JP3444258B2 - Adaptive filter, control method thereof, and storage medium storing program - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an adaptive filter, a control method thereof, and a storage medium storing a program,
In particular, the present invention relates to an echo canceller used for data transmission and an acoustic system, an automatic equalizer for digital data transmission, an adaptive filter generally used for identifying an unknown system, a control method thereof, and a storage medium storing a program.

[0002]

2. Description of the Related Art The principle of a conventional adaptive filter is shown in FIG.
And FIG. 5 will be described. FIG. 4 is a block diagram showing the principle of a conventional adaptive filter. In FIG. 4, a conventional adaptive filter 100 creates an estimated value of an unknown signal sequence from an input reference signal sequence to a known filter,
The parameters of the filter are updated based on the signal sequence of the error between the unknown signal sequence and the estimated value sequence to correctly identify the unknown system 101.

Noise at the time of normal observation is added to this unknown signal sequence. The adaptive filter 100 converges from an unlearned initial state to a final state. The unknown signal sequence is often given as a response to the input reference signal sequence of the unknown system 101, and the case where it is used as an echo canceller or an automatic equalizer corresponds to this.

The adaptive filter is often realized as a non-recursive (FIR) filter, and has a configuration using a plurality of stages of delay devices T as shown in FIG. In FIG. 5, k in the N-tap non-cyclic adaptive filter
The control circuit of the th tap weight is shown.

Here, the correlation value is obtained in the correlator using the sum of the error signal e _n and the additive noise ν _n and the filter input reference signal a _nk at that tap, and the kth tap weight C _k ⁽ⁿ⁾ is obtained. Control. In FIG. 5, n indicates the time, a _n indicates the input reference signal, and α _c indicates the step gain.

Now, as a tap weight control algorithm using a correlation value obtained by multiplying the sum of the error signal e _n and the additive noise ν _{n by} the input reference signal a _nk , the root mean square (LMS) shown below is used. : Least Mean Square) algorithm is well known and widely used in industrial applications. The kth tap weight is represented by the following equation. ^{_{C k (n + 1) =}} C k (n) + α c (e n + ν n) a nk (k = 0,1, ··· N-1) (1)

Next, as a tap weight control algorithm using the correlation value obtained by multiplying the polarity of the sum of the error signal e _n and the additive noise ν _{n by} the input reference signal a _nk , the following sine algorithm is known. ing. C _k ^{(n + 1)} = C _k ⁽ⁿ⁾ + α _c sgn (e _n + ν _n ) a _nk (2)

Here, sgn (.) Is a polarity function and outputs +1 (-1) if the argument is positive (negative).

On the other hand, the normalized LMS algorithm in which normalization is introduced into the LMS algorithm is expressed by the following equation as well known. ^{_{C k (n + 1) =}} C k (n) + α c (e n + ν n) a nk / ‖a (n) || ² (3)

Similarly, an algorithm obtained by performing the following normalization on the signature algorithm is called a normalized signature algorithm. C _k ^{(n + 1)} = C _k ⁽ⁿ⁾ + α _c sgn (e _n + ν _n ) a _nk / ‖a ⁽ⁿ⁾ ‖ ² (4)

[0011] Here, the denominator appearing in the right-hand side of equation (3) and ⁽⁴⁾ (‖a ⁽ⁿ⁾ || ²⁾ is input reference signal vector a
⁽ⁿ⁾ = [a _n , a _n-1 , ... a _{n-N + 1} ] ^T (T is
Transpose is shown) is the square of the norm.

Since the sine algorithm or the normalized sine algorithm uses the polarity of the sum of the error signal and additive noise, it has the advantage that the operation is stable even when a large disturbance such as impulse noise occurs in the additive noise. Have.

The normalized signature algorithm is described in, for example, S. Koike, "Analysis and Performance of Adaptiv.
e Filters Using Normalized Sign Algorithm, "Procee
dings IEEE ISPACS'96, pp.1326-1330, Singapore, No
As described in detail in v. 1996., it has a characteristic that it is insensitive to the fluctuation of the power value of the input reference signal, like the normalized LMS algorithm.

[0014]

However, the sine algorithm or the normalized sine algorithm is LM
When compared with the S algorithm or the normalized LMS algorithm, respectively, there is a drawback that the convergence of the filter is slow. Therefore, while taking advantage of the characteristics of the signature algorithm or the normalized signature algorithm as described above, L
It is desired to realize the fast convergence of the MS algorithm or the normalized LMS algorithm.

