JP3222646B2 - Adaptive filter - 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 whose characteristics can be changed by adaptive control.

[0002] More specifically, the present invention relates to an adaptive filter capable of realizing required characteristics by supplying an appropriate learning input signal.

[0003]

2. Description of the Related Art FIG. 10 shows a configuration of a normal adaptive filter. Here, 51 is a learning input signal source necessary for updating the filter coefficient, 52 is a filter whose coefficient can be controlled, and 53 is the same as an ideal output signal when the learning input signal is supplied to the filter 52. It is an ideal signal source that generates a signal. The learning input signal generated from the learning input signal source 51 preferably contains various signal components not only in the passband of the filter but also in the stopband in order to faithfully realize the filter characteristics. A white signal is preferred.

[0004]

However, when a white input signal is supplied as a learning input signal, it is necessary to calculate an output signal corresponding to the white input signal to generate an ideal output signal. There is a problem that a huge circuit is required.

Accordingly, an object of the present invention is to provide
An object of the present invention is to provide an adaptive filter in which a circuit size is reduced by selecting a specific learning input signal to reduce a manufacturing cost.

[0006]

In order to achieve the above object, according to the present invention, there is provided a filter having a controllable filter coefficient and a learning filter for supplying a learning input signal to the filter. A signal generation unit, an ideal output signal generation unit that generates an ideal output signal obtained when the learning input signal is supplied to the filter on which the adaptive control is completed, and an output signal of the filter and the ideal output signal. On the basis of,
An adaptive algorithm executing means for controlling a coefficient of the filter, wherein the learning signal generating means comprises: a frequency F _{0 at} which the phase shift amount of the filter becomes 0 degree; A means for generating a signal including a component of a frequency F ₀ ± F ₁ that is an integral multiple of 180 degrees
And the ideal output signal generating means operates in a frequency range of F ₀ ± F ₁ .
Ideal output signal by adding the wavenumber component in the frequency components of the F ₀
It is output as a signal . Also, in claim 2
According to the present invention described above, the ideal output signal generating means includes
₀ signal les with reversing the phase of the frequency components of ± F ₁
Attenuates the bell, the ideal is added to the frequency components of the F ₀
It is output as an output signal .

[0007]

According to the present invention, the learning input signals are F ₀ , F ₀ + F
_1, F ₀ -F serves from ₁ consisting of at least three kinds of frequencies of the signal, with respect to the signal of frequency F ₀ is the phase characteristic of the filter is zero, also the frequency F _{0 +} F ₁ at the output terminal of the filter And the phase characteristic of the signal of F ₀ −F ₁ is 1
Out ideal since the select F ₁ such that an integral multiple of 80 degrees
The force signal has inverted the phase of the frequency component of F ₀ ± F ₁
Ri, and or attenuate the signal level, the frequency components of the F ₀
And output as an ideal output signal.
An ideal output signal can be generated with a simple configuration.

[0008]

Embodiments of the present invention will be described below in detail.

Embodiment 1 FIG. 1 is a block diagram showing an entire embodiment of the present invention. In the figure, reference numeral 2 denotes an input signal source that outputs a learning input signal 2A, and 4 denotes a bandpass filter whose characteristics can be changed by an adaptive method. 6 is an amplifier or attenuator for amplifying or attenuating the signal level, and the phase is 18
An ideal output signal 6A consisting of an inverter for shifting by 0 degree
This is a circuit for generating. 8 is an ideal output signal 6A
Is a subtractor that calculates the difference between the output signal and the filter output signal 4A and outputs an error signal 8A.

FIG. 2 shows a specific circuit configuration for generating the learning input signal 2A and the ideal output signal 6A. In the figure, the frequency generating circuit for generating a signal of frequency f ₁ is 10, 11 is a frequency generator for generating a signal of frequency f _0. 12 is a multiplier, because it receives the signal of the frequency f ₀ and f _1, the signal of the frequency f ₀ + f ₁ and f ₀ -f ₁ appears at the output terminal.

