RU2722220C1 - Device for multichannel adaptive echo signal compensation - Google Patents

Abstract

FIELD: acoustics.

SUBSTANCE: invention relates to development of telecommunication systems for exchange of information by acoustic speech signals in conditions of external acoustic interference. Invention can be used in loudspeaker communication systems (LCS), in operational-command control telecommunication audio exchange systems characterized by presence of external acoustic noise - echo signals. To determine long-term echo parameters and number thereof, to determine the number of compensation channels, the compensation device calculates a cross-correlation function (CCF) of the digitized signal obtained at the output of the microphone, and reference signal of correlator obtained in frame generation unit and delays. By value of CCF and energy of reference signal R_uy(kT), R_uu(0), long-term echo parameters are estimated: estimation of echo delay

on k^th delay interval and estimation of echo attenuation

relative to energy of reference signal of correlator, applying local extrema allocation algorithm in long-term parameters calculation unit. Long-term parameters are used to form an echo signal model for generating a reference signal of the L^th order adaptive filter of the k^th channel. Adaptive echo compensation algorithm consists in formation of a matrix of a reference signal, correction of adaptive filter coefficients by an adaptation algorithm, correction of a matrix according to a minimized value of error estimation of echo delay - Δ_k, iterative minimization of error estimation of echo delay. When the minimized echo delay assessment error in the compensation device is the most complete compensation of the echo signal reflected with delay.

EFFECT: achieved technical result is increase in efficiency of echo-signals suppression in LCS telecommunication systems by criterion of level of compensation of reflected echo-signals, is provided by input to unit of compensation of unit from several adaptive filters, each of which provides compensation of echo signals only in its range of delays localized in the vicinity of the long-term delay parameter.

1 cl, 7 dwg

Description

The invention relates to the field of creating communication systems for exchanging audio information, relates to the field of creating telecommunication systems for the exchange of information by acoustic speech signals under external acoustic noise. The invention can be used in speakerphone systems, in operational-command and control telecommunication audio exchange systems.

The exchange of audio information by telecommunication systems of loud-speaking communication (GHS) is characterized by the presence of external acoustic noise and acoustic noise - echo signals. Therefore, in the design of GHS telecommunication systems, the task is to create algorithms for suppressing acoustic noise and compensating for acoustic echo signals.

A number of patents are known for solving the problem of suppressing interference of the form of echo signals. So in accordance with the patent for the invention No. 2495506 (13) C2 it is known "Device and method for calculating the control parameters of the echo cancellation filter and device and method for calculating the delay value" [1].

The invention relates to devices and methods for calculating control parameters of a blocking filter. The technical result is to improve the sound quality of acoustic systems by suppressing the reflected signals in proportion to the noise level. A device for calculating control parameters of a noise reduction filter for filtering a second audio signal in order to eliminate an echo signal based on a first audio signal includes a calculator having in its construction a determinant of value for calculating at least one energy indicator for a band signal, at least , two consecutive in time data blocks of at least one signal from a group of signals. The calculator also includes an average value determiner for determining at least one average value of at least one calculated energy indicator for the band signal. The calculator also includes a modifier for adjusting at least one energy indicator for the strip signal based on the calculated average value for the strip signal. The calculator also includes a control parameter calculator (270) for the suppression filter (210) based on at least one adjusted energy indicator.

The main reasons hindering the achievement of the result set out in the presented patent is that the presence of stationary or quasi-stationary interference and noise in the microphone signal in such cases can significantly affect the resulting sound quality of the system as a whole.

In accordance with the patent for invention RU 2520359 (13) C2, it is known “Acoustic Echo Cancellation Device and Front-End Conferencing Device” [2].

The invention relates to acoustic echo suppression means. The technical result consists in reducing computational complexity and increasing the efficiency of the acoustic echo cancellation process. The acoustic echo canceller includes an input interface means for extracting a downmix signal from an input signal containing downmix and overhead parametric information, which together represent a multi-channel signal; also includes a calculator for calculating the transmittance of the adaptive filter based on the downmix signal and the microphone signal or a signal derived from the microphone signal; and an adaptive filter of the microphone signal or a signal derived from the microphone signal, using the specified transmittance for suppressing the echo excited by the multi-channel signal in the microphone signal.

The device in question does not include a multichannel encoder designed to encode multiple input audio signals with subsequent conversion into a downmix signal and its accompanying parameter information, which together contain many input audio signals, one of which is a microphone signal from at least one microphone unit. For this reason, it is not intended to obtain a down-mix signal as a derivative of the microphone signal, which reduces the quality of processing of the transmitted data.

