JP2012049708A - Echo route characteristic estimation device, echo route characteristic estimation program and echo canceller - Google Patents

Abstract

PROBLEM TO BE SOLVED: To estimate characteristics of an echo route for high efficiency of adaptive control of an echo canceller having an adaptive filter even in the case of multiple reflection with one or a plurality of echo generation points.SOLUTION: The present invention extracts a first signal component in a predetermined band of a first input signal, shifts the first signal component to a predetermined low band and performs downsampling of the first signal component, and then extracts a second signal component in a predetermined band of a second input signal received through an echo route, shifts the second signal component to a predetermined low band and performs downsampling of the second signal component. Also, the present invention estimates an impulse response in the echo route, based on the downsampled first signal component and second signal component and estimates echo route characteristics of the echo route, based on the estimated impulse response waveform of the echo route.

Description

  The present invention relates to an echo path characteristic estimation apparatus, an echo path characteristic estimation program, and an echo canceller, and for example, an echo path characteristic estimation apparatus that estimates an echo path characteristic that can have one or a plurality of echo generation points (echo generation sources). The present invention can be applied to an echo path characteristic estimation program and an echo canceller.

  Echo cancellers used in voice communication are roughly classified into two types: acoustic echo cancellers and line echo cancellers.

  An acoustic echo canceller, for example, in a hands-free phone or conference device that uses a speaker and a microphone, a sound signal received from the communication partner is emitted as a sound wave from the speaker, and the sound wave wraps around the microphone and becomes an echo signal to the communication partner. It is a device that suppresses the deterioration of call quality by returning.

  A line echo canceller, for example, is equipped with an analog component called a 2-wire / 4-wire converter in a subscriber exchange or a VoIP (Voice Over Internet Protocol) router that accommodates an analog telephone, and the impedance of the analog component This is a device that suppresses deterioration of call quality due to leakage of a part of a voice signal received from a communication partner due to inconsistency as an echo signal into a transmission signal to the communication partner.

  Any type of echo canceller of the voice communication system cannot identify the echo path in advance using white noise before the start of communication like the echo canceller of the data communication system. Therefore, the echo canceller of the voice communication system must identify the echo path by regarding the voice signal as a test signal during voice communication. There are technical difficulties in this regard. In other words, there is room for improvement in this respect.

  In the case of acoustic echo, the number of echo paths through which sound waves travel from the speaker to the microphone is not limited to one. Usually, multiple echo paths exist simultaneously. For example, there are a path in which a sound wave directly goes from a speaker to a microphone through a space, and a path in which a sound wave is reflected at an interior wall / floor / ceiling / installed object one or more times. In such a case, the acoustic echo canceller needs to be designed in consideration of these multiple reflections.

  Also in the line echo, the number of echo paths is not limited to one, and multiple reflections may exist. For example, in a corporate network, voice communication may be performed via a multistage link such as a public-private network (public network-private network-public network) connection. Under such circumstances, as the analog line appears at a plurality of locations, the 2-wire / 4-wire converter also appears at a plurality of locations, so that a plurality of echo paths exist simultaneously. In such a case, the line echo canceller needs to be designed in consideration of these multiple reflections.

  When an echo path for generating an echo of a received signal from a communication partner is regarded as a linear circuit, the echo path is generally expressed by a finite impulse response (FIR) filter H (z).

Here, H (z) is a z function. Assuming an Nth-order FIR filter, H (z) can be expressed as equation (1-1). Note that h (k) is a tap coefficient.

  As shown in Expression (1-2), the tap coefficient h (k) in Expression (1-1) is also a time-domain impulse response h (n) obtained by inverse z-transforming H (z). Here, in Expression (1-2), j is a pure imaginary number, and C is a circular integration path including the origin of the z plane.

  In general, the echo path is expressed by the transfer function H (z) of Expression (1-1) in the frequency domain, and is expressed by the impulse response h (n) of Expression (1-2) in the time domain.

  Next, a specific example of the impulse response h (n) of the echo path is shown in FIG.

Here, time is converted into the number of samples (integer),
N _total : _Total length of h (n) N _d : Round-trip propagation delay time between the echo canceller and the echo path N _hyb : An element that exists in the echo path and generates an echo, for example, 2-wire / 4-line conversion The impulse response time of the instrument. Here, N _total = N _d + N _hyb .

  The impulse response of the echo path shown in FIG. 4 is the case where the echo is generated at one place. For example, in a corporate network or the like, due to a multi-stage link configuration such as public-private public connection, when analog lines exist in a plurality of locations, echo paths also exist in a plurality of locations, and echoes that have been multi-reflected return to the speaker.

  FIG. 5 shows an example of an impulse response when there are, for example, two echo paths.

In FIG.
N _total : _total length of h (n) N _d1 : round-trip propagation delay time between the echo canceller and the first echo path N _hyb1 : an element that exists in the first echo path and generates an echo, for example 2-wire / 4-wire transducer impulse response time N _d2 = round-trip propagation delay time between echo canceller and second echo path N _hyb2 = element present in the second echo path and generating an echo, eg The impulse response time of the 2-wire / 4-wire converter. Here, N _total = N _d2 + N _hyb2 .

