JP3131202B2 - Apparatus and method for separating and estimating transfer characteristics of multiple linear transfer paths and recording medium storing the method as a readable program - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to an N-input M-output linear FIR system (N = 2, 3,...; M = 1, 2,...)
The present invention relates to an apparatus and method for separating and estimating a transfer characteristic of a multi-linear transfer path for simultaneously estimating a number of signal transfer paths, and a recording medium storing the method as a readable program. Such systems include structures with multiple sensors and multiple actuators, multi-speakers,
There is a multi-microphone system and the like.

[0002]

2. Description of the Related Art With the progress of digital technology in recent years and the speeding up of arithmetic processing, sound signal processing such as sound pressure control and noise suppression, which was conventionally intended for one-input one-output systems, has been expanded to multi-input multi-output systems. It is getting. In such signal processing, a signal passed through a control filter is input to a multiple-input multiple-output system. Since the coefficients of the control filter are calculated from the characteristics of the target multi-input multi-output system, it is necessary to accurately obtain the characteristics.

A home theater is considered as an application example of such an acoustic signal processing. The sound of conventional video has been two-channel stereo reproduction, but it has been developed into a home theater by multi-channeling using four or six speakers. In order to provide sound closer to that of a movie theater, it is necessary to know the multiple sound transfer characteristics in order to correct the control filter for sound signal processing in accordance with the sound transfer characteristics of the room where the sound is actually heard.

Conventionally, an N-input M-output linear FIR system is decomposed into N 1-input M-output subsystems, and each transfer function of each subsystem is estimated by calculating a correlation between an input signal and an M output signal. Thus, the transmission characteristics of the NM signal transmission paths have been determined. In this method, transfer functions are sequentially determined for N subsystems. As such an example, Japanese Patent Application Publication No. 6-131003 shows a method for sequentially estimating the transfer function of a multiple-input multiple-output system that models the operating characteristics of a chemical plant and reducing the order of the identification model. I have. In this method, each time a signal u is supplied to each input terminal, response signals from a plurality of output terminals are obtained, and signals are not simultaneously supplied to a plurality of input terminals. Therefore, since simultaneous measurement cannot be performed, it is necessary to repeatedly measure response signals of a plurality of output terminals with respect to an input signal for each of all input terminals.

In order to avoid this, a method is proposed in US Pat. No. 5,661,813 in which an input signal is given uncorrelated fluctuations or N pseudo noises which are uncorrelated with each other are input and all transfer functions are simultaneously obtained. Have been. When estimating the transfer characteristics of an N-input, M-output linear FIR system using this method, it is necessary to confirm that the input signal is uncorrelated for all N × (N-1) / 2 combinations. Take it.
In addition, when a set having a large correlation is found, if it is corrected by giving a different variation to the input signal, it is necessary to check the correlation again for the other sets.

When an N-input M-output linear FIR system is driven by the same signal or a highly correlated signal, the prior art guarantees that the transfer characteristics of the NM multiplex transmission paths are reliably separated and estimated. Not. An example of such an example is a multi-input echo canceller device of a multi-channel communication conference system. In a multi-channel communication conference system, a single speaker's voice picked up by a plurality of microphones from a remote location is transmitted as a multi-channel signal, and the receiving side receives those signals, and a transmission multi-microphone is provided. Playback from the multi-speaker in the specified acoustic space. Since there is a large cross-correlation between multi-channel signals generated from the same speaker, the transfer function between the multi-speaker and multi-microphone at the receiving side is completely separated and estimated even if residual echo is canceled. Not necessarily.

[0007]

SUMMARY OF THE INVENTION It is an object of the present invention to separate and estimate the transfer characteristic of a multiple linear transmission path even if the simultaneous input signals to the multiple linear transmission path are signals having correlation with each other. It is an object of the present invention to provide an apparatus and method for separating and estimating a transmission path which does not need to check the correlation between input signals, and a recording medium which records the method as a readable program.

[0008]

According to the present invention, N input points of a linear FIR system, N is an integer of 2 or more, and M output points, and M is an integer of 1 or more, NM is defined between N and M. In the transfer characteristic measurement in which the transfer characteristics of the transfer paths are simultaneously separated and measured, the input N-channel signal is processed by N pre-filters in which all zeros have transfer characteristics different from each other, and the N-channel signal is processed before the N-channel. Generating a processed signal, and inputting the N-channel pre-processed signal to the N input points of the linear FIR system by N actuators, respectively;
At the M output point of the linear FIR system, the linear FI
A response signal from the R system is detected by M sensors, and the N · M transfer characteristics are estimated based on the N-channel pre-processed signal and the response signal from the M output point.

Thus, it is possible to simultaneously separate and estimate the transfer characteristics of NM multiplexed transfer paths from various types of input signals.

[0010]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Principle of the Invention In the present invention, an N-input M-output linear FIR (Finite Impulse) is used.
(e Response) A configuration is adopted in which N pre-filters having mutually different zeros are inserted before each input point of the system. Thereby, the N · M can be obtained from various input signals.
Simultaneous estimation of multiple multiplex transmission paths is possible as follows. The transfer characteristics of a linear FIR system can be represented by a z-polynomial, which is also referred to herein as a transfer function.

The N-input M-output linear FIR system has M sets of
Since it can be treated as an N input 1 output system,
Here, referring to FIG. 1, an N-input one-output linear FIR system
An example of the transfer characteristic measuring device in the case of (1) will be described. Unknown N input
1-output linear FIR system 11 _m Is transmitted by z-transform
Sex is H₁(z), ..., H_N(z) N transmission paths 11
H_{1 m}, ..., 11H_NmAnd the sum of the outputs of these transmission paths is
Adder 11A_mIt is represented by N input points 11S₁, ..., 1
1S_NIs an unknown system 11_m Represents N input means for
And adder 11A_mIs unknown system 11_m Output from
It constitutes an output means for taking out. Input point 11S₁, ..., 11S
_NAnd adder 11A_mIs, for example, system 11_m Is an acoustic hall
(Indoor sound field) is a speaker and a microphone,
In general, any unknown system can be activated
It can be called a heater and a sensor.

X ₁ (z),..., X _N (z) are N-channel input signals x
₁ (k), ..., a z-transform of _{x N (k), Y (} z) is the z transform of the output signal y (k). _{G 1 (z), ...,} G N (z) is prefilter 12 ₁ before being introduced by the present invention, ..., a transfer characteristic of 12 _N, H
₁ (z), ..., H _N (z) are the input points of the unknown system 11 _m
11S _1, ..., is the transfer characteristic of the transfer path to the adder 11A _m is the output point from the 11S _N. The input / output relationship of this system is z
The conversion is given by the following equation: Y (z) = X (z) zG ₁ (z) H ₁ (z) +... + G _N (z) H _N (z)｝ (1) Then, (a): the order of G ₁ (z), ..., G _N (z) is H ₁ (z), ..., H
_{If the order of N} (z) is greater than N-1 times and (b): G ₁ (z), ..., G _N (z) all zeros are different from each other, the above equation (1) is satisfied. There is only one set of H ₁ (z), ..., H _N (z). This means that if a pre-processed signal, which is an output signal of a pre-filter, is input, an N transmission path of the N-input one-output linear FIR system can be uniquely determined. The zero points of the filters G ₁ (z), ..., G _N (z) are different if the order of each of these filters is P and the transfer characteristic is

[0013]

(Equation 13) Means that the values of _anp for all pairs of (n, p) are different from each other. In other words, these transfer characteristics G ₁ (z),..., G _N (z) mean that they are relatively prime.

