CN106533500B - A method of optimization Echo Canceller convergence property - Google Patents
A method of optimization Echo Canceller convergence property Download PDFInfo
- Publication number
- CN106533500B CN106533500B CN201611060313.1A CN201611060313A CN106533500B CN 106533500 B CN106533500 B CN 106533500B CN 201611060313 A CN201611060313 A CN 201611060313A CN 106533500 B CN106533500 B CN 106533500B
- Authority
- CN
- China
- Prior art keywords
- data block
- echo
- signal
- far
- error
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000005457 optimization Methods 0.000 title description 8
- 238000001228 spectrum Methods 0.000 claims abstract description 28
- 239000013598 vector Substances 0.000 claims abstract description 24
- 230000003044 adaptive effect Effects 0.000 claims abstract description 20
- 238000001514 detection method Methods 0.000 claims description 18
- 230000001133 acceleration Effects 0.000 claims description 13
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 6
- 230000003595 spectral effect Effects 0.000 claims description 6
- 238000001914 filtration Methods 0.000 claims description 5
- 238000003491 array Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 abstract description 4
- 230000011664 signaling Effects 0.000 abstract 1
- 230000006870 function Effects 0.000 description 12
- 230000000694 effects Effects 0.000 description 8
- 230000008859 change Effects 0.000 description 5
- 238000013459 approach Methods 0.000 description 4
- 238000000354 decomposition reaction Methods 0.000 description 4
- 238000012937 correction Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 description 1
- 241000219498 Alnus glutinosa Species 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000000903 blocking effect Effects 0.000 description 1
- 239000000872 buffer Substances 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 210000005069 ears Anatomy 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 239000010977 jade Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 230000000979 retarding effect Effects 0.000 description 1
- 238000002945 steepest descent method Methods 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M9/00—Arrangements for interconnection not involving centralised switching
- H04M9/08—Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic
- H04M9/082—Two-way loud-speaking telephone systems with means for conditioning the signal, e.g. for suppressing echoes for one or both directions of traffic using echo cancellers
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B3/00—Line transmission systems
- H04B3/02—Details
- H04B3/20—Reducing echo effects or singing; Opening or closing transmitting path; Conditioning for transmission in one direction or the other
- H04B3/23—Reducing echo effects or singing; Opening or closing transmitting path; Conditioning for transmission in one direction or the other using a replica of transmitted signal in the time domain, e.g. echo cancellers
Landscapes
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Filters That Use Time-Delay Elements (AREA)
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
Abstract
The invention proposes a kind of methods for optimizing Echo Canceller (AEC) convergence property, it is used to constantly correct the weight coefficient of the transverse adaptive filter device of Echo Canceller, comprising: step a: carrying out registration process to the close of Echo Canceller, remote data；Step b: initialization filter frequency domain weight coefficient vectorFar end signal power spectrumFor null vector；Step c: the remote data block that i-th newly inputs is denoted asNear end data block is denoted asLength is M data sample, and the 2M data that the data block and previous data block adjacent thereto to each new input collectively form do Fourier transform and carry out single-ended call according to the amplitude spectrum first-order difference related coefficient of proximal end and remote signaling and detect；Step d: frequency-domain constraint and the estimation of convergence accelerated factor are carried out to obtain the estimation of time domain echo, and the error obtained after carrying out echo cancelltion near end signal exports；With step e: when AEC is in single-ended talking state, being updated according to error output to filter weight coefficient.
Description
Technical Field
The present invention relates to signal processing technology in the field of digital communication, and more particularly, to An Echo Cancellation (AEC) technology in speech signal processing.
