KR20030000347A - Apparatus and method for removing echo-audio signal using time-varying algorithm with time-varying step size - Google Patents

Abstract

PURPOSE: A sound echo canceling apparatus applying a time varying adaptive algorithm and a method therefor are provided to set a coefficient vector for generating a sound echo estimation signal on the basis of a correlation degree of a receiving sound signal and an estimation error signal. CONSTITUTION: A cross correlation degree calculator(103) receives two power values of a receiving sound signal and an estimation error signal inputted from a power calculator(102), the receiving sound signal corresponding to a vector value having L numbers of degrees through an A/D converter(101), and the estimation error signal, and calculates cross correlation degrees. The cross correlation degree calculator(103) outputs only the cross correlation degree of a specific factor among the calculated cross correlation degrees. An LPF(Low Pass Filter)(104) filters the cross correlation degree of the specific factor. A time varying adaptive constant calculator(105) calculates a time varying adaptive constant using the filtered cross correlation degree, and outputs the calculated time varying adaptive constant to a coefficient vector updating device(106). A time coefficient increasing device(107) increases a time coefficient whenever the coefficient vector updating device(106) calculates the coefficient factor.

Description

Acoustic echo canceller with time-varying adaptive algorithm and its method {APPARATUS AND METHOD FOR REMOVING ECHO-AUDIO SIGNAL USING TIME-VARYING ALGORITHM WITH TIME-VARYING STEP SIZE}

The present invention relates to an apparatus and method for removing an acoustic echo signal in a communication system for transmitting and receiving an acoustic signal between a far-end speaker and a near-end speaker.

More specifically, in order to prevent the received sound signal received from the other party (the speaker of the far end) through the sound output unit from interfering with the transmitted sound signal to be transmitted to the other party through the sound input unit, an acoustic echo canceling device to which a time-varying adaptive algorithm is applied; It's about how.

Speaker-phones, which are used in hand-free mobile phones and conference conferencing systems, are an additional device for providing convenience to users using telephony systems, especially portable telephony systems. Although there is an advantage in providing a sound signal, the sound echo is generated because the far-end speaker's voice signal (the other party's voice) output through the speaker is re-input into the microphone of the near-end speaker. These echoes degrade call quality and, in severe cases, make the call itself impossible. Therefore, in order to use a handfree mobile phone or a speakerphone more efficiently, an acoustic echo removing device for removing acoustic echo must be installed.

1 is a basic configuration of an apparatus for removing acoustic echo in a general telephony communication system, FIG. 2 is a detailed configuration of an echo path estimator of FIG. 1, and FIGS. 3A and 3B are normalized least mean square (NLMS) and Sum-LMS ( A detailed configuration of a conventional coefficient vector controller to which a Least Mean Square (Algorithm) algorithm is applied, respectively, and FIGS. 1 to 3b refer to components provided on the near-end speaker side, and corresponding components on the far-end speaker side are the same. It is provided.

1 is a speaker for outputting a received sound signal {x (k)} input from the speaker side of a far-end to a speaker side of the near end; Transmitted sound signal {d (k)} [acoustic echo signal {y (k)} generated by the near-end acoustic environment, surrounding noise {n (k)} including the sound signal and measurement noise on the speaker side near the end Included] is a microphone (microphone) is input; The received acoustic signal {x (k)} inputted from the speaker side of the far end and the estimated error signal {e (k)} outputted from the subtractor 30 are received, and the received acoustic signal outputted through the speaker is converted into an echo path ( echo echo signal for estimating the acoustic echo signal {y (k)} that is changed via the echo path { (k)} an adaptive filter for estimating the echo path, that is, the echo path estimator (10); An A / D converter 20 which is synchronized with a predetermined period and converts the micro-inputted transmission signal {d (k)} into digital data; And the digitally converted transmission sound signal {d (k)} and the output acoustic reflection signal { and a subtractor 30 for subtracting (k)}.

As shown in FIG. 2, the echo path estimator 10 receives the received acoustic signal {x (k)} input from the speaker side of the far end and the estimated error signal {e (k)} output from the subtractor 30. A coefficient vector adjuster (11) which receives a and adjusts a coefficient vector (W (k)) for estimating the acoustic echo signal {y (k)}; And the acoustic echo estimation signal by the coefficient vector adjusted as described above { and an echo path estimation signal generator 12 for generating (k)}.

When the NLMS algorithm is applied as a conventional acoustic echo cancellation method, the coefficient vector controller 11a is synchronized with a predetermined period as shown in FIG. An A / D converter 13a for converting the signal into a second one; A power calculator (14a) for calculating power of the digitally converted received sound signal {x (k)}; An adaptive constant calculator 15a for calculating an adaptive constant {mu (k)} using the calculated power σ ² _x and the received acoustic signal {x (k)}; A coefficient vector updater 16a for updating the coefficient vector W (k) by the calculated adaptive constant {mu (k)} and the estimated error signal {e (k)}; And a time coefficient increaser 17a that increases the time coefficient k by 1 each time the coefficient vector {W (k)} is calculated.