Reference numeral 13 denotes an adder. From the output end of the adder, output signals (f ₀ + f ₁ and f ₀ -f ₁ ) of the multiplier 12 are output.
And an addition signal 2A of the signal having the frequency f ₀ .
This addition signal 2A is used as a learning input signal of the filter 4. The input signal source 2 is configured by the above-described 10, 11, 12, and 13.

Next, a circuit configuration for obtaining the ideal output signal 6A will be described.

An inverter 14 shifts the phase of the signals of the frequencies f ₀ + f ₁ and f ₀ -f ₁ by 180 degrees. An attenuator 15 attenuates the output signal level of the inverter 14.

Reference numeral 16 denotes an adder.
_0, f ₀ + f _1, the ideal output signal 6A having a frequency component of f ₀ -f ₁ is output. Where f ₀ + f ₁ , f ₀
Since the frequency component of −f ₁ is attenuated by the attenuator 15, the level is lower by a predetermined amount than the frequency component of f ₀ .

FIG. 3 shows ideal characteristics of the bandpass filter (hereinafter referred to as BPF) 4 shown in FIG. Here, the horizontal axis of FIG. 3A represents frequency, and the horizontal axis represents gain. In FIG. 3B, the horizontal axis represents frequency, and the vertical axis represents the amount of phase shift. Here is the center frequency of BPF4 as f _0. Further, there is a phase of 180 degrees out of signal frequency to that of the center frequency as f ₀ + f ₁ and f ₀ -f _1. Further, the frequency f _{0 for} the signal of the center frequency f ₀
The gain difference between the signals + f ₁ and f ₀ −f ₁ is assumed to be A (dB).

In order to generate the ideal output signal 6A of the filter having such ideal characteristics, the frequency of the input signal must be f ₀
Since the and gain by inverting the phase of 180 degrees at the time of + f ₁ and f ₀ -f ₁ it is sufficient to attenuate by A, as shown in FIG. 2, it may be used an inverter 14 and an attenuator 15.
That is, when three types of frequency signals as shown in FIG. 3 are selected, it can be seen that the ideal output signal 6A of the filter can be easily generated by a simple circuit. And BPF
In the case of 4, there is an advantage that the center frequency, the bandwidth, and the gain at the center frequency can be substantially defined by these three frequencies f ₀ , f ₀ ± f ₁ .

Embodiment 2 FIG. 4 shows another embodiment of the present invention. In FIG.
1 is an ideal input signal source. 22 is the cutoff frequency ω
This is a Gm-C filter having L parameters as parameters relating to Q, M parameters as parameters relating to Q value, and N parameters as parameters relating to gain A.

[0018] 23 is a cut-off frequency omega _p gradient filter about 24 gradient filter for a certain Q value Q _q, 25 is the gain A _r Gm-C from the filter 22 a predetermined output as respective input signal with a gradient filter about Receiving a signal. The configuration of these gradient filters is substantially the same as the original Gm-C filters 23 to 25.

26 is an ideal output signal source, 27 is a subtractor, 2
8 to 30 are multipliers, and 31 to 33 are integrators acting as accumulators.

[0020] In this circuit shown in FIG. 4, the coefficient ω ₁ ... ω _L of all the cut-off frequency

[0021]

Ω _i (t + 1) = ω _i (t) −Δ ・ ε∂H (s) / ∂ω _p (t) (i = 1 to L) (1) Go. Here, ε is an error signal output from the subtractor 27, Δ is a predetermined coefficient, and p is an arbitrary value from 1 to L. Coefficient Q ₁ ... Q _M is about the same as Q value

[0022]

(2) Q _j (t + 1) = Q _i (t) −Δ ・ ε∂H (s) / ∂Q _q (t) (j = 1 to M) (2) Go. Here, q is an arbitrary value from 1 to M. Factor A ₁ ... A _N relates likewise gain A

[0023]

A _k (t + 1) = A _k (t) −Δ · ε · ∂H (s) / ∂A _r (t) (k = 1 to N) (3) Go. Here, r is an arbitrary value from 1 to N.

