JP3092647B2 - Adaptive filter device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to relates to the adaptive filter device is a kind of digital filter used in the identification or the like of noise canceling or system.

[0002]

2. Description of the Related Art A transversal adaptive filter has means for multiplying a plurality of input signals at different sampling points by coefficients. Each coefficient is updated according to a predetermined algorithm. A typical algorithm for updating coefficients is LM
There is an S (Least Mean Square) algorithm. A variant of this typical LMS algorithm is a normalized LMS
(NLMS) or an algorithm called learning identification method is described in, for example, "Measurement and Control," Vol. 7, No. 9, September 1968.
No. or "IEEE Transaction Automatic Control", Vol. AC-12, which is known in the paper of Noda and Nagumo in June 1967. This normalized LMS algorithm has the advantage that the convergence speed can be improved, but has the disadvantage that the amount of calculation for finding the coefficients is large. As a modification of the normalized LMS,
The above-mentioned paper describes a normalized LMS by the “coding method”. In this coding method, a sign of +1 is used when a plurality of input signals at different sampling points have a value larger than 0, a sign of −1 when the input signal is smaller than 0, and a sign of 0 when the input signal is 0.
, The amount of calculation for updating the coefficient is significantly reduced. However, the convergence speed becomes slow and the convergence error becomes large.

Accordingly, the present invention reduces the amount of calculation for updating the coefficients as in the case of the coding method, and reduces the convergence speed by the normalized LMS.
Another object of the present invention is to provide an adaptive filter device that can increase the speed as in the case of the above and can reduce the convergence error to an intermediate level between the normalized LMS and the coding method.

[0004]