According to the present invention, an N-channel input signal
Even if mutually correlated signals are used as X ₁ (z), ..., X _N (z), by designing the pre-filters 12 ₁ , ..., 12 _N to have different zeros from each other, , The transfer characteristic H of the transfer path from each input point 11S _n to M output points 11A _m (m = 1,..., M)
_n (z) can be uniquely determined. Transfer characteristic estimator 19 _m
Is a pre-processed signal U which is an input signal to the input points 11S ₁ , ..., 11S _N
1 (z) = X ₁ (z) G ₁ (z), ..., _UN (z) = X _N (z) G _N (z) and output signal Y (z)
From the transmission characteristic (or impulse response) of each transmission path. As a method for estimating the transfer characteristic, various methods conventionally used can be used. Regarding those methods, for example, U.S. Patent Nos. 5,272,695 and 5,40
It is introduced in 8.530.

The input signals X ₁ (z),..., X _N correlated with each other
An example of a typical model of the system that generates (z) is given below.
One. The first example is the correlation signal generation model shown in FIG. 2, in which, for example, one speaker 18S and a plurality of (N) microphones are arranged in a common sound field (for example, an acoustic hall) 18 and reproduced from the speaker 18S. A plurality of (N) microphones with one speaker's voice signal V (z) from the generated sound source 17
18A _1, ..., are collected by 18A _N, signals X ₁ to outputs of the microphones have a correlation with each other (z), ..., it is generated as X _N (z). An audio signal V (z) from the same sound source 17 is a speaker
Signals correlated with each other from 18S through acoustic paths 18 ₁ , ..., 18 _N whose transfer characteristics are represented by F ₁ (z), ..., F _N (z) respectively.
X ₁ (z), ..., are outputted as X _N (z), the prefilter 12 ₁ of the system of FIG. 1, ..., given to 12 _N.

In the case of FIG. 2, the correlated signals X ₁ (z),.
The generation of X _N (z) is modeled as a processing result F _n (z) V (z) by a one-input N-output linear filter F _n (z) (1 ≦ n ≦ N) of a single signal source V (z). Can be When this is combined with the system of FIG. 1, the input / output relationship is expressed by the following equation using z-transformation: Y (z) = V (z) ｛G ₁ (z) F ₁ (z) H ₁ (z) +. _N (z) F _N (z) H _N (z)｝ (3), and the following condition degG _n (z) F _n (z)> (N-1) degH ₁ (z), (N− 1) degH ₂ (z), ..., (N-1) degH _N (z) (4a) GCD ｛G ₁ (z) F ₁ (z), G ₂ (z) F ₂ (z) ,. .., G _N (z) F _N (z)｝ = 1 (4b), the M transmission path of the N-input / one-output linear FIR system 11 _m is unique as in the case of the same signal input in FIG. Is decided. Here, degF (z) represents the order of the polynomial F (z) of z. GCD (G ₁ (z), G ₂ (z)) is a polynomial G of z
Represents the greatest common polynomial of ₁ (z) and G ₂ (z).

The correlation signal generation model shown in FIG. 3, which is a second example, has a plurality of (J) speakers 18 in a common sound field 18.
S ₁ , ..., 18S The voices (uncorrelated with each other) of a plurality of (J persons) reproduced from _{J are} reproduced by a plurality of (N) microphones 18.
A ₁ ,..., 18A _N are picked up, and the outputs of those microphones are used as correlated signals X ₁ (z),..., X _N (z). In this case, the sound source 17 ₁ of J's, ..., all the audio signal V from the 17 _J
1 (z), ..., V _J (z) are input to each of the N microphones, so that J sound sources and N microphones 18A ₁ , ..., 18A
Between _N , JN transmission paths are defined. In FIG. 3, the transfer characteristics of these JN transfer paths are represented by F ₁₁ (z), F ₁₂ (z),..., F _1N (z), F ₂₁ (z), F ₂₂ (z),. .., F _2N (z) ‥‥‥‥‥‥‥‥‥‥‥ F _J-11 (z), F _J-12 (z), ..., F _J-1N (z), F _J1 ( z), F _J2 (z), ..., F _JN (z).

By treating the J inputs and N outputs as N sets of J inputs and 1 outputs, this correlation signal generation model is converted into signals V ₁ (z),... From _J sound sources 17 ₁ ,. J, 1-output linear filter F _jn (z) given., V _J (z) (1 ≦ j ≦ J, 1 ≦ n ≦ N)
The following equation, which is the result of processing N sets by

[0019]

[Equation 14] Can be modeled as In the case of the model of FIG. 3 in which highly correlated input signals are generated from a plurality of sound sources, the high-frequency signals generated from J sufficiently wideband and independent signals V _j (z) (j = 1,. The correlation signal X _n (z) (n = 1,..., N) is modeled as in equation (5). Therefore, if the N signals X _n (z) (n = 1,..., N), which are the z-transforms of the highly correlated signals, are described by a horizontal vector, the highly correlated signals X _n (z) and J The relationship with the sound source signal V _j (z) is expressed by the following equation.

[0020]

(Equation 15) Therefore, the model of FIG. 3 is combined with the system of FIG. 1 to provide a microphone (i.e., adder) 11A _m which is an output terminal of the N-input M-output linear FIR system 11 from the _J sound sources 17 ₁ ,.
The relationship up to the output signal is expressed by equation (7) shown in FIG. 4 using a matrix composed of transfer functions.

From the input / output relationship of the signal, the actual transfer characteristic H
_{In order for 1} (z), ..., H _N (z) to be accurately estimated, there exists a set of H ₁ (z), ..., H _N (z) that satisfies the above equation, and There must be only one. H ₁ (z), ..., H _N that satisfies Equation (7)
If there are multiple pairs of (z), the transfer characteristics H ₁ (z), ..., H _N (z) obtained from the relationship between the input and output signals do not always match the true transfer characteristics. Disappears. Therefore, conditions for obtaining the transfer characteristics H ₁ (z),..., H _N (z) from equation (7) must be considered.

Where:

[0023]

(Equation 16) Defines virtual transfer functions D ₁ (z),..., D _J (z). Rewriting equation (7) in Fig. 4 using the above equation gives the following equation

[0024]

[Equation 17] Becomes Assume that the J multiple signals that produce highly correlated signals are sufficiently wideband and independent. At this time, the digital signal processing theory guarantees that only D ₁ (z),..., D _J (z) can be obtained. Further, equation (8) is

[0025]

(Equation 18) Can be rewritten as The set of J corresponding elements of the vertical vector on both the left and right sides represents J equations. Each of these equations has the same form as equation (3). Therefore, similar to the conditions of equations (4a) and (4b), the following condition degG _n (z) F _jn (z)> (N-1) degH ₁ (z), ..., (N-1) degH _N (z) (11a) GCD ｛G ₁ (z) F _j1 (z), ..., G _N (z) F _jN (z)｝ = 1 (11b) where n = 1, ..., If N and j = 1,..., J hold, even if a highly correlated signal generated from a plurality of signal sources is input, the N input 1 output linear It is guaranteed that N transmission paths of the FIR system can be uniquely determined (Emura, Miyoshi, "Separation Estimation Method for Multiple Linear Paths", Advanced Technique EA98-62,1998-09, p
See p.25-32, Appendix).

