KR100902954B1 - Active noise control system and method in enclosed field of 3-dimension using c0rrelation filtered-x least mean squares algorithm - Google Patents

Abstract

An active noise control system and method in enclosed field of 3-dimension using correlation filtered-x least mean squares algorithm are provided to improve the convergence performance of the FXLMS algorithm of DSP. The Co-FXLMS algorithm comprises an adaptive filter, an output element and a addition route estimating filter. The adaptive filter adapts actively according to the ambient conditions. The output element outputs the first route transfer function and addition route transfer function. An estimating value filter inputs the estimating value of the addition route transfer function. The actuator obtains the error signal by adding the first route transfer function and addition route transfer function. The cross correlation least-mean-square(100) controls the coefficient vector of the adaptive filter. The cross correlation least-mean-square extracts the cross correlation of the reference signal in real time.

Description

ACTIVE NOISE CONTROL SYSTEM AND METHOD IN ENCLOSED FIELD OF 3-DIMENSION USING C0RRELATION FILTERED-X LEAST MEAN SQUARES ALGORITHM}

The present invention relates to an active noise control system and method of a three-dimensional hermetic system, and more particularly, to actively suppress internal noise generated in a three-dimensional hermetic system using a Corelation Filtered-X Least Mean Squares (Co-FXLMS) algorithm. The present invention relates to an active noise control technology in a three-dimensional closed system that can be controlled.

In particular, the present invention develops a Co-FXLMS algorithm that updates the convergence coefficient in real time by using a correlation between an error signal and a filtered reference signal based on the FXLMS (Filtered-X Least Mean Squares) algorithm. The present invention relates to a technique that can be extended to a three-dimensional hermetic system by improving the convergence performance of a changing reference signal and applying the FXLMS algorithm, which is mainly applied only to a conventional one-dimensional duct.

In general, attenuating noise from a noise source includes passive noise reduction techniques for installing sound absorbers, sound insulation materials, or resonators on the paths of noise generation, and active noise minimization using secondary noise sources. There is a noise reduction technique.

However, the passive noise reduction method has a high effect in the high frequency region of 500 Hz or more, but the effect is sharply reduced in the low frequency region of 500 Hz or less. Therefore, in order to reduce long-wave low-frequency noise, it is necessary to install several thick sound absorbers, sound insulation materials, or bulky resonators, but in this case, cost increases and space constraints are caused by the increase in volume.

In order to supplement the problems of the passive noise reduction method, research on active noise control (hereinafter referred to as 'ANC') that reduces noise by using a secondary source has been highlighted.

The ANC technique detects the noise signal generated from the noise source by the adaptive signal processing technique using the FXLMS algorithm in the Digital Signal Processor (DSP) for high-speed signal processing. In addition, by generating an additional signal of an antiphase to reduce the noise, it uses the principle of actively reducing the noise by using these wave interference. This is more effective than conventional passive noise reduction methods in reducing low frequency noise below 500 Hz, such as fan noise in heating, ventilating, and air conditioning (HVAC) systems and rotor noise in automobile engines. It is known.

This ANC technique can be divided into problems in the one-dimensional sound field that can handle the sound waves as plane waves and problems in the three-dimensional sound field that cannot be treated by the plane waves. ANCs in one-dimensional acoustic spaces such as (Exhaust System) and Intake System have been mainly studied.

However, it is unreasonable to apply the noise reduction device by the conventional ANC technique for noise control in a place such as an intake machine of a vehicle. The reason is that the FXLMS algorithm, which is used to make the control sound in the DSP, has its stability and convergence rate depending on the convergence coefficient and the filtered input vector, and also performs the adaptive control using the fixed convergence coefficient. In fact, the magnitude and frequency of the noise emitted from the intake system of a vehicle changes over time, and the emission noise includes various harmonic components. Therefore, when adaptive control is performed using the FXLMS algorithm, the following performance is greatly reduced. .

FIG. 3 is a block diagram schematically illustrating the FXLMS algorithm, and in particular, illustrates a block diagram of the front FXLMS algorithm.