Background
Echo cancellation is an indispensable element in a telephone communication system. The process of echo generation is shown in fig. 1: the downlink signal u (n) transmitted from the far end can become the sound which can be heard by human ears after being radiated from the near end loudspeaker. Meanwhile, after the sound radiated outside is reflected and transmitted by the near-end acoustic medium, the sound is finally picked up by the microphone as a part of the uplink signal and transmitted to the far end, so that the far-end speaker hears the sound of the far-end speaker and the echo is formed. The existence of echo objectively affects the quality of conversation and information exchange between the callers, and subjectively brings poor conversation feeling to the callers, but the existence of echo cannot be avoided in a hands-free conversation system. The current approach to this problem is mainly solved by adaptive filtering.
In fig. 1, s (n) is a near-end signal, y (n) is a far-end transmitted signal u (n) and an echo generated at the near-end after being radiated from a loudspeaker, [ w (n) ]0(n)w1(n)...wM-1(n)]For the weight coefficient of the transversal adaptive filter, the echo cancellation is to continuously modify the weight coefficient of the transversal adaptive filter according to a certain rule, so that the amplitude-frequency response can be close to the real echo path to the maximum extent, and the output y of the echo cancellation is used1(n) to approximate the true echo y (n) and then subtract y from the total signal s (n) + y (n) picked up by Mic1And (n) the purpose of canceling echo is achieved. The performance of the adaptive filter dominates the effect of echo cancellation.
Various types of algorithms for implementing the adaptive filter include:
1. LMS (Least-Mean-Square) Adaptive algorithms and their derivatives, such as nlms (normalized LMS), dlms (delayed LMS), alms (additive LMS), BLMS (block LMS), fast BLMS, PBFDAF (Partitioned block frequency-domain Adaptive Filter), pbd (Partitioned block indirect frequency-domain), etc.;
2. AP (Affine Projection algorithm) and its derivative algorithms, such as APRU (affinity Projection with reactive matrix updates), BAP (Block affinity Projection), etc.;
3. RLS (Recursive Least squares) algorithms and derivative algorithms, such as QR-RLS (QR-decomposition-based RLS, RLS based on QR decomposition), QR-LSL (QR-decomposition-based Least square lattice type based on QR decomposition), and the like;
the algorithms of NLMS, DLMS, ALMS, APRU and RLS are to update the filter coefficient in time domain by using single sample input as unit, while BLMS, fast BLMS, PBFDAF, PBUFD and BAP are to update the filter coefficient in time domain by using data block input as unit.
In the above several types of adaptive filtering algorithms, the RLS is best in terms of the convergence characteristic and tracking capability of the filter, but the final output effect of the RLS is good at the expense of high computational complexity, and for an actual echo channel with a long delay characteristic, some related optimization work and difficulty involved in the RLS when the RLS is applied to the actual AEC implementation are also self-evident; however, in terms of computational complexity, the least amount of computation required should be the LMS algorithm, but the LMS algorithm has poor convergence characteristics and takes a long time for the adaptive filter to enter a steady state.
Disclosure of Invention
The invention considers from the practical engineering application requirements of AEC, and deeply optimizes and improves the algorithm itself based on the fast BLMS for the defects of poor convergence characteristic, long time required by the adaptive filter to enter the steady state and unclean echo cancellation in the initial call stage when the LMS algorithm is applied to the AEC, thereby providing a method for optimizing the convergence characteristic of the echo canceller so as to enhance the echo cancellation effect after the initial call stage and the later filter enter the steady state.