The relationship between input and output signals in the combination of the highly correlated signal generation system based on the model of FIG. 3 and the N-input M-output linear FIR system whose transfer characteristics are unknown in FIG. 1 is made to correspond to equation (7) shown in FIG. It can be seen that the outputs of the partial terms V, F, G, H on the left side of equation (7) correspond to the respective signals as follows. That is, the term V of the input signal vector is J input signals V ₁ (z),.
_J (z), and the product VF of this and the transfer characteristic matrix F is the high correlation signal X ₁ (z),..., X _N (z) collected and synthesized by each microphone in the high correlation signal generation system. , And the product VFG of this and the term G of the transfer characteristic matrix of the prefilter is given by the prefilters 12 ₁ ,..., 12 provided according to the invention.
The output of _N, the preprocessing signal U applied to the unknown system
₁ (z),..., U _N (z), and the product of this and the transfer characteristic matrix H of the multiple linear paths, that is, the right side of the equation (7) is the microphone 11
A ₁ , ..., 11A Response output of unknown system detected by _M
Represents Y (z).

From this correspondence, an equation is obtained from the J input signals V ₁ (z),..., V _J (z) from the model sound source and the microphone output signal Y (z).
D ₁ (z), ..., D _J (z) defined in (9) are uniquely determined. Further, only one set of G ₁ (z) H ₁ (z),..., G _N (z) H _N (z) is determined from the highly correlated N input signal and the microphone output signal Y (z). Therefore, from the input / output relationship between the N input signal after passing through the pre-filter and the microphone output signal Y (z), it is clear that the only set of H ₁ (z),..., H _N (z) is determined. . Actual H ₁ (z), ..., H _N
For the estimation of (z), a method using an N-input / one-output adaptive filter described in the following embodiments and other methods are possible.

Although the case where the transfer characteristics in the system of FIG. 1 are estimated using the model sound source of FIG. 3 has been described above, the case where the model sound source of FIG. 2 is used is also established. First Embodiment FIG. 5 is a block diagram showing an embodiment of an apparatus for separately estimating transfer characteristics of NM transfer paths of an N-input M-output linear FIR system according to the present invention. This embodiment shows a case where the transfer characteristic of an unknown linear FIR system 11 is estimated by simulating the transfer characteristic using an adaptive filter. N input signal
x ₁ (k), ..., x _N (k) is the value of _N provided in accordance with the present invention.
Prefilters 12 ₁ , ..., 12 with different zeros
After being processed by _N , it is input to an N-input M-output linear FIR system 11 having unknown transfer characteristics to be measured. Transfer characteristic estimating unit 19 of M N inputs and one output adaptive filter 13 _1, ..., and 13 _M, the M subtractors 10 _1, ..., and a 10 _M. The pre-processed signals u ₁ (k),..., U _N (k) output from the _N pre-filters 12 ₁ ,..., 12 _N are M N-input one-output adaptive filters 13 ₁ , ... is given as an input signal to the 13 _M, M number of signals y ₁ is their processing results '(k), ..., y M'
(k) is output as an estimated signal (referred to as a replica signal simulating the signals y ₁ (k),..., y _M (k)). M each subtractor 10 _m
(m = 1, ..., M) subtracts the corresponding replica signal y _m '(k) from the corresponding response output y _m (k) of the system 11, respectively, and subtracts the result e _m (k) of the subtraction. providing the adaptive filter 13 _m corresponding as an error signal. The M N-input 1-output adaptive filters 13 ₁ ,..., 13 _M provide NM transmission paths 11 H _nm (z) (1 ≦ n ≦ N, 1 ≦ m ≦ M) of the system 11. The transfer characteristics are simultaneously estimated separately. In the following description, when it is not necessary to distinguish 1 to N and 1 to M, these are not added.

Next, the adaptive filters 13 ₁ ,..., 13 _M will be described. Although the adaptive filter itself is publicly known, a description of the publicly known adaptive filter will be made with reference to FIG. 6 by taking an example of estimation of an N-input one-output linear FIR system. FIG. 6 shows a set of N-input and one-output systems in the N-input and M-output linear FIR system 11 shown in FIG. 5 as 11 _m, and a subtracter 10 _m and an adaptive filter constituting a transfer characteristic estimator 19 _m associated therewith. 13 _m is indicated. In the following description, a discretized signal is used. When the transfer characteristic between the input pre-processing signal u (k) and the output response signal y (k) is linear, there is a relationship of Y (z) = H (z) X (z) using z-transform. Then, the adaptive filter 13 _m calculates the linear transfer characteristic H (z) from both the signals u (k) and y (k).
Is estimated.

Now, _assuming that the time is represented by k, N input pre-processed signals u _n (k) (k = 1, 2,...) Are input to the linear FIR system 11 _m .
(n = 1,..., N), and as a result, a response signal y _m (k) is output from the system 11 _m . N linear FIR systems constituting the linear FIR system 11 _m, that is, transfer paths having transfer characteristics H _1m (z),..., H _Nm (z), respectively.
11H _1m , ..., 11H _Nm impulse response is h _nm (k) (n = 1, ...,
N). At this time, the relationship between the input signal u _n (k) and the response output y (k) is expressed by

[0031]

[Equation 19] It is represented by The tap length of the impulse response is L, the following equation ^{_{h nm T = [h nm (}} L-1), ..., h nm (0)] input preprocessing signal and an impulse response (13) u _n ^T (k _{) = [u n (k-} L + 1), ..., u n (k)] (n = 1, ..., when expressed by a vector as n) (14), the relationship of input and output Next formula

[0032]

(Equation 20) Is described by the convolution operation. At this time, N L-dimensional vectors, that is, the next N vectors constituting one N-input one-output adaptive filter 13 _m , w _nm ^T (k) = [w _nm (L−1),. , w _nm (0)] (n = 1, ..., N) (16), and the difference between the replica signal y _m '(k) at time t = k and the response output y _m (k) of the system And the error signal e (k) as

[0033]

(Equation 21) Defined by Using the error signal e (k) and the pre-processing signal u (k), the coefficient of the adaptive filter is updated at each time k. This update method is proposed several, equation ^{_{w nm T (k + 1)}} = w nm T (k) + αe (k) u n T (k) (n = 1 as an example, ..., N ) There is an update method like (18). Here, α is an adjustment parameter. Note that the present invention is not limited to this updating method, and another updating method may be used.

When the signal x _n (k) has a sufficiently wide band, a vector w composed of adaptive filter coefficients is obtained after a sufficient time has passed.
_nm ^T (k) converges to a vector h _nm ^T consisting of the impulse response of the linear FIR system, ie, k → ∞, | h _nm ^T −w _nm ^T (k) | → [0, ..., 0] (n = 1, ..., N) (19) are known, and a vector w _nm ^T (k) consisting of adaptive filter coefficients is used as an estimate of a vector h _nm ^T consisting of an impulse response of a linear FIR system. be able to. That is, the transfer characteristic of the adaptive filter having the coefficient vector w _nm ^T (k) obtained at this time becomes equal to the transfer characteristic H _nm (z) of the N-input one-output linear FIR system to be measured.

