A numerically stable echo cancellation method and system for telephone communication
By constructing input vectors and calculating iterative forgetting factors, the weight vector of the echo canceller is updated in real time, which solves the problem of slow convergence speed of anti-pulse filters when related signals, and realizes efficient echo cancellation and anti-pulse performance improvement in telephone communications .
Patent Information
- Application Number
- CN202411035447.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-31
AI Technical Summary
The existing anti-pulse filters converge significantly slower when the input signal is a related signal, making it difficult to effectively handle the echo cancellation task in telephone communication.
By obtaining the input signal and the expected signal sample value, constructing the input vector and generating error signals, using nonlinear functions and variance estimation values to calculate the iterative forgetting factor, adjusting the sparse punishment intensity factor, updating the weight vector of the echo canceller in real time, and improving the algorithm's Robustness and pulse resistance.
The numerical stability and recognition performance of the echo cancellation algorithm are improved, the processing capability of related signals is enhanced, and rapid convergence and good pulse resistance are achieved.
Smart Images

Figure CN118573778B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of telephone communication, and in particular to a numerically stable echo elimination method and system for telephone communication. Background Art
[0002] In telephone communications, due to the impedance mismatch at each point in the transmission medium, echo is generated through electrical signals. This echo is called line or network echo. Echo cancellation is often implemented using an adaptive algorithm. At the same time, the echo channel is sparse, that is, most of the elements in the weight vector of the unknown system are zero or close to zero. Zero-point attraction theory and proportional adaptive strategy are often used to solve the identification problem of sparse systems. At present, research shows that the use of proportional updates and low-order norm regularization of weight vectors can effectively improve the convergence speed of adaptive filters.
[0003] In the general theoretical research of adaptive algorithms, it is assumed that a model with infinite word length (or infinite precision) is used to generate analog sampling values of input data and perform internal operations. However, when the adaptive filtering algorithm is implemented digitally in practice, both the input data and iterative data must be quantized to a preferred precision, and the quantization process will affect the performance of the digital implementation of the algorithm. At present, research shows that the use of QR decomposition with good numerical characteristics can solve the problem of numerical instability of the RLS algorithm.
[0004] In addition, in some special environments, the output signal of the unknown system may be contaminated by impulse noise. Therefore, in order to enhance the anti-impulse interference capability of the adaptive filter, a series of anti-impulse filters have been proposed, such as the error sign minimum mean square filter, the mixed norm filter, the filter based on maximum correlation entropy, etc. These filters have a good convergence effect when the input signal is a white signal, but when the input signal is a correlated signal, the convergence speed is significantly slowed down.
[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention
[0006] Therefore, the technical problem to be solved by the present invention is to overcome the prior art in order to enhance the anti-pulse interference capability of the adaptive filter, a series of anti-pulse filters are proposed, such as error sign minimum mean square filter, mixed norm filter, filter based on maximum correlation entropy, etc. These filters have a good convergence effect when the input signal is a white signal, but when the input signal is a correlated signal, the convergence speed is significantly slowed down.
[0007] In order to solve the above technical problems, the first aspect of the present invention provides a numerically stable echo cancellation method for telephone communication, the method comprising:
[0008] Obtaining input signal sampling values and expected signal sampling values;
[0009] Constructing an input vector according to the input signal sampling values;
[0010] generating an error signal according to the input vector and the expected signal sample value;
[0011] generating a nonlinear function and a variance estimate of the error signal based on the error signal;
[0012] Generate a measurement noise variance estimate value according to the expected signal sampling value and the variance estimate value;
[0013] generating an iterative forgetting factor value according to the measurement noise variance estimate and the variance estimate;
[0014] Iteratively generate an iterative sparse penalty strength factor value for the rear array block elements according to the iterative forgetting factor value, the nonlinear function, the input vector and the expected signal sampling value;
[0015] The weight vector of the echo canceller is updated according to the rear array block elements, the iterative sparse penalty strength factor value and the sparse penalty function.
[0016] In one embodiment of the present invention, the input vector is expressed as:
[0017] ;
[0018] in, for Time and its previous continuous The input signal sampling value at the moment, Indicates the weight vector length of the echo canceller, the superscript Represents a transpose operation.
[0019] In one embodiment of the present invention, the error signal is expressed as:
[0020] ;
[0021] in, Reason Time Echo Canceller tap weights The weight vector formed.
[0022] In one embodiment of the present invention, the formula for updating the weight vector of the echo canceller is:
[0023] ;
[0024] in, is the iterative sparse penalty strength factor value at time n, is a square-root lower triangular matrix defined by the robust weighted autocorrelation matrix of the input vector, , is the cross-correlation vector between the input vector and the expected signal observation value, is the sparse penalty function, is a symbolic function in vector form.
[0025] In one embodiment of the present invention, after iterating, the array block elements and The formula is:
[0026] ;
[0027] in, is the iterative forgetting factor value, represents the unitary rotation, for The expected signal sampling value at time, It is a nonlinear function The square root of The value of Calculation, parameters is the shape parameter that controls the nonlinear function and satisfies , is a regularization parameter used to overcome numerical computation difficulties. , and are the corresponding position elements of the rear array, is an all-zero vector.
[0028] In one embodiment of the present invention, the formula for iterative forgetting factor value is:
[0029] ;
[0030] in, For values between The upper bound of the forgetting factor is for Variable value at time The variance estimate of The square root of The value of calculate, , For values between The smoothing factor, for Timing Error Signal The variance estimate of The square root of calculate, for Measuring noise at all times Variance estimate The square root of is a smaller regularization parameter, For values between The adjustment parameters between .
[0031] In one embodiment of the present invention, the noise is measured The formula for estimating the variance of is:
[0032] ;
[0033] in, for The estimated variance of the expected signal sampling value at time is calculated based on calculate, for The estimated variance of the output signal at time , whose value is based on calculate.
[0034] In one embodiment of the present invention, the formula for the iterative sparse penalty strength factor value is as follows:
[0035] ;
[0036] in, Indicates the maximum value operation.
[0037] A second aspect of the present invention provides a numerically stable echo cancellation system for telephone communication, the system comprising: a vector construction module, an estimation value calculation module, a calculation module, and an update module;
[0038] The vector construction module is configured to: obtain input signal sampling values and expected signal sampling values; construct an input vector according to the input signal sampling values;
[0039] The estimated value calculation module is configured to: generate an error signal according to the input vector and the expected signal sampling value; generate a nonlinear function and a variance estimate of the error signal according to the error signal; generate a measurement noise variance estimate according to the expected signal sampling value and the variance estimate;
[0040] The calculation module is configured to: generate an iterative forgetting factor value according to the measurement noise variance estimate and the variance estimate; iteratively generate an iterative sparse penalty strength factor value for the rear array block elements according to the iterative forgetting factor value, the nonlinear function, the input vector and the expected signal sampling value;
[0041] The updating module is configured to update a weight vector of an echo canceller according to the rear array block elements, the iterative sparse penalty strength factor value and a sparse penalty function.
[0042] The above technical solution of the present invention has the following advantages compared with the prior art:
[0043] The numerically stable echo elimination method and system for telephone communication described in the present invention improve the robustness of the numerically stable algorithm through a nonlinear function, and at the same time adjust and update the filter iteration coefficient penalty intensity factor value and the coefficient penalty intensity factor value in real time, thereby further improving the recognition performance, and having good anti-pulse performance while accelerating the recognition speed. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings.
[0045] Figure 1 It is a flow chart of a numerically stable echo elimination method for telephone communication provided by the present invention;
[0046] Figure 2 It is a numerically stable echo cancellation method for telephone communication provided by the present invention and a normalized mean square deviation curve diagram of an adaptive filter when the impulse noise Pr=0.01 in the system;
[0047] Figure 3 It is a numerically stable echo cancellation method for telephone communication provided by the present invention and a normalized mean square deviation curve diagram of an adaptive filter when the impulse noise Pr=0.1 in the system;
[0048] Figure 4 It is an architecture diagram of a numerically stable echo cancellation system for telephone communication provided by the present invention. DETAILED DESCRIPTION
[0049] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.