In order to achieve the above object, the present invention provides a reference signal d (t) at a predetermined sampling period.
Signal supply means for supplying a reference signal, an input signal supply means for supplying an input signal x (t) related to the reference signal at a predetermined sampling period, the reference signal supply means and the input signal supply. And the input signal is sequentially delayed with one sampling period as a unit delay time, so that L samples from the 0th to the L-1st samples at different sampling times (where L is a positive integer greater than 2) ), X (t), x (t−1),... X (t
-L + 1) at the same time, and the 0th to L-1th coefficients w0 (t), w1 (t),
.. WL-1 (t) are respectively multiplied, and an output y (t) is obtained by adding the L multiplied outputs obtained thereby, and the output y (t) is obtained.
An error signal e (t) between the input signal x (t) and the error signal e (t) is updated based on the input signal x (t) and the error signal e (t). Value w0 (t + 1),
a digital filter signal processing means for determining w1 (t + 1)... wL-1 (t + 1), wherein the digital filter signal processing means includes the update coefficients w0 (t + 1) and w1 (t + 1). ),
.. WL-1 (t + 1), the coefficient updating means including a coefficient updating means for determining the updating coefficient w0 (t +
1), w1 (t + 1),... Wl-1 (t + 1), wi (t + 1) = wi (t) + β [e (t) / ｛Σ _{j = 0} ^L-1
x ′ (t−j) x (t−j)｝] x ′ (t−i) [where each symbol indicates the following. i and j are positive integers from 0 to L-1; t is a symbol indicating a sampling time; wi (t + 1) is L update coefficients w0 (t + 1);
w1 (t + 1),... wL-1 (t + 1); wi (t) are L coefficients w0 (t), w1 (t),.
Is an arbitrary constant; e (t) is an error signal at sampling time t; x (t-j) is L input signals x (t), x (t-1),. L +
1); x '(ti) and x' (tj) are L input signals x (t) and x (t-t) at different sampling times.
1),... A modified input signal x ′ corresponding to x (t−L + 1)
(T), x ′ (t−1),... X ′ (t−L + 1), where x ′ (t−i) and x ′ (t−j) are: | x (t−i) | ≧ (1 / L) Σ _{j = 0} ^L-1 | x (t−
j) | when sgn [x (ti)] (where sg
n [x (ti)] is +1 and x when x (ti)> 0.
-1 when (ti) <0, 0 when x (ti) = 0
Is shown. | X (t−i) | <(1 / L) Σ _{j = 0} ^L−1 | x (t−
j) is zero when | ｝] Is determined according to an adaptive filter device. Further, the invention according to claim 2 provides a reference signal supply means for supplying a reference signal d (t) at a predetermined sampling period, and an input signal x (t) related to the reference signal at a predetermined sampling period. An input signal supply unit for supplying the reference signal supply unit and the input signal supply unit, and sequentially delaying the input signal with one sampling period as a unit delay time, thereby starting from the 0th different sampling time. L-1 up to L-1th (however,
L is a positive integer greater than 2) input signals x (t), x
X (t-L + 1) are simultaneously obtained, and the 0th to Lth input signals w0 (t), w1 (t),... WL-1 are added to the 0th to Lth input signals. (t), and an output y (t) obtained by adding the L multiplied outputs obtained thereby.
To obtain an error signal e (t) between the output y (t) and the reference signal d (t), and obtain the input signal x (t) and the error signal e (t).
(t), the updated values w0 (t + 1), w1 (t + 1)... wL-1 (t +
1) a digital filter signal processing means for determining 1), wherein the digital filter signal processing means comprises:
1), w1 (t + 1),... WL-1 (t + 1). The coefficient updating means includes: a coefficient updating unit for determining the update coefficients w0 (t + 1), w1 (t + 1),.
(T + 1) is given by wi (t + 1) = wi (t) + βe (t) x ′ (t−
i) [where each symbol indicates the following: i is 0 to L-1
T is a symbol indicating a sampling time; w
i (t + 1) is L update coefficients w0 (t + 1), w1
(T + 1),... WL-1 (t + 1); wi (t) is the L number of coefficients w0 (t), w1 (t),... WL-1 (t) before updating; β is an arbitrary constant; t) is an error signal at sampling time t; x '(ti) is L input signals x (t), x (t-1),... x (t-L + 1) at different sampling times.
, X ′ (t), x ′ (t−1),
... x '(t-L + 1), where x' (ti)
Is | x (t−i) | ≧ (1 / L) Σ _{j = 0} ^L−1 | x (t−
j) | (j is a positive integer from 0 to L-1)
gn [x (t−i)] (where sgn [x (t−i)
i)] is +1 when x (t−i)> 0, and x (t−i) <
0 indicates −1, and x (t−i) = 0 indicates 0. | X (t−i) | <(1 / L) Σ _{j = 0} ^L−1 | x (t−
j) is zero when | ｝] Is determined according to an adaptive filter device. In the present application, Σ _{j = 0} ^L-1 is x ′ (t−
j) L in which j of x (t−j) is changed from 0 to L−1
値_{j = 0} ^L-1 | x
(T−j) | indicates that the sum of L values obtained by changing j of | x (t−j) | from 0 to L−1 is obtained.

[0005]

The basic operation method of the present invention is the same as that of the coding method, and sgn [x
(T-i)]. However, sgn [x (ti)] is not used for all input signals. The above (1 / L) Σ _{j = 0} ^L−1 | x (t−j) | is L input signals x (t) and x at different sampling times
The average value of the absolute values of (t-1),... X (t-L + 1) is shown. When the average value is used as a threshold value and the input signal is smaller than the threshold value, the value of x ′ (t−i) is calculated by the sign function Sg + [x (t (t)).
-I)] to zero without determination. By performing such processing, the convergence speed becomes faster than that of the conventional coding method. Also, the convergence error (adaptive error) is calculated using the conventional normalized LMS.
, But smaller than conventional coding. The amount of calculation is almost the same as that of the conventional coding method. Therefore,
The method according to the present invention may be such that the convergence error may be intermediate between the conventional normalized LMS and the conventional coding method. However, the method is suitable for an adaptive filter device that requires a small amount of calculation and a fast convergence speed. It is possible to provide a high-performance noise canceller or the like despite the price. Also,
The second aspect of the invention applies the same technical idea as the first aspect of the invention to LMS (Least Mean Square: Least Mean Square Algorithm), and further simplifies the operation than the first aspect of the invention. Can be.