FIG. 7 is a flowchart for explaining the operation of the embodiment of FIG. Hereinafter, the operation of the embodiment of FIG. 5 will be described with reference to the flowchart of FIG. In the following description, CONV [A, B] represents a convolution operation of the FIR filter A and the signal B. Input signal x _n (k) (n = 1, ..., N)
There is applied to the prefilter 12 _n, convolution of the input signal in the pre-filter 12 _n x _n (k) and pre-filter coefficient G _n (z)
u _n (k) = CONV [G _n (z), x _n (k)] is performed over a predetermined number of samples (step S1), and the calculation result is input to the N-input M-output linear FIR system 11 _m. (Step S
2), input to the N-input / one-output adaptive filter 13 _m (m = 1,..., M) (step S3).

In the transfer characteristic estimating section 19, N inputs and 1 output
Pre-processed signal u by adaptive filter 13m _n(k) and adaptive fill
Coefficient w_nmConvolution operation y of (z)_m'(k) = CONV [w_nm(z), u
_n(k)] and the replica signal y_mGet '(k). Subtractor 10
_m Is the system response signal y_m(k) and the replica signal y_m'(k)
Error e between_mIs

[0037]

(Equation 22) (Step S4). The power average _Perr of the error signal obtained in step S4 during a fixed time T is expressed by the following equation.

[0038]

(Equation 23) It is determined whether or not the value is _{equal to} or more than the predetermined expression threshold value _Eth (step S5). This error power average _Perr is equal to the threshold E _th
If more, the adaptive estimation of the filter 13 transmission characteristic H _nm by a factor of ^{_{m w nm T (z) (}} z) is determined not to be sufficiently converged, the adaptive filter 13 _m is the system input signal u _n and the error signal
Update the adaptive filter coefficient w _nm ^T (z) by the equation (18) from e _m (step S6), and the return to repeat the estimation process step S1.

[0039] determines that the error signal power average P _err is equal smaller than the threshold value E _th adaptive filter coefficient w _nm ^T (z) is sufficiently converged on the transfer characteristic H _nm (z) in step S5, w _nm ^T ( z)
The estimation result of H _nm (z) is used (step S7). FIG.
Steps S1 to S6 are executed in the same processing cycle, and the processing of steps S1 to S6 is repeated every time k is incremented. With the above processing, the filter coefficient described in step S7 in the flowchart of FIG. 7 is obtained as a processing result by the adaptive filter. Application Example 1 FIG. 8 shows, as a second embodiment of the present invention, _NM transmission paths H _nm (z) (1 ≦ n ≦ N, 1) of an N-input M-output acoustic system 11.
≤ m ≤ M). N input signals _{x 1 (k), ...,} x N (k) is, N pieces of common zeros without a prefilter 12 _1, ..., 12 after being processed by _N, N-number of the speaker 11S ₁ , ..., it is radiated into space sound field 11 by the 11S _N. N output from the _N pre-filters 12 ₁ , ..., 12 N
, U _N (k) are given as input signals to each of the _M N-input, one-output adaptive filters 13 ₁ ,..., 13 _M , and their pre-processing signals u ₁ (k),. output replica signal y _m 'and (k), the response signal y ₁ output from the microphone _{11A 1 ~11A M (k),} ..., y M
An error e _m (k) from (k) is calculated by the subtractor 10 _m , and the adaptive filter coefficient is adaptively updated so that the power average Perr of the error is minimized. The transfer characteristics of the NM transfer paths are simultaneously separated and estimated by the M N-input 1-output adaptive filters.

When measuring the sound transfer characteristics of a concert hall in a state where a spectator is present using such a sound system measurement method, the sound transfer characteristics between a plurality of musical instrument playing positions and a plurality of listening positions are simultaneously measured. Can be estimated.
According to this method, a signal having a sufficiently wide band and high correlation can be used as the estimation drive signals x ₁ (k),..., X _N (k).
It is possible to measure the acoustic characteristics of the concert hall in which the audience is located without using an unpleasant measurement-only signal such as a pseudo-noise signal that is uncorrelated with each other.

When the present invention is applied to a home theater having multi-speakers, by placing a measurement microphone near the ear of the listener, it is possible to simultaneously estimate sound transfer characteristics between a plurality of speakers and microphones from highly correlated real sound. is there. The sound transfer characteristics between the speaker and the microphone are affected by the reverberation characteristics of the listening room and the attitude of the listener.
This can be measured without using an unpleasant measurement-only signal such as the pseudo-noise signal. To implement the above-described method of estimating the transfer characteristic of the multiple linear transfer path, a computer such as an IC-ROM, a magnetic disk, a CD-ROM, and an MO disk may be used as a computer program capable of executing the processing procedure shown in FIG. By recording the program on a readable recording medium and executing the program using a computer, it is also possible to estimate a transfer characteristic of a target multiple linear transfer path. Second Embodiment The embodiment shown in FIG. 9 does not use the adaptive filters 13 ₁ ,..., 13 _M as in the first embodiment shown in FIGS. First-order simultaneous equations that obtain data of the response output of the system to application of a signal to the M-output linear FIR system over a predetermined number of time points (sample points) k and define transfer characteristics from the input / output signal data. Is generated, and a transfer characteristic is obtained by solving it. Prefilter 12 ₁ similar to the first embodiment, ..., 12 _N are provided, whereby the input signal x ₁ (k), ..., pretreatment obtained by processing the x _N (k) The signals u ₁ (k),..., U _N (k) are provided to an N-input M-output linear FIR system 11. The transfer characteristic estimating unit 19
It is composed of a multiplexed signal waveform storage section 14 and a multiplexed signal analysis section 15, and has M response outputs y from the system 11.
₁ (k),..., Y _M (k) are stored in the multiplexed signal waveform storage unit 14 for a predetermined number of time points of k. Based on these data, the multiplexed signal analyzer 15 generates the first-order simultaneous equations for obtaining the transfer characteristics as described below, and solves them to obtain the transfer characteristics of the system 11.

In FIG. 9, N input M output lines to be measured
The FIR system 11 has M sets of independent N input and 1 output lines.
It can be broken down into a FIR system. N inputs and mth output
Characteristics of N input 1 output linear FIR system consisting of
Is N impulse responses h _1m, ..., h_NmDescribed in
It is. Let the tap length of the impulse response be L,
In the channel, the preprocessing signal at the L point continuously from each point k
Number u_n(k), ..., u_nThe vector with (k + L-1) as the element at time k =
KL × L matrix arranged for 1, ..., KL (K is a positive number not less than N)
B_nIs

[0043]

(Equation 24) Defined by The relational expression between each input / output signal and transfer characteristics is
The following first-order system of equations for KL variables, each component y _m (k) of the impulse response

[0044]

(Equation 25)Given by Where the vector h_1m, ..., h_NmTo
In this case, the definition of Expression (13) is used as it is. Equation (23)
And B₁, ..., B_NIs the pre-processed signal u (k) (k =
1, ...). On the other hand, y_m(1), ..., y_m(KL) is cis
Measured as the response output signal of the system 11,
By solving equation (23), the impulse response h _1m, ...,
h_NmIs required. Z-transform each impulse response
The sound transfer characteristic H_nm(z) (n = 1, ..., N) is obtained.

Equation (23) is further modified to give the following equation

[0046]

(Equation 26) The correlation between the pre-processed signals u _n (k) is removed to remove the influence of noise to obtain impulse responses h _1m , ..., h _Nm , and transfer characteristics H _1m (z) ,. .., H _Nm (z) may be obtained.