As shown in FIG. 3, the FXLMS algorithm has an adaptive filter W (z) that actively adapts to the surrounding environment, a primary path transfer function P (z), and an additional path transfer function S (z). An output element for respectively outputting an output element, an additional path prediction value filter for inputting an estimated value S '(z) of the additional path transfer function, an actuator to which the first path transfer function and the additional path transfer function are added, and the actuator The minimum mean squarer that feeds back the error signal e (n) obtained by and the reference signal x '(n) compensated for by the additional path transfer function prediction value to adjust the coefficient vector of the adaptive filter ( LMS).

In the pre-feed FXLMS algorithm, the additional path transfer function S (z) is modeled off-line and the prediction is expressed as S '(z). Unwanted signals from the main sound source are observed by the reference signal sensor. The reference signal x (n) passes the adaptive filter W (z) and the additional path transfer function S (z) to minimize the error signal e (n) and control signal y '( n)). The reference signal x (n) is filtered by the additional path transfer function prediction value S '(z) filter, so that the additional path transfer function prediction value used to update the coefficient of the adaptive filter W (z) is compensated. Obtained reference signal x '(n). The adaptive filter W (z) is implemented as a finite response filter together with the additional path transfer function prediction value S '(z) filter.

From the block diagram, the error signal e (n) at time n is expressed by the following expression (1).

e (n) = d (n) -y '(n) = d (n) -s (n) * y (n) = d (n) -s (n) * [ w ^T (n) x (n )]

Where s (n) is the impulse response of the additional path transfer function S (z) at time n. The vector of the reference signal x (n) and the coefficient vector w (n) of the adaptive filter W (z) at time n are as shown in Equations 2 and 3 below.

x (n) = [x (n) x (n-1) ... x (n-L + 1)]

w (n) = [w ₀ (n) w ₁ (n) ... w _L-1 (n)]

Where L is the order of the adaptive filter W (z). The purpose of the adaptive filter W (z) is to minimize the instantaneous mean square error ξ (n) = e ² (n), and the steepest descent algorithm to achieve this objective is Is defined as:

w (n + 1) = w (n) -μ ▽ ξ (n) / 2

Where?? (N) is an instantaneous estimate of the mean square error gradient at time n and is defined as in Equation 5 below.

▽ ξ (n) = ▽ e ² (n) = 2 [▽ e (n)] e (n) = 2 [-s (n) * x (n)] e (n) = -2 x '(n e (n)

Substituting Equation 5 into Equation 4 yields an FXLMS algorithm, which is shown in Equation 6 below.

w (n + 1) = w (n) + μ x '(n) e (n)

Here, μ is the convergence coefficient, and in order for the adaptive filter W (z) to converge stably, μ must satisfy the following condition.

0 <μ <2 / LPx, Px = E [x ² (n)]

Where Px is the power of the reference signal.

As shown in Equation 6, convergence time and stability in the adaptation process of the FXLMS algorithm depend on μ and x '(n). In addition, since the FXLMS algorithm uses a fixed convergence coefficient, the filter order L should be large when the power of the reference signal is small, and the filter order L should be small when the power of the reference signal is large. However, when the power of the reference signal changes over time, there is a disadvantage in that the fixed convergence coefficient does not exhibit normal control performance.

Recently, with the development of DSP technology, one-dimensional and three-dimensional acoustic spaces are well adapted to the ANC technique in three-dimensional acoustic spaces such as in automobiles and airplanes, and when the surrounding environment such as noise characteristics, temperature, and humidity changes. There is an active research on the adaptive ANC technique.

Active noise control devices in the one-dimensional sound field commercialized or announced to date include digital sound cancellation systems of DIGISONIX in the US, electronic noise systems of Hitachi plant construction in Japan, and ANC systems for automobile exhaust systems of Noise Cancellation Technologies in the US. As a commercialized or announced three-dimensional active noise control device, a refrigerator equipped with an ANC system was sold by Toshiba, Japan in September 1990, and a vehicle equipped with an ANC system was sold by Nissan in November, 1991. It became.