The invention provides a method for optimizing the convergence characteristic of an echo canceller, which is used for continuously correcting the weight coefficient of a transverse self-adaptive filter of the echo canceller and comprises the following steps:
step a: performing alignment processing on near-end data and far-end data of an echo canceller;
step b: initializing filter frequency domain weight coefficient vectorsRemote signal power spectrumIs a zero vector;
step c: the ith newly input far-end data block [ u (iM), u (iM +1).. -, u (iM + M-1)]Is marked asNear-end data block [ x (iM),.., x (iM +1), x (iM + M-1)]Is marked asThe length is M data samples, 2M data formed by each newly input data block and the previous data block adjacent to the newly input data block are subjected to Fourier transform, and single-ended call detection is carried out according to the amplitude spectrum first-order difference correlation coefficient of the near-end signal and the far-end signal;
step d: according to the Fourier transform result, the single-end call detection result and the frequency domain weight coefficient of the current adaptive filterPerforming frequency domain constraint and convergence acceleration factor estimation to obtain time domain echo estimation, and obtaining error output after echo cancellation of near-end signalAnd
step e: outputting the error when the AEC is in single-end call statePerforming time domain constraint to obtainAnd according to the error after the time domain constraintAnd a diagonal matrix obtained from spectral components of the power spectrum of the remote signalThe filter frequency domain weight coefficients are updated.
The invention has the following advantages: the method for optimizing the convergence characteristic of the echo canceller is based on the fast BLMS algorithm, greatly shortens the time required by the echo canceller to enter a steady state and improves the convergence characteristic under the conditions of inheriting the advantage of low computational complexity and easy realization, overcomes the defects that the echo cancellation effect is poor in the initial call stage and the cancellation effect is not ideal enough after the echo canceller enters the steady state when the echo canceller is applied to AEC, and simultaneously bypasses the defects that the good convergence characteristic of the AP-class algorithm and the RLS-class algorithm is high and the computational complexity is high.
Drawings
Fig. 1 is a block diagram of a conventional echo canceller;
FIG. 2 is a comparison graph of the results of single-ended call detection;
FIG. 3 is a graph showing a comparison of the convergence characteristics of the BLMS algorithm before and after the improvement of the present invention;
fig. 4 is a flow chart of a method of optimizing the convergence characteristics of an echo canceller in accordance with the present invention.
Detailed Description
In order to better understand the spirit of the present invention, it is further described below with reference to some preferred examples of the present invention.
The method of the present invention is described in detail below with reference to fig. 4. The method for optimizing the convergence characteristic of the echo canceller continuously corrects the weight coefficient of the transverse adaptive filter of the echo canceller by the following steps. After the incoming near-end (upstream) and far-end (downstream) signals have passed through their respective buffers, the following steps are performed:
step a: performing alignment processing on near-end data and far-end data of an echo canceller;
step b: initializing filter frequency domain weight coefficient vectorsRemote signal power spectrumIs a zero vector; here, T denotes the transpose of the matrix, and M denotes the block data sample length0 in parentheses represents data of the 0 th frame; (this initialization step is not shown for simplicity of the flow chart);
step c: the ith newly input far-end data block [ u (iM), u (iM +1).. -, u (iM + M-1)]Is marked asNear-end data block [ x (iM),.., x (iM +1), x (iM + M-1)]Is marked asThe length is M data samples, 2M data formed by each newly input data block and the previous data block adjacent to the newly input data block are subjected to Fourier transform, and single-ended call detection is carried out according to the amplitude spectrum first-order difference correlation coefficient of the near-end signal and the far-end signal;
step d: according to the Fourier transform result, the single-end call detection result and the frequency domain weight coefficient of the current adaptive filterPerforming frequency domain constraint and convergence acceleration factor estimation to obtain time domain echo estimation, and obtaining error output after echo cancellation of near-end signalAnd
step e: outputting the error when the AEC is in single-end call statePerforming time domain constraint to obtainAnd according to the error after the time domain constraintAnd a diagonal matrix obtained from spectral components of the power spectrum of the remote signalThe filter frequency domain weight coefficients are updated.
The above steps are further explained below.