Referring to FIGS. 1 to 3A attached to the operation of the acoustic echo canceller applied to the conventional NLMS algorithm configured as described above, the received acoustic signal transmitted through the designated path from the speaker side of the far-end {x (k) } Is output through the speaker to the speaker near the end, and the received sound signal is converted into an acoustic echo signal y (k) through an echo path and then input into a microphone. At this time, the microphone is input together with the ambient signal {n (k)} including the voice signal of the near end speaker and the measurement noise. The micro-input sound signal {d (k)} is converted into digital data by the A / D converter 20 which is synchronized at a predetermined period and then input to the subtractor 30.

On the other hand, the echo path estimator 10 receives the received acoustic signal {x (k)} transmitted through the designated path from the speaker side of the far-end, and the A / D converter (a) in the coefficient vector controller 11a synchronized with a predetermined period ( 13a) converts the received sound signal {x (k)} into digital data. When the digitally converted received sound signal {x (k)} is a vector value having L orders (n = 0, 1, 2, ..., L-1), the received sound signal {x (k)} Is equal to Equation (1), and the power calculator 14a calculates the power (σ ² _x ) of the received sound signal from the vector value {X (k) of the digitally converted received sound signal as in Equation (2). The adaptive constant calculator 15a calculates the adaptive constant {μ (k)} as shown in Equation (3) using the power (σ ² _x ) of the received sound signal calculated as shown in Equation (2).

X (k) = x _n (k) = [x ₀ (k), x ₁ (k), ..., x _L-1 (k)] Equation (1)

X ^T (k) X (k) = Lσ ² _x Equation (2)

μ (k) = α / (Lσ ² _x ) Equation (3)

Here, k is a time coefficient increasing by 1 and α is an adaptive filter, that is, a normalized adaptive constant of the echo path estimator 10, and k and α described later have the same meaning and will not be described. In addition, X ^T (k) is a transpose vector of X (k) in which the arrangement of the rows and columns of the vector X (k) is changed.

The coefficient vector updater 16a calculates the coefficient vector {W (k)} as shown in equation (4) using the adaptive constant {μ (k)} calculated as shown in equation (3). The entire operation is repeatedly performed while increasing the time coefficient k by one, and the coefficient vector W (k) calculated by equation (4) is updated each time.

W (k + 1) = W (k) + μ (k) e (k) X (k) Equation (4)

Here, W (k + 1) corresponds to a coefficient vector generated at a time point t = k + 1, and e (k) is an estimation error signal and the same applies to the following description.

The echo path estimation signal generator 12 performs an acoustic echo estimation signal for estimating the echo path by the coefficient vector {W (k) calculated by updating as shown in equation (4). (k)} as shown in equation (5) and output to the subtractor 30, the subtractor 30 is a sound generated by equation (5) from the transmission sound signal {d (k)} input through the microphone Echo Estimation Signal { By subtracting (k)} as shown in equation (6), the estimated error signal {e (k)} is calculated.

(k) = W ^T (k) X (k) Equation (5)

e (k) = d (k)- (k) = d (k)-W ^T (k) X (k) Equation (6)

W ^T (k) is a transpose vector of W (k) that changes the arrangement of the rows and columns of the coefficient vector W (k).

The coefficient vector {W (k)} calculated by updating as described above is set as Equation (7) below by the above-mentioned equations, where the acoustic echo signal {y (k)} is an acoustic echo estimation signal { (k) = W ^T (k) X (k)}, i.e., acoustic echo estimation signal { Even though the coefficient vector {W (k)} is calculated so that (x)} accurately estimates the acoustic reflection signal {y (k)}, it is determined by n (k) that is variably input according to the acoustic environment of the near-end side. The problem is that the coefficient vector W (k) set as described above is distorted by the size of a term including n (k) in Equation (7). As described above, n (k) corresponds to noise from the perspective of the echopath estimator 10 whose purpose is to accurately calculate the coefficient vector, and as the value of the term containing n (k) is closer to 0, the coefficient vector {W ( k)} is an acoustic echo estimation signal { It can be seen that (x)} is calculated more in accordance with the degree of estimating the acoustic echo signal y (k)}.

