positions, so that the gain at the position where the new peak has grown and the position of the original peak are almost at the same level. As a result, a filter is obtained in which the peaks of the two side lobes are approximately at the same level and lower than the original peak.

  The above is a specific explanation of the first update, and the second time the filter is updated so that the gain becomes zero at the position of the two side lobes that are peaked. Since the side lobe becomes higher at another position, the gain is updated until the gain of the new peak position and the original two peak positions are almost the same level. As a result, the levels of the three side lobes are almost equal. At the same level, a filter having a lower value than the original two peaks is obtained.

  Similarly, by repeating the above processing, a filter can be obtained in which the sidelobe peaks are lowered at substantially the same level at a plurality of positions. However, the filter can be updated to a peak at the same time. Is a number obtained by subtracting 4 from the number of microphones (in the case of a three-dimensional filter), and by repeating a finite number of times, a filter in which the side lobe peak cannot be reduced no matter how the coefficient is updated is obtained. This is the principle by which a filter with a minimum maximum value can be obtained.

  By performing the above iterative process until the maximum value of the side lobe is not reduced, a filter that minimizes the maximum value of the side lobe can be designed, but in this case, the main lobe has a sum of squares of the size of the side lobe. Compared to the smallest filter, it spreads out. Therefore, using a filter with a narrow main lobe as an initial value, considering the balance between the value of the maximum value of the side lobe and the value of the sum of squares of the size of the main lobe, the sum of squares of the side lobe size is predetermined. It is also possible to design a filter in which the main lobe and the side lobe are balanced by stopping the processing repeatedly when it becomes the reference.

  In the above description, the algorithm for designing the filter that minimizes the side lobe according to the minimax rule of FIG. 2 is executed by the microcomputer provided in the beam forming apparatus according to the present invention. This microcomputer executes a filter design algorithm that minimizes the side lobe according to the minimax standard, and updates the filter coefficient of the filter with the calculation result.

In addition, the filter is specifically configured by a digital filter or an analog filter as the linear filter 3, and the linear filter 3 is updated with the calculation result according to the present invention, so that the linear filter 3 has a transfer characteristic G _1. (ω) to G _n (ω) are filters capable of minimizing the side lobe according to the minimax standard, and thereby a beam forming apparatus capable of suppressing the gain of the entire side lobe to be small can be specifically realized.

FIG. 3 shows an example in which two-dimensional beam forming is realized by arranging 31 microphones in a spherical microphone array having a radius of 130 [mm]. The transfer function H _{r, θ, φ n} (ω) is calculated from the theoretical formula of the sphere and calculated as the distance r = 1 [m], and the target direction (θ, φ) = (0,0), the sound source frequency Is calculated as 1000 [Hz]. (A) is a conventional delay-sum operation, (b) is a conventional operation using a linear filter that minimizes the sum of squares of the side lobe size, and (c) is a minimum of the side lobe according to the present invention in the minimax standard. The two-dimensional directivity characteristics in the case where beam forming is realized by calculation using a linear filter is shown.

  The horizontal axis represents the azimuth angle, the vertical axis represents the elevation angle, and the color represents the directivity gain. In the delay-sum operation of (a), the main lobe is wide, and in the linear filter operation (b) where the sum of squares of the side lobe size is the minimum, the side lobe is observed as an imaginary sound source although the main lobe is narrow and the directivity is small. Has been. On the other hand, in the linear filter operation in which the maximum value of the side lobe of (c) according to the embodiment of the present invention is minimized, the main lobe is wider than (b), but is narrower than (a). The imaginary sound source seen in a) and (b) is not observed, and it is clear that the beam forming for the purpose of sound source exploration is effective.

  In the above-described embodiment, the microphone for the sound source is described as an example of the input sensor. However, the present invention does not include these side lobes even in the case of the antenna for the radio wave or the light receiving sensor for the light source. The present invention can be similarly applied as a beam forming device for reduction.

  The present invention can be applied as a beam forming apparatus for a microphone, an antenna, or a light receiving sensor having a good directivity characteristic with a suppressed side lobe size.

FIG. 1 is a diagram showing an embodiment of the present invention. FIG. 2 is a diagram for explaining an algorithm for designing a filter in the embodiment of the present invention. FIG. 3 is a diagram comparing the characteristics of the embodiment of the present invention and the prior art.

Explanation of symbols

1 ... Sound source
2 ... Microphone
3 Linear filter
4 ... Adder

Claims (4)

In a beam forming apparatus in which a plurality of input sensors are arranged in an arbitrary space, and directivity characteristics are formed by computing input signals from the plurality of input sensors.
An arithmetic device that performs arithmetic processing based on the input signals from the plurality of input sensors so that the side lobe in the directivity characteristic is minimized by the minimax standard;
Means for updating a filter coefficient of a filter provided for each of the plurality of input sensors based on a calculation result of the calculation device;
The beam forming apparatus characterized in that the side lobe is minimized by the minimax standard as much as possible by updating the filter coefficient. In a beam forming apparatus that realizes beam forming by a linear filter,
An arithmetic unit for calculating an arbitrary filter whose target lobe main lobe gain is not 0 as an initial value so that the gain of the side lobe in the non-target direction of the directional characteristic formed thereby becomes lower in one or more directions; ,
A beam forming apparatus comprising: means for performing a process of updating a filter coefficient of a filter once or a plurality of times based on a calculation result of the calculation apparatus.   A method of forming a directional characteristic by arranging a plurality of input sensors in an arbitrary space and processing input signals from the plurality of input sensors, wherein the side lobe in the directional characteristic is a minimax standard. A beam forming method, wherein the filter coefficient is updated so as to be minimized. The beam forming is realized by a linear filter, and an arbitrary filter whose main lobe gain in the target direction is not zero is set as an initial value, and the gain of the side lobe in the non-target direction of the directional characteristic formed thereby is one. A beam forming method, wherein the process of updating to a filter that decreases in the above direction is performed once or a plurality of times.

Images