In accordance with the patent for utility model No. 148617 (13) U1, “Adaptive invariant echo canceller” is known [3].

The proposed utility model relates to the field of telecommunications and can be used to upgrade existing and develop new duplex information transmission systems.

This model continuously evaluates and generates compensating signals and subtracts them from the sum of the received and compensated echo signals, which ensures the adaptability of the invariant echo canceller.

The structural diagram of an adaptive invariant echo canceller consists of a control signal calculation loop, two echo compensation loops, and an adaptation module.

The principle of operation of this circuit involves a training phase, during which signals from the opposite side should be absent, and the transmitter transmits a training signal of unit amplitude recorded in the memory unit, and the echo path response to it is sent to the second memory unit.

In the message transmission mode, the time samples of the transmitted signals through the key are fed to the input of the memory block, where they are stored during the sampling time interval, and to the first input of the divider, which calculates the control signal equal to the ratio of the two time samples of the transmitter signal transmitted at the current and previous times . Due to the invariance property — maintaining the same ratio for the corresponding samples of the echo signal at the output of the differential system (i.e., at the output of the parasitic echo path) —the multiplication of the transmitted signal stored in the second memory block by the echo multiplier by the control signal gives the value echo from the currently transmitted transmitter signal. When subtracting the echo value calculated in this way from the sum of the real echo and the received signal, the echo is compensated by the first echo canceller circuit.

However, the first circuit, eliminating the echo signals, simultaneously changes the structure of the received signals. To restore it, a second (lower) circuit is used, which has a mirror structure consisting of an adder, a multiplier, a memory block.

As a result of changing the time-frequency characteristics of the used communication channel after the training stage, the differential system is unbalanced, as a result of which echo signals appear in the second circuit, repeating the shape of the transmitted signals, accurate to an unknown factor. To eliminate the arisen echoes, an adaptation module is used, containing a divider, integrator, divider, multiplier, and subtractor. In this case, by means of a divider, an integrator and a divider, a value estimate is calculated.

The disadvantage of an adaptive invariant echo canceller is the presence of an echo cancellation error due to interference coming to the input of the echo canceller from the communication channel.

In accordance with the patent for utility model No. 168793 (13) U1, “Correlation adaptive invariant echo canceller” is known [4].

The utility model relates to telecommunications and can be used to upgrade existing and create new duplex information transmission systems.

The purpose of the utility model is to increase the accuracy of compensation of echo signals introduced by the own transmitter into the received signal, when the differential system is unbalanced and the noise is affected. To this end, it is proposed to introduce additional blocks into the adaptive invariant echo canceller circuit that detect, calculate and eliminate echo signals arising from the imbalance of the differential system and the effects of interference. The device detects, calculates and eliminates echoes due to the imbalance of the differential system due to their similarity with the transmitted signals, and increasing the accuracy of echo cancellation is ensured by replacing the procedure for dividing the sum of the received signals, noise and echo signals with the transmitted signals by the procedure for calculating the correlation value of this sum with transmitted signals. When changing the parameters of the communication channel and when the differential system is unbalanced, and when the noise channel n (t) is simultaneously exposed from the channel, the output of the second echo canceller circuit, namely, at the output of the BP3 memory block, the signal of the total error of echo cancellation ΔSэк (t) appears: ΔSek (t) = kSper (t) + n (t), where k is the change in the echo path redistribution coefficient resulting from the imbalance of the differential system. The reduction of the echo cancellation error ΔSec (t) is carried out by performing operations implemented in the adaptation module.

The main reasons hindering the achievement of the result set in the presented patent is that the above system does not provide sufficient compensation accuracy for the purposes of operational-command telecommunication communication systems.

Acoustic echo cancellation compensation is one of the methods of suppressing this type of external interference in GHS telecommunication systems. The complexity of solving this problem largely depends on the values of the echo delay. If the delays in the echo propagation channels reach values of the order of 0.5 s or more, the task is complicated by the requirements for the adaptive filter order as part of the echo canceller, which in this case should have several thousand taps. The complication in this case is due, first of all, to the stability problem and the low rate of convergence of such filters.