Step c includes a step c1 of performing fourier transform on the far-end data block and the near-end data block, obtaining:
in step c, the single-ended call detection is performed after the fourier transform is performed on the data block, and the method further comprises the following steps:
step c 2: obtaining a first order difference of a magnitude spectrum of data of a distal blockAnd first order difference of near-end block data magnitude spectrumRespectively as follows:
wherein,respectively represent vectorsThe mth component of (a), and so on;
step c 3: according to the aboveAndcalculating the amplitude spectrum first-order difference correlation coefficient rho of the near-end signal and the far-end signal of the ith frame,wherein,is a sign function whenWhen the value is larger than a certain preset energy threshold value for generating echo, the value of the sign function is 1, otherwise, the value is 0;
step c 4: when the p calculated in the step c3 is greater than a preset single-ended call threshold value TSingleTalkThen, it can be determined that the AEC is in the single-ended call state.
Step d further comprises:
step d 1: preliminarily estimating an echo spectrum:diag is a mathematical symbol representing the generation of the corresponding diagonal matrix from the vector;
step d 2: calculating the error between the near-end signal magnitude spectrum and the estimated echo signal magnitude spectrum:
where M denotes the length of the current input data block;
wherein,are respectively vectorThe mth component of (1);
the above steps d1 and d2 are mainly performed for calculating the convergence acceleration factor and the time domain echo signal in the following step d3, which are not shown in fig. 4 for the sake of simplicity of the flowchart.
Step d 3: carrying out frequency domain constraint and convergence acceleration factor estimation to obtain an echo signal estimated in a time domain:
taking the next M elements;
wherein the vectorM component ofThe constant beta is calculated byGamma is a very small positive number, 1+ gamma → 1+，1-γ→1-(ii) a For ease of discussion, scalars are used hereinIs named as a convergence acceleration factor and is used as a convergence acceleration factor,named convergence acceleration factor vector;
step d 4: obtaining an error output after echo cancellation of the near-end signal:
wherein,in order to be the near-end signal,is an estimated echo signal obtained in the time domain.
In step e, updating the filter weight coefficients when the AEC is determined to be in the single-ended call state, further comprising:
step e 1: and (3) carrying out time domain constraint on the estimation error:a zero vector of dimension Mx 1;
step e 2: for far-end signal power spectrumEach spectral component ofFirst order lag filtering is performed:
wherein,is a pair of angular arrays, the elements of which areIs obtained by taking reciprocal of each component, eta is a constant, eta is more than 0 and less than 1, sigma is a very small positive number, sigma → 0+；
Step e 3: obtaining an error after constraint according to the time domainAnd diagonal matrixTo adaptive filter weight coefficientUpdating:
first M elements of
Wherein alpha is a step size factor, the value range of the step size factor needs to meet the requirement of a standard BLMS algorithm,representing a zero vector of dimension Mx 1.
Through the steps, the weight coefficient of the transverse self-adaptive filter of the echo canceller can be continuously corrected, and the convergence characteristic of the echo canceller is optimized, so that the echo cancellation effect at the initial stage of the call is enhanced.
The theoretical basis of the process of the invention is explained in detail below.