  FIG. 6 is a block configuration diagram around the echo canceller. In FIG. 6, an input signal x (n) is a signal received from a communication partner not shown in FIG. The input signal x (n) is supplied to the echo canceller 51 and the echo path 52.

  The input signal x (n) returns to the echo canceller 51 via the echo path 52 as an echo signal y (n). At the same time, the input signal x (n) is converted by the echo canceller 51 into an estimated echo signal y ^ (n) that is an estimated value of the echo signal y (n). In the subtractor 53, the estimated echo signal y ^ (n) is subtracted from the echo signal y (n) to calculate a residual echo signal e (n).

  Normally, the subtractor 53 is built in the echo canceller 51, but in FIG. 6, the subtractor 53 is drawn separately from the echo canceller 51 for convenience of explanation. This residual echo signal e (n) is fed back to the echo canceller 51 at the same time as it is sent to the communication partner.

The above is the outline of the adaptive control of the echo canceller. When the adaptive control of the echo canceller is expressed by a learning identification algorithm generally adopted, it is as follows.

Here, N is the order of the adaptive filter of the echo canceller, and h _n ^ (k) is the tap coefficient of the adaptive filter of the echo canceller at time n. Further, μ is a constant, and the range of μ is 0 <μ ≦ 1.

When the impulse response h (k) of the echo path is assumed, h _n ^ (k), 0 ≦ k ≦ N is an estimated value of h (k), 0 ≦ k ≦ N. That is, the echo canceller estimates the tap coefficient h (k) before estimating the echo signal y (n).

  When h (n) = 0, n> M, the echo canceller is known to exhibit the best performance when N = M.

  To achieve this, the device can measure the impulse response time (filter order) of the echo path during communication and dynamically change the order of the echo canceller's adaptive filter to its impulse response time value. Good to do.

  Conventionally, the adaptive filter order of the echo canceller is not dynamically changed, but there is a technique that assumes the maximum value of the impulse response time of the echo path in advance and sets the order of the adaptive filter of the echo canceller to the maximum value. .

The techniques described in Patent Document 1 and Patent Document 2 assume a maximum value of the impulse response time of the echo path in advance and wait until the adaptive filter converges. Then, after convergence of the adaptive filter, the adaptive filter interval is determined by analyzing the distribution of the coefficient of the adaptive filter. According to the techniques described in Patent Document 1 and Patent Document 2, for example, in the case of an echo path in which reflection occurs twice as illustrated in FIG. 5, the coefficient of the adaptive filter is approximately equal to this after the convergence of the echo canceller by the adaptive filter. The distribution will be similar to the impulse response. Therefore, the total length of the adaptive filter can be reset to a value corresponding to N _total . If the adaptive control section is limited to sections corresponding to N _hyb1 and N _hyb2 , the maximum effect can be obtained with the minimum resources.

  In the techniques described in Patent Document 3 and Patent Document 4, first, the order of the adaptive filter of the echo canceller is set to an appropriate value. The techniques described in Patent Literature 3 and Patent Literature 4 determine that the adaptive filter order is insufficient when an echo return loss improvement amount (ERLE: Echo Return Loss Enhancement), which will be described later, falls below a predetermined threshold. The order of the filter is increased. Thereafter, the process of increasing the order of the adaptive filter is repeated until ERLE becomes equal to or greater than the threshold.

Here, ERLE is a quantity defined as ERLE = <y ² (n)> / <e ² (n)> using the signal appearing in FIG. Here, <> represents a statistical average.

JP 2006-128758 A JP 2009-17029 A JP-T-2001-503217 JP-A-11-122144

  However, the following problems may occur in the above-described prior art and the techniques described in Patent Documents 1 to 4.

  The conventional technique for setting the order of the adaptive filter to the maximum value without dynamically changing the order of the adaptive filter of the echo canceller described above is based on the assumption that the impulse response length of the echo path is the maximum value assumed. To do. The order of the adaptive filter of the echo canceller is also made to correspond to the maximum value. Therefore, if the impulse response length of the actual echo path is smaller than the maximum value, the resources, processing, and power of the echo canceller corresponding to the difference between the impulse response length of the actual echo path and the maximum value are wasted. Is done. Furthermore, the convergence time of the adaptive filter is increased by an amount corresponding to this difference. This can lead to an increase in cost / size / power consumption of the device, particularly when the echo canceller is mounted in a multi-channel in one device.

  The technologies described in Patent Literature 1 and Patent Literature 2 realize optimal resource allocation after the adaptive filter convergence of the echo canceller. However, until the convergence of the adaptive filter is completed, the technologies described in Patent Literature 1 and Patent Literature 2 require resources equivalent to the above-described conventional technology. This reduces the ability of the number of simultaneous calls, particularly in the case of an apparatus that implements multi-channel echo cancellers. The number of simultaneous calls is the number of calls that can start communication at the same time, and is one of the important indexes representing the capability of the voice communication apparatus.