[0006]

First Embodiment Next, an adaptive filter device according to a first embodiment of the present invention will be described with reference to FIGS. FIG. 1 shows a corresponding filter device for modeling an unknown system. An unknown dynamic system 2 is connected to an input terminal 1 for supplying an input signal x (t) at a predetermined sampling period. . The unknown dynamic system 2 equivalently includes a second-order delay element 3, a noise source 4, and an adder 5 that adds the output of the delay element 3 and the noise of the noise source 4. This unknown dynamic system 2 outputs a desired response, that is, a reference signal d (t) at a predetermined sampling period.

The input terminal 1 is also connected to a transversal type adaptive filter 6. The transversal type adaptive filter 6 includes L-1 (three) delay elements 7, 8, 9 as delay means for sequentially delaying the input signal x (t) with one sampling period as a unit delay time. , A tap line 10 connected to the input terminal 1 and a delay element 7,
Tap lines 11, 1 connected to output terminals 8 and 9
2, 13 and L connected to each of the tap lines 10 to 13
(Four) coefficient multipliers 14, 15, 16, 17; adders 18, 19, 20 for sequentially adding the outputs of the multipliers 14, 15, 16, 17; Become.

The input signal x (t) at the sampling time t is sent to the first tap line 10 as it is. The output terminals of the delay elements 7, 8, 9 are connected to the input signal x (t-1) at the immediately preceding sampling time (t-1) and the input signal x (t) at the previous sampling time (t-2). -2) The input signal x (t-
3) is obtained. When there are L tap lines in total, a signal of x (t−L + 1) is obtained on the Lth line.

The multipliers 14, 15, 16 and 17 are connected to a coefficient updating means 21, and a coefficient w given from the multiplier is provided.
0 (t), w1 (t), w2 (t) and w3 (t) are converted into four input signals x (t), x (t-1), x (t−
2) and x (t−3) are respectively multiplied, and w0 (t) x (t),
w1 (t) x (t-1), w2 (t) x (t-2), w3 (t) x
(T-3) is transmitted.

The three adders 18, 19, and 20 determine the sum of the outputs of the four multipliers 14, 15, 16, and 17, and provide y (t) = w0 (t) x (t) + w1 (t ) x (t-1) + w2 (t) x (t-2) + w3 (t) x (t-3) Expression (1) is output.

An error operation circuit 22 connected to the unknown dynamic system 2 and the adder 20 executes an operation of d (t) -y (t) and outputs an error (error) signal e (t).

The coefficient updating means 21 is connected to the error calculation circuit 22, the input terminal 1, and the output terminals of the delay elements 7, 8, and 9 in order to obtain information used for coefficient determination.

In FIG. 1, the transversal type adaptive filter 6 is shown functionally (equivalently) by an individual circuit in order to make it easy to understand, but actually, the filter 6 and the error calculating circuit 22 are digital signals. It is composed of a processing circuit or a microprocessor (microcomputer).

The coefficient updating means includes tap lines 10 to 13
When the number of multipliers 14 to 17 is L in general, the following equation is calculated. wi (t + 1) = wi (t) + β [e (t) / ｛Σ _{j = 0} ^L-1
x ′ (t−j) x (t−j)｝] x ′ (t−i) Formula (2) Each symbol of this formula (2) is defined in the claims and the section of the means for solving the problem. It is on the street. In the embodiment, i and j take values of 0, 1, 2, and 3, respectively.