The stability and low convergence rate of filters can be solved by introducing a block of adaptive filters, each of which provides compensation only in its own range of delays localized in the vicinity of a certain characteristic value. In this case, echo signals at various delay intervals are considered mutually uncorrelated. As a result, the number of taps of these filters decreases to two to three hundred, the number of intervals of localized delays also decreases and is within a few units (from 1 to 3), thus, a noticeable gain occurs in the total number of taps of the compensation unit.

обозначает оценку задержки эха в k-канале,

- коэффициент, учитывающий оценку затухания в указанном канале и L - порядок фильтра, который устанавливается четным числом, х(n)=u₀(n)+y(n) оцифрованный сигнал на микрофонном выходе.An adaptive filter circuit of order L is shown in FIG. 1 on which the variable

denotes an estimate of the echo delay in the k-channel,

is the coefficient taking into account the attenuation estimate in the indicated channel and L is the order of the filter, which is set by an even number, x (n) = u ₀ (n) + y (n) the digitized signal at the microphone output.

и затухания

эхо-сигнала в k-канале.A block diagram of a multi-channel adaptive echo cancellation device is shown in FIG. 2. In accordance with the circuit of FIG. 2, the digitized signal at the microphone output in the form x (n) = u ₀ (n) + y (nD), consisting of the useful signal u ₀ (n) and the echo signal y (nD), is fed to the adder and to the calculation unit mutual correlation function (VKF) in the form of R _uy (kT). Based on the calculated VKF, the long-term delay parameters are evaluated

and attenuation

echo in the k-channel.

Moreover, the algorithm for estimating long-term parameters includes the following steps [5]:

1. Estimation of cross-correlation in the form

2. The calculation of the signal energy u ₀ (n), in accordance with the expression of the form

3. The allocation of local extremes

и

тогда

If

and

then

4. Evaluation of the long-term parameters of the echo channels

тогда

где β - порог, определяемый дисперсией шума в канале в виде β=mσ², m∈{2÷3}, [5].4.1. Estimation of attenuation if

then

where β is the threshold determined by the dispersion of noise in the channel in the form β = mσ ² , m∈ {2 ÷ 3}, [5].

4.2. Estimation of a significant amount of delay. If a = a _k , then

The obtained parameters are fed to the blocks of the formation of the reference model of the echo signals, which take into account the estimates of the echo delay during the formation of frames and estimates of the echo attenuation relative to the initial level of the emitted signal. The generated model of the echo signal forms the reference signal of the adaptive filter of the L order of the order, which is fed to its input. The adaptive echo cancellation algorithm includes the following steps:

1. The formation of the reference signal adaptive filter

2. The formation of the matrix U (n) = (h ₁ u (n) h ₂ u (n-1) ... h _L u (nL)),

3. Correction of adaptive filter coefficients in accordance with the rule of the form,

Where

4. Correction matrix:

- if Δ _k > 0, then

- if Δ _k <0, then

- if Δ _k = 0, then

5. Correction of the iteration number n = n + 1.

6. Unconditional transition to step 1.

Thus, the signal from the output of the adaptive filter after matrix correction is fed to the adder, providing compensation of the echo signal to the level of estimation error

Investigations of the determination of long-term parameters were carried out by modeling software in the Matlab environment.

As a signal model, the sum of the same fragments of the speech signal with different delays and attenuations was used. The initial segment of the signal model is shown in FIG. 3, a segment of a model characterized by the presence of fragments simulating echo signals is shown in FIG. 4. The calculation of the WKF function was carried out according to the formula of claim 1 in the algorithm for estimating long-term parameters.

In FIG. Figure 5 presents the result of studying the normalized WKF model of the speech signal depending on the delay of the nT echo signal model, which illustrates the application of the method of correlation-extreme estimation of long-term parameters of reflected echo signals in the development of echo cancellation algorithms in audio communication telecommunication systems [6, 7].

In studies simulating the degree of echo cancellation compensation by a multi-channel adaptive compensation device in accordance with the block diagram of FIG. 2. An estimate of the echo compensation level is represented by a histogram of the obtained compensation values in FIG. 6, which shows that the degree of compensation of the echo signal is in the range from 6 to 8.2 dB. Moreover, the maximum distribution of the compensation level is in the vicinity of 7.5 dB.

The dynamics of the settings of the device multi-channel adaptive echo cancellation is shown in Fig 7A. (curve 2). The average setup time can be estimated in the range from 0.2 to 0.4 seconds.