The present invention has been made in view of the above demands, and provides an adaptive filter for speeding up the convergence of a sine algorithm or a normalized sine algorithm, a control method therefor, and a storage medium storing a program. To aim.

More specifically, it is an object of the present invention to provide an adaptive filter that is insensitive to fluctuations in the power value of an input reference signal, is resistant to disturbances such as impulse noise, and has a high convergence speed, a control method thereof, and a storage medium storing a program. It is intended.

[0017]

In order to solve the above-mentioned problems, the invention according to claim 1 is provided with a non-recursive filter,
An adaptive method that multiplies the polarity of the sum of the error signal and additive noise by the value at each tap of the input reference signal to obtain a correlation value, and updates and controls the weighting factor of each tap by the product of the obtained correlation value and the step gain. An adaptive filter based on a filtering method,
The step gain is characterized by being obtained by averaging the absolute value of the sum of the error signal and additive noise.

According to a second aspect of the present invention, a non-recursive filter is provided, and the correlation value is obtained by multiplying the polarity of the sum of the error signal and additive noise by the value at each tap of the input reference signal, and the obtained correlation is obtained. An adaptive filter based on an adaptive filtering method in which the weighting coefficient of each tap is updated and controlled by the product of the normalized value obtained by dividing the value by the squared value of the norm of the input reference signal vector, and the step gain is , The absolute value of the sum of the error signal and the additive noise is averaged to obtain.

According to a third aspect of the present invention, a non-recursive filter is provided, the polarity of the sum of the error signal and additive noise is multiplied by the value at each tap of the input reference signal to obtain a correlation value, and the obtained correlation is obtained. A method of controlling an adaptive filter by an adaptive filtering method in which the weight coefficient of each tap is updated and controlled by the product of the value and the step gain, and the step gain is obtained by averaging the absolute value of the sum of the error signal and additive noise. It is characterized in that it is made to obtain.

According to a fourth aspect of the present invention, a non-recursive filter is provided, the polarity of the sum of the error signal and additive noise is multiplied by the value at each tap of the input reference signal to obtain a correlation value, and the obtained correlation is obtained. A method of controlling an adaptive filter by an adaptive filtering method for updating and controlling a weighting coefficient of each tap by a product of a normalization value and a step gain obtained by dividing a value by a square value of a norm of an input reference signal vector, The step gain is characterized by being obtained by averaging the absolute value of the sum of the error signal and additive noise.

According to a fifth aspect of the present invention, a non-recursive filter is provided, the polarity of the sum of the error signal and additive noise is multiplied by the value at each tap of the input reference signal to obtain a correlation value, and the obtained correlation is obtained. A storage medium that stores a control program for an adaptive filter based on an adaptive filtering method that updates and controls the weighting coefficient of each tap by the product of a value and a step gain, and the step gain is an absolute sum of error signals and additive noise. It is characterized in that a control program for averaging the values is stored.

According to a sixth aspect of the present invention, a non-recursive filter is provided, the polarity of the sum of the error signal and additive noise is multiplied by the value at each tap of the input reference signal to obtain a correlation value, and the obtained correlation is obtained. A memory that stores the control program of the adaptive filter that uses the adaptive filtering method to update and control the weighting factor of each tap by the product of the normalized value obtained by dividing the value by the square of the norm of the input reference signal vector and the step gain In the medium, the step gain is characterized by storing a control program which is obtained by averaging the absolute value of the sum of the error signal and additive noise.

[0023]

BEST MODE FOR CARRYING OUT THE INVENTION Next, a storage medium storing an adaptive filter, a control method thereof and a program according to an embodiment of the present invention will be described in detail with reference to the accompanying drawings. 1 to 3, there is shown an embodiment of an adaptive filter according to the present invention, a control method thereof, and a storage medium storing a program.

First, the principle of the adaptive filter according to the embodiment of the present invention will be described. In the adaptive filter according to the embodiment of the present invention, the case of processing using a more complicated normalized sine algorithm will be described.

In the present invention, in order to speed up the convergence of the normalized sine algorithm, the step gain in the above equation (4) is made variable and a leak accumulator as shown in the following equation is used. Control. α _C ^{(n + 1)} = (1-ρ) α _C ⁽ⁿ⁾ + ρ α _C eq | e _n + ν _n | (5)

Where α _C eq is the “equivalent” step gain and ρ is the leakage coefficient.