Expression (2) is expressed by the following expression in a vector format generally used in the field of adaptive filters. W (t + 1) = W (t) + β [e (t) / {X ′ ^T (t) X (t)}] X ′ (t) Equation (3) where t is a symbol indicating a sampling time, W (t) is a coefficient vector indicating L coefficients at the sampling time t, and W (t + 1) is L at the sampling time t + 1.
Coefficient vector indicating the number of coefficients, β is an arbitrary constant, e (t)
Is the error signal at sampling time t, X (t) is a vector indicating L input signals at different sampling times, and X ′ (t) is L input signals ｛x (t), x ( t-1)... x (t−L + 1)} L modified input signals {x ′ (t), x ′ (t−
1)... A vector representing x ′ (t−L + 1)} where an arbitrary modified input signal selected from the L modified input signals is x ′ (ti) (where i is 0 to L− X (t−i) | ≧ (1 / L) Σ _{j = 0} ^L−1 | x (t−
j) | (where j is a numerical value from 0 to L-1)
Is satisfied, x '(ti) = sgn [x (ti)] (where sgn [x (ti)] is x (ti)>
+1 when x is 0, -1 when x (t-i) <0, x (t-
i) The sign of 0 is shown when = 0. | X (t−i) | <(1 / L) Σ _{j = 0} ^L−1 | x (t−
j) | holds when x ′ (t−i) = 0. ｝ X ' ^T
(t) is the transposed vector of the vector X '(t).

The four coefficients w0 (t), w1 (t),
The update coefficients of w2 (t) and w3 (t) are shown below according to equation (2). w0 (t + 1) = w0 (t) + β [e (t) / A] x0 (t) w1 (t + 1) = w1 (t) + β [e (t) / A] x1 (t-1) w2 (t + 1) = W2 (t) + β [e (t) / A] x2 (t-2) w3 (t + 1) = w3 (t) + β [e (t) / A] x3 (t-3) Equation (4) A in equation (4) is ｛Σ _{j = 0} ^L−1 x ′ (t−
j) x (t-j)}.

FIG. 2 is a block diagram functionally (equivalently) showing the coefficient updating means 21 of FIG. 1 for executing the operation of equation (4). The coefficient updating means 21 includes four input terminals 31, 32, 33, and 34 connected to the input terminal 1 and the output terminals of the delay elements 7, 8, and 9 in FIG.
And an error signal input terminal 35 connected to the terminal. The average value calculation circuit 36 connected to the four input terminals 31, 32, 33, and 34 outputs the signals x (t) of the input terminals 31 to 34,
With x (t-1), x (t-2) and x (t-3) as inputs, an average value of these absolute values is calculated. That is, (1 /
L) Σ _{j = 0} ^L−1 | x (t−j) | = xa (t) Perform the operation.

The four decision circuits 37, 3 connected to the input terminals 31, 32, 33, 34 and the average value calculation circuit 36
8, 39 and 40 are input signals x (t), x (t-1), x
(T-2), x (t-3) is equal to or more than the average value xa (t),
Or if the input signal is greater than the average value, then determine whether the input signal is positive, negative or zero, and generate a sign of +1 or -1 or 0 . That is, the judgment circuits 37 to 40 determine x0 '(t), x1' (t),
a circuit for determining x2 '(t) and x3' (t), where x (t) ≥xa (t) x (t-1) ≥xa (t) x (t-2) ≥xa (t) x When (t-3) ≥xa (t) and x (ti) is a positive value, +1 is generated, and when x (ti) is a negative value, -1 is generated.
When (ti) is zero, a zero is generated. Also, x
When (ti) <xa (t), zero is generated.

The four multipliers 41, 42, 43 and 44 connected to the input terminals 31, 32, 33 and 34 and the decision circuits 37, 38, 39 and 40 respectively provide x '(t-
i) execute the operation of x (t−i), and x ′
(t) x (t), x '(t-1) x (t-1), x' (t-
2) Output x (t-2) and x '(t-3) x (t-3). Note that x '(t), x' (t-1), x '(t-
2) Since x '(t-3) is +1 or -1 or 0, this multiplication can be easily performed.