By applying the above processing to each of the N-input one-output linear FIR systems decomposed into M sets, NM signal transmission paths can be estimated. If the input signal is sufficiently wide band, since the pre-processed signal generated by applying the pre-filters 12 ₁ , ..., 12 _N is applied to the measurement target,
It is guaranteed that the solution of equation (24) is uniquely found. FIG. 10 is a flowchart showing a procedure for obtaining the transfer characteristics in the second embodiment of FIG.

Step S1: As in the case of the first embodiment, the processing on the input signal x _n (k) by the pre-filter is performed by the convolution operation u between the transfer characteristic G _n (z) of the filter and the signal x _n (k).
_n (k) = CONV [ _Gn (z), _xn (k)], (n = 1, ..., N). Step S2: Obtained pre-processed signal u _n (k) (n = 1,..., N; k
= 1, ..., KL + L-1) with N-input M-output linear FIR system 11
And fetched into the multiplexed signal waveform storage unit 14.

Step S3: The response signal y _m (k) (m = 1,..., M; k = 1,..., KL) of the linear FIR system 11 is taken into the multiplexed signal waveform storage section 14. Step S4: The multiplex signal analysis section 15, the input signal u _n
KL × L matrix _Bn of equation (22) is calculated from (1),..., U _n (KL + L−1). Step S5: KL expressed by equation (23) from the obtained matrix B _n
The impulse responses h _1m ,..., H _Nm (m = 1,..., M) are obtained by solving the simultaneous linear equations.

In the measurements of the first and second embodiments according to the present invention described above, the measurement procedure is recorded in advance on a recording medium readable as a program to be executed by a computer. The program may be executed by executing the program read from the computer on a computer. It is clear that the second embodiment described above can be applied to the measurement of the multiplex transfer characteristic in the same acoustic system as that described in FIG. 8 as an application example, and the description thereof will be omitted.

The transfer characteristic measurement principle of the multiple linear FIR system according to the present invention described above is not limited to the measurement in the acoustic system shown as the application example described above, but any system that can be modeled as an N-input M-output linear FIR system 11. Is also applicable. In that case, N input M
The output linear FIR system includes a medium whose transfer characteristic is to be measured, an actuator for inputting signals to a plurality of points on the medium, and a response signal to the input signal at a plurality of output points different from the input point. It consists of three elements, a sensor for detection.

For example, for a flexible space structure such as an antenna of an artificial satellite or a solar cell panel, a large marine structure, or the like, the response to an excitation signal given to a plurality of points is measured at the plurality of points to obtain the structure. Is measured, and from the result, it is possible to estimate what kind of vibration distribution occurs when an impact is applied to the structure.

Specifically, in the solar cell 20 shown in FIG. 11, the medium corresponds to the solar cell panel 21, the actuator corresponds to the control motor 22, and the sensor corresponds to the vibration detector 23. In the members shown in FIG. 12, the medium corresponds to the member 30 through which the vibration propagates, the actuator corresponds to the vibration source 31, and the sensor corresponds to the vibration detector 32.

A member for transmitting vibration with the N vibration sources 31 and the M vibration detectors 32 is also considered as such an N-input M-output linear FIR system.
In order to verify the effect of the present invention by a numerical simulation, in the embodiment of FIG.
The simulation results in two cases where 1 is a 2-input and 1-output and two correlation signals are generated and given by the sound source models of FIGS. 2 and 3 are shown below.

The input signal is actually measured by sampling at 8 kHz.
The room acoustic transfer characteristics cut into 512 taps were used. The reverberation time in this room was 200 ms. Adaptive filter 13 ₁ , 13 ₂
An ES algorithm (Exponentially weighted Step-size algorithm;
S. Makino & Y. Kaneda, Weighted Step-size Projection
Algorithm for Acoustic Echo Cancellers ", IEICE Tr
ans., Vol. E75-A, No. 11, pp. 1500-1508, Nov. 1992).

As the pre-filters 12 ₁ and 12 ₂ , a maximum phase filter and a minimum phase filter with a delay were used. The transfer function is given by the following equation. G ₁ (z ^-1 ) = 0.2 + 1.0z ^-L G ₂ (z ^-1 ) = 1.0z ^-L + 0.2z ^-2L L = 512 (25) This set of prefilters has the following properties The zero points (excluding the point at infinity) of each prefilter are mirror images of the unit circle on the z plane and are relatively prime.

The frequency amplitude characteristics of each prefilter are equal. The above conditions are common to the two numerical simulations. (1) Numerical simulation A Using the above pre-filter, the following three types of signals A1: uncorrelated white noise, A2: identical white noise, A3: single white noise and FIR filter with 512 taps The transfer characteristic estimation according to the present invention was verified for the generated correlated noise. FIGS. 13A, 13B, and 13C show the measurement results for the three types of signals A1, A2, and A3.
In the figure, curve (a) is the following equation without using pre-filtering.

[0058]

[Equation 27] In defined as an error | e | ^2, curve (b) is the error in the case of using the pre-filtering. Here, h ₁ and h _{2 are} true sound transfer characteristics expressed by a 512-tap FIR filter, and h∧ ₁ and h∧ ₂ are sound transfer characteristics estimated by an adaptive filter.

When mutually uncorrelated white noise (A1) was input (FIG. 13A), the estimation error decreased to -30 dB in one second regardless of the presence or absence of the prefilter. When the same white noise (A2) was input (FIG. 13B), the estimation error was significantly reduced by the pre-filter as compared with the case where the pre-filter was not used (a). Curve (a) saturates around -4 dB, while curve (b) continued to decline after reaching -20 dB in 5 seconds.
This tendency is observed when correlated noise (A3) is input (Fig. 1).
3C). Curve (a) saturates at about -9 dB, while curve (b) reaches -20 dB in 8 seconds. These results confirm the effectiveness of the multi-acoustic path estimation method using disjoint prefilters.

(2) Numerical Simulation B A multi-acoustic path according to the present invention using three types of two-channel input signals generated from two independent sound sources and four FIR filters simulating room acoustic transfer characteristics with 512 taps. The estimation method was verified. FIGS. 14A and 14B show changes in the estimation error of the adaptive filter defined by the above equation (26) for the case where the pre-filter is not used and the case where the pre-filter is used. If the amplitude ratio of two sound sources is 1:10 in curve (a) as three types of signals, the curve
FIG. 14B shows the case where the amplitude ratio is 3:10, the curve (c) shows the case where the amplitude ratio is 10:10, and FIG. 14A shows the case where the pre-filter is not used.
FIG. 14B shows the result when the pre-filter is used.

14A and 14B, the effect of introducing the pre-filter is not clear when the amplitude ratio of the two sound sources is 10:10. However, as the amplitude ratio decreases, the effectiveness of introducing a pre-filter becomes clear. The case where the adaptive filter estimation error continues to decrease without being saturated as shown in FIG. 14A when the pre-filter is not used can also be explained by the framework of the pre-filter. Assuming that the correlation between the two input signals of simulation B is sufficiently high,
When r is the amplitude ratio of two sound sources,

[0062]

[Equation 28] Gives a good approximation. Using the z-transform, the relationship between the input and output signals is expressed by the following equation: [H ₁ (z) J ₁ (z, r) + H ₂ (z) J ₂ (z, r)] X ₁ (z) = J ₁ (z, r) Y (z) (28) The tendency of the convergence speed to increase as the amplitude ratio approaches 1 is due to J ₁ (z, r) and J ₂ (z, R) defined by equation (27).
Performs the same function as the prefilter, suggests that the distance between zeros of J ₁ (z, r) and J ₂ (z, r) on the z-plane increases as r approaches 1. are doing.

[0063]

As described above, according to the apparatus and method for separating and estimating the transfer characteristic of a multiple linear transfer path according to the present invention, an N-input M-output linear FIR system is driven by the same signal or a signal having a high correlation. Even in the case where the output signals of these pre-filters are connected to each of the pre-stages of the respective input points by connecting N pre-filters designed so that a common zero does not exist, the M output signals are output. An adaptive filter that produces a replica of
It is possible to simultaneously estimate NM signal transmission paths of an N-input M-output linear FIR system.

[0064]

[Appendix] Consider simultaneous estimation of N acoustic paths between N speakers and one microphone. Assume that the order of the acoustic path is given by M-1 and that the N input signals are the same signal x (k). The signal picked up by the microphone is y (k)
And the transfer function from the input point of the signal x (k) to the output point of the signal y (k) is H ₀ (z). At this time, the relationship between the acoustic path and the prefilter is given by the following equation: G ₁ (z) H ₁ (z) + G ₂ (z) H ₂ (z) + ... + G _N (z) H _N (z) = H ₀ (z) (A-1). In order to separate and estimate N acoustic paths from the signal after passing through the pre-filter and the microphone signal y (k), H _n (z) satisfying the equation (A-1) must be uniquely determined .

Here, it is shown that there is a pre-filter for separating and estimating the acoustic path in N from the same N input signal x (k), and its order is given by (N-1) M. The order of the N pre-filter and L-1, the following equation _{G n (z) = g nL} -1 z L-1 + g nL-2 z L-2 + ... + g n0 z 0

[0066]

(Equation 29) Consider an NM × (L + M-1) matrix S (M) defined by H _n (z) = h _nM-1 z ^M-1 + h _nM-2 z ^M-2 + ... + h _n0 h _n ^T = [h _nM-1 , h _nM-2 , ..., h _n0 ] H ₀ (z) = h _{0L + M-1} z ^{L + M-1} + h _{0L + M-2} z ^{L + M-2} + ... + h ₀₀ h ₀ ^T = [h _{0L + M- 1} , h _{0L + M-2} , ..., h ₀₀ ] The relationship between the N acoustic paths and the prefilter is expressed by the following equation [h ₁ ^T , h ₂ ^T ,..., H _N T] S ( M) = h ₀ ^T (A-3) If the matrix S (M) is a square matrix and regular, [h ₁ ^T , h
₂ ^T , ..., h _N ^T ] is clearly determined uniquely from h ₀ ^T.
(A) The degree of G _N (z) is L−1.

(B) Two matrices [G ₁ (z),..., G _N−1 (z)] ^T and G _N
(z) is irreducible (ie, disjoint). Is satisfied, the rank of the matrix S (M) satisfies Equation (A-4).
(S.Kung, T.Kailath & M.Morf, "based on the discussion of the generalized resultant matrix obtained by the permutation of
A Generalized Resultant Matrix for Polynomial Matr
ices, "Proc. IEEE Conference on Decision and Contr
ol, pp. 892-895, Dec. 1976).

Rank S (M) = M + degG _N (z) = M + L−1 (A-4) Therefore, S (M) becomes a square matrix (that is, NM = M + L−1), and G
_{1 If} (z), ..., G _N (z) is designed to satisfy conditions (a) and (b), S (M) is a square matrix from equation (A-4) .
Therefore, the N acoustic paths H ₁ (z),..., H _N (z) are uniquely determined. At this time, the following equation L-1 = (N-1) M (A-5) holds for the order L-1 of the prefilter.

[Brief description of the drawings]

FIG. 1 is a system block diagram for explaining the principle of the present invention.

FIG. 2 is a diagram showing an example of a model for generating a plurality of correlation signals as input signals in FIG. 1 from a single sound source.

FIG. 3 is a diagram showing a model for generating a plurality of correlation signals from a plurality of sound sources.

FIG. 4 is a diagram for explaining an expression representing an input / output relationship when the model of FIG. 3 is coupled to the system of FIG. 1;

FIG. 5 is a block diagram showing a first embodiment of a transfer characteristic measuring device according to the present invention.

FIG. 6 is a diagram for explaining the operation of the adaptive filter focusing on one set of N-input / one-output linear FIR system 11m in FIG. 5;

FIG. 7 is a flowchart showing a processing procedure for transfer characteristic measurement in the embodiment of FIG. 5;

FIG. 8 is a block diagram of a transfer characteristic measuring device when the embodiment of FIG. 5 is applied to measurement of transfer characteristics in an acoustic system.

FIG. 9 is a block diagram showing a second embodiment of the transfer characteristic measuring device according to the present invention.

FIG. 10 is a flowchart showing a processing procedure for transfer characteristic measurement in the embodiment of FIG. 9;

FIG. 11 is a diagram showing a solar cell panel as a specific example to which the measurement method of the present invention can be applied.

FIG. 12 is a view showing a plate-shaped member as a specific example to which the measuring method of the present invention can be applied.

FIG. 13A is a graph showing a numerical simulation of an approximation error due to an adaptive filter coefficient of a transfer characteristic in the case of two inputs and one output, and FIG. 13B shows a case where the two input signals are uncorrelated white signals; For the same white signal, C is the correlated noise generated by the FIR filter from two single white noises.

FIG. 14A shows a numerical simulation of an approximation error by an adaptive filter coefficient of a transfer characteristic in a case of two inputs and one output without using a pre-filter, using a signal having a correlation generated from two independent sound sources. FIG. 7B is a diagram showing a numerical simulation of an approximation error caused by an adaptive filter coefficient when a pre-filter is used.

Continuation of the front page (58) Fields investigated (Int. Cl. ⁷ , DB name) H04B 3/20, 7/015 H04R 3/02 H04H 21/00 G01H 1/00-17/00

Claims (24)

(57) [Claims] 1. N input points of a linear FIR system, where N is 2
The above-mentioned integer, M output points, and M is an integer of 1 or more. M is a transfer characteristic measuring device for simultaneously separating and measuring the transfer characteristics of NM transfer paths defined between them. Respectively, and outputs a pre-processed signal, N pre-filters having transfer characteristics in which all zeros are different from each other, and the pre-processed signals from the N pre-filters are converted to the pre-processed signals of the linear FIR system. N input to each of N input points
Actuators and the linear FI at the M output point of the linear FIR system.
M sensors for respectively detecting response signals from the R system; N.M from the pre-processed signals output from the N pre-filters and the response signals detected by the M sensors
And a transfer characteristic estimating unit for calculating the transfer characteristics of the individual pieces. 2. The transfer characteristic measuring device according to claim 1,
The transfer characteristic estimating unit is provided with the pre-processed signals from the N pre-filters, and outputs M replica signals which are signals obtained by estimating the response signals from the linear FIR system. Output adaptive filter and M
The replica signals and the corresponding M response signals from the linear FIR system are provided, a difference therebetween is generated as an error signal, and the M error signals are converted to the corresponding M adaptive filters. And M adaptive filters that adaptively update the filter coefficients representing their transfer characteristics so that the error signal is minimized, and obtain the obtained filter coefficients. Means for obtaining the impulse response representing the transfer characteristic of the linear FIR system. 3. The transfer characteristic measuring device according to claim 1,
The transfer characteristic estimator includes M response signals from the linear FIR system and N response signals from the N pre-filters.
And a plurality of input / output waveform holding means for capturing and holding them over a predetermined number of time points, and a vector having the captured response signal as an element, And a multi-input / output signal analyzing means for solving the simultaneous equations obtained on the assumption that the matrix is equal to the product of the transfer characteristics of the linear FIR system to obtain the transfer characteristics. 4. The transfer characteristic measuring device according to claim 2,
The input signal of the N channel is x₁(k), .... x_N(k), its z
Convert X₁(z), ..., X_N(z) and the output of the prefilter
The preprocessing signal₁(k), ..., u_N(k), and its z-transform is U
₁(z), ..., U_N(z), and the outputs of the M sensors are y
₁(k), ..., y_M(k) and its z-transform is Y₁(z), ..., Y_M(z) and
The transfer characteristics of the N prefilters are given by G₁(z), ..., G
_N(z) of the NM transmission paths of the linear FIR system
H for transfer characteristics_nm(z) (n = 1, ..., N; m = 1, ..., M)_n(z) = X_n(z) G_n(z) to calculate the pre-processed signal u_n(k), each adaptive filter has a tap number L, and its impulse
Response w_nm(0), ..., w _nm(L-1)
Ruta is given byTo calculate the replica signal y_m'(k) and subtract the above
The container has the following formula e_m(k) = y_m(k) −y_m'(k) to generate the error signal,
The filter is u_n ^T(k) = [u_n(k-L + 1), ..., u_n(k)], n =
1, ..., N and impulse of the above adaptive filter at time k
Vector w_nm(k) to w_nm ^T (k) = [w_nm(L-1), ..., w_nm(0)], the error signal and the output of the prefilter
The following equation w_nm ^T (k + 1) = w_nm ^T(k) + αe (k) u_n ^TBy repeating the calculation of (k) at each time point k,
Update the impulse response of the adaptive filter.
These are the determined adjustment parameters. 5. The transfer characteristic measuring device according to claim 4,
The adaptive filter terminates the updating process when the power of the error signal becomes equal to or less than a predetermined value, and replaces the obtained filter coefficient of the adaptive filter with the linear FIR.
Means for obtaining an impulse response representing the transfer characteristics of the system. 6. The transfer characteristic measuring device according to claim 3,
The input signals of the N channels are x ₁ (k),... X _N (k),
Let X ₁ (z), ..., X _N (z) be the transform, and let u ₁ (k), ..., u _N (k) the pre-processed signal output by the prefilter U
₁ (z), ..., U _N (z), and the outputs of the above M sensors are y
₁ (k), ..., y _M (k), and its z-transform is Y ₁ (z), ..., Y _M (z),
The transfer characteristics of the N prefilters are denoted by G ₁ (z), ..., G
_N (z), the transfer characteristics of the _NM transfer paths of the linear FIR system are H _nm (z) (n = 1, ..., N; m = 1, ..., M), The pre-filter calculates the following equation U _n (z) = X _n (z) G _n (z) to generate the pre-processed u _n (k) signal, and the multi-input / output signal analysis means Assuming that the number of taps of the impulse response representing the transfer characteristic H _nm (z) is L, the impulse response vector and the preprocessing signal vector of the transfer characteristic H _nm (z) are expressed by the following equation: h _nm ^T = [h _nm (L-1 _{), ..., h nm (0} )] u n T (k) = [u n (k-L + 1), ..., u n (k)] n = 1, ..., in N KL × L continuous pre-processing signals at each time point Using the matrix defined by and the above impulse response vector, The impulse response h _1m ,... Expressing the transfer characteristic of the linear FIR system by solving the system of linear equations given by
means for obtaining h _Nm . 7. The transfer characteristic measuring device according to claim 6,
The multi-input and output signal analyzing means for taking the correlation between the left side of the input signal components of the simultaneous equations, matrix both sides [B _1, ...,
B _N ] ^T multiplied by The transfer characteristic of the linear FIR system is determined by solving the simultaneous equations defined by 8. The transfer characteristic measuring apparatus according to claim 1, wherein the linear FIR system is an acoustic hall, the N actuators are N speakers, and the M sensors are There are M microphones. 9. The N input point of the linear FIR system, where N is 2
A transfer characteristic measuring method for simultaneously separating and measuring transfer characteristics of N · M transfer paths defined between the above integers and M output points, where M is an integer of 1 or more, including the following steps: a) the input N-channel signal is processed by N pre-filters having all zeros having different transfer characteristics to generate an N-channel pre-processed signal; (b) the N-channel pre-process A signal is input to the N input points of the linear FIR system by N actuators, respectively. (C) A response signal from the linear FIR system is detected by the M sensors at the M output points of the linear FIR system. (D) Estimating the N · M transfer characteristics based on the N-channel pre-processed signal and the response signal from the M output point. 10. The transfer characteristic measuring method according to claim 9, wherein the step (d) inputs the N-channel pre-processed signals to M N-input one-output adaptive filters, respectively.
Generating M replica signals, which are signals obtained by estimating the M response signals from the linear FIR system, and calculating the difference between each of the M replica signals and the corresponding M response signals from the linear FIR system; As an error signal,
A step of adaptively updating a filter coefficient representing a transfer characteristic of the adaptive filter so that the M error signals are minimized. 11. The transfer characteristic measuring method according to claim 9, wherein in the step (d), the response output of the linear FIR system and the N-channel pre-processed signal are acquired and retained over a predetermined number of time points, respectively. Means for obtaining the transfer characteristic by solving a simultaneous equation obtained by assuming that the vector of the acquired response signal is equal to the product of the matrix of the pre-processed signal and the vector of the transfer characteristic of the linear FIR system. . 12. The transfer characteristic measuring method according to claim 10, wherein the N-channel input signals are x ₁ (k),..., X _N (k), and their z-transforms are X ₁ (z),. ., X _N (z), the pre-processed signal output from the pre-filter is u ₁ (k), ..., u _N (k), and its z-transform is U
₁ (z), ..., U _N (z), and the outputs of the above M sensors are y
₁ (k), ..., y _M (k), and its z-transform is Y ₁ (z), ..., Y _M (z), and the transfer characteristics of the N prefilters are G ₁ ( z), ..., G _N (z),
Assuming that the transfer characteristics of the _NM transfer paths of the linear FIR system are H _nm (z) (n = 1, ..., N; m = 1, ..., M), the above step (a) Calculates the following equation U _n (z) = X _n (z) G _n (z) to generate the pre-processed signal. In the step (d), the number of taps of the adaptive filter is L, and the impulse response is _Let w _nm (0), ..., w _nm (L-1) be the following equation: The calculations to 'generate (k), the error signal equation _{e m (k) = y m} (k) -y m' said replica signal y _m computed by ^{_{(k), u n T (}} k ) = [u _n (k−L + 1),..., u _n (k)], n = 1,..., N, and a vector w _nm comprising the impulse response of the adaptive filter at time k If (k) is w _nm ^T (k) = [w _nm (L-1), ..., w _nm (0)], the following equation w _nm ^T (k + comprising the step of updating the impulse response of the adaptive filter by repeatedly executing ^{_{1) = w nm T (k}} ) + calculation of ^{_{αe (k) u n T (}} k) to each time point k, alpha is predetermined Adjustment parameters. 13. The transfer characteristic measuring method according to claim 12, wherein in said step (d), the power of said error signal is obtained at each time point, and when the value becomes equal to or less than a predetermined value, the updating process is terminated. And obtaining the impulse response of the adaptive filter obtained at that time as an impulse response representing the transfer characteristic of the linear FIR system. 14. The transfer characteristic measuring method according to claim 11, wherein said N input signals are x ₁ (k),... X _N (k), and their z-transforms are X ₁ (z),. ., X _N (z), the pre-processed signal output from the pre-filter is u ₁ (k), ..., u _N (k), and its z-transform is U
₁ (z), ..., U _N (z), and the outputs of the above M sensors are y
₁ (k), ..., y _M (k), and its z-transform is Y ₁ (z), ..., Y _M (z),
The transfer characteristics of the N prefilters are denoted by G ₁ (z), ..., G
_N (z), the transfer characteristics of the _NM transfer paths of the linear FIR system are H _nm (z) (n = 1, ..., N; m = 1, ..., M), said step (a) generates the following formula _{U n (z) = X n} (z) calculates the G _n (z) the preprocessed signal u _n (k), the step (d) of the transfer characteristic When Hnm taps of the impulse response representing the (z) is L, the impulse response vector and the preprocessing signal following equation vector h _nm ^T = [h _nm of the transfer characteristic Hnm (z) (L-1 ), .. _{., h nm (0)]} u n T (k) = [u n (k-L + 1), ..., u n (k)] n = 1, ..., defined by n, the The following equation is obtained from KL × L consecutive preprocessing signals at each time point. Using the matrix defined by The impulse response h _1m ,... Expressing the transfer characteristic of the linear FIR system by solving the system of linear equations given by
obtaining h _Nm . 15. The transfer characteristic measuring method according to claim 14, for taking the correlation between the left side of the input signal components of the simultaneous equations, both sides [B _1,. . . , B _N ] ^T multiplied by a matrix as follows: The transfer characteristic of the linear FIR system is determined by solving the simultaneous equations defined by 16. The method according to claim 9, wherein the linear FIR system is an acoustic hall, the N actuators are N speakers, and the M sensors are: There are M microphones. 17. The linear FIR system has N input points, N is an integer of 2 or more, and M output points, M is an integer of 1 or more, and simultaneously separates transfer characteristics of NM transfer paths defined between the input points. A readable recording medium on which a measuring procedure is recorded as a program to be executed by a computer, the program including the following steps: (a) transmitting an input N-channel signal, wherein all zeros have transfer characteristics different from each other; (B) inputting the N-channel pre-processed signals to the N-input points of the linear FIR system by the N actuators. ,
(c) a response signal from the linear FIR system is detected by the M sensors at the M output points of the linear FIR system, and (d) a pre-processed signal of the N channel and a response signal from the M output point. Based on the above, the NM transfer characteristics are estimated. 18. The recording medium according to claim 17, wherein said step (d) comprises inputting said N-channel pre-processed signals to M N-input one-output adaptive filters, respectively, and outputting said M-channels from said linear FIR system. M replica signals, which are signals obtained by estimating the response signal of the linear FIR system, are obtained as error signals, and the differences between the M replica signals and the corresponding M response signals from the linear FIR system are obtained. M
Adaptively updating a filter coefficient representing a transfer characteristic of the adaptive filter so that the number of error signals is minimized. 19. The recording medium according to claim 17, wherein in the step (d), the response output of the linear FIR system and the pre-processed signal of the N channel are respectively captured and held over a predetermined number of times, and the captured data are captured. Means for obtaining the transfer characteristic by solving a simultaneous equation obtained by assuming that the vector composed of the response signal is equal to the product of the matrix composed of the pre-processed signal and the vector composed of the transfer characteristic of the linear FIR system. 20. The recording medium according to claim 18, wherein said N-channel input signals are x ₁ (k),..., X _N (k), and their z-transform X _1.
(z), ..., X _N (z), the pre-processed signal output from the pre-filter is u ₁ (k), ..., u _N (k), and its z-transform is U ₁ (z ), ..., U
_{Let N} (z) be the output of the above M sensors y ₁ (k), ..., y
_M (k) and its z-transform is Y ₁ (z), ..., Y _M (z), and the transfer characteristics of the N prefilters are G ₁ (z), ..., G _N ( z), the transfer characteristics of NM transmission paths of the linear FIR system are represented by H
_nm (z) (n = 1, ..., N; m = 1, ..., M), the above step (a) is performed by the following equation U _n (z) = X _n (z) G _n ( z) to generate the pre-processed signal. In the step (d), the number of taps of the adaptive filter is L, and the impulse response is w _nm (0), ..., w _nm (L-1 ), The following equation: The calculations to 'generate (k), the error signal equation _{e m (k) = y m} (k) -y m' said replica signal y _m computed by ^{_{(k), u n T (}} k ) = [u _n (k−L + 1),..., u _n (k)], n = 1,..., N, and a vector w _nm comprising the impulse response of the adaptive filter at time k If (k) is w _nm ^T (k) = [w _nm (L-1), ..., w _nm (0)], the following equation w _nm ^T (k + 1) = it includes a step of updating the impulse response of the adaptive filter by the calculation of ^{_{w nm T (k) + αe}} (k) u n T (k) is repeated at each time point k, the α is previously These are the determined adjustment parameters. 21. The recording medium according to claim 20, wherein in the step (d), the power of the error signal is obtained at each time, and when the value becomes equal to or less than a predetermined value, the updating process is terminated. Obtaining the impulse response of the adaptive filter obtained at that time as an impulse response representing the transfer characteristic of the linear FIR system. A recording medium 22. The method of claim 19, the N input signals _{x 1 (k), .... x} N (k), the z-transform X
₁ (z), ..., X _N (z), the pre-processed signal output from the pre-filter is u ₁ (k), ..., u _N (k), and its z-transform is U
₁ (z), ..., U _N (z), and the outputs of the above M sensors are y
₁ (k), ..., y _M (k), and its z-transform is Y ₁ (z), ..., Y _M (z),
The transfer characteristics of the N prefilters are denoted by G ₁ (z), ..., G
_N (z), the transfer characteristics of the _NM transfer paths of the linear FIR system are H _nm (z) (n = 1, ..., N; m = 1, ..., M), said step (a) generates the following formula _{U n (z) = X n} (z) calculates the G _n (z) the preprocessed signal u _n (k), the step (d) of the transfer characteristic Assuming that the number of taps of the impulse response representing Hnm (z) is L, the impulse response vector and the pre-processed signal vector of the transfer characteristic Hnm (z) are expressed by the following equation: h _nm ^T = [h _nm (L-1),. _{., h nm (0)]} u n T (k) = [u n (k-L + 1), ..., u n (k)] n = 1, ..., defined by n, the The following equation is obtained from KL × L consecutive preprocessing signals at each time point. Using the matrix defined by and the above response signal, The impulse response h _1m ,... Expressing the transfer characteristic of the linear FIR system by solving the system of linear equations given by
obtaining h _Nm . 23. The recording medium according to claim 22, wherein the correlation between the input signal components on the left side of the simultaneous equations is obtained.
[B _1, on both sides. . . , B _N ] ^T as the following equation The transfer characteristic of the linear FIR system is determined by solving the simultaneous equations defined by 24. The recording medium according to claim 17, wherein the linear FIR system is an acoustic hall, the N actuators are N speakers, and the M sensors are M sensors. Microphone.