W (k) = W (k-1) + α / {L (σ ² _x )} {y (k) -W ^T (k) X (k)} + α / {L (σ ² _x )} X (k) n (k) Equation (7)

Accordingly, the Sum-LMS algorithm is applied to solve the problem of the NLMS algorithm. In this case, the coefficient vector controller 11b is synchronized with a predetermined period as shown in FIG. 3B to receive the received sound signal {x (k). } An A / D converter 13b which receives an input and converts it into digital data; A power calculator (14b) for calculating the powers of the digitally converted received sound signal (x (k)) and the estimated error signal (e (k)) output from the subtractor (30); An adaptive constant calculator 15b for calculating an adaptive constant {mu (k)} from the calculated power σ ² _x , σ ² _e and the received sound signal {x (k)}; A coefficient vector updater 16b for updating the coefficient vector W (k) by the calculated adaptive constant {mu (k)} and the estimated error signal {e (k)}; And a time coefficient increaser 17b that increases the time coefficient k by 1 each time the coefficient vector {W (k)} is calculated. The power calculator 14b includes the power of the estimated error signal ( σ ² _e ) is further calculated, and the adaptive constant calculator 15b calculates the power σ ² _x , σ ² _{e of} each of the received acoustic signal {x (k)} and the estimated error signal {e (k)}. The adaptive constant is calculated as shown in equation (3-1).

μ (k) = α / {L (σ ² _x + σ ² _e )} Equation (3-1)

W (k) = W (k-1) + α / {L (σ ² _x + σ ² _e )} {y (k) -W ^T (k) X (k)} + α / {L (σ ² _x + σ ² _e )} X (k) n (k)

Formula (7-1)

Therefore, the coefficient vector updater 16b uses the acoustic echo estimation signal {by the adaptive constant {mu (k)} calculated by equation (3-1). (k)} is calculated by updating the coefficient vector {W (k)} for generating, and the coefficient vector {W (k)} calculated as in Equation (7-1) is outputted from the subtractor 30. Further adaptation is made to the power magnitude of the error signal e (k). Therefore, compared with the case where the NLMS algorithm is applied, the magnitude of the term including n (k) is reduced by the power σ ² _e of the estimated error signal {e (k)} as in Equation (7-1). , The coefficient vector {W (k)} calculated therefrom is an acoustic echo estimation signal { (k)} is set to more closely match the degree to which the acoustic echo signal y (k)} is estimated.

However, by setting the size of the adaptive constant {mu (k)} proportional to the magnitude of the coefficient vector {W (k)} as described above, the amount of change of the updated coefficient vector {W (k)} is also small, and thus The convergence time until the coefficient vector {W (k)} converges to a predetermined value is longer than in the case where the NLMS algorithm is applied, and as a result, the convergence speed estimated by the acoustic echo signal to the acoustic echo signal is slowed. Problems arise.

Accordingly, an object of the present invention was created to solve the above problems, and a time-varying adaptive algorithm for setting a coefficient vector for generating an acoustic echo estimation signal based on a correlation between the received acoustic signal and an estimated error signal is provided. The present invention provides an acoustic echo canceling device and a method thereof.

Another object of the present invention is to provide an acoustic echo canceling device and a method to which a time-varying adaptive algorithm is applied so that the coefficient vector takes the shortest time to converge to a predetermined value.

Another object of the present invention is to provide an acoustic echo canceling method to which a time-varying adaptive algorithm is applied to minimize the amount of computation required to calculate the coefficient vector.

1 is a basic configuration of an acoustic echo canceller in a general telephone communication system,

FIG. 2 is a detailed configuration of an echo path estimator of FIG. 1;

3A is a detailed configuration of a conventional coefficient vector controller to which the NLMS algorithm is applied,

3b is a detailed configuration of a conventional coefficient vector controller to which the Sum-LMS algorithm is applied,

4 is a detailed configuration of a coefficient vector adjuster according to an embodiment of an acoustic echo cancellation device to which a time-varying adaptive algorithm according to the present invention is applied,

5 is a flowchart of an acoustic echo removing method to which a time-varying adaptive algorithm according to the present invention is applied;

6A is a graph of cross-correlation diagrams calculated in accordance with the present invention;

FIG. 6B is a graph showing a comparison between the time-varying adaptive constants low-pass filtered and the time-varying adaptive constants low-filtered according to the present invention,

7 (a) to 7 (e) show the results of experiments for removing acoustic echoes by the time-varying adaptive algorithm according to the present invention, compared with the results of the above experiments by conventional NLMS and Sum-LMS algorithms. Will,

8A is a configuration of an echo echo estimator calculated according to the present invention and an acoustic echo canceller in which an echo path measured by an experiment is implemented;

Figure 8b shows a comparison between the coefficient misadjustment index to which the algorithm according to the present invention is applied and the coefficient misadjustment index to which the conventional algorithm is applied, respectively.