As shown in figure 1 of the drawings, in which,is the tap input of the transversal filter at n time, y (n) is the echo generated after Speaker extension, s (n) is the near-end signal,the tap weight coefficient of the transverse filter at the moment n; echo cancellation is to continuously modify the weight coefficient of the transversal filter according to a certain rule to obtain an echo path that makes the amplitude-frequency response approach the real maximum, and output y by using the echo path1(n) to approximate the true echo y (n) and then subtract y from the echo signal picked up by Mic and the near-end signal1(n):
Assume that s (n) is a mean of 0 and a variance of σ2White noise of (2); based on the minimum mean square error criterion, a cost function is now considered
It is the transversal filter weight vectorA continuous differentiable function of; the significance of the above formula is: at a defined time n1To n2Given the expected value s (n) + y (n) and the input value u (n), an optimal weight coefficient needs to be foundMake itCan be minimized; according to the adaptive Filter principle (fourth edition) (Zhengbao jade et al, electronics industry Press, 2003.4), the cost function is divided intoAt the current weightThe first order taylor expansion is performed to obtain:
in the above formulaWhen the size of the particles is smaller, the effect is achieved,is composed ofIn thatA gradient vector of (1), order
μ is a small positive number, called the step factor; substituting equation (4) into equation (3) can obtain:
the symbol | | represents a modulo operation; as can be seen from the formulae (3) to (5): when inCorrection amount of (A)Along the edgeWhen the direction of the negative gradient is advanced, the cost function is gradually reduced and finally approaches to the minimum value J with the increase of the correction timesmin(ii) a The above is the theoretical basis of the steepest descent method based on the LMS and the derivative algorithms thereof such as NLMS, DLMS, BLMS and the like;
if it will beIn thatPerforming second-order Taylor expansion:
h (n) is a Hessian matrix,the upper type is toTaking the derivative and making the reciprocal 0, we can get:
the above formula is a Newton method iterative formula, and one-step updating can be realized to enable the cost function to reach the minimum value;
now, the following changes are made to equation (4):
wherein, the formula beta → 1 and (8) are substituted into the formula (3) to obtain:
the final term of which is derived from equation (9) with an accelerating or retarding cost functionTo the direction ofThe speed of approach, namely:
(I) (beta-1) andwhen the signs are different, the acceleration function is realized;
(II) (beta-1) andwhen the numbers are the same, the effect of blocking is achieved.
Therefore, only need to be in each pairThe beta is well controlled in the process of correction, so that a large amount of matrix operations related to a Newton iterative formula expressed by the formula (7) can be effectively avoided, and meanwhile, the matrix operations can be performedThe initial convergence speed of the filter is obviously accelerated, which is the theoretical basis for realizing the improved method of the invention. The comparison of the convergence characteristics of the echo canceller after the optimization by the method of the present invention with the convergence characteristics before the optimization is shown in the upper diagram of fig. 3; it can be seen from the figure that the method of the present invention has better convergence characteristics after being optimized; the lower graph of fig. 3 is a graph of the change of the output cost function with time, and it can be seen from the graph that after the optimization by the method of the present invention, the change characteristic of the value of the cost function with time is better than that before the optimization (i.e. the value of the cost function after the optimization is much smaller than that before the optimization).
When a sample with the data length L being 1 is input in fig. 1, the weight coefficient of the filter is updated once, and the corresponding algorithm is LMS/NLMS, etc.; when the weight coefficient of the filter is updated once after the input data length is a section of data block with L >1, the corresponding algorithm is BLMS.
The frequency domain constraints are explained below.
In FIG. 1, the total signal picked up by Mic is s (n) + y (n), y1(n) is an estimate of the echo signal y (n); let S (k), Y1(k) Respectively are s (n), y1(n) a fourier transform; since it is an estimation, there is an estimation error inevitably; from the frequency domain, when:
|Y1(k)|＞|S(k)+Y(k)| (10)
i.e. the estimated amplitude spectrum of the echo signal is larger than the amplitude spectrum of the original total signal, this is not reasonable, so in the implementation of the new algorithm, each time the echo signal is estimated, the following frequency domain constraints are imposed:
|E(k)|≤|S(k)+Y(k)| (11)
|U(k)W(k)|≤|S(k)+Y(k)| (12)
u (k), w (k) are the fourier transforms of the filter input and weight coefficients, respectively; the frequency domain constraint can directly influence the error, and in turn can further influence the weight value adjusting mechanism, and simultaneously ensure that the far end can not hear suddenly increased and intolerable noise when the performance of the filter is still unstable or the performance of the filter is temporarily deteriorated or even diverged due to various reasons such as echo channel change and the like.
Single-ended talk detection is described below.