The increase in the tuning speed of the multi-channel adaptive echo cancellation device is confirmed by the results of an experimental study of the dependence of the adaptive filter tuning time on the number of adjustable coefficients, which are presented in FIG. 7b. From FIG. 7b it can be seen that the setup time of the adaptive filter in the multi-channel adaptive echo cancellation device with the evaluation of long-term parameters is reduced by more than two times compared to the setup time of the echo canceller with an adaptive filter with the number of adjustable coefficients L> 1000.

Thus, in the presented device of multi-channel adaptive echo cancellation, the tuning speed of adaptive filters is more than doubled compared to the tuning speed in known echo cancellers, the degree of compensation of echo signals is increased to -6 ÷ -8 dB, which is 1-4 dB more compared to the known results.

The technical result of the invention is:

- to minimize the error in estimating the delay in the k-th channel according to the data within the final time window, due to iterative correction of the coefficients of adaptive filters of the echo canceller, as well as by reducing the number of taps of the above filters and increasing the tuning speed;

- introducing into the device multi-channel adaptive echo cancellation, developed on the basis of the method of correlation-extreme estimation of long-term parameters, an algorithm for estimating long-term parameters of the acoustic echo signal model, which allowed the creation of a multi-channel echo canceller algorithm with a reduced number of compensation channels equal to the number of allocated echo channels with significant long-term delays;

- also the technical result is to increase the efficiency of the telecommunication systems of audio exchange and HGS in the conditions of intensive exposure to external acoustic noise - echo signals.

Thus, the proposed block diagram in FIG. 2 implementation of the method of multi-channel adaptive compensation of echo signals provides the possibility of more efficient transmission of information in telecommunication systems, audio exchange, in GHS systems.

SOURCES OF INFORMATION

1. RF patent No. 2495506 (13) C2 (IPC G10L 21/02 (2013.01), Н04М 9/08 (2006.01), G11B 20/24 (2006.01)). Published 10/10/2013. Bull. No. 28.

2. RF patent No. 2520359 (13) C2 (IPC G10L 21/02 (2013.01), Н04М 9/08 (2006.01) Published on June 20, 2014. Bull.

3. RF patent No. 148617 (13) U1 (IPC Н04В 3/20 (2006.01)). Published December 10, 2014. Bull. Number 34.

4. RF patent No. 168793 (13) U1 (IPC Н04В 3/20 (2006.01). Published on 02.21.2017. Bull. No. 6.

5. Ermolaev V.A., Kropotov Yu.A. On correlation estimation of parameters of models of acoustic echo signals // Problems of Radioelectronics. 2010. T. 1. No. 1. S. 46-50.

6. Kropotov Y.A., Belov A.A. Method of correlation-extreme parameter estimation of acoustic echoes in telecommunication audio exchange systems. 2015 International Siberian Conference on Control and Communications, SIBCON 2015 - Proceedings. 7147108.

7. Kropotov Yu.A., Ermolaev V.A. Modeling and research methods of acoustic signals, noise and interference in telecommunication systems: monograph. - M. - Berlin: Direct Media, 2016 .-- 256 p.

Claims (1)

A device for multi-channel adaptive compensation of echo signals in telecommunication systems for the exchange of information by voice signals in loud-speaking communication systems in operational-command and information-control telecommunication audio-exchange systems containing an adder, the first input of which is a digitized signal from the microphone output, consisting of a useful signal and an echo -signal, at the same time the same signal from the microphone output is fed to the first input of the mutual correlation function calculation unit (VKF), delayed frames of the speech signal from the output of the frame formation and delay unit are fed to the second input of the VKF calculation unit, the calculated VKF is fed to the VKF value calculation unit at k = 0 and to the unit for calculating the estimates of long-term echo parameters, the calculated estimates of the echo parameters, the estimate of the value of the echo attenuation

and estimation of echo delay value

fed to the adaptive filter reference signal generating unit, in which the adaptive filter reference signal is generated in the form of an echo signal model with an estimate of the echo attenuation and an estimate of the echo delay and the initial delay

where L is the order of the adaptive filter, the output reference signal of the adaptive filter is supplied to the first input of the adaptive filter, the error signal of estimation of the delay echo is fed from the adder output to the second input of the adaptive filter

from the output of the adaptive filter, the generated echo estimate, taking into account the minimized error in estimating the delay of the echo, is multiplied by a coefficient i = -1 and fed to the second input of the adder of the adaptive compensation device; from the output of the adder, a speech signal cleared of the echo signal in gain is fed to the speakerphone.

Images