The three adders 45, 46 and 47 connected to the multipliers 41 to 44 are for obtaining A in the equation (4), where A is x '(t) x (t) + x' (t- 1) (t-
1) + x '(t-2) x (t-2) + x' (t-3) x
(T−3) is sent to the divider 48.

A β multiplier 49 connected to the error signal input terminal 35 multiplies the error signal e (t) by a constant β and sends the result to a divider 48. β is, for example, 0.5.

The divider 48 calculates β [e (t) / A] in the equation (2).
Is sent to the multipliers 50, 51, 52, 53.

The multipliers 50, 51, 52 and 53 are connected to the decision circuits 37, 38, 39 and 40 and also to the divider 48, and the second term x '(t) on the right side of the equation (2) −
i) is multiplied to obtain ｛βe (t) / A｝ x '(t) ｛βe (t) / A｝ x' (t-1) ｛βe (t) / A｝ x '(t-2 ) ｛Βe (t) / A｝ x '(t-3) is output.

The four memories 54, 55, 56 and 57 store the coefficients at the current time point t. That is, equation (2)
W0 (t), w1 (t), w2 (t),
w3 (t) is stored.

The four adders 58, 59, 60, 61 correspond to the outputs of the multipliers 50, 51, 52, 53 and the memories 54, 5, 53.
5, 56, and 57, and outputs an added value of these outputs to output terminals 62, 63, 64, and 65. The four output terminals 62, 63, 64, and 65 have w0 (t +) corresponding to the update coefficient wi (t + 1) in equation (2).
1), w1 (t + 1), w2 (t + 2), w3 (t +
3), which are the multipliers 14, 15, 1 in FIG.
6 and 17 are sent as coefficients at the next sampling time (t + 1).

FIG. 3 shows a procedure executed by the coefficient updating means 21 according to a predetermined program for coefficient updating.
First, when starting in block 70, in block 71, input signals x (t), x (t−1), x (t−2), x
(T-3) is obtained. Next, in block 72, the average value xa (t) of the absolute values of the input signals is obtained. .

Next, as shown in block 73, it is determined whether or not the absolute value | x (t) | of the input signal is equal to or greater than the average value xa (t). If this value is greater than the average value xa (t), it is determined whether x (t) is greater than or less than zero, as shown in block 74. When x (t)> 0, +1 is output as x '(t) as shown in block 75, and when x (t) <0, -1 is output as x' (t) as shown in block 76. . On the other hand, when it is determined in block 73 that | x (t) | is smaller than the average value xa (t), zero is output as x '(t) as shown in block 77. The value of x '(t) is temporarily stored in a memory.

As shown in blocks 78, 79 and 80,
The values of x '(t-1), x' (t-2), and x '(t-3) are sequentially obtained in the same manner as in blocks 73 to 77.

Next, it is determined whether or not x '(t) = 1, as indicated by a block 81 in FIG. When x '(t) = 1, the value of x' (t) x (t) is x
(t) is obtained. On the other hand, if x '(t) is not 1, it is determined at block 83 whether x' (t) =-1 or not, and YES
In this case, -x (t) is obtained as x '(t) x (t) as shown in block 84. When the output of the block 83 is NO
In this case, zero is obtained as x '(t) x (t) as shown in block 85.

Next, as shown in blocks 86, 87 and 88, the value of x '(t-1) x (t-1), x' (t-
2) The value of x (t-2) and the value of x '(t-3) x (t-3) are obtained in the same manner as in blocks 81 to 85.

Next, x 'obtained in blocks 81 to 85
(t) The value A is obtained by adding the value of x (t) and the values obtained in blocks 86, 87 and 88 as shown in block 89.

Next, as shown in block 90, βe (t)
/ A. Next, as shown in block 91, βe
x '(t), x' (t-1), x 'for (t) / A
(T-2), x '(t-3) multiplied by the value (+1, -1 or 0), and βe (t) / A {x' (t)}, βe (t) /
A {x '(t-1)}, βe (t) / A {x' (t-
2) Find {, βe (t) / A {x '(t-3)}.