In AEC, the filter tap coefficients can only be modified in single-ended speech situations, i.e. in equation (1), the filter coefficients can only be updated if s (n) does not contain near-end speech; when s (n) includes near-end speech, the main component constituting the cost function j (w) is speech, and if the filter coefficients are updated again to make j (w) tend to be minimum, the near-end speech component is inevitably affected by different degrees, and even the filter is diverged in severe cases; the double-end call detection mainly comprises a Geigel algorithm, an ERLE algorithm, a normalized cross-correlation detection algorithm, a system cache double-filtering method and the like, wherein the normalized cross-correlation detection algorithm is relatively common and stable.
The speech is characteristic, and the echo signal sounds in subjective perception since it can be compared with the reference signal, which indicates that the two have commonality in the essential characteristics of the speech; from the perspective of frequency domain, the difference and the relative change of voice are mainly reflected on the difference and the relative change of each spectral component of a spectrogram, and in view of the current voice frequency domain characteristic, namely Mel frequency cepstrum coefficient, which is successfully applied in the field of voice recognition, the invention detects the single-ended call mode by comprehensively calculating the correlation of the first-order difference of the amplitude spectrum of the near-end frame signal and the far-end frame signal and the energy of the far-end signal.
Fig. 2 is a comparison between the normalized cross-correlation detection algorithm and the single-ended call detection result obtained by processing the actual signal by the method of the present invention, and it can be seen that: in the figure, the part of the former marked by an arrow in the time period of 0.5 second to 1.2 seconds generates false detection (namely the correlation coefficient calculated by the former is small and the single-ended call state is not detected), and the part of the former marked by the arrow in the time period of 4 seconds to 4.5 seconds generates false detection (namely the correlation coefficient calculated by the former is large and the non-single-ended call state is mistakenly detected to be the single-ended call state), but the single-ended call detection of the invention has good performance and better robustness.
While the foregoing has been with reference to the disclosure of the present invention, it will be appreciated by those skilled in the art that various changes and modifications may be made without departing from the spirit and scope of the invention as defined by the appended claims. Therefore, the protection scope of the present invention should not be limited to the disclosure of the examples, but should include various alternatives and modifications without departing from the invention, which are covered by the claims of the present patent application.
Claims (5)
1. A method for optimizing the convergence characteristics of An Echo Canceller (AEC), said method for continuously modifying the weight coefficients of a transversal adaptive filter of an echo canceller, comprising the steps of:
step a: performing alignment processing on near-end data and far-end data of an echo canceller;
step b: initializing filter frequency domain weight coefficient vectorsRemote signal power spectrumIs a zero vector;
step c: the ith newly input far-end data block [ u (iM), u (iM +1).. -, u (iM + M-1)]Is marked asNear-end data block [ x (iM),.., x (iM +1), x (iM + M-1)]Is marked asThe length is M data samples, 2M data formed by each newly input data block and the previous data block adjacent to the newly input data block are subjected to Fourier transform, and then single-end call detection is carried out on the data block subjected to Fourier transform according to the amplitude spectrum first-order difference correlation coefficient of the near-end signal and the far-end signal;
step d: according to the Fourier transform result, the single-end call detection result and the frequency domain weight coefficient of the current adaptive filterPerforming frequency domain constraint and convergence acceleration factor estimation to obtain time domain echo estimation, and obtaining error output after echo cancellation of near-end signalAnd
step e: outputting the error when the AEC is in single-end call statePerforming time domain constraint to obtainAnd according to the error after the time domain constraintAnd a diagonal matrix obtained from spectral components of the power spectrum of the remote signalThe filter frequency domain weight coefficients are updated.
2. The method according to claim 1, wherein step c comprises a step c1 of calculating a fourier transform of the far-end data block and the near-end data block, resulting in:
3. the method of claim 2 wherein the step c of performing single-ended call detection on the fourier transformed data block comprises:
step c 2: calculating the first order difference of the magnitude spectrum of the far-end data blockAnd first order difference of magnitude spectrum of near-end data blockRespectively as follows:
wherein,respectively represent vectorsThe mth component of (1), and so on, the following;
step c 3: according to the aboveAndcalculating the amplitude spectrum first-order difference correlation coefficient rho of the near-end signal and the far-end signal of the ith frame,wherein,is a sign function whenWhen the value is larger than a certain preset energy threshold value capable of generating echo, the value of the sign function is 1, otherwise, the value is 0;
step c 4: when the p calculated in the step c3 is greater than a preset single-ended call threshold value TSingleTalkThen, it can be determined that the AEC is in the single-ended call state.