In Korea, ANC Technologies Inc. and Korea Maritime Institute of Technology began joint development in 1989 and succeeded in producing prototypes in late 1990. The company began developing and mass-producing the first active silencer, an ANC system, in three dimensions. Research on ANCs in hermetic systems is still underway, and there are no published products.

The present invention develops the Co-FXLMS algorithm which changes the convergence coefficient of the FXLMS algorithm in real time by using the correlation of the error signal and the filtered reference signal, thereby improving the convergence performance of the DSP's FXLMS algorithm. It is to provide an active noise control system and method of a three-dimensional closed system using the Co-FXLMS algorithm, which can be applied to the three-dimensional closed system.

In order to achieve the above object, the present invention provides an adaptive filter W (z) that actively adapts to the surrounding environment, a primary path transfer function P (z), and an additional path transfer function S (z). An output element for outputting each, and an additional path prediction value S '(z) for inputting the predicted value S' (z) of the additional path transfer function, wherein the primary path transfer function and the additional path transfer function In the ANC system equipped with the FXLMS algorithm configured to feed back the reference signal compensated by the added error signal e (n) and the additional path transfer function prediction value to adjust the coefficient vector of the adaptive filter. The correlation vector of the reference signal x '(n) compensated for the error signal and the additional path transfer function prediction value is extracted in real time, and the convergence coefficient of the FXLMS algorithm is adjusted based on the extracted cross-correlation. Equipped with a cross-correlation least average squarer A Co-FXLMS algorithm provides an active noise control system in a three-dimensional hermetic system that performs active noise control.

The present invention also provides an adaptive filter W (z) that actively adapts to the surrounding environment, an output element for outputting a primary path transfer function P (z) and an additional path transfer function S (z), respectively. And an additional path prediction value filter for inputting a prediction value S '(z) of the additional path transfer function, wherein the error signal e (n) obtained by adding the first path transfer function and the additional path transfer function An ANC method using an FXLMS algorithm configured to feed back a reference signal x '(n) compensated for by the additional path transfer function prediction to adjust a coefficient vector of an adaptive filter. Extracting a correlation between the error signal and the reference signal compensated for the additional path transfer function prediction value in real time; Adjusting a convergence coefficient of the FXLMS algorithm by the extracted cross-correlation; Performing active adaptive control in real time by adjusting the coefficient vector of the adaptive filter in real time with the convergence coefficient of the adjusted FXLMS algorithm, and executing active noise control by the Co-FXLMS algorithm in which the convergence coefficient is adjusted in real time. Provides an active noise control method of the three-dimensional hermetic system.

According to the present invention, by implementing the Co-FXLMS algorithm that can correct the convergence coefficient of the FXLMS algorithm in real time by using the cross-correlation of the reference signal and the error signal compensated for the prediction of the additional path transfer function, power over time Active noise control can be implemented in real time for all changing enclosures.

In addition, the present invention is to form a plurality of channels for one reference signal and to adjust the convergence coefficient of the FXLMS algorithm for each of the channels to each of the real-time, the advantage that can be applied to all products having a three-dimensional closed system There is this.

These objects, features and advantages of the present invention will be more readily understood by reference to the accompanying drawings and the following detailed description.

1 is a block diagram showing an embodiment of a Co-FXLMS algorithm applied to an ANC system of a three-dimensional hermetic system according to the present invention, illustrating a Co-FXLMS algorithm implemented as one single channel for one reference signal. Doing.

In FIG. 1, the Co-FXLMS algorithm includes an adaptive filter (W (z)), a primary path transfer function (P (z)), and an additional path transfer function (S (z)) that are actively adapted to the surrounding environment. And an additional path predictive value filter for inputting an output element for outputting each other and an estimated value S '(z) of the additional path transfer function. In addition, the primary path transfer function and the additional path transfer function are added by the actuator to obtain an error signal e (n). A cross-correlation least average squarer 100 for feeding back the reference signal x '(n) compensated for the error signal and the additional path transfer function prediction value with an adaptive filter to adjust the coefficient vector of the adaptive filter It is composed. The cross-correlation least average squarer (100) extracts in real time a cross-correlation (R (n)) of an error signal e (n) and a reference signal x '(n) compensated for an additional path transfer function prediction value. Co-FXLMS algorithm for adjusting the convergence coefficient (μ) of the FXLMS algorithm by the extracted cross-correlation to adjust the coefficient vector (w (n)) of the adaptive filter in real time.

As a result, the cross-correlation least average squarer 100 has a cross-correlation between the reference signal x '(n) and the error signal e (n) compensated for the additional path transfer function prediction value S' (z). R (n)) is extracted in real time. The coefficient vector w (n) of the adaptive filter W (z) can be adjusted in real time by adjusting the convergence coefficient μ of the FXLMS algorithm by the extracted cross-correlation.

Figure 2 is a block diagram showing another embodiment of the Co-FXLMS algorithm applied to the ANC system of the three-dimensional hermetic system according to the present invention, the figure illustrates a multi-channel Co-FXLMS algorithm consisting of four channels, Multi-channel configuration may be implemented by selectively setting the number of channels in accordance with the characteristics of the closed system applied to the ANC system.

In order to implement the Co-FXLMS algorithm according to the embodiment of FIG. 2, the ANC system according to another embodiment of the present invention includes four adaptive filters (W ₁ (z) to W ₄ (z)) and an additional path transfer function. (S ₁₁ (z) to S ₁₄ (z)) Four additional path transfer function predictive value filters (S ₁₁ ) for inputting the output elements and their respective predicted values (S ₁₁ '(n) to S ₁₄ ' (n)). '(z) ~ S ₁₄ ' (z)) and cross-correlation least average squarers 101-104, so that four channels can be configured for one reference signal x (n). do.

That is, four adaptive filters (W ₁ (z) to W ₄ (z)) and additional path transfer functions (S ₁₁ (z) to S ₁₄ (z)) output elements provided in four channels, respectively, Four additional path transfer function prediction filters (S ₁₁ '(z) to S ₁₄ ' (z)) for inputting respective prediction values (S ₁₁ '(n) to S ₁₄ ' (n)) for the additional path transfer function. And the error of each reference signal x _'11 (n) to x' ₄₁ (n) filtered by the additional path transfer function predictor filter S ₁₁ '(z) to S ₁₄ ' (z) of each channel. Cross-correlation of the signals e ₁ (n) are extracted in real time to adjust the coefficient vectors of the adaptive filters W ₁ (z) to W ₄ (z), respectively. Is implemented.

Thus, the additional path transfer function prediction values S ₁₁ '(n) to S ₁₄ ' for each of the four signals and the error signal e ₁ (n) by the cross-correlation least average squarer 101-104 of each channel. (n)) is extracted in real time the cross-correlation (R ₁ (n) ~ R ₄ (n)) of each reference signal (x ₁₁ '(n) ~ x ₄₁ ' (n)) is compensated. By adjusting the convergence coefficient (μ) of the FXLMS algorithm by the extracted cross-correlation, the coefficient vectors w ₁ (n) to w ₄ (n) of the respective adaptive filters W ₁ (z) to W ₄ (z) are adjusted. ) Can be adjusted in real time.

Hereinafter, the Co-FXLMS algorithm according to each embodiment of the present invention will be described in detail.

First, as shown in Equation 6, the Co-FXLMS algorithm is used to improve the disadvantage that the normal control performance does not appear when the power of the reference signal varies with time during the adaptation process of the FXLMS algorithm using the fixed convergence coefficient. will be.

In order to compensate for this drawback, the Co-FXLMS algorithm needs to normalize the convergence coefficient as shown in Equation 8 with respect to the power of the reference signal x '(n) of FIG.

μ (n) = α / LP'x (0 <α <2)

Where α is the normalized convergence coefficient and P'x is an estimate of the power of x '(n). The simplest way to estimate P'x is to use a running-average filter for x ' ² (n). Equation 9 below is the M-order running-average filter for x ' ² (n).

If the order of the running-average filter is equal to the order of the adaptive filter, that is, L = M, P'x is expressed by Equation 10 below.

P'x = x ' ^T (n) x ' (n) / L

Substituting Equation 10 into Equation 8 yields Equation 11 below.

μ (n) = α / x ' ^T (n) x ' (n)

In Equation 11, the convergence coefficient is diverged by the adaptive filter W (z) when x '(n) = 0, so that when the small amount δ is added to the denominator of Equation 11, Equation 12 is obtained.

μ (n) = α / {δ + x ' ^T (n) x ' (n)}

When w (n), the coefficient vector of the adaptive filter W (z), converges to the optimal coefficient vector w ⁰ , the cross-correlation R of the error signal e (n) and the reference signal x '(n) with compensation of the additional path transfer function prediction are compensated. (n) is 0 is the basic concept of the Co-FXLMS algorithm and is expressed as in Equation 13 below.

R (n) = E [e (n) x '(n)] = 0

The cross-correlation R (n) of Equation 13 is equal to the expected value of e (n) x '(n) and is shown in Equation 14 below.

E [e (n) x '(n)] ≒ E [{d (n) -w (n) x ' (n)} x '(n)]

= E [d (n) x '(n) -x ' (n) x ' ^T (n) w (n)] = E [d (n) x ' (n)]-E [ x '(n) x ' ^T (n)] w (n)

In Equation 14, when the vector w (n) converges to the optimal vector w ⁰ , the expected value E [e (n) x '(n)] becomes 0, thereby demonstrating the basic concept of the Co-FXLMS algorithm. If w (n) is far from w ⁰ , the cross-correlation and convergence coefficients become relatively large and cross-correlation and convergence coefficients become relatively small as w (n) approaches w ⁰ . ) And the convergence coefficient μ (n) are proportional to each other. Therefore, the convergence coefficient of Equation 12 may be expressed as Equations 15 and 16 below.

R (n) = λ R (n-1) + (1-λ) x '(n) e (n)

Where C is a constant and λ is a constant between 0 and 1. Therefore, Co-FXLMS algorithm for a single channel as shown in Figure 2 is the following equation (17).

Meanwhile, when the single channel Co-FXLMS algorithm is applied to a multi-channel Co-FXLMS algorithm composed of one reference signal, four multichannels (control speakers), and one error microphone as shown in FIG. Equations 18a to 18d are as follows.

_{w 1 (n + 1) =} w 1 (n) + μ (n) x '11 (n) e 1 (n) = w 1 (n) + μ (n) [s' 11 (n) * x ( n)] e ₁ (n)

w ₂ (n + 1) = w ₂ (n) + μ (n) x _'21 (n) e ₁ (n) = w ₂ (n) + μ (n) [s' ₁₂ (n) * x ( n)] e ₁ (n)

w ₃ (n + 1) = w ₃ (n) + μ (n) x _'31 (n) e ₁ (n) = w ₃ (n) + μ (n) [s' ₁₃ (n) * x ( n)] e ₁ (n)

w ₄ (n + 1) = w ₄ (n) + μ (n) x _'41 (n) e ₁ (n) = w ₄ (n) + μ (n) [s' ₁₄ (n) * x ( n)] e ₁ (n)

)(only,

)

According to the present invention as described above, the cross-correlation R (n) of the reference signal x '(n) compensated for the error signal e (n) and the additional path transfer function prediction value is extracted in real time, and the convergence coefficient μ (n) is based thereon. By using the Co-FXLMS algorithm that corrects in real time, it is possible to control the convergence speed with a large convergence coefficient at the beginning of control with large correlation, and to reduce the convergence speed with a small convergence coefficient in the latter control with small correlation. As a result, the stability in the latter half of the control can be improved to reduce the possibility of divergence. Therefore, when the ANC for the three-dimensional hermetic system is executed using the Co-FXLMS algorithm according to the present invention, the convergence speed can be improved while ensuring the stability of the ANC compared to the conventional FXLMS algorithm.

In addition, the ANC technology that Co-FXLMS algorithm is required as described in the present invention can be extended to a car, a washing machine, a refrigerator for low noise, etc., and thus, it is expected to contribute to strengthening the competitiveness of low noise products with a lot of consumer interest.

As described above, the preferred embodiment of the present invention has been described, but the present invention may use various changes, modifications, and equivalents. Therefore, the above description does not limit the scope of the invention defined by the limitations of the following claims.

1 is a block diagram illustrating an embodiment of a Co-FXLMS algorithm applied to an ANC system of a three-dimensional closed system according to the present invention.

Figure 2 is a block diagram showing another embodiment of the Co-FXLMS algorithm applied in the ANC system of the three-dimensional hermetic system according to the present invention,

3 is a block diagram schematically illustrating the FXLMS algorithm.

<Description of the symbols for the main parts of the drawings>

100, 101 ~ 104: Cross-correlation minimum mean square

Claims (6)

An adaptive filter W (z) that actively adapts to the surrounding environment, an output element for outputting a primary path transfer function P (z) and an additional path transfer function S (z), respectively, and the additional path An additional path prediction value filter for inputting a predicted value S '(z) of a transfer function, wherein the error signal e (n) obtained by adding the first path transfer function and the additional path transfer function and the additional path transfer In the ANC system equipped with the FXLMS algorithm configured to adjust the vector coefficient of the adaptive filter by feeding back the reference signal whose function prediction value is compensated for, In the adaptation process of the FXLMS algorithm, the cross-correlation of the reference signal compensated for the error signal and the additional path transfer function prediction value is extracted in real time, and the convergence coefficient of the FXLMS algorithm is adjusted based on the extracted cross-correlation to adjust the coefficient vector of the adaptive filter. Active noise control system in a three-dimensional closed system using the Co-FXLMS algorithm, characterized in that the active noise control by the Co-FXLMS algorithm having a cross-correlation least average squarer for real-time control. The method of claim 1, For time n, the cross-correlation R (n) between the reference signal x '(n) and the error signal e (n), in which the additional path transfer function prediction value is compensated, R (n) = λR (n-1) + (1-λ) x '(n) e (n) (λ is a constant between 0 and 1) The convergence coefficient (μ (n)) of the FXLMS algorithm is

(C is a constant) The coefficient vector w (n) of the adaptive filter W (z) is

Active noise control system in a three-dimensional closed system using the Co-FXLMS algorithm. The method of claim 1, Using the Co-FXLMS algorithm, the adaptive filter, the additional path transfer function, and the cross-correlation least mean square are composed of a plurality of channels for one reference signal, and are configured to be actively adaptively controlled in real time for each of the plurality of channels. Active noise control system in 3D enclosed system. An adaptive filter W (z) that actively adapts to the surrounding environment, an output element for outputting a primary path transfer function P (z) and an additional path transfer function S (z), respectively, and the additional path An additional path prediction value filter for inputting a predicted value S '(z) of a transfer function, the error signal e (n) obtained by adding the first path transfer function and the additional path transfer function, and the additional path transfer In the ANC method using the FXLMS algorithm configured to feed back the reference signal compensated for function prediction to adjust the vector coefficient of the adaptive filter, In the adaptation process of the FXLMS algorithm, extracting in real time a correlation between the error signal and the reference signal compensated for the additional path transfer function prediction value; Adjusting a convergence coefficient of the FXLMS algorithm by the extracted cross-correlation; Real-time active adaptive control by adjusting the coefficient vector of the adaptive filter in real time with the convergence coefficient of the adjusted FXLMS algorithm, and executing active noise control by the Co-FXLMS algorithm in which the convergence coefficient is adjusted in real time. An active noise control method in a three-dimensional hermetic system using the Co-FXLMS algorithm. The method of claim 4, wherein For time n, the cross-correlation R (n) between the reference signal x '(n) and the error signal e (n), in which the additional path transfer function prediction value is compensated, R (n) = λR (n-1) + (1-λ) x '(n) e (n) (λ is a constant between 0 and 1) The convergence coefficient (μ (n)) of the FXLMS algorithm is

(C is a constant) The coefficient vector w (n) of the adaptive filter W (z) is

Active noise control method in a three-dimensional hermetic system using the Co-FXLMS algorithm. The method of claim 4, wherein The adaptive filter, the additional path transfer function, and the cross-correlation minimum mean square are composed of a plurality of channels with respect to one reference signal, so that the real-time active adaptive control is performed on each of the plurality of channels. Active noise control method in closed system.

Images