JP4705893B2 - Echo canceller - Google Patents

Description

  The present invention relates to an echo canceller for canceling echo generated by wraparound of a received voice from a speaker to a microphone or wraparound of a received signal to a transmission side in a hybrid transformer of two-wire / four-wire conversion.

FIG. 2 is a configuration diagram of a conventional echo canceller.
In the echo canceller, the received signal x (t) input from the far end is converted into an analog signal by a DAC (digital-to-analog converter) 1 and output from the near-end speaker 2 to wrap around the microphone 3 as an acoustic signal. The analog signal is converted into a digital signal by an ADC (analog / digital converter) 4 and transmitted to the far end as an echo.

  The echo canceller 5 calculates a reception signal x (t) at a discrete time t with a tap coefficient H (t) and generates a pseudo echo r (t), and a transmission signal y ( The adder 5b outputs a residual signal e (t) by subtracting the pseudo echo r (t) from t). The residual signal e (t) is output to the transmission line and is given to the adaptive filter 5a as a tap coefficient update signal.

In addition to the audio signal s (t), the background noise n (t) and the acoustic echo d (t) from the speaker 2 are input to the microphone 3, and these are converted into digital signals by the ADC 4 to be transmitted signals y (t ) Is output. Assuming that the impulse response of the echo path of the acoustic echo d (t) can be approximated by an mth-order FIR (finite impulse response) filter, the tap coefficient H (t) of the adaptive filter 5a at the discrete time t and this The received signal x (t) input to the adaptive filter 5a is expressed as follows.
H (t) = [h1 (t), h2 (t),..., Hm (t)] ^T
X (t) = [x (t), x (t−1),..., X (t−m + 1)] ^T
However, T represents transposition of a vector.

Thus, the pseudo echo r (t) and the residual signal e (t) are expressed as follows.
r (t) = H (t) ^T X (t)
e (t) = y (t) -r (t)

  That is, the residual signal e (t) is a signal obtained by subtracting the pseudo echo r (t) from the transmission signal y (t) and canceling the echo.

The update of the tap coefficient H (t) is generally expressed as follows.
H (t + 1) = H (t) + μΔH (t)

  Here, μ is a step size for adjusting the convergence speed of the tap coefficient, and the update vector ΔH (t) varies depending on the type of the adaptive algorithm.

As an adaptive algorithm, an LMS algorithm (least mean square method), an RLS algorithm (recursive least square method), and the like are widely known, but an NLMS algorithm (learning identification) that has a relatively small amount of calculation and exhibits good convergence characteristics Method) is often used. Update of the tap coefficient H (t) by the NLMS algorithm is expressed as follows.
H (t + 1) = H (t) + μ [e (t) X (t)] / ‖X (t) ‖ ²
In this case, μ is 0 <μ <2.

  The tap coefficient H (t) is updated only in a single talk state where the received signal x (t) is present but the voice signal s (t) is not present, and the received signal x (t) and the voice signal are updated. Updating is not performed when the double talk state where both s (t) exist and when the reception signal x (t) does not exist. In general, the amount of echo cancellation decreases as the background noise n (t) increases.

JP 2002-94419 A JP-A-10-301600

  However, in the echo canceller, when the received signal x (t) input to the adaptive filter 5a is a colored signal such as speech (a signal having a nonuniform spectrum distribution), the convergence speed is significantly reduced. there were. As a solution to this problem, Patent Document 1 proposes a method of increasing the convergence speed by converting an input signal into a white signal (a signal having a uniform spectral distribution) using an inverse filter. However, the configuration of the proposed method has a problem that the accuracy of the echo cancellation process is lowered when the echo delay time is large or when the input signal is originally a white signal. Further, since the speech spectrum is generally inclined (about -6 dB / Oct in the case of vowels), the inverse filter has a high frequency emphasis characteristic. If background noise is included in the transmission signal, the noise is reduced. There was a problem that the component was remarkably amplified and the amount of echo cancellation decreased.

  An object of the present invention is to provide an echo canceller having a high convergence speed regardless of the characteristics of a received signal.