  The problem of the technology described in Patent Literature 3 and Patent Literature 4 is to change the order of the adaptive filter in accordance with the value of ERLE of the echo canceller. There is no problem in the so-called single talk state in which only the voice from the opposite speaker exists. However, ERLE decreases when the voices of the speakers at both ends exist at the same time, that is, in a so-called double talk state. At this time, it is difficult for the echo canceller to strictly discriminate whether the order of the adaptive filter is insufficient and ERLE is lowered or whether ERLE is lowered because of the double talk state. Furthermore, although it is in the double talk state, when it is erroneously determined that the ERLE has decreased and the adaptive control is continued, the coefficient of the adaptive filter diverges. This not only makes echo cancellation impossible, but in the worst case, noise may be generated.

  Therefore, even in the case of one echo generation point or multiple reflections, the echo path characteristic can be estimated to improve the efficiency of adaptive control of the echo canceller having the adaptive filter. An estimation device, an echo path characteristic estimation program, and an echo canceller are required.

  In order to solve such a problem, the echo path characteristic estimation apparatus according to the first aspect of the present invention (1) extracts a first signal component of a predetermined band from a received first input signal, and the first signal First frequency shifting means for shifting the component to a predetermined low frequency; (2) first downsampling means for downsampling the first signal component from the first frequency shifting means; and (3) echo. A second frequency shift means for extracting a second signal component of a predetermined band from the second input signal received via the path and shifting the second signal component to a predetermined low frequency; (4) Second down-sampling means for down-sampling the second signal component from the second frequency shift means; and (5) down-sampling by the first down-sampling means and the second down-sampling means. And (6) an impulse response waveform of the echo path estimated by the impulse response estimation means, based on the first signal component and the second signal component that have been shifted. And an echo path characteristic estimating means for estimating an echo path characteristic of the echo path.

  The echo path characteristic estimation program according to the second aspect of the present invention is a computer that extracts (1) a first signal component of a predetermined band from a received first input signal, and the first signal component is reduced to a predetermined low level. First frequency shifting means for shifting to a frequency range, (2) first downsampling means for downsampling the first signal component from the first frequency shifting means, and (3) received via an echo path A second frequency shift means for extracting a second signal component of a predetermined band for the second input signal and shifting the second signal component to a predetermined low frequency; (4) from the second frequency shift means Second down-sampling means for down-sampling the second signal component, (5) down-sampled by the first down-sampling means and the second down-sampling means (6) an impulse response estimating means for estimating an impulse response of the echo path based on the signal component of 1 and the second signal component, and (6) an echo response of the echo path based on the impulse response waveform of the echo path estimated by the impulse response estimating means. It is made to function as an echo path characteristic estimation means for estimating the echo path characteristic.

  The echo canceller of the third aspect of the present invention is an echo canceller that generates a pseudo echo by an adaptive filter and subtracts the second input signal from the first input signal to eliminate the echo component. An echo path characteristic estimation device is provided.

  According to the present invention, it is possible to estimate the characteristics of an echo path in order to improve the efficiency of adaptive control of an echo canceller having an adaptive filter even in the case of one echo generation point or multiple reflections. it can.

It is a block diagram which shows the internal structure of the multiple reflection estimator of embodiment, and its periphery structure. It is explanatory drawing explaining the downsampling process of embodiment. It is a figure which shows the actual impulse response of the echo path | pass in the case of setting it as the operating condition of FIG. 7 of embodiment. It is explanatory drawing explaining the specific example of the impulse response h (n) of an echo path | route. It is explanatory drawing which shows the example of an impulse response when the conventional echo path exists in two places. It is a block block diagram which shows the structure of an echo canceller periphery. It is a figure which shows an example of the operating condition of the multiple reflection estimation process of embodiment.

(A) Embodiment Hereinafter, embodiments of an echo path characteristic estimation apparatus, an echo path characteristic estimation program, and an echo canceller according to the present invention will be described with reference to the drawings.

  The present invention relates to a physical phenomenon called echo of an analog signal which is a continuous value system. Today, however, digital signal processing, which is far superior in cost and performance to analog signal processing, has become widespread. The multiple reflection estimator and method of the present invention is also assumed to be implemented by digital signal processing.

  In the following description of the embodiment, even if it is a phenomenon or processing of an analog signal, it should be noted that it is described in a discrete value system. In fact, from the viewpoint of an echo canceller realized by digital signal processing, the outside world can be seen only as a discrete value system.

(A-1) Configuration of Embodiment FIG. 1 is a configuration diagram showing the internal configuration of the multiple reflection estimator of this embodiment and the configuration around it.

  In FIG. 1, the case where the multiple reflection estimator 1 is disposed between the echo canceller 51 and the echo path 52 is illustrated. Note that the multiple reflection estimator 1 may be incorporated in the echo canceller 51 as long as it is positioned in front of the echo canceller 51.

  The multiple reflection estimator 1 is an echo path characteristic estimation device that estimates one or a plurality of echo reflection points (echo generation points) in the echo path 52 based on an input signal x (n) and an echo signal y (n). is there. In addition, the multiple reflection estimator 1 provides the echo canceller 51 with estimated echo path information coe_info up to each estimated echo reflection point. Thereby, the echo canceller 51 can specify the section of each echo reflection point based on the estimated echo path information coe_info from the multiple reflection estimator 1, and can perform appropriate adaptive control.

  The echo canceller 51 deletes the echo signal y (n) that has passed through the echo path 52. The echo canceller 51 has an adaptive filter and a subtracter (not shown). The echo canceller 51 generates an estimated echo signal y ^ (n) obtained by estimating the echo signal y (n) based on the input signal (x) received from the opposite speaker. The echo canceller 51 subtracts the estimated echo signal y ^ (n) from the echo signal y (n) to calculate a residual echo signal e (n). Further, the echo canceller 51 receives the estimated echo path information coe_info from the multiple reflection estimator 1 and performs adaptive control using an adaptive filter based on the estimated echo path information coe_info.

  The echo path 52 is an echo path including one or a plurality of echo generation points. The input signal x (n) is a signal received from an opposite speaker (not shown). The input signal x (n) is reflected at each echo generation point through the echo path 52. The reflected echo signal y (n) is supplied to the multiple reflection estimator 1 and the echo canceller 51.

  In FIG. 1, a multiple reflection estimator 1 includes a band pass filter (BPF) 11, a multiplier (MPY) 12, a low pass filter (LPF) 13, a decimation unit (DCM) 14, a band pass filter (BPF). ) 11, multiplier (MPY) 22, low-pass filter (LPF) 23, decimation unit (DCM) 24, sine wave generator (SGEN) 31, single talk detector (SGL) 32, adaptive digital filter (ADF) ) 33 and a subtractor (SUB) 34.

The band pass filter (BPF) 11 extracts only a predetermined band frequency component from the input echo signal y (n). The band pass filter (BPF) 11 supplies the extracted band frequency component to the multiplier (MPY) 12 as an output signal y ₁ (n).

The multiplier (MPY) 12 calculates the product of the output signal y ₁ (n) and the sine wave s (n) from the sine wave generator (SGEN) 31. The multiplier (MPY) 12 supplies the calculated signal as an output signal y ₂ (n) to the low-pass filter (LPF) 13.

The low pass filter (LPF) 13 extracts only a predetermined low frequency component from the output signal y ₂ (n) received from the multiplier (MPY) 12. The low-pass filter (LPF) 13 supplies the extracted low-frequency component to the decimation device (DCM) 14 as the output signal y ₃ (n).

The decimation device (DCM) 14 decimates the output signal y ₃ (n) from the low-pass filter (LPF) 13 at a predetermined interval. The decimation device (DCM) 14 gives the signal after decimation to the subtracter (SUB) 34 as an output signal y ₄ (r).

The band pass filter (BPF) 21 extracts only a predetermined band frequency component from the input signal x (n) received from the opposite speaker. The band pass filter (BPF) 21 supplies the extracted band frequency component to the multiplier (MPY) 22 as an output signal x ₁ (n).

The multiplier (MPY) 22 calculates the product of the output signal x ₁ (n) and the sine wave s (n) from the sine wave generator (SGEN) 31. The multiplier (MPY) 22 supplies the calculated signal to the low-pass filter (LPF) 23 as the output signal x ₂ (n).

The low pass filter (LPF) 23 extracts only a predetermined low frequency component from the output signal x ₂ (n) received from the multiplier (MPY) 22. The low-pass filter (LPF) 23 supplies the extracted low-frequency component to the decimation device (DCM) 24 as the output signal x ₃ (n).

The decimation device (DCM) 24 decimates the output signal x ₃ (n) from the low-pass filter (LPF) 23 at a predetermined interval. The decimation unit (DCM) 24 gives the signal after decimation to the adaptive digital filter (ADF) 33 as the output signal x ₄ (r).

The subtracter (SUB) 34 subtracts the estimated signal y ₄ ^ (r) from the adaptive digital filter (ADF) 33 from the output signal y ₄ (r) output from the decimation unit (DCM) 14. ₄ (r) is calculated. The subtracter (SUB) 34 provides the estimated error signal e ₄ (r) to the adaptive digital filter (ADF) 33.

  The sine wave generator (SGEN) 31 generates a sine wave signal s (n) having a single frequency and supplies the sine wave signal s (n) to the multiplier (MPY) 12 and the multiplier (MPY) 22. It is.

  The single talk detector (SGL) 32 receives the input signal x (n) and the echo signal y (n) of the opposite speaker, and based on the input signal x (n) and the echo signal y (n) This is to detect a single talk indicating that only the speaker is busy. The single talk detector (SGL) 32 gives the detection result of the single talk to the adaptive digital filter (ADF) 33 as a coefficient update enable signal update_en.

  The adaptive digital filter (ADF) 33 estimates the impulse response of the echo path. The adaptive digital filter (ADF) 33 has the same function as the adaptive filter that the echo canceller 51 has.

The adaptive digital filter (ADF) 33 receives the output signal x ₄ (r) from the decimation unit (DCM) 24 and receives the estimated error signal e ₄ (r) from the subtracter (SUB) 34 to output the DCM. and it outputs an estimated signal _y 4 ^ (r) of the signal _y 4 (r).

  The adaptive digital filter (ADF) 33 estimates the impulse response only in the case of single talk based on the coefficient update enable signal update_en.

  Further, the adaptive digital filter (ADF) 33 has an analysis unit 331. The analysis unit 331 analyzes the round-trip propagation delay time and impulse response time to one or a plurality of echo reflection points based on the impulse response of the echo path by the adaptive digital filter (ADF) 33. The adaptive digital filter (ADF) 33 gives information on the filter coefficient, which is the analysis result of the analysis unit 331, to the echo canceller 51 as estimated echo path information coe_info.

(A-2) Operation of Embodiment (A-2-1) Outline of Embodiment Before describing the specific operation of this embodiment, the outline of the embodiment will be described.

In determining the impulse response length of the echo path 52, it is not necessary to measure the exact impulse response. The reason for this is that, generally, the impulse response of a plurality of echo generating elements (for example, a 2-wire / 4-wire transducer and a wall) existing in the echo path 52 is initially abrupt as shown in FIG. Stands up and then shows damped oscillation. Further, the fluctuation of the impulse response time of the echo generating element itself (corresponding to N _hyb1 and N _hyb2 in FIG. 5) corresponds to the round-trip propagation delay time of the echo path where multiple reflections exist (corresponding to N _d1 and N _d2 in FIG. 5). ) Is very small compared to the variation.

  For example, it is known that the impulse response time of a 2-wire / 4-wire converter is distributed between 3 ms and 12 ms. On the other hand, the round-trip propagation delay time varies from several ms to several hundred ms depending on the layer 2 protocol and network configuration used for voice communication. Therefore, if it is possible to measure how the initial peaks of the individual echo generating elements are distributed in the entire echo path, the impulse response length of the entire echo path can be determined.

  Therefore, the multiple reflection estimator 1 of this embodiment extracts some band components from each of the opposite speaker signal x (n) and the echo signal y (n) input to the echo canceller 51. Then, the multiple reflection estimator 1 shifts the frequency of each extracted band component to the low band side. Next, the multiple reflection estimator 1 thins out the samples of these audio signals. Then, the multiple reflection estimator 1 estimates the impulse response length of the echo path 52 with an adaptive filter similar to the echo canceller 51.

  The estimation result by the multiple reflection estimator 1 is given to the echo canceller 51 used in voice communication. As a result, the order of the adaptive filter of the echo canceller 51 can be matched with the impulse response of the actual echo path.

(A-2-2) Operation Description Hereinafter, processing in the multiple reflection estimator 1 of the embodiment will be described with reference to the drawings.

(A) Downsampling First, an input signal x (n) from the opposite speaker is input to the echo canceller 51 and the multiple reflection estimator 1.

  In the multiple reflection estimator 1, a part of the band component of the input signal x (n) is down-sampled. Hereinafter, a method for down-sampling a part of the band component of the input signal x (n) will be described.

In the multiple reflection estimator 1, an input signal x (n) is input to a band pass filter (BPF) 21. When the input signal x (n) passes through the band pass filter (BPF) 21, an output signal x ₁ (n) that is a part of the band component of the input signal x (n) is output.

  FIG. 2A is an explanatory diagram for explaining extraction of a part of the band components of the input signal x (n) by the band-pass filter (BPF) 21. FIG. 2A illustrates a case where the Nyquist frequency related to the sample of the audio signal is 4 kHz. The pass band of the band pass filter (BPF) 21 is 350 Hz to 800 Hz.

In FIG. 2A, the dotted line indicates the input signal x (n), and the solid line indicates the output signal x ₁ (n) passing through the band-pass filter (BPF) 21. As shown in FIG. 2A, when the input signal x (n) passes through the band pass filter (BPF) 21, the band component of the input signal in the 450 Hz band from 350 Hz to 800 Hz is extracted.

Next, x ₁ (n) and a sine wave signal s (n) having a single frequency fc are input to the multiplier (MPY) 22.

  Here, the sine wave generator (SGEN) 31 may be realized by, for example, a second-order infinite impulse response filter having a pole corresponding to a desired frequency on a unit circle, or stores a reference sine wave signal as a memory. It may be realized by preparing in advance in a table and reading the table while thinning it out appropriately in order to obtain a sine wave having a desired frequency.

The multiplier (MPY) 22 calculates the product of x ₁ (n) and s (n). Here, according to the relationship of the formula (2-1) from the triangular formula, a signal x ₂ (n) frequency-shifted in two places above and below is obtained with the frequency fc as the central axis of the mirror image.
sin α sin β = {cos (α−β) −cos (α + β)} / 2 (2-1)
FIG. 2B is an explanatory diagram for explaining the frequency shift. The multiplier (MPY) 22 calculates the product of x ₁ (n) and a sine wave signal s (n) having a single frequency fc (= 300 Hz) according to Expression (2-1). As a result, the component (x ₁ (n)) of 350 Hz to 800 Hz is shifted to a frequency band of 650 Hz to 1100 Hz and a frequency band of −50 Hz to −500 Hz with fc = 300 Hz as the central axis of the mirror image. At this time, from the sampling theorem, the component shifted to a negative frequency is folded back to the positive side with DC as the central axis of the mirror image. Therefore, the component shifted to the frequency −50 Hz to −500 Hz band is folded back to the frequency 50 Hz to 500 Hz band with the frequency f = 0 as the central axis. The frequency shifted component of the frequency 650 Hz to 1100 Hz and the component of the frequency 50 Hz to −500 Hz are x ₂ (n).

Next, x ₂ (n) frequency-shifted by the multiplier (MPY) 22 is input to the low-pass filter (LPF) 23. When x ₂ (n) passes through the low-pass filter (LPF) 23, x ₃ (n), which is a predetermined low-frequency signal component, is output.

FIG. 2C is an explanatory diagram for explaining extraction of a low-frequency signal component. In FIG. 2C, the pass band of the low-pass filter (LPF) 23 is 500 Hz or less. When x ₂ (n) from the multiplier (MPY) 22 passes through the low-pass filter (LPF) 23, a component of 50 Hz to 500 Hz that is 500 Hz or less is extracted as x ₃ (n).

Next, x ₃ (n) is input to the decimation unit (DCM) 24. The decimation unit (DCM) 24 decimates x ₃ (n) samples from the low-pass filter (LPF) 23 at regular intervals. That is, the decimation device (DCM) 24 converts the discrete time index from n to r. As a result, the signal x ₄ (r) from which the samples are thinned is output to the adaptive digital filter (ADF) 33.

  Here, the thinning-out rate performed by the thinning-out device (DCM) 24 will be described. The thinning rate is determined by the ratio between the Nyquist frequency of the signal before thinning and the passband of the low pass filter LPF24.

  For example, in the example of FIG. 2, the Nyquist frequency is 4 kHz, and the low-pass band of the low-pass filter (LPF) 23 is 500 Hz. Therefore, in this case, the thinning rate is 1/8 (= 500/4000).

  As described above, the multiple reflection estimator 1 down-samples the input signal x (n) by shifting a part of the band component to the low frequency side and further thinning out the signal samples.

  The multiple reflection estimator 1 also performs downsampling for the echo signal y (n). The downsampling method for the echo signal y (n) is the same as the downsampling method for the input signal x (n). Here, the downsampling for the echo signal y (n) will be briefly described.

  Note that the pass band of the band pass filter (BPF) 11 and the low pass filter (LPF) 13 and the decimation rate of the decimation unit (DCM) 14 are the band pass filter (BPF) 21 for the input signal x (n). The pass band of the low-pass filter (LPF) 13 is the same as the decimation rate of the decimation unit (DCM) 24.

First, the echo signal y (n) is input to the band pass filter (BPF) 11. When the echo signal y (n) passes through the band pass filter (BPF) 11, an output signal y ₁ (n) that is a part of the band component of the echo signal y (n) is output.

Next, y ₁ (n) and a sine wave signal s (n) having a single frequency fc are input to the multiplier (MPY) 12. In the multiplier (MPY) 12, the product of y ₁ (n) and s (n) is made by the equation (2-1), and the frequency shift to the low frequency side is performed.

Y ₂ (n) frequency-shifted to the low frequency side by the multiplier (MPY) 12 is input to the low frequency filter (LPF) 13. When y ₂ (n) passes through the low-pass filter (LPF) 13, y ₃ (n) that is a predetermined low-frequency signal component is output.

Then, y ₃ (n) is input to the decimation unit (DCM) 14. The decimation unit (DCM) 14 decimates samples of y ₃ (n) from the low-pass filter (LPF) 13 at regular intervals. As a result, the signal y ₄ (r) is output from the decimation unit 4 to the subtracter (SUB) 34.

(B) Processing of Adaptive Digital Filter (ADF) 33 and Subtracter (SUB) 34 Next, operations of the adaptive digital filter (ADF) 33 and the subtracter (SUB) 34 will be described.

  Signals input to the adaptive digital filter (ADF) 33 and the subtracter (SUB) 34 are thinned out. For this reason, the signal processing speed of the adaptive digital filter (ADF) 33 and the subtracter (SUB) 34 is reduced, which is equal to the thinning rate.

The adaptive digital filter (ADF) 33 uses the signal x ₄ (r) obtained by thinning out a part of the band of the input signal x (n) from the opposite speaker as an input signal, and the echo estimated by the adaptive digital filter (ADF) 33. When the impulse response of the path 52 is h ₄ ^ (k), the convolution operation is performed according to the equation (2-2), and the signal y ₄ (r) is estimated by thinning out a part of the band of the echo signal y (n). Create the value y ₄ ^ (r).

Also, the subtracter (SUB) 34 receives y ₄ (r) from the decimation unit (DCM) 14 and estimates y ₄ (r) estimated by the adaptive digital filter (ADF) 33 y ₄ ^ Enter (r).

Subtractor (SUB) 34, according to equation _(2-3), calculates _y 4 and (r) _y 4 ^ a (r) estimated error signal _e 4 and (r). The subtracter (SUB) 34 gives the calculated estimated error signal e ₄ (r) to the adaptive digital filter (ADF) 33.

The adaptive digital filter (ADF) 33 uses the estimated error signal e ₄ (r) received from the subtractor (SUB) 34 and x ₄ (n) from the decimation unit (DCM) 24 to update the filter coefficient. Do.

  Here, the filter coefficient update algorithm of the adaptive digital filter (ADF) 33 is not particularly limited, and various algorithms can be widely applied. In this embodiment, the case where the learning identification method generally used well is applied is illustrated.

The adaptive digital filter (ADF) 33 calculates the filter coefficient h _{4, r + 1} ^ (k) at time r + 1 according to the arithmetic expression of the learning identification method shown in Expression (2-4).

Here, μ ₄ is a constant, and its range is 0 <μ ₄ ≦ 1.

  The adaptive digital filter (ADF) 33 performs filter coefficient update processing only when the coefficient update enable signal update_en from the single talk detector (SGL) is active.

  Here, the single talk detection method by the single talk detector (SGL) 32 is not particularly limited and can be widely applied. For example, a single talk detector (SGL) receives an input signal x (n) and an echo signal y (n) which are voice signals of the opposite speaker, and monitors the call state in both directions. Then, when only the input signal x (n) exists, that is, when the one-way speech state is established, the single talk detector (SGL) makes the coefficient update enable signal update_en active.

For example, the single talk detector (SGL) determines that it is single talk when the condition of Expression (2-5) is satisfied.

Here, <> represents a statistical average.

Convergence of the adaptive digital filter (ADF) 33 is regarded as convergence when the echo return loss improvement amount (ERLE) defined by the following equation (2-6) reaches a certain threshold value.

In the adaptive digital filter (ADF) 33, the analysis unit 331 analyzes the estimated h ₄ ^ (k) when the output signal y ₄ ^ (r) converges.

For example, assume that the distribution of h ₄ ^ (k) becomes as shown in FIG. 5 after convergence. FIG. 5 shows an impulse response when there are two echo reflection points in the echo path 52. In this case, the analysis unit 331 of the adaptive digital filter (ADF) 33 includes the round-trip propagation delay time N _d1 and the impulse response time N _{hyb1 up} to the first reflection point, and the round-trip propagation delay time N _d2 and the impulse up to the second reflection point. Response time N _hyb2 is calculated. The adaptive digital filter (ADF) 33 then sends the round-trip propagation delay time N _d1 and impulse response time N _hyb1 to the first reflection point and the round-trip propagation delay time N _{d2 to} the second reflection point with respect to the echo canceller 51. And the impulse response time N _hyb2 is supplied.

Thereby, the echo canceller 51 configures the N _d2 section and the N _d2 −N _d1 −N _hyb2 section with the delay elements, and _sets the N _hyb1 section and the N _hyb2 section as targets of adaptive control.

If the echo canceller 51 does not have the ability to make detailed settings so far, the adaptive digital filter (ADF) 33 may deliver only N _total .

Since the adaptive digital filter (ADF) 33 only has to supply the position of each reflection point to the echo canceller 51, the round-trip propagation delay time N _d1 to the first reflection point and the round-trip propagation delay to the second reflection point. The time N _d2 may be transferred and the impulse response times of the first reflection point and the second reflection point may be transferred as a predetermined value (for example, a constant).

(C) Operation Example FIG. 7 shows an example of operation conditions of the multiple reflection estimation process of the embodiment. FIG. 3 is a diagram showing an actual impulse response of the echo path when the operating conditions of FIG. 7 are used.

  FIG. 3A shows the actual impulse response of the echo path. FIG. 3B shows the impulse response estimated by the multiple reflection estimator 1 of this embodiment. In FIG. 3B, since the thinning rate is 1/8, the time resolution is 1 ms.

  Comparing FIG. 3 (A) and FIG. 3 (B), the estimated impulse response waveform shown in FIG. 3 (B) is thinned out from the actual impulse response shown in FIG. 3 (A). Therefore, the details are different. However, it can be seen that the position of the impulse response peak is almost captured. That is, the multiple reflection estimator 1 captures the position of the echo reflection point included in the echo path.

  The echo canceller 51 makes it possible to cancel the echo by reporting the three peaks supplied from the multiple reflection estimator 1 and subjecting each of the predetermined sections to a subject of adaptive control from a few ms before.

(A-3) Effects of Embodiment As described above, according to this embodiment, the following effects can be obtained.

(A-3-1) Effective use of voice frequency distribution For example, a voice signal band used in an analog telephone network is limited to 300 Hz to 3400 Hz. Therefore, even if the low frequency of the speech signal is simply extracted by the low pass filter and the impulse response of the echo path is estimated, there is no signal of 300 Hz or less. Therefore, the accuracy and estimated speed of the estimated value of the impulse response of the echo signal are deteriorated. However, this embodiment extracts an arbitrary band of a voice signal (for example, a frequency band where voice energy is concentrated), shifts the frequency band to a low frequency side, and down-samples the analog telephone. The impulse response of the echo path can also be estimated for the voice signal on the network.

(A-3-2) Reduction of processing amount of adaptive digital filter In this embodiment, since both the opposite speaker's voice signal and its echo signal are down-sampled, the adaptive digital filter process for estimating the impulse response of the echo path The amount can be reduced.

(A-3-3) Improvement of estimation speed of adaptive digital filter In the coefficient update expression of the adaptive digital filter exemplified in Expression (2-4), μ ₄ is a constant that determines the convergence speed and the steady-state error. The mu ₄ convergence speed closer to zero is slower but steady-state error becomes smaller and the convergence speed becomes faster when steady-state error is increased close to 1. In consideration of this, a normal speech echo canceller does not set μ ₄ to a value close to 1. On the other hand, in the case of the embodiment, the purpose is not to accurately measure the impulse response of the echo path, but it is only required to be able to be obtained within a certain allowable deviation to the echo propagation point. Therefore, the multiple reflection estimator of the embodiment can set μ ₄ to a value close to 1, and can make the estimated speed of the impulse response of the echo path faster than the conventional technique.

(A-3-4) Economicalization of voice echo canceller Propagation delay time to a plurality of echo occurrence points inherent in the echo path estimated by the adaptive digital refilter according to the embodiment, and impulse response time of each echo occurrence point ( This may be treated as a constant) to the echo canceller, so that the echo canceller can specify the total length of the echo path and the interval to be adaptively controlled with respect to the impulse response of the echo path. Therefore, it is possible to reduce the processing amount, circuit implementation scale, cost, and power of the echo canceller. In particular, this effect is significant in the case of a VoIP gateway device in which an echo canceller is mounted in multiple channels.

(B) Other Embodiments The multi-reflection estimator of the above-described embodiment is, for example, a line echo canceller that cancels echoes generated on an analog telephone communication line, or a hands-free phone or conference system that uses a speaker and a microphone. The present invention can be applied to an acoustic echo canceller that cancels generated echoes.

  DESCRIPTION OF SYMBOLS 1 ... Multiple reflection estimator, 11 ... Band-pass filter (BPF), 12 ... Multiplier (MPY), 13 ... Low-pass filter (LPF), 14 ... Decimator (DCM), 21 ... Band-pass filter (BPF), 22 ... multiplier (MPY), 23 ... low pass filter (LPF), 24 ... decimator (DCM), 31 ... sine wave generator (SGEN), 32 ... single talk detector (SGL) 33, an adaptive digital filter (ADF), 34, a subtracter (SUB).

Claims (6)

A first frequency shift means for extracting a first signal component of a predetermined band from the received first input signal and shifting the first signal component to a predetermined low frequency;
First downsampling means for downsampling the first signal component from the first frequency shift means;
Second frequency shift means for extracting a second signal component of a predetermined band for the received second input signal via the echo path, and shifting the second signal component to a predetermined low band;
Second downsampling means for downsampling the second signal component from the second frequency shift means;
Impulse response estimation means for estimating an impulse response of the echo path based on the first signal component and the second signal component down-sampled by the first down-sampling means and the second down-sampling means. When,
An echo path characteristic estimation device comprising: echo path characteristic estimation means for estimating an echo path characteristic of the echo path based on an impulse response waveform of the echo path estimated by the impulse response estimation means.   The echo path characteristic estimation means estimates a round-trip propagation delay time between at least one or a plurality of echo occurrence points in the echo path based on one or a plurality of peak positions of the impulse response waveform of the echo path. The echo path characteristic estimation apparatus according to claim 1, wherein: Comprising sine wave generating means for generating a single frequency sine wave signal;
The first frequency shift unit and the second frequency shift unit are configured to calculate the first signal component and the second frequency component by multiplying the sine wave signal by the first signal component and the second signal component. The echo path characteristic estimation apparatus according to claim 1 or 2, wherein the two signal components are frequency-shifted to a low frequency range. A single talk detecting means for monitoring reception of the first input signal and the second input signal and detecting a single talk state;
The impulse response estimation means estimates the impulse response of the echo path when a single talk state is detected by the single talk detection means. Echo path characteristic estimation device. Computer
A first frequency shift means for extracting a first signal component of a predetermined band from the received first input signal and shifting the first signal component to a predetermined low frequency;
First downsampling means for downsampling the first signal component from the first frequency shift means;
A second frequency shift means for extracting a second signal component of a predetermined band for the received second input signal via the echo path and shifting the second signal component to a predetermined low frequency;
Second downsampling means for downsampling the second signal component from the second frequency shift means;
Impulse response estimation means for estimating an impulse response of the echo path based on the first signal component and the second signal component down-sampled by the first down-sampling means and the second down-sampling means. ,
An echo path characteristic estimation program that functions as echo path characteristic estimation means for estimating an echo path characteristic of the echo path based on an impulse response waveform of the echo path estimated by the impulse response estimation means.   5. An echo path characteristic according to claim 1, wherein a pseudo echo is generated by an adaptive filter, and an echo component is canceled by subtracting the second input signal from the first input signal. An echo canceller comprising an estimation device.

Images