JPH0993088A - Method and apparatus for estimating and controlling adaptive filter coefficient - Google Patents

Abstract

PROBLEM TO BE SOLVED: To accelerate convergence and further to provide high estimation accuracy after convergence. SOLUTION: A coefficient arithmetic part 100 successively and cumulatively operates the coefficient update quantity of filter coefficient of non-recursive filter by the addition normalizing LMS method. Based on the calculated result of coefficient arithmetic part 100, the coefficient of non-recursive filter is updated. An observation part 120 observes coefficient update quantity proximated to the estimation error of filter coefficient of non-recursive filter. A block length control part 130 controls the length of a block, for which the coefficient update quantity at the coefficient arithmetic part 100 is operated, according to the transition of coefficient update quantity. Then, the coefficient arithmetic part 100 calculates quantity An(m) and Pn required for coefficient updating in the addition normalizing LMS method. The observation part 120 observes the transition of coefficient update quantity An(m)/Pn. The block length control part 130 decides the block length based of the transition of coefficient update quantity An(m)/Pn.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a filter coefficient setting control in a non-recursive filter, and more particularly, to transmitting a reference signal to a signal transmission system having an unknown characteristic and obtaining a response output of the signal transmission system. Thus, the present invention relates to an adaptive filter coefficient estimation control method and apparatus suitable for accelerating the convergence of the coefficient estimation processing of the system for estimating the coefficient of the adaptive acyclic filter that simulates the characteristics of the signal transmission system.

[0002]

Adaptation for sending a reference signal to a signal transmission system having an unknown characteristic and simulating the response characteristic of the signal transmission system based on the response of the signal transmission system to the reference signal and the reference signal. A system for estimating the filter coefficient of an acyclic filter is known. Such a system can simulate the characteristics of the signal transmission system with high accuracy, but it is necessary to consider the balance between the time required for estimating the filter coefficient and the obtained accuracy.

FIG. 9 shows a generalized schematic configuration of a system expected to be improved in operation by applying the present invention. However, in FIG. 9, an adder 230 for adding the response g _j (Equation 9) and the disturbance N _j of the signal transmission system 200 whose characteristics are unknown, and its output Y _j (Equation 10) and the acyclic type (FIR). ) The subtractor 240 that generates the difference E _j (Equation 12) from the pseudo response G _j (Equation 11) synthesized by the filter 210 may not be given as a concrete device or circuit.

Here, the response g _j of the signal transmission system 200 is j: sample time index (iteration) indicating the number of times the sample is repeated corresponding to time, h _j (i): the signal transmission system 200 I-th sample value of impulse response h _j , X _j (i): signal obtained by delaying sample value of reference signal X _j by i sampling period, I: maximum delay of impulse response of signal transmission system 200, Σ: i = 1 to I, which is expressed by Equation 9.

[0005]

[Equation 9]

The output Y _j of the adder 230 is the equation 10, the pseudo response G _j synthesized by the acyclic filter 210 is the equation 11,
Then, the difference E _j is expressed by the equation 12.

[0007]

[Formula 10] Y _j = g _j + N _j

[0008]

[Equation 11]

[0009]

[Equation 12] E _j = Y _j −G _j There is also an example in which another unknown signal transfer system exists between the difference E _j and the coefficient updating circuit 220. In that case, in the coefficient updating circuit 220, On the input side of a certain reference signal X _j , another filter simulating the characteristic of the unknown signal transmission system is placed in front.

The coefficient updating circuit 220 uses the coefficient H _j of the _acyclic filter 210 so that the difference E _j is minimized. For example, when the learning identification method is adopted as the adaptive algorithm, the coefficient of the m-th tap is used. Are updated according to the equation 13.

[0011]

(Equation 13)

However, K is a step gain (0 <K <2)
And has a function of determining the speed at which the coefficient H _j of the non-recursive filter 210 converges to the impulse response h _j of the unknown signal transmission system 200 and the magnitude of the response residual shown in Expression 14 obtained after the convergence. .

[0013]

D _j = g _j −G _j For example, the convergence speed is fastest when K = 1, and conversely, the response residual D _j obtained after convergence is smaller as the step gain K is closer to 0. Become. It goes without saying that it is desirable for the system that the convergence be fast and that the response residual obtained after the convergence be small. However, these cannot be realized simultaneously with one fixed step gain.

Therefore, the present invention intends to provide a high-speed estimation system for adaptive filter coefficients, which realizes high-speed convergence and high estimation accuracy at the same time, by appropriately controlling this coefficient updating process.