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

10, 100: echo path estimator 11, 11a, 11b, 110: coefficient vector controller

12: echo path estimation signal generator 13a, 13b, 20, 101: A / D converter

14a, 14b, 102: power calculator 15a, 15b: adaptive constant calculator

16a, 16b, 106: coefficient vector updater 17a, 17b, 107: time coefficient increaser

30: subtractor 103: cross-correlation calculator

104: low pass filter 105: time-varying adaptive constant calculator

200: echo path

Acoustic echo canceller applied to the time-varying adaptive algorithm according to the present invention for achieving the above object, the sound output unit for outputting the received sound signal received from the far end speaker to the near end speaker, and the far end from the far end speaker A communication system comprising an audio input unit for inputting a transmission sound signal to be transmitted to a speaker of a communication system, the communication system having an acoustic echo path from the audio output unit to the audio input unit, wherein the received acoustic signal is changed by passing through the acoustic echo path. And an echo path estimating means for generating an acoustic echo estimating signal for estimating an echo signal, wherein the acoustic echo estimating signal is an estimated error signal generated by subtracting a previously generated acoustic echo estimating signal from the transmission acoustic signal. And a time varying correlation between the received sound signal and the received sound signal. A.

The acoustic echo canceling device to which the time-varying adaptive algorithm is applied comprises: calculating means for calculating a correlation between the received acoustic signal to be output from the acoustic output unit and the estimated error signal; And generating means for generating the acoustic echo estimation signal in response to the calculated cross-correlation.

The acoustic echo canceling device to which the time-varying adaptive algorithm according to the present invention is applied further includes power measuring means for measuring the power of the received acoustic signal and the estimated error signal to be output, wherein the calculating means comprises: It is characterized by further calculating the cross-correlation based on power.

The acoustic echo cancellation apparatus to which the time-varying adaptive algorithm according to the present invention is applied further comprises a low-pass filter for low-passing the calculated cross-correlation, wherein the generating means is configured to perform the sound in response to the low-pass cross-correlation. It is characterized by further generating an echo estimation signal.

In the acoustic echo cancellation apparatus to which the time-varying adaptive algorithm according to the present invention is applied, the calculated cross-correlation is a vector value, and extraction means for extracting a specific cross-correlation for a specific coefficient from the calculated cross-correlation. It is configured to further include, wherein the generating means is characterized in that for generating the acoustic echo estimation signal further in response to the extracted specific cross-correlation.

Acoustic echo cancellation device to which the time-varying adaptive algorithm according to the present invention is applied comprises: calculating means for calculating a time-varying adaptive constant based on the calculated cross-correlation; And resetting means for resetting a predetermined coefficient vector in response to the calculated time-varying adaptive constant, wherein the generating means generates the acoustic echo estimation signal based on the reset coefficient vector.

In addition, the acoustic echo cancellation method to which the time-varying adaptive algorithm according to the present invention is applied, the sound output unit for outputting the received sound signal received from the far end speaker to the near end speaker, and transmits from the near end speaker to the far end speaker A method for removing acoustic echo of a communication system having an acoustic echo path from an acoustic output section to an acoustic input section, comprising: an acoustic echo input section for inputting a transmission acoustic signal to be transmitted; And generating an acoustic echo estimation signal for estimating the changed acoustic echo signal, wherein the acoustic echo estimation signal comprises: an estimated error signal generated by subtracting a previously generated acoustic echo estimation signal from the transmission acoustic signal; It is characterized in that it is generated in response to the degree of time-varying correlation between the received sound signals.

An acoustic echo canceling method to which a time-varying adaptive algorithm according to the present invention is applied includes: calculating a correlation between a received acoustic signal to be output from the acoustic output unit and the estimated error signal; And generating the acoustic echo estimation signal in response to the calculated cross-correlation.

The acoustic echo canceling method to which the time-varying adaptive algorithm according to the present invention is applied further comprises a power measuring step of measuring power of the received acoustic signal and the estimated error signal to be output, wherein the calculating step comprises the step of measuring each of the measured signals. It is characterized by further calculating the cross-correlation based on power.

The acoustic echo cancellation method to which the time-varying adaptive algorithm according to the present invention is applied further comprises a low pass step of low-passing the calculated cross-correlation, wherein the generating step is performed in response to the low-pass cross-correlation. It is characterized by further generating an echo estimation signal.

In the acoustic echo removal method to which the time-varying adaptive algorithm according to the present invention is applied, the calculated cross-correlation is a vector value, and an extraction step of extracting a specific cross-correlation for a specific coefficient from the calculated cross-correlation The method may further include generating the acoustic echo estimation signal in response to the extracted specific cross-correlation.

Acoustic echo cancellation method to which the time-varying adaptive algorithm according to the present invention is applied, calculating a time-varying adaptive constant based on the calculated cross-correlation; And a resetting step of resetting a predetermined coefficient vector according to the calculated time-varying adaptive constant, wherein the generating step is characterized by generating the acoustic echo estimation signal based on the reset coefficient vector.