Subsequently, from the equation (5), the delay in the leak accumulator is removed, and the following equation (6) is given. α _C ⁽ⁿ⁾ ≈ α _C eq | e _n + ν _n | (6)

Therefore, by substituting the equation (6) into the equation (4), it can be seen that the same equation as the equation (3) which is the tap weight updating equation of the normalized LMS algorithm can be obtained. That is,
The normalized sign algorithm realizes a convergence speed close to that of the normalized LMS algorithm.

An embodiment of the present invention based on the above principle will be described below. FIG. 1 is a block configuration diagram showing a schematic configuration of an adaptive filter according to an embodiment of the present invention.
In FIG. 1, an input signal 1 to the adaptive filter according to the embodiment of the present invention is represented by one time slot delay device 11, 12,
The delayed signals 2, 3 and 4 are delayed by 13 and 14, respectively. In particular, the delay signal 3 is the input reference signal _ank at the k-th tap.

The multipliers 21, 22, 23 and 24 have tap weights 31, 3 for the input reference signal a _nk at each tap.
The products obtained by multiplying by 2, 33, and 34 respectively are added by the adder 25, and become the adaptive filter output.

The subtractor 26 subtracts the adaptive filter output from the filter input 5 to obtain the error 6. In general, additive noise is added (e _n + ν _n ) to the error 6.

The error 6 is first given to the polarity detector 111, its polarity (+1 or -1) is detected, and the square norm (| a ⁽ⁿ⁾ calculated by the norm calculator 110 from the input reference signal at each tap. ⁾ ‖ ² ) The divider 11 by 141
Divide by 2 and multiply the quotient by the tap gain 142 separately calculated by the multiplier 113 to obtain the common part output 7.

On the other hand, the absolute value calculator 121 acquires the absolute value (| e _n + ν _n |) of the error 6 and multiplies it by the multiplier 122.
And the leakage coefficient 132 are sequentially multiplied by the multiplier 123 and the multiplier 123, respectively, and input to the adder 125.

The output from the adder 125 is delayed by the one time slot delay device 126 and tap gain (α _C ⁽ⁿ⁾ )
142 is obtained. The tap gain 142 is multiplied by the leakage coefficient complement (1−ρ) 133 by the multiplier 124, and the product is used as the other input of the adder 125.

In FIG. 1, the portion surrounded by the broken line is commonly used for each tap. For the k-th tap, the common part output 7 is multiplied by the input reference signal 3 in the multiplier 41, and the product, that is, the normalized correlation value is input to the adder 42, and the output from the adder 42 is one time. The slot weight is delayed by the slot delay unit 43 to obtain the tap weight 33. The tap weight 33 is used as the other input of the adder 42. The other taps have the same configuration.

When the adaptive filter of the present invention is implemented by a sine algorithm, the polarity detector 1 is not provided without the norm calculator 110 and the divider 112 in FIG.
The polarity detected by 11 is multiplied by the tap gain 142 which is separately calculated by the multiplier 113 to obtain the common part output 7.

FIG. 2 shows the simulation result of the convergence process of the adaptive filter according to the embodiment of the present invention, and shows the case where the normalized sine algorithm is used. In FIG. 2, the broken line shows the convergence of the root mean square error (MSE: Mean Square Error) when the step gain is fixed, and the solid line shows the convergence process when the step gain is variably controlled according to the present invention. . The dotted line shows the convergence when the normalized LMS algorithm is used.

As is apparent from FIG. 2, according to the adaptive filter and its control method of the embodiment of the present invention, the convergence is remarkably accelerated even when the normalized sine algorithm is used, and the normalized LMS algorithm is A close convergence speed can be obtained.

FIG. 3 is a block diagram showing the case where the step gain control of the adaptive filter according to the embodiment of the present invention is performed by the CPU. In FIG. 3, the adaptive filter 100
In the step gain control of, the CPU 300 reads the control program stored in the storage medium 200 and executes the control program.

As the storage medium 200, for example, R
OM, RAM, flash memory, memory card, optical disk, magneto-optical disk, magnetic storage medium, etc. can be used.

The above-described embodiment is a preferred embodiment of the present invention, and various modifications can be made without departing from the gist of the present invention.

[0042]

As is apparent from the above description, according to the adaptive filter of the present invention, the control method thereof, and the storage medium storing the program, the step gain of the adaptive filter is calculated as the sum of the error signal and the additive noise. By averaging the absolute values, the convergence of the sine algorithm or the normalized sine algorithm can be speeded up.