An echo canceller according to claim 1 of the present invention includes a delay buffer that delays a received signal and outputs a delayed received signal, a linear prediction unit that calculates a reflection coefficient and a linear prediction coefficient from the received signal, A first inverse filter for whitening the delayed received signal using the linear prediction coefficient to generate a whitened received signal, and the delayed received signal and the whitened received signal are each calculated with the same tap coefficient and simulated. An echo and whitened pseudo echo are output, an adaptive filter that updates the tap coefficient according to a given residual signal, an echo delay time is estimated based on the tap coefficient of the adaptive filter, and the linear prediction unit A delay control unit that outputs the output linear prediction coefficient by delaying the delay time, and a transmission signal using the linear prediction coefficient output from the delay control unit. A second inverse filter for colorizing and generating a whitened transmission signal; a first adder for subtracting the pseudo echo from the transmission signal to output a first residual signal; and A second adder that subtracts the whitened pseudo echo and outputs a second residual signal ; a wideband signal determination unit that determines whether the received signal is a white signal based on the reflection coefficient; and the tap When the coefficient is not converged and the received signal is not a white signal, the second residual signal is selected, and when the tap coefficient is converged or the received signal is a white signal, the second residual signal is selected. select 1 of the residual signal to the residual signal switching unit providing the selected residual signal to the adaptive filter, comprising the.
An echo canceller according to a second aspect of the present invention is the echo canceller according to the first aspect of the present invention, further comprising a narrowband signal determining unit that determines whether the received signal is a tone signal based on the reflection coefficient, When the received signal is determined to be a tone signal, updating of the tap coefficient in the adaptive filter is stopped.
The echo canceller according to a third aspect of the present invention is the echo canceller according to the first or second aspect, wherein the transmission signal and the first residual signal, or the whitened transmission signal and the second residual signal are used. A signal-to-noise ratio calculating unit that calculates a signal-to-noise ratio that is a ratio between the echo included in the transmission signal and the background noise included in the transmission signal, and based on the signal-to-noise ratio The step size at the time of updating the tap coefficient in the adaptive filter is adjusted.
An echo canceller according to a fourth aspect of the invention is the echo canceller according to the first, second or third aspect, wherein the linear prediction coefficient calculated by the linear prediction unit is written, and in response to a request from the delay control unit A first-in first-out buffer for reading out and outputting the linear prediction coefficients in the order of writing is provided.
The echo canceller of the invention according to claim 5 uses the delay buffer that delays the received signal and outputs the delayed received signal, the linear prediction unit that calculates the linear prediction coefficient from the received signal, and the linear prediction coefficient The delayed reception signal is whitened, the first inverse filter that generates the whitened reception signal, and the delayed reception signal and the whitened reception signal are calculated with the same tap coefficient, and a pseudo echo and a whitened pseudo echo are output. And an adaptive filter that updates the tap coefficient according to a given residual signal, an echo delay time is estimated based on the tap coefficient of the adaptive filter, and the linear prediction coefficient output from the linear prediction unit is A delay control unit that outputs the signal after being delayed by the delay time, and a whitened transmission signal using the linear prediction coefficient output from the delay control unit A second inverse filter to be generated; a first adder that subtracts the pseudo echo from the transmission signal to output a first residual signal; and a white adder that subtracts the whitened pseudo echo from the whitened transmission signal. A second adder that outputs a residual signal of 2 and the second residual signal when the tap coefficient has not converged, and the first residual signal when the tap coefficient has converged And an echo canceller comprising: a residual signal switching unit that supplies the selected residual signal to the adaptive filter, wherein the transmission signal and the first residual signal or the whitened transmission signal And a signal-to-noise ratio calculation unit that calculates a signal-to-noise ratio that is a ratio between the echo included in the transmission signal and the background noise included in the transmission signal based on the second residual signal, Based on the signal-to-noise ratio, Characterized by being configured to adjust the step size for updating the tap coefficients in the filter.

According to the first aspect of the present invention, when the tap coefficients of the adaptive filter are not converged and the received signal is not a whiteness signal, the whitened pseudo echo is subtracted from the whitened transmission signal. The first residual obtained by subtracting the pseudo echo from the transmitted signal when the tap coefficient of the adaptive filter is converged or the received signal is a whiteness signal. Since the tap coefficient is updated using the signal, the tap coefficient can be quickly converged regardless of the characteristics of the received signal.
In particular, in the second inverse filter that generates the whitened transmission signal, the linear prediction coefficient output from the linear prediction unit is delayed by the delay time of the echo estimated based on the tap coefficient of the adaptive filter. Since the transmission signal is whitened using the linear prediction coefficient, a residual signal in consideration of the echo delay time can be obtained, and the tap coefficients can be converged quickly and accurately. In addition, since it has a high-band signal determination unit, when the received signal is a whiteness signal, the adaptive filter tap coefficient is immediately updated without waiting for the switching timing by the residual signal switching unit. To switch to a stable state.

According to the second aspect of the present invention, the narrowband signal determination unit is provided, and when the received signal is determined to be a tone signal, the updating of the tap coefficient in the adaptive filter is stopped. Divergence of tap coefficients can be prevented when the signal is a tone signal.
According to the invention of claim 3, the S / N (signal-to-noise ratio) calculation unit is provided, and the tap coefficient is updated according to the S / N that is the ratio of the echo and the background noise included in the transmission signal. Since the configuration is such that the step size is adjusted, for example, when there is a lot of background noise in the transmission signal, the update speed of the tap coefficient is decreased to reduce the convergence speed. The influence of the high-frequency noise component amplified by the inverse filter can be reduced, and the reduction in the amount of echo cancellation can be reduced. Further, when there is not much background noise in the transmission signal, the convergence rate can be improved by increasing the update amount of the tap coefficient.
According to the invention of claim 4, since the FIFO (first-in first-out) buffer is provided, the tap coefficients can be accurately converged regardless of the echo delay time.

According to the fifth aspect of the invention, as in the first aspect of the invention, the delay of the echo estimated based on the tap coefficient of the adaptive filter by the delay control unit using the linear prediction coefficient output from the linear prediction unit. Since the transmission signal is whitened by the second inverse filter using the delayed linear prediction coefficient, the residual signal taking into account the echo delay time is obtained, and taps quickly and accurately. The coefficient can be converged. Furthermore, since the S / N calculation unit is provided, as in the invention according to claim 3, for example, when there is a lot of background noise in the transmission signal, the update rate of the tap coefficient is reduced to reduce the convergence speed. However, the influence of the high-frequency noise component amplified by the second inverse filter can be reduced, and the reduction in the echo cancellation amount can be reduced. Further, when there is not much background noise in the transmission signal, the convergence rate can be improved by increasing the update amount of the tap coefficient.

  The above and other objects and novel features of the present invention will become more fully apparent when the following description of the preferred embodiment is read in conjunction with the accompanying drawings. However, the drawings are for explanation only, and do not limit the scope of the present invention.

  FIG. 1 is a configuration diagram of an echo canceller showing Embodiment 1 of the present invention. The solid line in the figure indicates the flow of the audio signal, and the broken line indicates the flow of the control signal.

  In the echo canceller, the received signal input from the far end is converted into an analog signal by the DAC 1, output from the near-end speaker 2, and circulates as an acoustic signal to the microphone 3, and converted to a digital signal by the ADC 4 to be distant as an echo. This prevents transmission to the end.

  The echo canceller 10 outputs a reception signal x (t) obtained by delaying the reception signal x (t + N) input from the far end by a discrete time N, and a p-th order from the reception signal x (t + N). The linear prediction unit 12 for calculating the reflection coefficient and the linear prediction coefficient LPC1. The received signal x (t) output from the delay buffer 11 is supplied to the DAC 1 and is also supplied to the inverse filter 13 and the adaptive filter 14. The linear prediction coefficient LPC 1 output from the linear prediction unit 12 is the same as that of the inverse filter 13. The delay control unit 15 is provided.

  The inverse filter 13 generates a reception signal xw (t) obtained by whitening the reception signal x (t) using the linear prediction coefficient LPC1, and supplies the reception signal xw (t) to the adaptive filter 14. The adaptive filter 14 calculates pseudo echoes r (t) and rw (t) by calculating received signals x (t) and xw (t) with tap coefficients H (t), respectively. Further, the delay control unit 15 estimates the echo delay time based on the tap coefficient H (t) of the adaptive filter 14, and linearly predicts the input linear prediction coefficient LPC1 at a timing delayed by the estimated delay time. This is output as the coefficient LPC2.

  Further, the echo canceller 10 includes an inverse filter 16 and an adder 17 to which a transmission signal y (t) is given from the ADC 4. The inverse filter 16 generates a transmission signal yw (t) obtained by whitening the transmission signal y (t) using the linear prediction coefficient LPC2 provided from the delay control unit 15 and supplies the transmission signal yw (t) to the adder 18.

  The adder 17 subtracts the pseudo echo r (t) generated by the adaptive filter 14 from the transmission signal y (t) and outputs a residual signal e (t). The residual signal e (t) It is output to the transmission line and given to one input side of the changeover switch 19. The adder 18 subtracts the whitened pseudo echo rw (t) generated by the adaptive filter 14 from the transmission signal yw (t) and outputs a residual signal ew (t). A signal ew (t) is given to the other input side of the changeover switch 19.

  The changeover switch 19 selects one of the residual signals e (t) and ew (t) based on the changeover signal SW given from the changeover control unit 20, and gives it to the adaptive filter 14 as a tap coefficient update signal. Is. After the operation permission signal EN is given to the echo canceller 10, the switching control unit 20 provides a residual signal ew to the changeover switch 19 until the tap coefficient of the adaptive filter 14 is updated a predetermined number of times. (T) is selected, and thereafter, a switching signal SW for selecting the residual signal e (t) is output.

  The switching control unit 20 is configured by, for example, a counter, and counts a common clock signal for processing (not shown) and outputs a constant count value when a single talk state is output by the detection signal WT of the double talk detection unit 21. The switch 19 is configured to switch from the residual signal ew (t) to e (t). The double talk detector 21 detects the state of single talk or double talk by comparing and determining the levels of the received signal x (t) and the residual signal e (t). Although not shown, the tap coefficient updating operation of the adaptive filter 14 is performed only in the single talk state based on the detection signal WT of the double talk detector 21.

Next, the operation will be described.
The p-th order linear prediction coefficient LPC1 of the linear prediction unit 12 is a0, a1,..., Ap, and the inverse filters 13 and 16 are as follows.

  When N received signals x (t), x (t + 1),..., X (t + N−1) corresponding to analysis frame units are input to the linear predictor 12, the linear predictor 12 receives these signals. The reflection coefficients c1, c2,..., Cp are calculated from the signal using the autocorrelation method, covariance method, Burg method, etc., and the linear prediction coefficient LPC1 (a0, a1,..., Ap) is calculated based on the calculated reflection coefficients. And the linear prediction coefficient LPC1 is set in the inverse filter 13.

  Further, the N received signals x (t) to x (t + N−1) are delayed by N samples by the delay buffer 11 and given to the inverse filter 13. As a result, the received signals x (t) to x (t + N−1) are inversely filtered using the linear prediction coefficient LPC1 calculated from these received signals, and the received signal xw (t) that has been whitened with high accuracy is obtained. Output from the inverse filter 13.

  Further, the linear prediction coefficient LPC1 calculated by the linear prediction unit 12 is provided to the delay control unit 15, and is set in the inverse filter 16 as the linear prediction coefficient LPC2 at a timing delayed by the echo delay time by the delay control unit 15. . The delay control unit 15 estimates the echo delay time based on the position of the maximum absolute value of the tap coefficient H (t) of the adaptive filter 14. For example, if the sampling frequency is 8 kHz and the 80th tap coefficient has the maximum absolute value, the delay control unit 15 sets the echo delay time to 10 ms (= 0.125 ms × 80) as shown in FIG. Judgment is made, and the linear prediction coefficient LPC2 is set in the inverse filter 16 with a delay of 10 ms from the timing when the linear prediction coefficient LPC1 is given.

  On the other hand, in addition to the audio signal s (t), the background noise n (t) and the acoustic echo d (t) are input to the microphone 3, and these are converted into digital signals by the ADC 4 and used as the transmission signal y (t). Is output. The transmission signal y (t) is given to the inverse filter 16 and the adder 17.

  In the inverse filter 16, the transmission signal y (t) is whitened using the linear prediction coefficient LPC 2, and the transmission signal yw (t) is generated and given to the adder 18. The adder 17 subtracts the pseudo echo r (t) output from the adaptive filter 14 from the transmission signal y (t), and outputs a residual signal e (t). Further, the adder 18 subtracts the pseudo echo rw (t) output from the adaptive filter 14 from the transmission signal yw (t), and outputs a residual signal ew (t).

  The residual signal e (t) is output to the far end through the transmission line. Further, the residual signals e (t) and ew (t) are given to the changeover switch 19, and one of them is selected according to the changeover signal SW from the changeover control unit 20, and the adaptive filter 14 is used for updating the tap coefficient. Given as a signal.

  After the operation of the echo canceller 10 is started by the operation permission signal EN, the residual signal ew is sent from the switch control unit 20 to the switch 19 until the tap coefficient of the adaptive filter 14 is updated a predetermined number of times. A switching signal SW for selecting (t) is output.

At this time, the tap coefficient H (t) of the adaptive filter 14 at the discrete time t and the input signal to the adaptive filter 14, that is, the signal Xw (t) output from the inverse filter 13 are respectively expressed as follows. The
H (t) = [h1 (t), h2 (t),..., Hm (t)] ^T
Xw (t) = [xw (t), xw (t−1),..., Xw (t−m + 1)] ^T

Thus, the pseudo echo rw (t) and the residual signal ew (t) are expressed as follows.
rw (t) = H (t) ^T Xw (t)
ew (t) = yw (t) -rw (t)

The update of the tap coefficient H (t) is expressed as follows.
H (t + 1) = H (t) + μ [ew (t) Xw (t)] / ‖Xw (t) ‖ ²

  Here, since xw (t) and yw (t) are white signals from which autocorrelation has been removed, the tap coefficient H (t) converges at a high speed.

  When the number of updates of the tap coefficient of the adaptive filter 14 reaches a predetermined number, the tap coefficient H (t) is considered to have converged, and the residual signal e (t) is sent from the switching control unit 20 to the changeover switch 19. A switching signal SW to be selected is output.

As a result, the echo canceller 10 has the same configuration as the conventional echo canceller 5, and the tap coefficient H (t) of the adaptive filter 14 at the discrete time t and the input signal to the adaptive filter 14, that is, the received signal X ( t) is expressed as follows.
H (t) = [h1 (t), h2 (t),..., Hm (t)] ^T
X (t) = [x (t), x (t−1),..., X (t−m + 1)] ^T

The pseudo echo r (t) and the residual signal e (t) are expressed as follows.
r (t) = H (t) ^T X (t)
e (t) = y (t) -r (t)

Then, the update of the tap coefficient H (t) is expressed as follows as in the conventional case.
H (t + 1) = H (t) + μ [e (t) X (t)] / ‖X (t) ‖ ²

  As described above, the echo canceller 10 according to the first embodiment is provided with the delay buffer 11 for buffering the received signals for the samples corresponding to the analysis frame of the linear prediction unit 12, and the linearity calculated by the linear prediction unit 12 is provided. Since the reception signal x (t) output from the delay buffer 11 using the prediction coefficient LPC1 is inversely filtered by the inverse filter 13, the whitening accuracy of the inverse filter 13 is improved. Furthermore, a delay control unit 15 is provided for controlling the timing of the linear prediction coefficient LPC2 set in the inverse filter 16 according to the echo delay time estimated from the position of the maximum absolute value of the tap coefficient H (t) of the adaptive filter 14. Therefore, the whitening accuracy of the inverse filter 16 is improved. Therefore, at the initial stage when the operation of the echo canceller 10 is started, the tap of the adaptive filter 14 is performed using the whitened reception signal xw (t) and transmission signal yw (t) output from the inverse filters 13 and 16. By updating the coefficient H (t), the tap coefficient H (t) can be converged at high speed and with high accuracy.

  Further, after the tap coefficient H (t) has converged, the tap coefficients of the adaptive filter 14 are used by using the reception signal x (t) and the transmission signal y (t) as usual without using the inverse filters 13 and 16. Since the changeover switch 19 is switched so as to update H (t), there is an advantage that the accuracy of echo cancellation can be kept high.

  FIG. 4 is a block diagram of an echo canceller showing Embodiment 2 of the present invention, and common elements to those in FIG. 1 are denoted by common reference numerals.

  The echo canceller 10A holds the linear prediction coefficient LPC1 output from the linear prediction unit 12 with respect to the echo canceller 10 of FIG. 1, and reads and outputs the held linear prediction coefficient in accordance with a read request from the delay control unit 15A. A FIFO (first-in first-out) buffer 22 is provided. In addition to the same function as the delay control unit 15 in FIG. 1, the delay control unit 15A reads the FIFO buffer 22 after a predetermined delay time from the output timing of the linear prediction coefficient LPC1 output from the linear prediction unit 12, A function of outputting to the inverse filter 15 as a linear prediction coefficient LPC2 is added. Other configurations are the same as those in FIG.

FIG. 5 is an explanatory diagram for comparing the operations of the echo cancellers of FIG. 1 and FIG.
The p-th order linear prediction coefficient LPC1 calculated in the i-th analysis frame in the linear prediction unit 12 is expressed as follows.
Ai = [a0i, a1i, ..., api]

  In the echo canceller 10 of FIG. 1, as shown in FIG. 5A, the linear prediction coefficient LPC1 set in the inverse filter 13 is set in the inverse filter 16 as the linear prediction coefficient LPC2 with a delay of the echo delay time TD. I have control. For this reason, when the echo delay time TD is longer than the time TF of the analysis frame, the linear prediction coefficient LPC1 is updated to the linear prediction coefficient for the next analysis frame, and the correct linear prediction coefficient LPC2 is set in the inverse filter 16. Will not be able to.

  On the other hand, in the echo canceller 10A according to the second embodiment, as shown in FIG. 5B, the linear prediction coefficient LPC1 is temporarily stored in the FIFO buffer 22 and read according to the read request from the delay control unit 15A. Thus, the inverse filter 16 is controlled as the linear prediction coefficient LPC2. Thereby, even if the echo delay time TD is longer than the time TF of the analysis frame, the correct linear prediction coefficient LPC2 can always be set in the inverse filter 16.

  Therefore, the echo canceller 10A of the second embodiment has the advantage that the tap coefficient H (t) can be accurately converged regardless of the echo delay time TD in addition to the same advantages as the echo canceller 10 of FIG. There is.

  FIG. 6 is a block diagram of an echo canceller showing Embodiment 3 of the present invention. Elements common to those in FIG. 1 are denoted by common reference numerals.

  The echo canceller 10B is obtained by adding a wideband signal determination unit 23 and an OR (logical sum) gate 24 to the echo canceller 10 of FIG. The broadband signal determination unit 23 compares the second-order reflection coefficient c2 output from the linear prediction unit 12 with the threshold value TH1, and when the reflection coefficient c2 is smaller than the threshold value TH1, the spectrum of the received signal x (t) Are detected and the characteristics are close to those of a white signal, and a detection signal DW for switching to select the residual signal e (t) is output. The OR gate 24 selects the logical sum of the signal for switching so as to select the residual signal e (t) output from the switching control unit 20 and the detection signal DW output from the wideband signal determination unit 23 as a selection signal SW. Is given to the changeover switch 19.

  In this echo canceller 10B, the second-order reflection coefficient c2 derived in the calculation process of the linear prediction unit 12 is compared with the threshold value TH1. The second-order reflection coefficient c2 represents the degree of sparse / dense spectrum, that is, the degree of dispersion / concentration with respect to the entire band. Therefore, when the value of the reflection coefficient c2 is smaller than the preset threshold value TH1, it can be determined that the spectrum is dispersed and the characteristic is close to that of a white signal. Thus, when the detection signal DW is output from the wideband signal determination unit 23, the changeover switch 19 is switched to select the residual signal e (t) regardless of the signal from the switching control unit 20.

  As described above, the echo canceller 10B according to the third embodiment determines that the received signal x (t) has characteristics close to those of the white signal based on the secondary reflection coefficient c2 output from the linear prediction unit 12. Sometimes, it has a broadband signal determination unit 23 for switching to select the residual signal e (t). As a result, in addition to the same advantages as the first embodiment, the third embodiment immediately adapts the adaptive filter without waiting for the switching timing by the switching control unit 20 when the received signal x (t) is close to a white signal. There is an advantage that the update of the 14 tap coefficients H (t) can be switched to a normal state to shift to a stable state.

  FIG. 7 is a configuration diagram of an echo canceller showing Embodiment 4 of the present invention, and common elements to those in FIG. 1 are denoted by common reference numerals.

  The echo canceller 10C is obtained by adding an S / N detector 25 to the echo canceller 10 of FIG. 1 and providing an adaptive filter 14A having a function added in place of the adaptive filter 14. The S / N detector 25 calculates an S / N (signal-to-noise ratio) from the transmission signal y (t) and the residual signal e (t) and outputs a detection signal SN. On the other hand, the adaptive filter 14A is configured such that the step size μ in the updating process of the tap coefficient H (t) is controlled according to the detection signal SN given from the S / N detection unit 25. Other configurations are the same as those in FIG.

In this echo canceller 10C, first, the signal level SLV (t) of the echo included in the transmission signal y (t) is shifted by the S / N detection unit 25 according to the following equation: the absolute value of the transmission signal y (t) It is obtained from the average.
SLV (t) = α × SLV (t−1) + (1−α) × | y (t) |
However, 0 <α <1.

  Next, in the S / N detection unit 25, the background noise level included in the transmission signal based on the long-term average and the short-term average of the level of the residual signal e (t), for example, by the method described in Patent Document 2 above. NLV (t) is determined.

From the signal level SLV (t) and the background noise level NLV (t) included in the transmission signal thus obtained, the S / N (t) of the transmission signal y (t) is calculated according to the following equation.
S / N (t) = 20 × log {SLV (t) / NLV (t)}

  The S / N (t) calculated by the S / N detection unit 25 is given to the adaptive filter 14A as the detection signal SN. The reason for obtaining the background noise level NLV (t) not from the transmission signal y (t) but from the residual signal e (t) is that the acoustic echo among the signals included in the transmission signal y (t) in FIG. This is because the ratio of d (t) and background noise n (t) is calculated as S / N (t). That is, when the background noise level NLV (t) is calculated from the transmission signal y (t), the acoustic echo d (t) contains a lot of noise components even in an environment where the background noise n (t) hardly exists. This is because S / N shows a small value.

  In addition, the above S / N (t) expression does not consider the audio signal s (t) included in the transmission signal y (t), but the audio signal s (t) is included in the transmission signal y (t). Since the update of the tap coefficient H (t) of the adaptive filter 14A is stopped by the double talk detector 21, the size of the step size μ when the audio signal s (t) exists is There is no particular need to consider.

In the adaptive filter 14A, the step size μ for adjusting the convergence speed of the tap coefficient H (t) is adjusted according to the detection signal SN given from the S / N detection unit 25. That is, when the transmission signal y (t) has little background noise and S / N (t) is large, the step size μ is set to a large value. When the transmission signal y (t) has a lot of background noise and the S / N (t) is small, the step size μ is set to a small value.

  As described above, according to the echo canceller 10C of the fourth embodiment, the size of the step size μ for updating the tap coefficient H (t) is adjusted according to the S / N of the transmission signal y (t). It is configured. As a result, when there is a lot of background noise in the transmission signal y (t), the update rate of the tap coefficient H (t) is reduced to reduce the convergence speed but the high frequency band amplified by the inverse filter 16 The influence of the noise component can be reduced to reduce the echo cancellation amount. Further, when there is not much background noise in the transmission signal y (t), there is an advantage that the convergence rate can be improved by increasing the update amount of the tap coefficient H (t).

  FIG. 8 is a block diagram of an echo canceller showing a fifth embodiment of the present invention. Elements common to those in FIGS. 6 and 7 are given common reference numerals.

  The echo canceller 10D is configured to control switching of the selector switch 19 by the detection signal DW of the wideband signal determination unit 23 and adjust the step size μ of the adaptive filter 14A by the detection signal SN of the S / N detection unit 25. is doing. Other configurations are the same as those in FIGS. 6 and 7. Therefore, this echo canceller 10D has the advantages of both the third embodiment and the fourth embodiment.

  FIG. 9 is a block diagram of an echo canceller showing Embodiment 6 of the present invention. Elements common to those in FIG. 8 are denoted by common reference numerals.

  The echo canceller 10E is obtained by adding a narrowband signal determination unit 26 to the echo canceller 10D of FIG. The narrowband signal determination unit 26 compares the second-order reflection coefficient c2 output from the linear prediction unit 12 with the threshold value TH2, and performs an update operation of the tap coefficient H (t) of the adaptive filter 14A according to the magnitude determination result. A detection signal DN for control is output.

  The narrowband signal determination unit 26 uses a preset threshold value for the secondary reflection coefficient c2 output from the linear prediction unit 12 in order to prevent the tap coefficient H (t) from diverging due to learning by the tone characteristic signal. When the reflection coefficient c2 is larger than the threshold value TH2 as compared with TH2, it is determined that the spectrum of the reception signal x (t) is concentrated and is close to the tone signal, and the tap coefficient H ( The detection signal DN for stopping the update operation of t) is output.

  As described above, the echo canceller 10E according to the sixth embodiment narrows down the updating operation of the tap coefficient H (t) of the adaptive filter 14A when the received signal x (t) has characteristics close to the tone signal. Since the band signal determination unit 26 is provided, in addition to the same advantages as in the fifth embodiment, it is possible to prevent the tap coefficient H (t) from diverging when the reception signal x (t) is close to the tone signal. There is an advantage that you can.

FIG. 10 is a block diagram of an echo canceller showing Embodiment 7 of the present invention.
The echo canceller includes an adaptive filter 31, an adder 32, and an S / N detector 33.

  The adaptive filter 31 calculates the reception signal x (t) at the discrete time t with the tap coefficient H (t) to generate a pseudo echo r (t), and also detects the detection signal SN given from the S / N detection unit 33. The tap coefficient H (t) is updated with the step size μ corresponding to the value.

  The adder 32 subtracts the pseudo echo r (t) from the transmission signal y (t) and outputs a residual signal e (t).

  For example, the S / N detection unit 33 sets the echo signal level SLV (t) included in the transmission signal y (t) and the background noise level NLV (t) in the same manner as the S / N detection unit 25 in FIG. Based on this, S / N (t) is calculated and given to the adaptive filter 31 as the detection signal SN.

  In this echo canceller, in addition to the audio signal s (t), background noise n (t) and acoustic echo d (t) are converted into digital signals and input as transmission signals y (t).

On the other hand, the received signal x (t) is input to the adaptive filter 31 from the far end. Here, it is assumed that the impulse response H of the echo path of the acoustic echo d (t) can be approximated by an m-th order FIR filter. The tap coefficient H (t) of the adaptive filter 31 at the discrete time t and the received signal x (t) input to the adaptive filter 31 are respectively expressed as follows.
H (t) = [h1 (t), h2 (t),..., Hm (t)] ^T
X (t) = [x (t), x (t−1),..., X (t−m + 1)] ^T

Thus, the pseudo echo r (t) output from the adaptive filter 31 and the residual signal e (t) output from the adder 33 are expressed as follows.
r (t) = H (t) ^T X (t)
e (t) = y (t) -r (t)

The update of the tap coefficient H (t) of the adaptive filter 31 is expressed as follows in the case of the NLMS algorithm.
H (t + 1) = H (t) + μ [e (t) X (t)] / ‖X (t) ‖ ²

  The tap coefficient H (t) is updated only in a single talk state in which the received signal x (t) is present but the audio signal s (t) is not present, under the control of a double talk detector (not shown) or the like. The update operation is stopped when the double talk state in which both the reception signal x (t) and the audio signal s (t) exist and when the reception signal x (t) does not exist.

  The S / N detector 33 calculates S / N (t) based on the echo signal level SLV (t) included in the transmission signal y (t) and the background noise level NLV (t), and the detection signal SN As given to the adaptive filter 31.

In the adaptive filter 31, the step size μ for adjusting the convergence speed of the tap coefficient H (t) is adjusted according to the detection signal SN provided from the S / N detection unit 33. That is, when the transmission signal y (t) has little background noise and S / N (t) is large, the step size μ is set to a large value, and the transmission signal y (t) has much background noise and S / N (t). Is small, the step size μ is set to a small value.

  As described above, according to the echo canceller of the seventh embodiment, the step size μ for updating the tap coefficient H (t) is adjusted according to the S / N of the transmission signal y (t). is doing. As a result, when there is a lot of background noise in the transmission signal y (t), the update rate of the tap coefficient H (t) can be reduced to reduce the convergence speed, but the reduction of the echo cancellation amount can be reduced. I can do it. Further, when there is not much background noise in the transmission signal y (t), there is an advantage that the convergence rate can be improved by increasing the update amount of the tap coefficient H (t).

In addition, this invention is not limited to the said Example, A various deformation | transformation is possible. Examples of this modification include the following.
(1) In FIG. 1 and the like, the switching control unit 20 is configured to output the switching signal SW according to the number of tap coefficient updates of the adaptive filter 14, but according to the actual tap coefficient convergence state. The switching signal SW can be output. In this case, for example, the power level of the residual signal e (t) at the time of single talk is monitored, and when the power level falls below a certain level, the switching signal SW for selecting the residual signal e (t) is set. Configure to output.
(2) In FIG. 7 and the like, the S / N detector 25 calculates the S / N from the transmission signal y (t) and the residual signal e (t), but the transmission signal yw (t) and the residual are calculated. The S / N may be calculated from the signal ew (t).
(3) In FIG. 7 and the like, the adaptive filter 14A is configured to adjust the step size μ when the tap coefficient H (t) is updated based on the detection signal SN. Instead of using the unit 25 and the detection signal SN, the step size μ may be switched between two stages using the switching signal SW.
(4) A FIFO buffer 22 similar to that shown in FIG. 4 may be provided in FIGS.

It is a block diagram of the echo canceller which shows Example 1 of this invention. It is a block diagram of the conventional echo canceller. It is explanatory drawing of the echo delay time estimated from a tap coefficient. It is a block diagram of the echo canceller which shows Example 2 of this invention. FIG. 5 is an explanatory diagram for comparing operations of the echo cancellers of FIG. 1 and FIG. 4. It is a block diagram of the echo canceller which shows Example 3 of this invention. It is a block diagram of the echo canceller which shows Example 4 of this invention. It is a block diagram of the echo canceller which shows Example 5 of this invention. It is a block diagram of the echo canceller which shows Example 6 of this invention. It is a block diagram of the echo canceller which shows Example 7 of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Delay buffer 12 Linear prediction part 13, 16 Inverse filter 14, 14A Adaptive filter 15 Delay control part 17, 18 Adder 19 Changeover switch 20 Switching control part 21 Double talk detection part 22 FIFO buffer 23 Broadband signal determination part 25 S / N Detection unit 26 Narrow band signal determination unit

Claims (5)

A delay buffer that delays the received signal and outputs a delayed received signal;
A linear prediction unit that calculates a reflection coefficient and a linear prediction coefficient from the received signal;
A first inverse filter for whitening the delayed received signal using the linear prediction coefficient to generate a whitened received signal;
An adaptive filter that computes the delayed received signal and the whitened received signal with the same tap coefficient to output a pseudo echo and a whitened pseudo echo, and updates the tap coefficient according to a given residual signal;
A delay control unit that estimates a delay time of an echo based on the tap coefficient of the adaptive filter and outputs the linear prediction coefficient output from the linear prediction unit by delaying the delay time;
A second inverse filter for whitening a transmission signal using the linear prediction coefficient output from the delay control unit and generating a whitened transmission signal;
A first adder for subtracting the pseudo echo from the transmission signal to output a first residual signal;
A second adder for subtracting the whitened pseudo echo from the whitened transmission signal to output a second residual signal;
A broadband signal determination unit that determines whether the received signal is a whiteness signal based on the reflection coefficient;
When the tap coefficient is not converged and the received signal is not a white signal, the second residual signal is selected, and when the tap coefficient is converged or the received signal is a white signal A residual signal switching unit that selects the first residual signal and provides the selected residual signal to the adaptive filter ;
Echo canceller, characterized in that it comprises a. A narrowband signal determination unit that determines whether or not the received signal is a tone signal based on the reflection coefficient, and updates the tap coefficient in the adaptive filter when the received signal is determined to be a tone signal. The echo canceller according to claim 1, wherein the echo canceller is configured to stop. Based on the transmission signal and the first residual signal, or the whitened transmission signal and the second residual signal, a ratio between the echo included in the transmission signal and background noise included in the transmission signal A signal-to-noise ratio calculating unit for calculating a signal-to-noise ratio is provided, and the step size at the time of updating the tap coefficient in the adaptive filter is adjusted based on the signal-to-noise ratio. The echo canceller according to claim 1 or 2. 2. A first-in first-out buffer for writing linear prediction coefficients calculated by the linear prediction unit and reading and outputting the linear prediction coefficients in the order of writing in response to a request from the delay control unit. 2. The echo canceller according to 2 or 3. A delay buffer that delays the received signal and outputs a delayed received signal;
A linear prediction unit for calculating a linear prediction coefficient from the received signal;
A first inverse filter for whitening the delayed received signal using the linear prediction coefficient to generate a whitened received signal;
An adaptive filter that computes the delayed received signal and the whitened received signal with the same tap coefficient to output a pseudo echo and a whitened pseudo echo, and updates the tap coefficient according to a given residual signal;
A delay control unit that estimates a delay time of an echo based on the tap coefficient of the adaptive filter and outputs the linear prediction coefficient output from the linear prediction unit by delaying the delay time;
A second inverse filter for whitening a transmission signal using the linear prediction coefficient output from the delay control unit and generating a whitened transmission signal;
A first adder for subtracting the pseudo echo from the transmission signal to output a first residual signal;
A second adder for subtracting the whitened pseudo echo from the whitened transmission signal to output a second residual signal;
When the tap coefficient has not converged, the second residual signal is selected, and when the tap coefficient has converged, the first residual signal is selected, and the selected residual signal is sent to the adaptive filter. A residual signal switching section to provide,
An echo canceller with
Based on the transmission signal and the first residual signal, or the whitened transmission signal and the second residual signal, a ratio between the echo included in the transmission signal and background noise included in the transmission signal A signal-to-noise ratio calculating unit for calculating a signal-to-noise ratio is provided, and the step size at the time of updating the tap coefficient in the adaptive filter is adjusted based on the signal-to-noise ratio. Echo canceller.

Images