FIG. 4 is a detailed configuration of a coefficient vector controller according to an embodiment of an acoustic echo cancellation apparatus to which a time-varying adaptive algorithm is applied according to the present invention. The received acoustic signal {x (k)} and estimation received from the power calculator 102 are shown in FIG. Corresponds to the two power values σ ² _x , σ ² _{e of the} error signal {e (k)} and a vector value having L orders converted into digital data by the A / D converter 101 synchronized with a predetermined period. Calculate the cross-correlation of the received sound signal {x (k)} and the estimated error signal {e (k)}, and output only the cross-correlation of a specific coefficient among the calculated cross-correlation. Group 103; A low pass filter 104 for low-passing the cross-correlation of the output specific coefficients; A time-varying adaptive constant calculator 105 for calculating a time-varying adaptive constant using the low-pass filtered correlation and outputting it to the coefficient vector updater 106; And a time coefficient increaser 107 which increases the time coefficient k by 1 every time the coefficient vector is calculated by the coefficient vector updater 106.

Hereinafter, with respect to the operation of the embodiment of the acoustic echo canceling device to which the time-varying adaptive algorithm according to the present invention configured as described above, in detail in parallel with FIG. 5 which is a flowchart of the acoustic echo removing method to which the time-varying adaptive algorithm according to the present invention is applied. Explain.

First, as shown in FIG. 1, which is a basic configuration of an acoustic echo canceller in a general telephone communication system, a received acoustic signal {x (k)} received along a designated line from the speaker side of the far end is output to the near end through a speaker. The output acoustic signal {x (k)} is converted into an acoustic echo signal {y (k)} through an echo path from a speaker to a microphone, and then corresponds to the acoustic environment of the near-end side. The microphone is input together with the speaker's voice signal or noise. The micro-inputted sound signal {d (k)} is input to the A / D converter 20, and the A / D converter 20 synchronized with a predetermined period receives the input sound signal {d (k)}. Convert to digital data. The subtractor 30 receives the acoustic echo estimation signal from the echo path estimator 10 { (k)}, and then the received acoustic echo estimation signal {from the transmitted acoustic signal {d (k)} converted into digital data as described above { (k)} is subtracted to output the estimated error signal {e (k)}.

On the other hand, the received sound signal {x (k)} received from the speaker side of the far end is input to the A / D converter 101 which is synchronized at a predetermined period, as shown in FIGS. 4 and 5, and A / D. The converter 101 detects the received sound signal {x (k)} every predetermined period (S10) and then converts the detected signal into digital data (S11). The power calculator 102 is configured to calculate the electric power σ ² _x , σ from the received sound signal {x (k)} converted into digital data and the estimated error signal e (k) input from the subtractor 30. ² _e ) (S12), and the cross-correlation calculator 103 calculates the cross-correlation between the received sound signal {x (k)} and the estimated error signal e (k) {δ _n (k). } Is calculated as shown in Equation (8) and normalized to the cross-correlation calculated using the product of the two powers (σ ² _x , σ ² _e ).

δ _n (k) = βδ _n (k-1) + (1-β) x _n (k) e (k) / (σ ² _x + σ ² _e ) ^1/2 Equation (8)

Here, n = 0,1,2, ..., L -1 is the received acoustic signal {x (k)} and the cross-correlation is also _{{δ n (k), δ} n-1 (k)} is the L-order Β is an arbitrary constant inserted for coefficient adjustment of each term, and the same applies to the following description.

In addition, the cross-correlation calculator 103 at the time when t = k as in Equation (8) to normalize the range of the instantaneous cross-correlation {x _n (k) e (k)} within a certain value The cross-correlation of X (x _n (k) e (k)) is divided by the sum of the respective power values.

The cross-correlation degree δ _n (k) between the received acoustic signal {x (k)} and the estimated error signal e (k)} calculated by Equation (8) is precisely (t = 1) to (t = It is a cumulative cross-correlation diagram that adds all the cross-correlation diagrams up to the point k), and the calculation amount can be reduced by setting the calculation formula as shown in Equation (8).

By calculating the cross-correlation {δ _n (k)} by Equation (8), the correlation between the received sound signal {x (k)} and δ _n (k), δ _n-1 (k)} Since L is a vector value having L orders as mentioned above, assuming that L = 1024, a large amount of calculation as described later is required when t = k.

n = 0, δ ₀ (k) = βδ ₀ (k-1) + (1-β) x ₀ (k) e (k) / (σ ² _x + σ ² _e ) ^1/2

n = 1, δ ₁ (k) = βδ ₁ (k-1) + (1-β) x ₁ (k) e (k) / (σ ² _x + σ ² _e ) ^1/2

...

n = 1023, δ ₁₀₂₃ (k) = βδ ₁₀₂₃ (k-1) + (1-β) x ₁₀₂₃ (k) e (k) / (σ ² _x + σ ² _e ) ^1/2

However, the coefficients {δ ₀ (k), δ ₁ (k), of the cross-correlation {δ _n (k)} calculated from (n = 0) to (n = 1023) at the time t = k. .. _,, δ ₁₀₂₃ (k)} tends to be the same, so that only the cross-correlation {δ ₁ (k)} for a specific coefficient (n = 1) is calculated (S13) and the cross-correlation of the calculated specific coefficient {δ By applying ₁ (k) = δ (k)} to the remaining coefficients as it is, it is not necessary to calculate each coefficient one by one, and the calculation amount can be greatly reduced. Equivalent to that, and corresponds to the graph shown in Figure 6a.