Therefore, it is possible to control the tap weight of the adaptive filter which is insensitive to the fluctuation of the power value of the input reference signal, is resistant to the disturbance such as impulse noise, and has a high convergence speed.

[Brief description of drawings]

FIG. 1 is a circuit configuration diagram showing a step gain control method using an adaptive filter according to an embodiment of the present invention.

FIG. 2 is a diagram showing a simulation result of a convergence process by the adaptive filter according to the embodiment of the present invention.

FIG. 3 is a block diagram showing another embodiment of the adaptive filter according to the present invention.

FIG. 4 is a block diagram showing the principle of a conventional adaptive filter.

FIG. 5 is a diagram showing a tap weight control algorithm of an acyclic adaptive filter.

[Explanation of symbols]

1 Input reference signal 2-4 Delayed signal 5 Filter input 6 error 7 Common section output 11-14 One time slot delay device 21-24 Multiplier 25 adder 26 Subtractor 41 Multiplier 42 adder 43 One time slot delay 110 norm calculator 111 Polarity detector 112 divider 113 multiplier 121 Absolute value calculator 122-124 multiplier 125 adder 126 One Time Slot Delay 131 Equivalent step gain 132 Leakage coefficient 133 Complement of Leakage Coefficient 141 Squared norm 142 step gain

─────────────────────────────────────────────────── ─── Continuation of the front page (58) Fields surveyed (Int.Cl. ⁷ , DB name) H03H 17/00-21/00 H04B 3/04-3/23

Claims (6)

(57) [Claims] 1. A non-recursive filter is provided, wherein a correlation value is obtained by multiplying the polarity of the sum of the error signal and additive noise by the value at each tap of the input reference signal, and the obtained correlation value and step gain are calculated. An adaptive filter according to an adaptive filtering method for updating and controlling a weighting coefficient of each tap by a product, wherein the step gain is obtained by averaging an absolute value of a sum of the error signal and the additive noise. An adaptive filter characterized by the following. 2. A non-recursive filter, wherein a correlation value is obtained by multiplying the polarity of the sum of the error signal and additive noise by the value at each tap of the input reference signal, and the obtained correlation value is obtained as the input reference signal. An adaptive filter according to an adaptive filtering method for updating and controlling a weighting factor of each tap by a product of a normalization value obtained by dividing by a squared value of a vector norm and a step gain, wherein the step gain is the error. An adaptive filter characterized in that it is obtained by averaging the absolute value of the sum of a signal and the additive noise. 3. A non-recursive filter is provided, and the polarity of the sum of the error signal and additive noise is multiplied by the value at each tap of the input reference signal to obtain a correlation value, and the correlation value obtained and the step gain are obtained. A method of controlling an adaptive filter by an adaptive filtering method for updating and controlling a weighting coefficient of each tap by a product, wherein the step gain is obtained by averaging absolute values of a sum of the error signal and the additive noise. A method for controlling an adaptive filter, characterized in that 4. A non-recursive filter is provided, and a correlation value is obtained by multiplying a polarity of a sum of an error signal and additive noise by a value at each tap of the input reference signal, and the obtained correlation value is obtained as the input reference signal. A method of controlling an adaptive filter by an adaptive filtering method for updating and controlling a weighting coefficient of each tap by a product of a normalization value obtained by dividing by a squared value of a vector norm and a step gain, wherein the step gain is A method for controlling an adaptive filter, characterized in that the absolute value of the sum of the error signal and the additive noise is averaged and obtained. 5. A non-recursive filter is provided, and the polarity of the sum of the error signal and additive noise is multiplied by the value at each tap of the input reference signal to obtain a correlation value, and the correlation value obtained and the step gain are obtained. A storage medium storing a control program for an adaptive filter by an adaptive filtering method for updating and controlling a weighting factor of each tap by a product, wherein the step gain is an absolute value of a sum of the error signal and the additive noise. A storage medium storing a program characterized by storing a control program that is obtained by averaging. 6. A non-recursive filter is provided, and a correlation value is obtained by multiplying a polarity of a sum of an error signal and additive noise by a value at each tap of the input reference signal, and the obtained correlation value is obtained as the input reference signal. A storage medium storing a control program for an adaptive filter according to an adaptive filtering method for updating and controlling the weighting coefficient of each tap by a product of a normalization value obtained by dividing by a squared value of a vector norm and a step gain. A storage medium storing a program, wherein the step gain stores a control program that is obtained by averaging an absolute value of a sum of the error signal and the additive noise.