By doing so, the cutoff frequencies ω and Q
All the circuits for correcting the value and the gain A can be shared, and as a result, the circuit scale can be greatly reduced.

Next, the specific operation of the modification of the present embodiment will be described for a fourth-order BPF circuit.

Here, description will be made using ω _i (i = 1... L). For example, the initial value of each ω _i is ω ₁ , ω ₂ , ω ₃ , ω ₄
And The values of the ideal values of each ω are denoted by ω ₀₁ , ω ₀₂ ,
FIG. 5A shows gain frequency characteristics of this ideal filter as ω ₀₃ and ω ₀₄ . The positions of these ωs are shown in FIG.
(A) is represented by coordinates (marked with a circle) on the X-axis. Further, FIG. 5B shows the characteristics and the position of the ω value when each ω value is larger than the ideal value by a certain magnitude.

Under the situation as shown in FIG. 5B, the circuit shown in FIG. 4 performs an adaptive operation. In this case, irrespective of which parameter ω ₁ to ω _L is used as a parameter representing the gradient filter 23, the ω value is reduced based on the correction algorithm, thereby causing the ω value to approach the characteristic of the ideal filter, and finally, Can make all the ω values coincide with those of the ideal value.

The ideal filter characteristic shown in FIG.
For sex, ω₁ ~ Ω_Four In the ω ₁ Only better than ideal
Small ω_Two ~ Ω_Four Is greater than the ideal value and ω₁ ~ Ω_Four 
Is larger than the average value of the ideal value.

FIG. 6B shows the concept of the frequency characteristic and the location of each ω value at this time. That is, ω ₁ is shifted to the low frequency side, and ω ₂ to ω ₄ are shifted to the high frequency side.
The width of the PF is widened, and the center frequency is closer to the high frequency side.

At this time, due to the adaptive operation of the circuit shown in FIG. 4, ω ₁ moves in a smaller direction in order to bring the overall characteristics of the filter closer to the ideal form even though the initial value is originally smaller than the ideal value. The average value of ω _{1 to} ω ₄ decreases, and the filter characteristics become as shown in FIG. Finally, the average values of ω ₁ to ω ₄ almost coincide with ideal ones.

In such a case, as can be seen from FIG. 6C, the bandwidth becomes wider than the ideal type shown in FIG. 6A. However, this adaptive filter performs adaptive control not only on ω but also on the Q value and the gain A at the same time. Accordingly, the average of the Q value becomes larger than the initial value by the Q value control algorithm, and finally approaches the shape of an ideal filter, and the characteristic as shown in FIG. 6D is obtained.

In the above example, the gradient filter 23
_{Although 1} was used, the result is almost the same even if ω ₂ is used. The reason is that no matter which of the gradient filters of ω ₁ and ω ₂ is used, the result of the adaptive operation is to increase or decrease the average value of ω. This is because, as long as the final goal is to make the error signal ε of ω become zero, the result does not change even if ω ₁ , ω _2, ω ₃ , or ω ₄ is used.

In practice, the final result will not be completely the same, because the phases of the outputs of the gradient filters are slightly different due to the selection of ω ₁ to ω ₄ , respectively. That is,
The updating algorithm of the adaptive filter in the present embodiment is as follows.
By bringing the frequency characteristics closer to the characteristics of the ideal filter, as a result, the average value of the coefficient group of the cutoff frequency Σω
The average value ΣQ _j / M of the coefficient group of _i / L and the Q value and the average value ΣA _k / N of the coefficient group of the gain A are brought close to ideal ones.

As described above, this filter replaces the originally required gradient filter with a gradient filter having the same characteristics. This can greatly reduce the circuit scale,
On the other hand, it may not be possible to completely match the desired filter.

However, the characteristics after learning (after the adaptive processing) are much closer to the ideal values as compared with the initial value characteristics.
In most cases, desired specifications can be satisfied.

In this embodiment, the center frequency f ₀ and f ₀ ± f ₁ located on both sides of the center frequency f ₀ are used as the frequencies included in the input signal for learning. As shown in FIG. 7, if the cutoff frequency of the filter shifts, the gain of f ₀ ± f ₁ shifts with respect to all of the phases, so that the frequency component ω is changed to a correct value as a result of the correction.