δ (k) = βδ (k-1) + (1-β) x (k) e (k) / (σ ² _x + σ ² _e ) ^1/2 Equation (9)

As shown in FIG. 6A, when the correlation coefficient {δ (k)} is updated and calculated such that the time coefficient k is close to t = 5000, the graph is represented as a received acoustic signal {x (k)}. The cross-correlation between the estimated error signals e (k) {δ (k)} decreases exponentially with orthogonal principle over time and becomes a stable state, and its value is significantly It can be seen that the smaller.

By the way, if the cross-correlation degree δ (k) calculated by Equation (9) decreases exponentially too quickly as in Fig. 6A, the cross-correlation degree δ (k) is determined to converge stably. Since the frequency of out of range becomes frequent and the convergence speed at the initial stage of adaptation is significantly lowered, it is necessary to keep the cross-correlation degree δ (k) somewhat larger in order to accelerate the convergence speed at the initial stage of adaptation. Accordingly, the low pass filter 104 passes the cross-correlation degree {δ (k) calculated by Eq. (9) as low-pass as in Eq. ), So that the convergence speed is faster than the cross-correlation degree (δ (k)) of Eq. (9) that does not pass low.

c (k) = βc (k-1) + (1-β) δ (k) equation (10)

That is, in Equation (10), the low-pass cross-correlation degree {c (k-1)} is multiplied by β at the time t = k-1, and the instantaneous cross-correlation degree {δ ( k)} by multiplying (1-β) by randomly adjusting the value of β, or weighting the cumulative value {c (k-1)} of the low-passed cross-correlation calculated up to the last; It is possible to control whether or not to give specific gravity to the cross-correlation degree {δ (k)} at the time point. The adjustable range of β value is a value between 0 and 1 and the value of β is set to be close to 1 The cross-correlation calculated by) has a low pass filter and usually the β value is set to 0.98.

The time-varying adaptive constant calculator 105 receives the received acoustic signal {x (k)} and the estimated error signal {e (k)} based on the low pass cross-correlation degree {c (k)} calculated from Equation (10). A time-varying adaptive constant {mu (k)} that varies depending on the degree of cross-correlation of is calculated as in Equation (11) (S15).

μ (k) = α│c (k) │ / {L (σ ² _x + σ ² _e )} (11)

In this case, the absolute value of the cross-correlation degree {c (k)} is taken to prevent the coefficient vector {W (k)} calculated therefrom from becoming unstable when the adaptive constant is smaller than 0. Dividing the time-varying adaptive constant {μ (k)} by the sum of the powers of {x (k)} and the estimated error signal {e (k)} results in the time-varying adaptive constant {μ (k) calculated by Equation (11). This is to normalize the value of}} within a predetermined range.

6B shows the time-varying adaptive constant {μ (k)} calculated by substituting the low-pass cross-correlation degree {c (k)} into Equation (11) and the low-pass adaptive constant calculator 105. A graph showing the comparison of the time-varying adaptation coefficient {μ (k)} obtained by substituting the uncorrelated correlation {δ (k)} into Equation (11). Shows that the low-pass time-varying adaptive constant {μ (k)} is somewhat delayed and softer than the low-pass time-varying adaptive constant {μ (k)}. You can check it. Therefore, by using the low-pass time-varying adaptation constant, it is possible to maintain a fast convergence speed when the initial adaptation and the path change.

The time-varying adaptive constant {mu (k)} output from the time-varying adaptive constant calculator 105 is input to a coefficient vector updater 106, and the coefficient vector updater 106 receives the input time-varying adaptive constant {μ ( k)} is substituted into equation (12) to calculate the coefficient vector {W (k)} (S16). Subsequently, the time coefficient increaser 107 increases the time coefficient k by 1 (S17), and the echo path estimation signal generator 12 shown in FIG. 2 calculates the coefficient vector {W (k (k) calculated from equation (12). ), An acoustic echo estimation signal for estimating the echo path from the speaker to the microphone { (k)} is generated by updating (S18), the acoustic echo signal {y (k)} included in the estimated error signal {e (k)} to be transmitted to the speaker side of the far-end is removed as much as possible, so that the estimated error signal { e (k)} will be minimized.