Here, what is important is that the control by the step gain becomes unnecessary when the convergence value of the estimation accuracy of the adaptive filter coefficient obtained when the step gain is fixed to 1 becomes a desired value or more. Is. That is, it should be noted that such control is effective only when the step gain needs to be less than 1 in order to obtain a desired estimation accuracy.

[0016]

As the adaptive algorithm applied to the coefficient updating circuit 220, the learning identification method described above has been researched more than before regarding its control method. For example, reference (1) Yamamoto R. and Kitayama S.:'An Adaptive
Echo Canceller with Variable Step Gain Method ', IE
CE Trans., E65, 1, pp. 1-8 (1979-01). References (2) Harris RW, Chabries DM and Bishop F.
A .: 'A Variable Step (VS) Adaptive Filter Algorithm
m ', IEEE Trans. Acoust., Speech & Signal Process.,
ASSP-34, 2, pp. 309-316 (1986-04). References (3) Tomoko Sawabe, Minoru Matsui, Masafumi Hagiwara, Masao Nakagawa: "Acceleration of Variable Step Adaptive Digital Filters", Theoretical theory
(A), J70-A, 2, pp. 1858-1860 (1977-11). References (4) Ryo Taguchi, Nozomu Hamada: "A Proposal of Variable Step Learning Identification Method", Theological Theory (A), J71-A, 8, pp. 1663-1665 (19)
88-08). References (5) Akihito Kato, Hajime Kubota: "Proposal of variable-step algorithm using fuzzy control", Theory of Communication (A), J73-
A, 11, pp. 1844-1850 (1990-11). References (6) Kensaku Fujii, Toshiro Oga: "Optimum control of modification constants in learning identification method", Theory of Communication (A), J75-A, 6, pp. 975-9
83 (1992-06). There are documents such as.

Each of these documents is related to a step gain control method in which the step gain K is set to a value close to 1 in the initial stage of convergence and then gradually reduced to improve the convergence characteristic.

Problems with such control methods are described in the literature.
It has already been pointed out in (6). That is, reference (6)
According to, the optimum control value of the step gain is shown in Equation 13.
Response residual D_j And disturbance N_j Root mean square of
It has been shown to be a function of value. However, that
Response residual D_j Accurately observe each time in the presence of disturbance
It is difficult to do in reality. Therefore, the above-mentioned literature
In general, including the method proposed in (1) to (6),
As shown in 5, residual E including disturbance_j The response residual D _j Close to
Similar to this residual E_j By using as control information
A method is used to determine the magnitude of the step gain.

[0019]

[Equation 15] E _j = g _j −G _j + N _j There is a problem here. That is, (i) the optimum control value of the step gain is almost 1 until the coefficient updating process begins to saturate.
Since it is fixed to, it is not necessary to control the step gain at the initial stage of coefficient update (see Ref. (6)), and
(ii) Such control becomes effective only after the coefficient updating process starts to saturate, and at that stage, the proportion of the disturbance N _{j in} the residual E _j is large, and the response residual required for the control is large. D
It is pointed out that observation of _j will be difficult.

In order to solve these problems, devise a control method that makes it unnecessary to observe the response residual D _j , or
Alternatively, it is necessary to find a method that enables observation of the response residual D _j . As a method belonging to the former, there is a method proposed in Ref. (6). That is, the difference E _j
The saturation of coefficient update is determined from the power ratio of the response Y _j including the disturbance N _j and the saturation is observed, and then the theoretically calculated optimum step gain is set. However, with this method, the logarithm of the power ratio required for determining saturation, and
It is necessary to calculate the optimum step gain value, and an increase in calculation amount cannot be avoided.

Therefore, although the present invention seeks to provide a method classified into the latter, one of the methods enabling the observation has already been proposed by the present inventors in the literature (7). References (7) Kensaku Fujii, Toshiro Oga: "'Individual Normalization' LMS Method Suitable for Fixed-Point Arithmetic and Its Application to Active Noise Control", IEEJ (A), J78-A, 6, pp. 669. -677 (1995-06).

It adapts the adaptive algorithm to an individual normalization L
MS method ('Individually' Normarized Least Mean Squar
e Algorithm), the number 1 of the individual normalized LMS method
The convergence characteristic is controlled by adjusting the number of addition terms forming the coefficient update equation shown in 6, that is, the block length J. Here, the individual normalized LMS method, by performing coefficient updates every J sampling periods, i.e. is characterized in that a block processing, the H _n m by coefficient update in the n-th block from the start of the operation The obtained coefficient is shown.

[0023]

(Equation 16)

[0024]

[Equation 17]

[0025]

(Equation 18)

The control method proposed by the present inventors for this individual normalized LMS method is the second term A _n (m) / P _{n of the equation} (16).
(M) is an average value [h (m) -H _n
(M)], and its variance can be suppressed by setting the number J of addition terms to be large.

[0027]

[Equation 19]

[0028]

Equation 20] _{Δ n (i) = h (} i) -H n (i) ( where, in the equation 19, Σ _m: i = 1~ except i = m
I) That is, the estimation error [h
(M) −H _n (m)] begins to be buried in the disturbance N _j when the coefficient update begins to saturate. At this point, the operation of increasing the number of addition terms is the estimation error [h (m) − H _n
(M)] has the effect of increasing the calculation accuracy. With this effect, the individual normalized LMS method makes it possible to control the convergence characteristic from the observation of the convergence situation, which was difficult with the learning identification method. It can be easily determined that the estimation error starts to be buried in the disturbance N _j by changing A _n (m) / P _n (m) from the monotonous decrease to the vibration.

The problem in this case is that in such an individual normalization LMS method, the number of addition terms J is set to an integer multiple of the number of taps of the filter in order to keep the amount of calculation as small as the learning identification method. There are restrictions. As a matter of course, such control of the convergence characteristic is unavoidable due to the above-mentioned restrictions. In a system with a large number of taps, the convergence speed of the adaptive filter becomes slow, so that it is necessary to improve the convergence speed. The aforementioned problems are not desirable in such a system.

The present invention has been made to solve the above-mentioned problem, and an object of the present invention is to realize an adaptive filter coefficient which realizes speeding up of convergence and can obtain high estimation accuracy after convergence. To provide an estimation control method and device.

[0031]

A first adaptive filter coefficient estimation control method according to the present invention is a reference signal sent to a signal transmission system having an unknown signal transmission characteristic, and the signal transmission for the reference signal. In estimating the filter coefficient of the non-recursive filter having the filter characteristic that simulates the characteristic of the signal transmission system based on the response output of the system, the coefficient of the filter coefficient of the non-recursive filter is calculated by the addition normalization LMS method. A coefficient calculation step for sequentially and cumulatively calculating the update amount, an observation step for observing the coefficient update amount approximated to the estimation error of the filter coefficient, and a block length for calculating the coefficient update amount in the coefficient calculation step. Is adjusted according to the transition of the coefficient update amount, and based on the calculation result of the coefficient calculation step, It is characterized by having a coefficient update step of performing coefficient updating of filter [Claim 1].

The block length adjusting step includes an extension step of extending the block length for that time until the coefficient update amount observed in the observing step becomes equal to or less than the coefficient update amount by the previous coefficient update calculation. [Claim 2]

In the extension step, a preset lower limit value of the block length may be set as an initial value of the block length, and the coefficient update amount may be determined every time from the initial value [claim 3].

A second adaptive filter coefficient estimation control method according to the present invention provides a reference signal sent to a signal transfer system having an unknown signal transfer characteristic and a response output of the signal transfer system to the reference signal. On the basis of this, in estimating the filter coefficient of the non-recursive filter having the filter characteristic simulating the characteristic of the signal transmission system, the coefficient of the m-th tap is updated to H by the coefficient update in the n-th block.
_n (m), step gain K, number of addition terms J, difference between response output including disturbance and pseudo response E _j , reference signal X _j
As

[0035]

(Equation 21)

[0036]

(Equation 22)

[0037]

(Equation 23)

A coefficient calculation step for sequentially and cumulatively calculating the coefficient update amount A _n (m) / P _n of the filter coefficient of the non-recursive filter, and the coefficient approximated to the estimation error of the filter coefficient. update amount a _n (m) / P and observation step of observing _n, the coefficient updating quantity a _n (m) / P _n is summations number J which is calculated in the coefficient calculating step the coefficient updating quantity a _n ( m) / P _n , a block length adjusting step of adjusting according to the transition, and a coefficient updating step of updating the coefficient of the acyclic filter based on the calculation result of the coefficient calculating step [claim 4] ].

In the block length adjusting step, the observation
Coefficient update amount A observed in step _{n + 1}(M) / P
_{n + 1}Is the coefficient update amount A from the previous coefficient update calculation_n(M)
/ P_nExtend the block length for that time until
An extension step may be included [Claim 5].

In the extension step, a preset lower limit value of the block length may be set as an initial value of the block length, and the coefficient update amount A _n (m) / P _n may be determined from the initial value every time. Claim 6].

A ratio R _{n obtained} by squaring the coefficient update amount A _n (m) / P _n and summing all taps of the non-recursive filter for monitoring and determination of the coefficient update amount in the observation step and the block length adjusting step.

[0042]

(Equation 24)

It is also possible to use [Claim 7]. A first adaptive filter coefficient estimation control apparatus according to the present invention generates a reference signal that gives a reference signal to a signal transfer system having an unknown signal transfer characteristic and a non-recursive filter for simulating the characteristic of the signal transfer system. A unit, an output detection unit that detects a response output of the signal transmission system, and a coefficient calculation unit that sequentially and cumulatively calculates the coefficient update amount of the filter coefficient of the non-recursive filter by the addition normalization LMS method. Observing means for observing the coefficient update amount approximated to the estimation error of the filter coefficient, and adjusting a block length for calculating the coefficient update amount in the coefficient calculating means according to the transition of the coefficient update amount It is characterized by comprising block length adjusting means and coefficient updating means for updating the coefficient of the acyclic filter based on the calculation result of the coefficient calculating means. Claim 8].

The block length adjusting means includes an extending means for extending the block length at that time until the coefficient update amount observed by the observing means becomes less than or equal to the coefficient update amount by the previous coefficient update calculation. It may be [claim 9].

The extending means may be means for making a lower limit value of a preset block length an initial value of the block length and determining the coefficient update amount from the initial value every time.
0].

A second adaptive filter coefficient estimation control apparatus according to the present invention provides a reference signal to a signal transmission system having an unknown signal transmission characteristic and a non-recursive filter for simulating the characteristic of the signal transmission system. A reference signal generation unit, an output detection unit that detects a response output of the signal transmission system, a coefficient of the m-th tap by coefficient update in the n-th block of the acyclic filter, H _n (m), With step gain K, number of addition terms J, difference between response output including disturbance and pseudo response E _j , and reference signal X _j , equations 21-
A coefficient calculation means for sequentially and cumulatively calculating the coefficient update amount A _n (m) / P _n of the filter coefficient of the non-recursive filter based on equation 23, and the approximation error approximated to the estimation error of the filter coefficient and observation means for observing the coefficient updating quantity _{a n (m) / P n} , the coefficient updating quantity of the coefficient updating quantity a _n (m) / P _n is summations number J which is calculated in the coefficient calculating means A block length adjusting means for adjusting according to the transition of A _n (m) / P _n ; and a coefficient updating means for updating the coefficient of the acyclic filter based on the calculation result of the coefficient calculating means. Characterized by [claim 1
1].

In the block length adjusting means, the coefficient update amount An ₊₁ (m) / P _{n + 1} observed by the observing means is the coefficient update amount A _n (m) / P _n obtained by the previous coefficient update calculation. An extension means may be included to extend the block length of the time until the following is true (claim 12).

The extension means may be means for setting a lower limit value of a preset block length as an initial value of the block length and determining the coefficient update amount A _n (m) / P _n from the initial value every time. [Claim 13].

The observing means and the block length adjusting means monitor the coefficient update amount and determine the coefficient update amount A.
_A means for using the ratio R _n of the formula 24 obtained by squaring _n (m) / P _n and summing up all the taps of the non-recursive filter may be used.

In the adaptive filter coefficient estimation control method and apparatus according to the present invention, the coefficient update amount of the filter coefficient of the non-recursive filter is successively cumulatively calculated by the addition normalization LMS method and approximated to the estimation error of the filter coefficient. While observing the coefficient update amount, the block length in which the coefficient update amount is calculated is adjusted according to the transition of the coefficient update amount, and the coefficient update of the non-cyclic filter is performed based on the calculation result of the coefficient update amount. By performing the above, it becomes possible to realize high-speed convergence and to obtain high estimation accuracy after convergence.

[0051]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment of an adaptive filter coefficient estimation control system according to the present invention will be described below with reference to the drawings.

[Embodiment 1] FIG. 1 shows the configuration of the first embodiment of the principle of the adaptive filter coefficient estimation control system according to the present invention.

The adaptive filter coefficient setting control system shown in FIG. 1 uses the characteristic estimation system shown in FIG. 9 and automatically estimates the filter coefficient of the non-recursive filter 210 based on the information of each part of the system. It is a system to do. The estimation control system includes a coefficient computing unit 100, a coefficient updating unit 110, an observing unit 120, and a block length adjusting unit 13.
It has 0.

The coefficient calculator 100 sequentially and cumulatively calculates the coefficient update amount of the filter coefficient of the non-recursive filter 210 by the addition-normalized LMS method. The coefficient update unit 110 uses the acyclic filter 2 based on the calculation result of the coefficient calculation unit 100.
Update the coefficient of 10. The observing unit 120 observes the coefficient update amount approximated to the estimation error of the filter coefficient of the acyclic filter 210. Then, the block length adjusting unit 130
Adjusts the block length in which the coefficient update amount is calculated in the coefficient calculation unit 100 according to the transition of the coefficient update amount.

Now, the principle adopted in this system will be described. In this system, as the adaptive algorithm, the addition-normalized LMS method capable of adjusting the number of addition terms is adopted as in the individual normalization LMS method described above.

Additive Normalized LMS Method ('Summational' Norm
The arized Least Mean Square Algorithm) is a method proposed by the present inventors by reference (8). References (8) Kensaku Fujii, Toshiro Oga: "Continuing update of adaptive filter coefficients in non-voice noise section" IEICE Technical Report, EA94-7
8, 1994-12.

In the addition-normalized LMS method, equations 21 and 22
And Equations 25, 26 and 27, which are similar to Equations 23 and 23, respectively.

[0058]

(Equation 25)

[0059]

(Equation 26)

[0060]

[Equation 27]

That is, in this addition-normalized LMS method,
There is no constraint that the number of addition terms J is an integer multiple of the number of taps. Therefore, in the system shown in FIG.
Is the amount A required for coefficient updating in the addition-normalized LMS method.
Calculate _n (m) and P _n . The coefficient updating unit 110 executes the coefficient updating process according to Equation 25. Observation section 120
_Observes the transition of the coefficient update amount A _n (m) / P _n . The block length adjustment unit 130 determines the block length based on the transition of the coefficient update amount A _n (m) / P _n .

First, in the addition-normalized LMS method, the number J of addition terms has the same effect as the step gain on the convergence characteristic. FIG. 2 is a diagram for comparing the convergence characteristics (the number of taps is I = 512) based on the difference in the number J of addition terms. Here, the effect of the number of addition terms on the convergence characteristic is
For example, doubling the number J of addition terms is a step gain K.
Is equal to 1/2, and therefore, when the response ratio of the unknown signal transmission system and the power ratio of the disturbance are 20 dB, the step gain of the learning identification method necessary to secure the estimation accuracy of 40 dB or more is Characteristics when size K = 1/128
(c) is K = 128, J = 163 in the addition normalized LMS method.
Characteristics when 84 is set (a), or K = 32, J = 4096
It is equivalent to the characteristic (b) when it is said. That is, K in the learning identification method is equal to K / in the addition normalized LMS method.
Equivalent to J.

Also in the addition-normalized LMS method, coefficient updating is performed once per J sampling period as in the above case. Then, the amount added to the coefficient H _n (m) at the time of the (n + 1) _th update, that is, the coefficient update amount A _n (m) / P _n is given by Equations 28 and 29,
A random variable having an average value of Δ _n (m) / I is constructed.

[0064]

[Equation 28]

[0065]

(Equation 29)

As a matter of course, the update amount that should be added to the current coefficient H _n (m) is, as shown in Expression 30, the current coefficient H _n (m) and the impulse response h of the signal transmission system.
The difference from (m) is Δ _n (m).

[0067]

Equation 30] _{Δ n (m) = h (} m) -H n (m) Therefore, reduced the difference Δ _n (m) as a result of the coefficient update, when the proportion of the Q _j (m) is increased, the number The calculation accuracy of the difference Δ _n (m) included in 28 falls, and the coefficient H _n (m)
The update process of will be saturated. At the same time, this means that the coefficient H _n (m) is calculated from the ratio of both components included in Eq.
It means that the convergence situation of can be observed, and that the operation of increasing the number of addition terms can reduce the ratio of Q _j (m) and postpone the saturation of the update process.

FIG. 3 shows the result of observing the transition of the ratio R _n (same as the equation 24) shown in the equation 31 obtained by squaring the coefficient update amount given in the equation 28 and summing up for all taps.

[0069]

[Equation 31]

However, in FIG. 3, K = 64, J = 8192
And other conditions have the same values as in FIG. FIG.
More clearly, the monotonic decrease of the ratio R _n means that the coefficient updating process is still in the initial stage where Δ _n (m) is sufficiently large compared to Q _j (m), on the contrary, the oscillation is , It can be used to judge that it is in a saturation stage where it is necessary to suppress the ratio of Q _j (m) by setting a large number of addition terms.

In this embodiment of the present invention, the number of addition terms is adjusted from the observation of the ratio R _n by utilizing the above-mentioned characteristics, and the convergence speed and high estimation accuracy are simultaneously realized. To do.

A specific control method for determining the number of addition terms from the observation of the ratio R _n will be described below. [Embodiment 2] Here, the ratio R _n obtained by updating the coefficient last time and the ratio R _{n + 1} this time are compared, and the coefficient update can be performed as follows according to the result. it can.

(1) If R _{n + 1} ≤R _n , it is determined that the process is in a monotonous decrease, and coefficient updating is executed without adjusting the number of addition terms. (2) When R _{n + 1} > R _n , it is judged that saturation has started, and coefficient updating is not executed immediately, but the coefficient is increased after increasing the number of addition terms until R _{n + 1} ≦ R _n. Update. That is,
Since the first term of the equation 20 can be kept constant until the coefficient is updated, the decrease of the ratio R _{n + 1} due to the increase of the number of addition terms is the decrease of the second term. On the other hand, since the amount of modification of the coefficient by one update is small, R _{n + 1} and R _n
It can be said that the first term of Formula 20 is almost equal to _n . Therefore, if the number of addition terms is increased until R _{n + 1} ≦ R _n and then the coefficient is updated, the above-mentioned difference Δ _n
The calculation accuracy of (m) will not deteriorate. As a result, the convergence characteristic is not saturated. If the number of addition terms increases automatically when the convergence characteristic is about to reach saturation, the initial value of the number of addition terms can be given small. Since the coefficient convergence is fast with a small number of addition terms, it can be expected to improve the convergence characteristics by such an operation. In this way, the second embodiment of the present invention aims to improve the convergence characteristic.

FIG. 4 shows the configuration of the second embodiment of the adaptive filter coefficient estimation control system according to the present invention. The adaptive filter coefficient estimation control system in FIG. 4 includes a coefficient calculation unit 100, a coefficient update unit 110, and an observation unit 120 similar to those in FIG. 1, and includes a block length adjustment unit 132 slightly different from the case in FIG. ing. 4, the same parts as those in FIG. 1 are designated by the same reference numerals,
The detailed description is omitted here.

That is, the block length adjusting unit 132 of FIG.
Is the ratio R _n observed by the observing unit 120, comparing the previous R _n and the present R _{n + 1,} and until the condition of R _{n + 1} ≦ R _n is reached, the number of addition terms is calculated by the coefficient calculating unit 100. Is linearly and sequentially increased to perform the coefficient calculation, and when R _{n + 1} ≦ R _n , the coefficient update section 110 gives the count value calculated by the coefficient calculation section 100 to update the coefficient. Control.

FIG. 5 shows an example of the convergence characteristic calculated by giving 512 as the initial value of the number of addition terms in the configuration of FIG. 4 and giving other conditions as in the case of FIG. In this case, unlike the individual normalization LMS method in which the increment of the number of addition terms is an integer multiple of the number of taps, the number of addition terms is increased by one until R _{n + 1} ≤R _n. is there. From FIG. 5, it can be seen that the convergence characteristic is greatly improved as compared with the learning identification method in which the step gain is fixed.

[Third Embodiment] In the application of the system according to the present invention, there may actually be an application example in which the power of disturbance changes. In such a case, in the method according to the second embodiment having the configuration shown in FIG. 4, the number of addition terms increased after the increase of the disturbance power does not return to the original value, and hence the improvement of the convergence speed cannot be expected thereafter. The problem arises. Therefore, such an example can be dealt with by adding the following modification. That is, a lower limit may be set for the number of addition terms, and the comparison between R _{n + 1} and R _n may be changed so that the number of addition terms that is the lower limit is again used as the starting point. In this case, if the power of the disturbance increases, the number of addition terms automatically increases, and if the power of the disturbance decreases, the number of addition terms automatically decreases.
In this way, the third embodiment of the present invention rationally improves the convergence speed even when the power of the disturbance changes.

FIG. 6 shows the configuration of the third embodiment of the adaptive filter coefficient estimation control system according to the present invention. The adaptive filter coefficient estimation control system in FIG. 6 includes a coefficient calculation unit 100, a coefficient update unit 110, and an observation unit 120 similar to those in FIG. 1, and is a block length adjustment unit that is slightly different from that in FIG. 1 or 4. 134 is provided. 6, parts similar to those in FIGS. 1 and 4 are denoted by the same reference numerals, and detailed description thereof will be omitted here.

The block length adjusting unit 134 is provided in the block of FIG.
In the length adjusting unit 132, R_{n + 1} And R _n Compare with the previous
Whereas the number of terms to be added was used as the starting point, the previous addition
Regardless of the number of terms, the initial value of the preset number of addition terms is
As a limit value, the repetition ratio starts from this lower limit value each time.
Make a comparison.

FIG. 7 shows an example of the convergence characteristic when the lower limit value of the number of addition terms, that is, the initial value is 512, and the increment / decrement unit is 256. On the way, j = 256 × 256
In the interval of ~ 512x256, the power of disturbance is increased four times and the operation of returning to the original power is added at j = 512x256.
According to FIG. 7, the number of addition terms automatically increases and decreases even in the presence of such a disturbance, and the convergence speed temporarily decreases when the disturbance increases, but when the disturbance decreases, the convergence speed decreases. It can be seen that is starting to increase again. Moreover, it is confirmed that the estimation accuracy is improved without deterioration even in the increase section of the disturbance. Here, in the learning identification method shown for comparison, the convergence characteristic does not deteriorate. The convergence value of the learning identification method obtained for this increased disturbance is the estimation accuracy obtained at this time. Because it is much higher than

In order to clarify this point, FIG. 8 shows an example in which the power of the disturbance is increased at the time when the learning identification method is almost converged. In the case of the learning identification method, the convergent value becomes smaller with respect to this increased disturbance amount, so that a decrease occurs as in the example of FIG. However, in the method according to the present invention, the decrease does not occur, the convergence speed only slightly decreases, and after that, the estimation accuracy starts to increase again.

In this embodiment, the convergence speed is effectively improved even if the disturbance is increased.

[0083]

As described above, according to the present invention,
The coefficient update amount of the filter coefficient of the non-recursive filter is sequentially cumulatively calculated by the addition normalization LMS method, the coefficient update amount approximated to the estimation error of the filter coefficient is observed, and the coefficient update amount is calculated. The block length to be adjusted is adjusted according to the transition of the coefficient update amount, and the coefficient of the acyclic filter is updated based on the calculation result of the coefficient update amount. An adaptive control coefficient estimation control method and apparatus capable of obtaining high estimation accuracy can be provided.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of an adaptive filter coefficient estimation control system according to a first embodiment of the present invention.

FIG. 2 is a diagram for explaining a difference in convergence characteristic due to a difference in the number of addition terms in the system of FIG.

FIG. 3 is a diagram for explaining a system between a convergence characteristic of estimation accuracy and a ratio R _n in the system of FIG.

FIG. 4 is a block diagram showing a configuration of an adaptive filter coefficient estimation control system according to a second embodiment of the present invention.

5 is a diagram for explaining a system between a convergence characteristic of estimation accuracy and a ratio R _n in the system of FIG.

FIG. 6 is a block diagram showing a configuration of an adaptive filter coefficient estimation control system according to a third embodiment of the present invention.

FIG. 7 is a diagram for explaining an example of a convergence characteristic of estimation accuracy when the disturbance increases in the system of FIG.

8 is a diagram for explaining another example of the convergence characteristic of the estimation accuracy when the disturbance increases in the system of FIG.

FIG. 9 is a block diagram showing a configuration of an example of a system for estimating characteristics of a signal transmission system.

[Explanation of symbols]

 100 ... Coefficient computing unit 110 ... Coefficient updating unit 120 ... Observing unit 130, 132, 134 ... Block length adjusting unit

Claims (14)

[Claims] 1. A filter characteristic for simulating a characteristic of the signal transmission system based on a reference signal sent to a signal transmission system having an unknown signal transmission characteristic and a response output of the signal transmission system with respect to the reference signal. In estimating the filter coefficient of the non-recursive filter having, a coefficient calculation step of sequentially and cumulatively calculating the coefficient update amount of the filter coefficient of the non-recursive filter by an addition-normalized LMS method, and an estimation error of the filter coefficient An observation step of observing the coefficient update amount that is approximated to, a block length adjustment step of adjusting a block length in which the coefficient update amount is calculated in the coefficient calculation step according to a transition of the coefficient update amount, and the coefficient calculation step And a coefficient updating step for updating the coefficient of the non-cyclic filter based on the calculation result of Estimation method of controlling adaptive filter coefficients. 2. The block length adjusting step includes an extending step of extending the block length for that time until the coefficient update amount observed in the observing step becomes equal to or less than the coefficient update amount obtained by the previous coefficient update operation. The method for estimating and controlling adaptive filter coefficients according to claim 1, wherein: 3. The adaptation according to claim 2, wherein in the extending step, a lower limit value of a preset block length is set as an initial value of the block length, and the coefficient update amount is determined from the initial value every time. Method for estimating and controlling filter coefficient. 4. A filter characteristic for simulating the characteristic of the signal transmission system based on a reference signal sent to a signal transmission system having an unknown signal transmission characteristic and a response output of the signal transmission system with respect to the reference signal. In estimating the filter coefficient of the non-recursive filter having H, the coefficient of the m-th tap by the coefficient update in the n-th block is set to H
_n (m), step gain K, number of addition terms J, difference between response output including disturbance and pseudo response E _j , reference signal X _j
As, [Equation 2] (Equation 3) A coefficient calculation step of sequentially and cumulatively calculating the coefficient update amount A _n (m) / P _n of the filter coefficient of the non-cyclic filter, and the coefficient update amount A approximated to the estimation error of the filter coefficient. an observation step of observing _n (m) / P _n , and the coefficient update amount A in the coefficient calculation step
a block length adjusting step of adjusting the number of addition terms J by which _n (m) / P _n is calculated according to the transition of the coefficient update amount A _n (m) / P _n ; and the block length adjusting step based on the calculation result of the coefficient calculating step. And a coefficient updating step for updating the coefficient of the non-cyclic filter. 5. The block length adjusting step comprises:
Coefficient update amount A observed in the measurement step_{n + 1}(M) / P_{n + 1}
Is the coefficient update amount A from the previous coefficient update calculation _n(M) /
P_nExtend the block length for that time until
Suitable according to claim 4, characterized in that it comprises long steps.
Method for estimating and controlling adaptive filter coefficient. 6. The extension step uses a preset lower limit value of the block length as an initial value of the block length, and determines the coefficient update amount A _n (m) / P _n from the initial value every time. The adaptive filter coefficient estimation control method according to claim 5. 7. The monitoring and determination of the coefficient update amount in the observing step and the block length adjusting step,
The ratio R _n (equation 4) obtained by squaring the coefficient update amount A _n (m) / P _n and summing all taps of the acyclic filter The adaptive filter coefficient estimation control method according to any one of claims 4 to 6, wherein: 8. A signal transfer system having an unknown signal transfer characteristic and a reference signal generator for applying a reference signal to a non-recursive filter for simulating the characteristic of the signal transfer system, and a response output of the signal transfer system. An output detection unit for detecting; a coefficient calculation unit for sequentially and cumulatively calculating the coefficient update amount of the filter coefficient of the non-recursive filter by an addition-normalized LMS method; An observing unit for observing a coefficient update amount; a block length adjusting unit for adjusting a block length for calculating the coefficient update amount in the coefficient calculating unit according to a transition of the coefficient update amount; An adaptive filter coefficient estimation control apparatus comprising: a coefficient updating unit for updating a coefficient of the acyclic filter based on a calculation result. 9. The block length adjusting means extends the block length at that time until the coefficient update amount observed by the observing means becomes less than or equal to the coefficient update amount by the last coefficient update calculation. 9. The method according to claim 8, further comprising:
15. An adaptive filter coefficient estimation control device according to claim 1. 10. The extension means is means for determining a coefficient update amount each time from a preset lower limit value of the block length as an initial value of the block length. An adaptive filter coefficient estimation control device as described. 11. A signal transfer system having an unknown signal transfer characteristic and a reference signal generator for supplying a reference signal to a non-recursive filter for simulating the characteristic of the signal transfer system, and a response output of the signal transfer system. The output detection unit for detecting, and the coefficient of the m-th tap by coefficient update in the n-th block in the non-recursive filter are set to H.
_n (m), step gain K, number of addition terms J, difference between response output including disturbance and pseudo response E _j , reference signal X _j
As, (Equation 6) (Equation 7) Coefficient calculation means for sequentially and cumulatively calculating the coefficient update amount A _n (m) / P _n of the filter coefficient of the non-cyclic filter, and the coefficient update approximated to the estimation error of the filter coefficient. Observing means for observing the amount A _n (m) / P _n , and the coefficient updating amount A _n (m) / in the coefficient calculating means
Wherein the addition term number J of P _n is calculated coefficient updating quantity A _n (m)
/ P _n , the block length adjusting means for adjusting the coefficient, and the coefficient updating means for updating the coefficient of the acyclic filter based on the calculation result of the coefficient calculating means. Coefficient estimation controller. 12. The method of claim 11, wherein the block length adjusting means, the coefficient updating quantity A _{n + 1} is observed by the observation means (m) / P n + ₁ is the coefficient updating quantity A _n from the previous coefficient update operation (m) 12. The adaptive filter coefficient estimation control device according to claim 11, further comprising extension means for extending the block length at that time until it becomes / P _n or less. 13. The extension means is means for setting a lower limit value of a preset block length as an initial value of the block length and determining a coefficient update amount A _n (m) / P _n from the initial value every time. The adaptive filter coefficient estimation control device according to claim 12. 14. The observing means and the block length adjusting means square the coefficient update amount A _n (m) / P _n to monitor and determine the coefficient update amount, and add the ratios for all taps of the non-recursive filter. R _n [Equation 8] 14. The method according to claim 11, characterized in that
2. An adaptive filter coefficient estimation control device according to claim 1.