If the pass band width is shifted as shown in FIG. 8, the phase and gain of the signal f ₀ ± f ₁ are shifted, so that the Q value is changed to a correct value.

Further, when only the gain is shifted as shown in FIG. 9, only A is changed to a correct value because all the gains are shifted.

As described above, since the number of variable parameters required for the adaptive operation of the filter is the same as the number of frequency components of the input signal for learning, the ideal signal generating circuit 6 (see FIG. 2) is different from the circuit shown in FIG. ) Is applied to operate optimally and efficiently. Further, by using the circuit as shown in FIG. 4, the circuit scale can be further reduced.

[0040]

As described above, according to the present invention, the learning input signal is composed of signals of at least three kinds of frequencies of F ₀ , F ₀ + F ₁ , and F ₀ -F _1.
For a signal of ₀ , the phase characteristic of the filter is zero, and at the output terminal of the filter, the frequencies F ₀ + F ₁ and F ₀ -F
The phase characteristic with respect to ₁ of the signal is configured so as to select the F ₁ as an integer multiple of 180 degrees, making it possible to circuit scale reduction in generating the learning input signal and the ideal output signal of the filter it can.

[Brief description of the drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention.

FIG. 2 is a circuit diagram for generating a learning input signal and an ideal output signal in the present embodiment.

FIG. 3 is a diagram illustrating ideal characteristics of a filter used in the present embodiment.

FIG. 4 is a block diagram showing another embodiment of the present invention.

FIG. 5 is a diagram showing the filter characteristics of an ideal filter and the ω value distribution (A) at that time, and the ω value distribution when the ω value is shifted and the filter characteristics (B) at that time.

FIG. 6 shows the distribution of ω values when the ω value deviates from the ideal filter characteristics (A) and the filter characteristics at that time (B)
And the filter characteristic (C) when the ω value is corrected by the method to which the present invention is applied and the filter characteristic (D) when the ω value and Q value are corrected by the method to which the present invention is applied. FIG.

FIG. 7 is a diagram illustrating filter characteristics when the ω value is shifted.

FIG. 8 is a diagram showing filter characteristics when the Q value is shifted.

FIG. 9 is a diagram illustrating filter characteristics when an A (gain) value is shifted.

FIG. 10 is an explanatory diagram of a conventionally known adaptive filter.

[Explanation of symbols]

 Reference Signs List 2 input signal source 2A learning input signal 4 filter (BPF) 6A ideal output signal 8 subtractor 8A error signal 10,11 frequency generation circuit 12 multiplier 13 adder 14 inverter 15 attenuator 16 adder 21 ideal input signal source 22 Gm-C filter 23, 24, 25 Gradient filter 26 Ideal output signal source 28, 29, 30 Multiplier 31, 32, 33 Integrator

Claims (2)

(57) [Claims] 1. A filter whose coefficient is controllable, a learning signal generating means for supplying a learning input signal to the filter, and when the learning input signal is supplied to the filter for which adaptive control is completed. An adaptive filter comprising an ideal output signal generating means for generating an ideal output signal obtained from the filter and an adaptive algorithm executing means for controlling coefficients of the filter based on the output signal of the filter and the ideal output signal. The learning signal generating means generates a signal including a frequency F _{0 at} which the phase shift amount of the filter is 0 degree and a signal including a frequency F ₀ ± F _{1 at which} the phase shift amount is an integral multiple of 180 degrees. Generating means for generating the ideal output signal , wherein the ideal output signal generating means has a frequency of F ₀ ± F ₁ .
And the ideal output signal by adding the components to the frequency component of the F ₀
An adaptive filter characterized in that it is output as a result. Wherein said ideal output signal generating means, the F ₀
Signal level causes inverting the phase of the frequency components of ± F ₁
Attenuates the Le, out ideally by adding the frequency components of the F ₀
The adaptive filter according to claim 1, wherein the adaptive filter outputs a force signal .