4. The method of claim 3, wherein step d further comprises:
step d 1: preliminarily estimating an echo spectrum:wherein diag represents the mathematical sign that produces the corresponding diagonal matrix from the vector;is the adaptive filter weight coefficient;
step d 2: calculating the error between the near-end signal magnitude spectrum and the estimated echo signal magnitude spectrum:
where M denotes the length of the current input data block; wherein,are respectively vectorThe mth component of (1);
step d 3: carrying out frequency domain constraint and convergence acceleration factor estimation to obtain an echo signal estimated in a time domain:
taking the next M elements;
wherein the vectorM component ofThe constant beta is calculated byGamma is a very small positive number, 1+ gamma → 1+，1-γ→1-(ii) a Here will be a scalar quantityIs named as a convergence acceleration factor and is used as a convergence acceleration factor,named convergence acceleration factor vector;
step d 4: obtaining an error output after echo cancellation of the near-end signal: in order to be the near-end signal,is the estimated echo signal obtained in the time domain in step d 3.
5. The method of claim 4, wherein step e further comprises:
step e 1: and (3) carrying out time domain constraint on the estimation error: a zero vector of dimension Mx 1;
step e 2: for far-end signal power spectrumEach spectral component ofFirst order lag filtering is performed:
wherein,is a pair of angular arrays, the elements of which areEach component of (a) is obtained by taking the reciprocal of the component of (b), eta is a constant,
0 < eta < 1, sigma is a very small positive number, sigma → 0+；
Step e 3: obtaining an error after constraint according to the time domainDiagonal matrixTo adaptive filter weight coefficientUpdating:
first M elements of
Wherein alpha is a step size factor, the value range of the step size factor needs to meet the requirement of a standard BLMS algorithm,representing a zero vector of dimension Mx 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611060313.1A CN106533500B (en) | 2016-11-25 | 2016-11-25 | A method of optimization Echo Canceller convergence property |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611060313.1A CN106533500B (en) | 2016-11-25 | 2016-11-25 | A method of optimization Echo Canceller convergence property |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106533500A CN106533500A (en) | 2017-03-22 |
| CN106533500B true CN106533500B (en) | 2019-11-12 |
Family
ID=58357451
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611060313.1A Active CN106533500B (en) | 2016-11-25 | 2016-11-25 | A method of optimization Echo Canceller convergence property |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106533500B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10880440B2 (en) * | 2017-10-04 | 2020-12-29 | Proactiv Audio Gmbh | Echo canceller and method therefor |
| CN108172233B (en) * | 2017-12-12 | 2019-08-13 | 天格科技（杭州）有限公司 | The echo cancel method of signal and error signal regression vectors is estimated based on distal end |
| CN108154885A (en) * | 2017-12-15 | 2018-06-12 | 重庆邮电大学 | It is a kind of to use QR-RLS algorithms to multicenter voice signal dereverberation method |
| CN108696648B (en) * | 2018-05-16 | 2021-08-24 | 上海小度技术有限公司 | Method, device, equipment and storage medium for processing short-time voice signal |
| CN108711433B (en) * | 2018-05-18 | 2020-08-14 | 歌尔科技有限公司 | Echo cancellation method and device |
| CN109754813B (en) * | 2019-03-26 | 2020-08-25 | 南京时保联信息科技有限公司 | Variable step size echo cancellation method based on rapid convergence characteristic |
| CN110995951B (en) * | 2019-12-13 | 2021-09-03 | 展讯通信（上海）有限公司 | Echo cancellation method, device and system based on double-end sounding detection |