W (k) = W (k−1) + α│c (k) │ / {L (σ ² _x + σ ² _e )} {y (k) -W ^T (k) X (k)} + α C (k) / {L (σ ² _x + σ ² _e )} X (k) n (k) Equation (12)

7 (a) to 7 (e) show the results of experiments for removing acoustic echoes by the time-varying adaptive algorithm according to the present invention, compared with the results of the above experiments by conventional NLMS and Sum-LMS algorithms. Where (a) is a received sound signal {x (k)}, (b) is a near-end speaker signal n (k)}, and (c), (d) and (e) are NLMS, respectively. Each estimated error signal e (k) is obtained by removing acoustic echo by the algorithm, the Sum-LMS algorithm and the time-varying adaptive algorithm according to the present invention. As shown, the estimated error signal (e) from which the acoustic echo signal is removed by the time-varying adaptive algorithm according to the present invention is estimated with the estimated error signal {(c) from which the acoustic echo signal is removed by conventional algorithms (NLMS, Sum-LMS). Compared with (d)}, the voice signal {n (k)} on the speaker side near the end is well preserved, and the removal effect is remarkable after the voice signal {n (k)} on the speaker near the end is generated. Can be.

FIG. 8A is a configuration of an echo echo estimator calculated according to the present invention and an acoustic echo canceller in which an echo path measured by an experiment is implemented, and for measuring characteristics of an acoustic echo signal input through a echo path through a microphone, the present invention; In the space having the same acoustic environment as the echo path estimator 100 calculated by the time-varying adaptive algorithm according to the present invention, an acoustic signal corresponding to a single impulse is generated and the acoustic echo signal input to the microphone by the generated acoustic signal It is configured to include the echo path 200 means the acoustic characteristics between the speaker and the microphone by actually measuring the, and the echo path 200 of the acoustic echo signal {y (k)} actually measured as described above Coefficient misregistration for confirming that the coefficient vector {H (k)} and the coefficient vector {W (k)} of the echo path estimator 100 implemented in FIGS. 4 and 5 mentioned above coincide. The positive index is calculated as shown in equation (13).

Coefficient misadjustment index = 10 · log [{W (k) -H (k)} ² / H ² (k)] Equation (13)

FIG. 8B is a graph illustrating a coefficient misalignment index indicating whether coefficient coefficients of the echo path estimator 100 and the echo path 200 correspond to each other by the time-varying adaptive algorithm according to the present invention. The echo path estimator 10 is compared with the coefficient misadjustment index indicating whether the coefficient vectors of the echo paths (not shown) coincide with each other. As shown in Fig. 8B, when the difference between the two coefficient vectors {H (k), W (k)} is large, the coefficient misalignment index value calculated by equation (13) is large and the acoustic echo signal {y (k)}. Is not properly removed and the two coefficient vectors {H (k), W (k)} match, the coefficient misadjustment index value calculated by Eq. (13) is small and the acoustic echo signal {y (k)} is sufficiently removed. As can be appreciated, it can be seen that the coefficient misalignment index (proposed) to which the time-varying adaptive algorithm according to the present invention is applied exhibits a smaller value than the coefficient misalignment index (NLMS, SumLMS) to which the conventional algorithm is applied. .

The present invention is not limited to the embodiments described above, and various modifications and changes can be made by those skilled in the art, which are included in the spirit and scope of the present invention as defined in the appended claims.

As described above, the present invention has an effect that the acoustic reflection generated in the hand-free mobile phone or the speakerphone is cleanly removed regardless of other noise components.

In addition, the present invention has the effect that the operation to remove the acoustic echo is made more quickly.

In addition, the present invention has the effect of making a more efficient call using a hand-free mobile phone or speakerphone.

Claims (20)