Next, as shown in a block 92, w0 (t), w0 (t-1), w0
(T-2) and w0 (t-3) are added to obtain an update coefficient w0
(T + 1), w1 (t + 1), w2 (t + 1), w3
(T + 1) is obtained and stored in a memory as shown in a block 93, and is used as a new coefficient at time t + 1 in the multipliers 14 to 17 in FIG. Then, block 7 in FIG.
The operation returns to 1 and the operation in the next sampling cycle is repeated.

As is apparent from the above description, in the present embodiment, the average value xa of the absolute values of the input signals at different sampling times.
(t) is calculated, and the calculation of the average value xa (t) is relatively easily achieved. That is, when calculating the average value, it is necessary to calculate the sum of the four input terminals 31, 32, 33, and 34, but the four input signals x (t) and x (t− 1), x (t-2), x (t-
This averaging is easily accomplished because the oldest one of the input signals x (t-3) of 3) is only replaced by the next new input signal. Also, x '(t), x'
Since (t-1), x '(t-2), and x' (t-3) are +1 or -1 or 0, the calculation on the right side of Expression (2) can be easily achieved. That is, the amount of calculation for updating the coefficient in the method according to the present invention is substantially equal to that in the conventional coding method.

Next, the method of the present invention and the conventional normalized LMS
And the relationship with the encoding method will be described in more detail. An arithmetic expression for updating the coefficient of the conventional normalized LMS corresponding to Expression (3) is as follows. W (t + 1) = W (t) + β [e (t) / {X (t)} ² }
X (t) In this operation expression, there is an operation of {X (t)} ² , so that the operation amount is larger than that of the method of the present invention. Further, an arithmetic expression for updating a coefficient in the conventional coding method is as follows. W (t + 1) = W (t) + β [e (t) / ｛X ′ ^T (t) X
(t)｝ X ′ (t)] where X ′ (t) = sgnx (ti), and
It is +1 when x (t−i)> 0, −1 when x (t−i) <0, and 0 when x (t−i) = 0. Comparing the coding method with the method of the present invention, in the present invention, an average value xa (t) is obtained, and when the input signal is smaller than the average value xa (t), a step of setting X '(t) to zero is added. The two are different in that they have been introduced. Therefore, as described above, there is not much difference between the present invention and the coding method in the amount of calculation.

In order to confirm the effect of the adaptive filter according to the present invention, white noise is added to the input terminal 1 of the apparatus of FIG.
When the convergence characteristic was examined by adding white noise also from the noise source 4, the result shown in FIG. 5 was obtained. In FIG. 5, C1
Is the characteristic of the method according to the invention, C2 is the conventional normalized LMS
And C3 indicate the characteristics of the conventional coding method. As is clear from the results, according to the present invention, the convergence, that is, the estimation error (convergence error) is almost equal to that of the normalized LMS, and is slightly improved over the conventional coding method. In FIG. 5, the abscissa indicates the sample number, and the ordinate indicates E = ｛‖Hw (t).
‖ ^2} / ‖H‖ indicating an estimated error indicated by ^2. However,
H in the above equation represents an unknown parameter, W represents its estimated value, and the initial value of the estimated value is zero. In addition, white noise is applied to the input terminal 1 of FIG. 1 by 1 / (1−0.95).
When the convergence was examined in the same manner as in FIG. 5 by adding the colored noise obtained through the IIR type digital filter represented by z ^-1 ), the result shown in FIG. 6 was obtained. As is clear from FIG. 6, the method of the present invention has better convergence than the conventional coding method when the input signal is colored noise.

The method according to the invention, a conventional normalized LMS,
For each of the conventional coding methods, the mean square value of the adaptive error was obtained after convergence in 32 measurements. In the case of white noise, 0.013, 0.010, 0.0
22, 0.012, 0.0 for colored noise
11, and 0.018. As is apparent from these results, in the adaptive error, the method of the present invention is almost equivalent to the conventional normalized LMS, and is superior to the conventional coding method.

[0038]

Second Embodiment Next, referring to FIG. 7, a second embodiment of the present invention will be described.
The coefficient updating means and method according to the embodiment will be described. The basic configuration of the adaptive filter device of the second embodiment is the same as that of FIG. 1, and the coefficient updating means 21 is modified as shown in FIG. In the coefficient updating unit of FIG. 7, the same parts as those of FIG. As is clear from the comparison between FIG. 2 and FIG. 7, the circuit of FIG.
This corresponds to a circuit in which the multipliers 41, 42, 43, 44, adders 45, 46, 47, and a divider 48 are omitted from the circuit of FIG.

The coefficient updating means shown in FIG. 7 is configured to obtain Wi (t + 1) = Wi (t) + βe (t) x ′ (ti). Each symbol in this formula is defined in the claims. In order to obtain the above, terminals 31 to 34 are assigned to the coefficient calculating means as x (ti).
X (t), x (t-1), x (t-2), x (t-
3) is input, and e (t) is input to the terminal 35.

The coefficient updating in the second embodiment is based on LMS (Least Mean Square Algorithm), and the operation is further simplified than in the first embodiment. Therefore, the first
As in the case of the third embodiment, it is possible to provide a noise canceller and the like with good performance despite its low price.

[0041]

[Modifications] The present invention is not limited to the above-described embodiment, and for example, the following modifications are possible. (1) The adaptive filter can be configured by a combination of the individual circuits shown in FIG. (2) It is also possible to adopt a configuration in which the unknown dynamic system 2 is not connected to the input signal 1.

[Brief description of the drawings]

FIG. 1 is a block diagram functionally showing an adaptive filter device according to a first embodiment of the present invention.

FIG. 2 is a block diagram functionally showing a coefficient updating unit of FIG. 1;

FIG. 3 is a flowchart showing a coefficient updating operation.

FIG. 4 is a flowchart showing a coefficient updating operation following FIG. 3;

FIG. 5 is a diagram showing convergence when white noise is input.

FIG. 6 is a diagram showing convergence when colored noise is input.

FIG. 7 is a block diagram functionally showing a coefficient updating unit according to the second embodiment.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Input terminal 7-9 Delay element 22 Error calculation circuit

Continuation of the front page (56) References JP-A-55-52646 (JP, A) JP-A-62-105535 (JP, A) JP-A-1-151830 (JP, A) JP-A-2-65310 (JP) JP-A-5-160676 (JP, A) JP-A-5-145379 (JP, A) IEICE Transactions A J76-A [3] (1993-3) p. 297-302 (58) Field surveyed (Int. Cl. ⁷ , DB name) H03H 21/00 G06F 17/10 H03H 17/06 635 JICST file (JOIS)

Claims (2)

(57) [Claims] 1. A reference signal d at a predetermined sampling period.
(t), input signal supply means for supplying an input signal x (t) having a relationship with the reference signal at a predetermined sampling cycle, and the reference signal supply means and the reference signal supply means. It is connected to the input signal supply means, and sequentially delays the input signal with one sampling period as a unit delay time, so that L samples from the 0th to the L-1st at different sampling times (where L is 2
Input signals x (t), x (t-
1),... X (t−L + 1) are simultaneously obtained, and the 0th to L−1th coefficients w0 are added to the 0th to Lth input signals.
(t), w1 (t),..., wL-1 (t) are respectively multiplied, and an output y (t) obtained by adding the L multiplied outputs obtained thereby is obtained. An error signal e (t) from the reference signal d (t) is obtained, and the input signal x (t) and the error signal e (t) are obtained.
, W1 (t + 1),... WL-1 (t + 1), the updated values w0 (t + 1), w1 (t + 1),.
A adaptive filter device having a digital filter signal processing means for determining, said digital filter signal processing means, the update coefficient
w0 (t + 1), w1 (t + 1),... wL-1 (t + 1)
Coefficient updating means for determining the coefficient
The stages include the update coefficients w0 (t + 1), w1 (t + 1),.
-1 (t + 1) is given by wi (t + 1) = wi (t) + β [e (t) / ｛Σ _{j = 0}
^L−1 x ′ (t−j) x (t−j) ′] x ′ (t−i) [where each symbol indicates the following. i and j are positive integers from 0 to L-1; t is a symbol indicating a sampling time; wi (t + 1) is L update coefficients w0 (t + 1), w1
(T + 1),... WL-1 (t + 1); wi (t) is the L coefficients w0 (t), w1 (t),.
wL-1 (t); β is an arbitrary constant; e (t) is an error signal at sampling time t; x (tj) is L input signals x (t), x (t- X '(ti) and x' (tj) are L input signals x (t), x (t-1), ... x at different sampling times.
A modified input signal x ′ (t) corresponding to (t−L + 1),
x ′ (t−1),... x ′ (t−L + 1), where x ′ (t−i) and x ′ (t−j) are: | x (t−i) | ≧ (1 / L) Σ _{j = 0} ^L-1 | x (t−
j) | when sgn [x (ti)] (where sg
n [x (ti)] is +1 and x when x (ti)> 0.
-1 when (ti) <0, 0 when x (ti) = 0
Is shown. | X (t−i) | <(1 / L) Σ _{j = 0} ^L−1 | x (t−
j) is zero when | Shall be determined in accordance}]
Adaptive filter device, characterized in that it. 2. A reference signal d at a predetermined sampling period.
(t), an input signal supply means for supplying an input signal x (t) having a relationship with the reference signal at a predetermined sampling cycle, and the reference signal supply means and the reference signal supply means. It is connected to the input signal supply means, and sequentially delays the input signal with one sampling period as a unit delay time, so that L samples from the 0th to the L-1st at different sampling times (where L is 2
Input signals x (t), x (t-
1),... X (t−L + 1) are simultaneously obtained, and the 0th to L−1th coefficients w0 are added to the 0th to Lth input signals.
(t), w1 (t),..., wL-1 (t) are respectively multiplied, and an output y (t) obtained by adding the L multiplied outputs obtained thereby is obtained. An error signal e (t) from the reference signal d (t) is obtained, and the input signal x (t) and the error signal e (t) are obtained.
, W1 (t + 1),... WL-1 (t + 1), the updated values w0 (t + 1), w1 (t + 1),.
, WL-1 (t + 1), w1 (t + 1),..., W1 (t + 1),... WL-1 (t + 1).
, WL (t + 1),... WL-
1 (t + 1) is given by wi (t + 1) = wi (t) + βe (t) x ′ (t−
i) [where each symbol indicates the following: i is a positive integer from 0 to L-1; t is a symbol indicating a sampling time; wi (t + 1) is L update coefficients w0 (t + 1), w1
(T + 1),... WL-1 (t + 1); wi (t) is the L coefficients w0 (t), w1 (t),.
wL-1 (t); β is an arbitrary constant; e (t) is an error signal at sampling time t; x '(ti) is L input signals x (t), x (t −1),... X (t−L + 1), corresponding to modified input signals x ′ (t), x ′ (t−1),.
x ′ (t−L + 1), where x ′ (t−i) is | x (t−i) | ≧ (1 / L) _{} j = 0} ^L−1 | x (t−
j) | (j is a positive integer from 0 to L-1)
gn [x (t−i)] (where sgn [x (t−i)
i)] is +1 when x (t−i)> 0, and x (t−i) <
0 indicates −1, and x (t−i) = 0 indicates 0. | X (t−i) | <(1 / L) Σ _{j = 0} ^L−1 | x (t−
j) is zero when |適 応], the adaptive filter device being determined in accordance with [1].