| CN110992923B (en) * | 2019-12-19 | 2022-06-07 | 浙江大华技术股份有限公司 | Echo cancellation method, electronic device, and storage device |
| CN113192527B (en) * | 2021-04-28 | 2024-03-19 | 北京达佳互联信息技术有限公司 | Method, apparatus, electronic device and storage medium for canceling echo |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6570986B1 (en) * | 1999-08-30 | 2003-05-27 | Industrial Technology Research Institute | Double-talk detector |
| CN102377453A (en) * | 2010-08-06 | 2012-03-14 | 联芯科技有限公司 | Method and device for controlling updating of self-adaptive filter and echo canceller |
-
2016
- 2016-11-25 CN CN201611060313.1A patent/CN106533500B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6570986B1 (en) * | 1999-08-30 | 2003-05-27 | Industrial Technology Research Institute | Double-talk detector |
| CN102377453A (en) * | 2010-08-06 | 2012-03-14 | 联芯科技有限公司 | Method and device for controlling updating of self-adaptive filter and echo canceller |
Non-Patent Citations (2)
| Title |
|---|
| 《FrequencyDomainAdaptiveFilterusingFFTAlgorithmforAcousticEchoCancellation》;K.G.Gunale等;《IEEEcomputersociety》;20101231;第582-587页 * |
| 《基于LMS算法的自适应滤波器在声学回声消除中的应用》;倪骁宁;《计算机时代》;20111030;第42-45页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106533500A (en) | 2017-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106533500B (en) | A method of optimization Echo Canceller convergence property | |
| CN110838300B (en) | Echo cancellation processing method and processing system | |
| US8498423B2 (en) | Device for and a method of processing audio signals | |
| US8594320B2 (en) | Hybrid echo and noise suppression method and device in a multi-channel audio signal | |
| US7693291B2 (en) | Multi-channel frequency-domain adaptive filter method and apparatus | |
| US8934620B2 (en) | Acoustic echo cancellation for high noise and excessive double talk | |
| WO2009130513A1 (en) | Two microphone noise reduction system | |
| CN103369162B (en) | A kind of listener's echo self adaptive elimination method of low complex degree | |
| CN110191245B (en) | An Adaptive Echo Cancellation Method Based on Time-varying Parameters | |
| CN110572525B (en) | An adaptive communication echo cancellation method for voice communication | |
| CN104601837A (en) | Robust convex combination type adaptive phone echo canceling method | |
| Van Vaerenbergh et al. | Kernel-based identification of Hammerstein systems for nonlinear acoustic echo-cancellation | |
| EP2930917A1 (en) | Method and apparatus for updating filter coefficients of an adaptive echo canceller | |
| CN113409806A (en) | Zero attraction echo cancellation method based on arc tangent function | |
| US6694020B1 (en) | Frequency domain stereophonic acoustic echo canceller utilizing non-linear transformations | |
| JP2006067127A (en) | Method and apparatus of reducing reverberation | |
| CN117238306A (en) | Voice activity detection and ambient noise elimination method based on double microphones | |
| EP1055292B1 (en) | Echo canceller | |
| US8144863B2 (en) | Method and apparatus for echo cancellation | |
| CN106170049A (en) | A kind of normalization sub-band adaptive echo cancel method having offset compensation | |
| CN110767245B (en) | Voice communication self-adaptive echo cancellation method based on S-shaped function | |
| JP4041770B2 (en) | Acoustic echo cancellation method, apparatus, program, and recording medium | |
| JP3616341B2 (en) | Multi-channel echo cancellation method, apparatus thereof, program thereof, and recording medium | |
| CN113824846A (en) | Howling suppression method and equipment for trunking communication system | |
| The et al. | Spectral Mask-Based Technique for Improving Generalized Sidelobe Canceller Beamformer's Evaluation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |