JP2002095097A - Adaptive signal processing system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable an adaptive signal processing system to be improved in convergence characteristics and made simple in structure. SOLUTION: A pre-processing filter 12 having the same characteristics as an adaptive inverse filter 14 is provided to the input side of a circuit composed of a series connection circuit consisting of an unknown transfer function 13 and the adaptive inverse filter 14 and a filter 15 connected in parallel with the series connection circuit. An initial-value memory 11 which stores the initial values of coefficients is provided. When the adaptive signal processing system starts operating, the initial value of a coefficient is transferred from the initial value memory 11 to the pre-processing film 12 and the adaptive inverse filter 14, and the coefficient values of the adaptive inverse filter 14 are transferred to the coefficient values of the pre-processing filter 12 when adaptation is made. The output of the band filter 15 is subtracted from the output of the adaptive reverse filter 14 by a subtractor 16, the difference between the outputs is inputted as error signals into a coefficient update calculation circuit 17. The coefficient update calculation circuit 17 calculates updated coefficients for updating the coefficients of the adaptive inverse filter 14 based on the error signals and output signals of the unknown transfer function 13.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an adaptive signal processing system for obtaining an inverse characteristic of an unknown transfer function adaptively at a high speed and stably by giving an initial value to a coefficient of an adaptive inverse filter. This is effective in adaptive signal processing for adaptively finding the inverse characteristic of the external auditory canal transfer characteristic of the listener in localization of the external sound image.

[0002]

2. Description of the Related Art A conventional adaptive signal processing system is described in, for example, "Adaptive Signal Prism" by B. Widrow, SDStearns.
ocessing, "Prentice-Hall (ISBN0-13-004029-0) pp.28
0-295 (1985). FIG.
It is a block diagram of a conventional example. FIG. 3 is an explanatory diagram for explaining the operation of the conventional example.

In the configuration shown in FIG. 2, an input signal is filtered and input to a coefficient update operation circuit.
Commonly known as ered-X. Further, since the input signal is a color signal such as voice, an affine projection algorithm is applied. Under this condition, it is shown that convergence can be achieved by using a block processing method that performs an operation for each of a plurality of samples. Tsutsui et al., "Study on the relationship between adaptive inverse filter estimation and identification problems," This is described in the Technical Report of the Communication Society of Japan, EA98-91, pp.9-16 (1998).

Therefore, a block projection algorithm (hereinafter, referred to as a BOP algorithm) is applied to the configuration shown in FIG. 2 to explain the prior art. This algorithm is generally
It is known that the input signal of color system converges faster than the LMS algorithm, and a detailed explanation is given by Shigeo Tsujii,
"Adaptive signal processing," described in Shokodo, pp. 70-97.

In FIG. 2, 1 is an unknown transfer function, 2 is an adaptive inverse filter, 3 is a bandpass filter, 4 is a subtractor, 5 is a reference transfer function, and 6 is a coefficient update operation circuit.

The adaptive inverse filter 2 for inversely identifying the characteristic of the unknown transfer function 1 often uses an FIR filter configuration.
The FIR filter characteristic, by giving the coefficients h _M from the outside, it is possible to obtain easily the desired characteristics in a stable operation. Now, M samples of the input signal x (n) are represented by x
_M (n), the output signal y of the adaptive inverse filter 2
The relationship of (n) is expressed as in equation (1) of FIG. Here, T indicates transposition of a matrix, and x _M (n) = [x (n),
x (n−1),..., x (n−M + 1)] ^T , h (n) =
[H ₀ (n), h ₁ (n),..., H _M−1 (n)] ^T. Furthermore, x (n) The elements of the vector _x M (n), _{h i}
(N) is the element of the vector h (n), i is the tap position of the FIR filter configuration, M is the number of taps of the adaptive inverse filter 2,
n represents time.

The output signal y (n) of the adaptive inverse filter 2
Is input to unknown transfer function 1. The output signal z (n) of the unknown transfer function 1 is obtained by calculating the impulse response of the unknown transfer function 1 by c
_{If N} , it is represented by equation (2) in FIG. Where y _N =
[Y (n), y (n-1), ..., y (n-N +
1)] ^T , c _N = [c ₀ , c ₁ ,..., C _N−1 ] ^T Note that y _N (n) is the y represented by the equation (2).
(N) is an output vector of the adaptive inverse filter 2 of N elements from the present to the past.

On the other hand, the output signal z (n) of the unknown transfer function 1 is subtracted by the subtractor 4 from the output signal d (n) of the bandpass filter 3 which has been subjected to band limitation using the input signal x (n) as an input. Is an error signal e (n). Further, a hat (∧) of the output signal u (n) when the input signal is input to the reference transfer function 5 having a transfer function characteristic substantially equal to that of the unknown transfer function 1 is obtained. Here, if the impulse response vector of the reference transfer function 5 is a hat (∧) of c _N , c
The hat (∧) of _N is used to obtain the hat (∧) of the signal u (n), and has a certain degree of correlation with the impulse response vector c _N of the unknown transfer function 1.

The number of taps of the hat (ｃ) of c _N and c _N is N, and the number of taps of the adaptive inverse filter h (n) is M (≧
2N), and the hats (∧) of the output signal u (n) of the hats (∧) of the error signals e (n) and c _N are respectively expressed as Expressions (3) and (4) in FIG. Can be. In FIG. 2, the BOP algorithm for adaptive inverse filter estimation is given by Expressions (5) and (6) in FIG.

Where w is the impulse response vector of the final value of the inverse filter of the impulse response c _N of the unknown transfer function, μ is the relaxation coefficient or step gain, j is the block number, L is the block length, and L = N + r− Given by 1. R is the order in the BOP algorithm,
This is the number of input signal vectors used to correct the estimated vector. At this time, u (j) (w−h) in equation (6)
(J)) is an r-order error signal vector e (j) = (e (j
L-1), e (jL-2),..., E (jL-r)) ^T. The matrix (represented by the hat (∧) of U (j) ⁺ ) is the Moore-Penrose of the hat (∧) of the matrix U (j).
Represents the general inverse of the matrix, and the hat (∧) of the matrix U (j)
Is a matrix in which each element has a hat (∧) of u (n), and is represented as Expression (7) in FIG. Further, w of the band-pass filter 3 is an inverted filter of unknown impulse response c _N, c _N and a linear convolution results band w filter 3
The adaptive inverse filter is modified so as to obtain the target impulse response b. That is, as shown in Expression (8) in FIG.
_N · w = b or w = c _N ⁻¹ · b. The length of b is N + M-1. In this way, h (n) can be determined sequentially from Equations (7) and (5).

FIG. 4 is a block diagram of another conventional example.
FIG. 5 is an explanatory diagram of an operation for explaining the operation of another conventional example. This example is a method devised for achieving stable convergence at high speed. This method is described in E. Bjarnason, “Acti
ve noise cancellation using a modified form of the
filered-X LMS algorithm, ”Proc. EUSIPCO'92, Signa
l ProcessingVI, Brussels, Belgium, vol.2, pp.1053-
1056 (1992).

The illustrated example shows an unknown transfer function 1, an adaptive inverse filter 2, a bandpass filter 3, a subtractor 4, a reference transfer function 5a,
5b, a coefficient update operation circuit 6, a pre-processing filter 7, an adder 8, and a subtractor 9. In this example, the unknown transfer function 1 and the reference transfer function 5a are connected in parallel, and the respective outputs are input to the subtractor 4 and the adder 8. The input of the unknown transfer function 1 and the reference transfer function 5a is provided with a pre-processing filter 7. Further, the input signal is provided to the adaptive inverse filter 2 via the reference transfer function 5b. The subtractor 4 subtracts the output z (n) of the unknown transfer function 1 from the output d (n) of the bandpass filter 3, and the adder 8 outputs the output e (n) of the subtractor 4 and the reference transfer function. The adder / subtractor 9 for adding the hat (∧) of the output z (n) of 5a subtracts the hat (∧) of the output y (n) of the adaptive inverse filter 2 from the output of the adder 8 and outputs the result. This is a subtractor that gives γ (n) to the coefficient update operation circuit 6.

Hat (∧) of hat signal (信号) of impulse response c of reference transfer function 5a connected in parallel with impulse response c of unknown transfer function 1, hat (∧) of output signal z (n), output signal of adaptive inverse filter 2 The hat (∧) of y (n) and the error signal γ (n) are expressed by equations (9), (10),
It is represented by (11). Further, since the coefficient value of the adaptive inverse filter 2 is copied to the coefficient value of the pre-processing filter 7, z
(N) can be modified as in equation (12) in FIG. 5, as in equation (2). Further, if Δz (n) = (hat (∧) of z (n)) − z (n) is defined, the expression (1) in FIG.
Obtain 3). From the above, the error signal γ (n) is expressed as in Equation (14) in FIG. 5, and eventually, d (n) + Δz (n)
Is identified by a hat (∧) of y (n). At this time, as shown in FIG. 4, the coefficient value of the pre-processing filter 7 is updated by copying the coefficient of the adaptive inverse filter 2. Here, the BOP algorithm is calculated by the above equation (5).
Similarly to the equation (6), it is defined by the equation (15) in FIG. Here, the hat (∧) and γ (n) of u in the equation are given by equations (16) and (17) in FIG. 5, respectively.

[0014]

However, the two types of prior art described above have the following problems, so that there is no technically practical one. (1) In the block processing, the convergence speed is slow, and a new memory processing for each block length is required. (2) Although an algorithm for adaptively finding an inverse filter and improving the convergence speed has been reported, the improved block configuration method of FIG. 4 has a large number of functional blocks and is complicated, as is apparent from the figure.

[0015]

The present invention employs the following structure to solve the above-mentioned problems. <Configuration 1> An adaptation in which an unknown transfer function is connected in series to form an inverse characteristic of the unknown transfer function, a coefficient value is obtained based on a given coefficient update value, and the filter characteristic is variably configured by the coefficient value. An inverse filter, an unknown transfer function and an adaptive inverse filter are connected in parallel to a series circuit,
A band-pass filter having a filter characteristic such that an output signal passed through a series connection circuit and a signal passed through itself become equal, an initial value memory storing a specific initial value as a coefficient value, and an unknown transmission of the output signal A pre-processing filter that is connected to be given to the function and the band-pass filter and has the same characteristics as the adaptive inverse filter, and the filter characteristics are variably configured by the given coefficient values. A subtracter for subtracting the output signal of the filter; and a coefficient update circuit for calculating a coefficient update value of the adaptive inverse filter from the output signal of the unknown transfer function and the output signal of the subtractor. The initial value of the coefficient value is transferred to the pre-processing filter and the adaptive inverse filter, respectively, and at the time of adaptation, the coefficient value of the adaptive inverse filter is pre-processed. Adaptive signal processing system characterized by being configured to forward to the coefficient values of the filter.

<Structure 2> In the adaptive signal processing system according to Structure 1, at the time of adaptation, the update coefficient value obtained from the coefficient update operation circuit is transferred to the pre-processing filter. An adaptive signal processing system configured to determine a coefficient value.

<Configuration 3> In the adaptive signal processing system according to Configuration 1 or 2, the initial value is an inverse impulse response sample of a centroid vector expressing the impulse response of an unknown transfer function as a vector, or an unknown value closest to the centroid point. An adaptive signal processing system, which is an inverse impulse response sample of a transfer function.

[0018]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in detail with reference to specific examples. First, prior to the description of the specific example, the initial value problem of the adaptive inverse filter, which is the premise of the present invention, will be briefly described.

Horiuchi, Hokari, Shimada, "A study on the construction of an adaptive inverse filter applied to out-of-head sound localization," IEICE Technical Report, EA99-93, pp.1-8 (1999) If there is some correlation between the unknown transfer function and the reference transfer function, or if the variation of the transfer function is known within a certain range, the center of variation of the inverse impulse response of the unknown transfer function as a vector It is described that by setting the vector or the inverse impulse response of the sample closest to the center of gravity to the coefficient value of the adaptive inverse filter as an initial value, the adaptive inverse filter is stabilized and the convergence time is shortened. That is, it is reported that if the range of variation of the unknown transfer function is known to some extent in advance, a characteristic with a short convergence time can be obtained by setting an initial value at the center. Therefore, in this specific example, the initial value of the coefficient value of the adaptive inverse filter is set to the centroid vector of the variation expressing the inverse impulse response of the unknown transfer function as a vector, or the inverse impulse response of the sample closest to the centroid point. Like that.