A sound output unit for outputting a received sound signal received from the far end speaker to the near end speaker, and a transmission sound signal to be transmitted from the near end speaker to the far end speaker. In the communication system having a sound input unit for input, and having an acoustic echo path from the sound output unit to the sound input unit, And echo path estimating means for generating an acoustic echo estimation signal for estimating the acoustic echo signal changed by the received acoustic signal passing through the acoustic echo path, wherein the acoustic echo estimation signal is generated from the transmission acoustic signal. And a time-varying correlation algorithm between the estimated error signal generated by subtracting the estimated acoustic echo signal and the received acoustic signal. The method of claim 1, wherein the echo path estimation means, Calculating means for calculating a correlation between the received sound signal to be output from the sound output unit and the estimated error signal; And And an generating means for generating the acoustic echo estimation signal in response to the calculated cross-correlation. The method of claim 2, wherein the echo path estimation means, And a power measuring means for measuring the power of the received sound signal and the estimated error signal to be output, wherein the calculating means further calculates the cross-correlation based on each of the measured powers. Acoustic echo canceller applied to a time-varying adaptive algorithm. The method of claim 2 or 3, wherein the echo path estimation means, And a low pass filter means for low-passing the calculated cross-correlation diagram, wherein the generation means further generates the acoustic echo estimation signal in response to the low-pass cross-correlation diagram. Applied acoustic echo canceller. The method of claim 2 or 3, wherein the echo path estimation means, The calculated cross-correlation is characterized in that the vector value, and further comprising an extraction means for extracting a specific cross-correlation for a specific coefficient of the calculated cross-correlation, wherein the generating means is the extracted specific And an acoustic echo estimation signal further generated according to the cross-correlation. The method of claim 5, wherein the echo path estimation means, And a low pass filtering means for low-passing the extracted specific cross-correlation, wherein the generating means further generates the acoustic echo estimation signal in response to the low-passing specific cross-correlation. Acoustic echo canceller with algorithm. The method of claim 2 or 3, wherein the echo path estimation means, Calculating means for calculating a time-varying adaptation constant based on the calculated cross-correlation degree; And And resetting means for resetting a predetermined coefficient vector in response to the calculated time-varying adaptive constant, wherein the generating means generates the acoustic echo estimation signal based on the reset coefficient vector. Applied acoustic echo cancellation device. The method of claim 7, wherein the echo path estimation means, And a low pass filter means for low-passing the calculated cross-correlation diagram, wherein the generation means further generates the acoustic echo estimation signal in response to the low-pass cross-correlation diagram. Applied acoustic echo canceller. The method of claim 7, wherein the echo path estimation means, The calculated cross-correlation is characterized in that the vector value, and further comprising extraction means for extracting a specific cross-correlation for a specific coefficient of the calculated cross-correlation, wherein the calculation means An acoustic echo cancellation device to which a time-varying adaptive algorithm is applied, further comprising time-varying adaptive constants in response to cross-correlation. The method of claim 9, wherein the echo path estimation means, And a low pass filtering means for low-passing the extracted specific cross-correlation, wherein the generating means further generates the acoustic echo estimation signal in response to the low-passing specific cross-correlation. Acoustic echo canceller with algorithm. A sound output unit for outputting a received sound signal received from the far end speaker to the near end speaker, and a transmission sound signal to be transmitted from the near end speaker to the far end speaker. A method for removing acoustic echo of a communication system having an acoustic input unit for inputting and having an acoustic echo path from the acoustic output unit to the acoustic input unit, And generating an acoustic echo estimation signal for estimating the acoustic echo signal changed by the received acoustic signal passing through the acoustic echo path, wherein the acoustic echo estimation signal is an acoustic echo previously generated from the transmission acoustic signal. And a time varying correlation algorithm between the estimated error signal generated by subtracting an estimated signal and the received sound signal. The method of claim 11, Calculating a correlation between the received sound signal to be output from the sound output unit and the estimated error signal; And And a generation step of generating the acoustic echo estimation signal in response to the calculated cross-correlation. The method of claim 12, And a power measuring step of measuring power of the received sound signal and the estimated error signal to be output, respectively, wherein the calculating step further calculates the cross-correlation based on each of the measured powers. Acoustic echo cancellation method using time-varying adaptive algorithm. The method according to claim 12 or 13, And a low pass filtering step of low-passing the calculated cross-correlation, wherein the generating step further comprises generating the acoustic echo estimation signal in response to the low-passing cross-correlation. Applied acoustic echo cancellation method. The method according to claim 12 or 13, The calculated cross-correlation is characterized in that the vector value, and further comprising an extraction step of extracting a specific cross-correlation for a specific coefficient of the calculated cross-correlation, wherein the generating step is the extracted specific The acoustic echo canceling method to which the time-varying adaptive algorithm is further generated according to the cross-correlation. The method of claim 15, And a low pass filtering step for lowpassing the extracted specific cross-correlation, wherein the generating step further generates the acoustic echo estimation signal in response to the low-passing specific cross-correlation. Acoustic echo cancellation method with algorithm. The method according to claim 12 or 13, A lower calculation step of calculating a time-varying adaptation constant based on the calculated cross-correlation; And And a resetting step of resetting a predetermined coefficient vector in response to the calculated time-varying adaptive constant, wherein the generating step generates the acoustic echo estimation signal based on the reset coefficient vector. Applied acoustic echo cancellation method. The method of claim 17, And a low pass filtering step of low-passing the calculated cross-correlation, wherein the generating step further comprises generating the acoustic echo estimation signal in response to the low-passing cross-correlation. Applied acoustic echo cancellation method. The method of claim 17, The calculated cross-correlation is characterized in that the vector value, and further comprising an extraction step of extracting a specific cross-correlation for a specific coefficient of the calculated cross-correlation, wherein the calculating step is the extracted specific A method for canceling acoustic echo using a time-varying adaptive algorithm, characterized in that further calculating a time-varying adaptive constant according to the cross-correlation. The method of claim 19, And a low pass filtering step for lowpassing the extracted specific cross-correlation, wherein the generating step further generates the acoustic echo estimation signal in response to the low-passing specific cross-correlation. Acoustic echo cancellation method with algorithm.