<< Specific Example >><Structure> FIG. 1 is a block diagram showing a specific example of the adaptive signal processing system of the present invention. The illustrated system includes an initial value memory 11, a pre-processing filter 12, an unknown transfer function 13,
Adaptive inverse filter 14, band-pass filter 15, subtractor 16,
And a coefficient update operation circuit 17.

The initial value memory 11 is a memory for storing a specific initial value as a coefficient value. In this specific example, an inverse impulse response sample of a centroid vector expressing the impulse response of the unknown transfer function 13 as a vector, or The inverse impulse response sample of the unknown transfer function 13 closest to the center of gravity has an initial value. The pre-processing filter 12
This filter can change the filter characteristics with the coefficient value from the initial value memory 11 or the coefficient value from the adaptive inverse filter 14, and has the same characteristics as the adaptive inverse filter 14. The pre-processing filter 12 is configured to receive an input signal x (n) and output the output u (n) to the unknown transfer function 13 and the bandpass filter 15.

The adaptive inverse filter 14 calculates the unknown transfer function 13
Is a filter that has the inverse characteristic of the above, and whose characteristics can be sequentially changed by the coefficient update value Δh (n) from the coefficient update operation circuit 17 with the output signal z (n) of the unknown transfer function 13 as an input. The bandpass filter 15 is connected in parallel with a circuit of the unknown transfer function 13 and the adaptive inverse filter 14 connected in series, and is a filter on which a delay and a band limit necessary for obtaining an inverse characteristic of the unknown transfer function 13 are applied. The output signal u (n) of the preprocessing filter 12 is input, and the output d (n) is connected to the subtractor 16. Subtractor 16
Is a subtractor that subtracts the hat (∧) of the output y (n) of the adaptive inverse filter 14 from the output d (n) of the bandpass filter 15 and outputs an error signal e (n). The coefficient update operation circuit 17 is a circuit for calculating the coefficient update value Δh (n) of the adaptive inverse filter 14 from the output signal z (n) of the unknown transfer function 13 and the output signal e (n) of the subtracter 16. .

<Operation> First, at the start of the operation, the coefficient values of the initial value memory 11 are copied as the coefficient values of the pre-processing filter 12 and the adaptive inverse filter 14. Next, when the input signal x (n) is input to the pre-processing filter 12, the coefficient previously transferred from the initial value memory 11 and the input signal x (n) are input.
The result of the convolution operation u (n) with (n) is output. The output signal u (n) of the preprocessing filter 12 is an unknown transfer function 1
3 and the output signal z of this unknown transfer function 13
(N) is output. Output signal z of unknown transfer function 13
(N) is input to the adaptive inverse filter 14, and a hat (∧) of the output signal y (n) is output. On the other hand, the output signal u (n) of the pre-processing filter 12 is input to the bandpass filter 15, and the output signal d (n) is output. Then, the subtractor 16 outputs the output signal d of the bandpass filter 15.
The hat (∧) of the output signal y (n) of the adaptive inverse filter 14 is subtracted from (n). The output is referred to as an error signal e (n), and the output signal z (n) of the unknown transfer function 13 and the error signal e (n).
The coefficient is updated by the coefficient update operation circuit 17 having (n) as an input.

FIG. 6 is an explanatory diagram of the arithmetic operation in which the above explanation is expressed by mathematical expressions. However, x _M (n), u _N (n), z
_M (n) is _xM (n) = [x (n), x (n-1), ..., x (n-
_{^{M + 1)] T u N}} (n) = [u (n), u (n-1), ..., u (n-
M + 1)] ^T z _M (n) = [z (n), z (n−1),..., Z (n−
M + 1)] is a time vector defined by ^T.

Here, in the LMS algorithm, as is well known, Δh (n) = μz (n) e (n). As described above, in this specific example, the inverse filter is replaced with a problem of identifying a transfer function obtained by dividing the transfer function of the bandpass filter 15 by the unknown transfer function 13,
The coefficient update operation circuit 17 includes NLMS other than LMS, affine projection (AP), and recursion (RLS), which are ordinary sequential algorithms.
It is clear that is applicable. Accordingly, the description of the coefficient update operation circuit 17 for calculating Δh (n) will be omitted.

<Effects> Next, in order to clarify the effects of the above specific example, a case where the present invention is applied to out-of-head sound image localization will be specifically described. First, an out-of-head sound image localization will be described. FIG. 7 is a diagram illustrating the principle of realizing the out-of-head sound image localization. In FIG.
(A) shows listening with a speaker, and (b) shows listening with a binaural earphone. In the figure, 100 is a sound source signal, 101 is a speaker, 102 is a listener, 103 is a digital filter, 104 is an earphone, and 105 is a microphone installed in the ear canal of the listener. The subscripts _L and _R, such as SSTF _L and SSTF _R , indicate the left and right sides.

The principle of out-of-head sound image localization will be described below with reference to FIG. The sound source in the space (the speaker 10 in FIG.
The same stimulus as the sound stimulus from 1) to the eardrum is applied to the earphone 10
By giving to the eardrum from 4, the sound source can be perceived as being outside. However, since it is extremely dangerous to capture the vibration stimulation signal on the eardrum by the sound wave from the living body as an electric signal, the sound source signal 100 shown in FIG.
The transfer function of the electrical signal from the ear to the eardrum cannot be measured exactly.

Therefore, a very small microphone 105 is attached to the external auditory canal of both ears, and a transfer function from the sound source signal 100 radiated from the speaker 101 to the output of the microphone 105, that is,
Spatial acoustic transfer function (SSTF: Spacia)
l Measure Sound Transfer Function).

Here, literature: Jens Brauelt,
Masayuki Morimoto, Toshiyuki Goto “Spatial Sound” Kashima Publishing p. 20 ~
p. As described in the sound wave propagation in the ear canal as described in Section 2.4 of Section 28, the sound wave in the ear canal is represented by a one-dimensional model. It is self-evident that the electric signal caused by the vibration of FIG. Therefore, even if the ultra-small microphone 105 is fixed to the ear canal and the transfer function, that is, the impulse response in the time domain is measured, the result is equivalent.

However, as pointed out in the above literature, since the speaker 101 has frequency characteristics, the true transfer function of the electric signal from the input of the speaker 101 to the output of the microphone 105 is If the transfer function is LSTF (Loud Speaker Transfer Function), it is SSTF / LSTF. Just this value is the head related transfer function HRTF (Head Related Transfer Fu
nction). LSTF is the speaker 1
This is a speaker transfer function for correcting the frequency characteristic of No. 01.

HRTF is the ratio of the transfer function when the head is present to the transfer function when the head is not present, as defined in the above document. Since this is the product of the function and the LSTF, and the spatial propagation is only the amount of delay, HRTF ≒ SSTF / LSTF.

On the other hand, in FIG. 7B, in order to create a transfer function equivalent to the binaural earphone (or stereo headphone) 104 by using the binaural earphone 104, a microphone attached to the external auditory canal from the input of the binaural earphone 104 The transfer function up to the output of 105, that is, the ear canal transfer function (ECTF: Ear Cana)
l Transfer Function), and if the transfer function of the product of the ECTF and the transfer function of the digital filter 103 matches the transfer function SSTF / LSTF, the same listening signal as the speaker listening is placed at the location of the microphone 105 installed in the ear canal. Can be played.

As described above, the transfer function of the digital filter 103 can be obtained by a computer. That is, this transfer function (out-of-head sound image localization transfer function) is represented by S
LTF (Soundimage Localization Transfer Functio
n)

SLTF = ｛SSTF / (ECTF · LS
TF)｝ ≒ HRTF / ECTF, and all the terms on the right-hand side of this equation can be obtained by measurement. After that, by performing mathematical operations, SLTF can be obtained.

However, even if the above SLTF can be obtained, the transfer functions ECTF and HRTF are:
The size differs depending on the size of the ear canal, the size of the ear, and the size of the face of the listener. In other words, if the transfer function does not match the shape of the face of the individual, an accurate sound image cannot be localized outside the head, and in the worst case, the front and rear judgment of the sound image cannot be performed or the head cannot be localized outside the head.

[0036] For this reason, unless the localization is performed outside the head due to individual differences in the transfer function, it is not generally used for an unspecified number of listeners. Therefore, it is hard to select several types of transfer functions measured in advance. There is a need for a solution or a system that can measure the transfer function for each individual. But,
The above-described hardware solution increases the amount of hardware, and the method using a system that measures the transfer function for each individual is very expensive.

By the way, studies on the individuality of the transfer functions ECTF and SSTF have already been made. For example, S.Yano, H.Hokari, andS.Simada: "A Study o
n Personal Difference in the Transfer Functions of
Sound Localization ", AES the 106th Convention, NO.4
922, May 8-11 (1999) concludes that the transfer function ECTF has greater individuality than the transfer function SSTF.

Therefore, by adopting an adaptive signal processing technique, an ECT that matches one's own while listening to musical sounds and voices without measuring the transfer function of the ECTF for each individual.
A method for obtaining the transfer function of F has already been disclosed in Japanese Patent Application No. 10-26.
No. 1292, "Earphone and out-of-head sound localization device"
Has been filed.

FIG. 8 is a functional block diagram of only the one ear side in the out-of-head sound image localization apparatus. The system shown is shown in FIG.
FIG. 3 shows a block on only one ear side for realizing the function of (b), in which only the inverse transfer function of ECTF is to be adaptively equalized. The illustrated system includes a digital filter 201, an adaptive filter 202, an earphone (speaker) 203, a microphone 204, a bandpass filter 205, and a subtractor 206.

Further, in order to clarify a part of the typical characteristics of the ECTF, the impulse response characteristics and the frequency amplitude characteristics in the time domain of the ECTF are shown. FIG. 9 shows the ECT
FIG. 9 is an explanatory diagram showing an impulse response characteristic and a frequency amplitude characteristic in a time domain of F. According to these characteristics, the impulse response length is approximately 5 msec or less, and is effective within 48 kHz. This characteristic is mainly due to the outer ear characteristics of the listener and the characteristics of the earphone. Therefore, this ECTF corresponds to the unknown transfer function 13 of the present invention shown in FIG.

The effect of applying this embodiment to the above-described out-of-head sound image localization technique will be described below. It is important to show convergence characteristics in order to clarify the performance of this example and the conventional example. Therefore, the estimation error ε (n) is defined. FIG. 10 is an explanatory diagram of the calculation of the estimation error ε (n). Where B (f), C (f), H (f, n)
Are the frequency characteristics of the bandpass filter 15, the unknown transfer function 13, and the adaptive inverse filter 14, respectively. That is, Expression (23) indicates the estimation error at each time at the time of convergence, and indicates that the product of the unknown transfer function 13 and the transfer function of the adaptive inverse filter 14 approaches the bandpass filter 15 which is the target transfer function. ing.

FIG. 11 is an explanatory diagram showing a comparison between the convergence characteristics of the conventional example and this specific example. In the figure, (a) shows the convergence characteristics of the conventional example shown in FIG. 2, (b) shows the convergence characteristics of the other conventional example shown in FIG. 4, and (c) shows the convergence characteristics of this specific example shown in FIG. Also,
Block processing is used in the conventional example, and sequential processing is used in the other conventional examples and this specific example. Furthermore, when the affine projection (AP) algorithm is applied to the coefficient update operation when a white signal and an audio signal are input. This represents the estimation error.

As shown in the figure, it can be seen that the conventional example (a) and the other conventional example (b) clearly have lower convergence characteristics than the present specific example (c). In addition, in the other conventional example of FIG. 4 in the voice input signal, the case where μ = 0.05 is shown because the divergence is unstable when the relaxation coefficient μ = 0.5. In this specific example, the coefficient value of each of the pre-processing filter 12 and the adaptive inverse filter 14 as an initial value is added to the center of gravity vector of the variation expressed by the inverse impulse response vector of the unknown transfer function 13, or the sample closest to the center of gravity ( ECTF) is set. As can be seen from these, the total transfer function of the pre-processing filter 12 and the unknown transfer function 13 is
Since the characteristics are almost the same as those of the bandpass filter 15, the signal z
(N) is a signal having substantially the same spectrum as the input signal x (n).

This is a configuration substantially different from the Filtered-X configuration method. That is, in the Filtered-X configuration method, a signal obtained by filtering the input signal x (n) with characteristics (reference transfer function) equivalent to the unknown transfer function is used for updating the coefficients. Is large, the convergence characteristics are degraded. This is the same in the other conventional example of FIG. On the other hand, since the reference transfer function is not required in the configuration method of this specific example, it can be seen that the calculation result is almost the same as the calculation result for identifying the ordinary unknown transfer function.

In this specific example, the coefficient value of the adaptive inverse filter 14 may be copied to the pre-processing filter 12, but the coefficient update value Δh (n) of the coefficient It is sent to the pre-processing filter 12 (see the dashed arrow in FIG. 1), and the pre-processing filter 12
A configuration may be adopted in which a result calculated directly from h (n) is substituted for a coefficient value.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a specific example of an adaptive signal processing system according to the present invention.

FIG. 2 is a block diagram showing a conventional adaptive signal processing system.

FIG. 3 is an explanatory diagram for explaining the operation of the conventional example.

FIG. 4 is a block diagram showing another conventional adaptive signal processing system.

FIG. 5 is an explanatory diagram for explaining the operation of another conventional example.

FIG. 6 is an explanatory diagram for explaining the operation of a specific example of the adaptive signal processing system according to the present invention.

FIG. 7 is a diagram illustrating the principle of realizing out-of-head sound image localization.

FIG. 8 is a functional block diagram of only the one ear side in the out-of-head sound image localization apparatus.

FIG. 9 is a diagram illustrating characteristics of an ear canal transfer function.

FIG. 10 is an explanatory diagram of calculation of an estimation error ε (n).

FIG. 11 is an explanatory diagram showing a comparison between convergence characteristics of a conventional example and this specific example.

[Explanation of symbols]

 Reference Signs List 11 Initial value memory 12 Preprocessing filter 13 Unknown transfer function 14 Adaptive inverse filter 15 Bandpass filter 16 Subtractor 17 Coefficient update arithmetic circuit

 ──────────────────────────────────────────────────続 き Continuing from the front page (71) Applicant 597032206 Haruhide Hokari 1603-1 Kamitomicho-cho, Nagaoka-city, Niigata Pref. Inside the University of Technology (72) Inventor Yasuo Aoyagi 1-7-12 Toranomon, Minato-ku, Tokyo Oki Electric Industry Co., Ltd. (72) Inventor Katsuto Miwa 1-4-4 Nishikicho, Tachikawa, Tokyo (72) Inventor Masaharu Shimada 1603-1 Kamitomiokacho, Nagaoka City, Niigata Prefecture Inside Nagaoka University of Technology (72) Inventor Haruhide Hokari 1603-1 Kamitomiokamachi, Nagaoka City, Niigata Prefecture ) Inventor Shunji Horiuchi 1603-1 Kamitomiokacho, Nagaoka City, Niigata Prefecture F-term in Nagaoka University of Technology 5D020 CE04 5D062 AA65 BB04

Claims (3)

[Claims] 1. An unknown transfer function is connected in series to form an inverse characteristic of the unknown transfer function, and a coefficient value is obtained based on a given coefficient update value, and the filter characteristic is variably configured by the coefficient value. The adaptive inverse filter is connected in parallel to the series connection circuit of the unknown transfer function and the adaptive inverse filter, and the output signal passing through the series connection circuit and the signal passing through itself become equal. A band filter having filter characteristics, an initial value memory for storing a specific initial value as a coefficient value, and an output signal connected to the unknown transfer function and the band filter, and equal to the adaptive inverse filter. A pre-processing filter having a characteristic and a filter characteristic variably configured by a given coefficient value; and A subtracter for subtracting an output signal of the filter; and a coefficient update circuit for calculating a coefficient update value of the adaptive inverse filter from the output signal of the unknown transfer function and the output signal of the subtractor. An initial value of a coefficient value is transferred from the initial value memory to each of the pre-processing filter and the adaptive inverse filter, and at the time of adaptation, a coefficient value of the adaptive inverse filter is transferred to a coefficient value of the pre-processing filter. An adaptive signal processing system characterized by the following. 2. The adaptive signal processing system according to claim 1, wherein at the time of adaptation, an update coefficient value obtained from said coefficient update operation circuit is transferred to said preprocessing filter, and said preprocessing filter updates the coefficient. An adaptive signal processing system configured to obtain a coefficient value from a value. 3. The adaptive signal processing system according to claim 1, wherein the initial value is an inverse impulse response sample of a centroid vector in which an impulse response of an unknown transfer function is represented by a vector, or an unknown transmission closest to a centroid point. An adaptive signal processing system, which is an inverse impulse response sample of a function.