[0050] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.
[0051] Reference Figure 1 As shown, the present invention provides a numerically stable echo cancellation method for telephone communication, the method comprising:
[0052] S100, obtaining input signal sampling values and expected signal sampling values;
[0053] In step S100, the input signal from the remote microphone is sampled to obtain the current time n and the previous continuous The input signal sampling value at the moment , sample the near-end microphone and obtain the expected signal sampling value at the current time n with echo .
[0054] S200, constructing an input vector according to the input signal sampling value;
[0055] In step S200, the input vector is represented as:
[0056] (1);
[0057] in, for Time and its previous continuous The input signal sampling value at the moment, Indicates the weight vector length of the echo canceller, the superscript Represents a transpose operation.
[0058] In practical application scenarios, the nth moment and its previous continuous The input signal sampling value at the moment To construct the input vector , where the superscript T represents the transpose operation.
[0059] S300, generating an error signal according to the input vector and the expected signal sampling value;
[0060] In step S300, the error signal is expressed as:
[0061] (2);
[0062] in, Reason Time Echo Canceller tap weights The weight vector formed.
[0063] In actual application scenarios, the input signal and the expected signal at time n are used to calculate the error signal vector through equation (2), and then the error signal vector at the current time n is sent back to the remote microphone.
[0064] S400, generating a nonlinear function and a variance estimate of the error signal according to the error signal;
[0065] In step S400, a nonlinear function is generated according to the error signal and is expressed as:
[0066] (3);
[0067] Among them, the parameters is the shape parameter that controls the nonlinear function and satisfies , is a regularization parameter used to overcome numerical computation difficulties.
[0068] The formula for generating an error signal variance estimate based on the error signal is as follows:
[0069] (4);
[0070] in, For values between The smoothing factor is calculated according to formula (3): Nonlinear function , which is used to improve the robustness of the algorithm, and use the robust weighted error signal to estimate the variance of the error signal vector. The estimation formula is shown in formula (4).
[0071] S500, generating a measurement noise variance estimate value according to the expected signal sampling value and the variance estimate value;
[0072] In step S500, the noise is measured The formula for estimating the variance of is:
[0073] (5);
[0074] in, for The estimated variance of the expected signal sampling value at time is calculated based on calculate, for The estimated variance of the output signal at time , whose value is based on calculate.
[0075] In actual application scenarios, the variance estimate of the expected signal sampling value and the error signal is substituted into formula (5) to obtain the variance estimate of the measurement noise.
[0076] S600, generating an iterative forgetting factor value according to the measurement noise variance estimate and the variance estimate;
[0077] In step S600, the formula for iterating the forgetting factor value is:
[0078] (6);
[0079] in, For values between The upper bound of the forgetting factor is for Variable value at time The variance estimate of The square root of calculate, , For values between The smoothing factor, for Timing Error Signal The variance estimate of The square root of calculate, for Measuring noise at all times Variance estimate The square root of is a smaller regularization parameter, For values between The adjustment parameters between .
[0080] In actual application scenarios, the estimated value of the measurement noise variance and the estimated value of the error signal variance are substituted into formula (6) to obtain the iterative forgetting factor value.
[0081] S700, iteratively generating an iterative sparse penalty strength factor value for the rear array block elements according to the iterative forgetting factor value, the nonlinear function, the input vector and the expected signal sampling value;
[0082] In step S700, the array block elements are iterated. and The formula is:
[0083] (7);
[0084] in, is the iterative forgetting factor value, represents the unitary rotation, for The expected signal sampling value at time, It is a nonlinear function The square root of The value of Calculation, parameters is the shape parameter that controls the nonlinear function and satisfies , is a regularization parameter used to overcome numerical computation difficulties. , and are the corresponding position elements of the rear array, is an all-zero vector. The formula for the iterative sparse penalty strength factor value is as follows:
[0085] (8);
[0086] in, Indicates the maximum value operation.
[0087] In practical application scenarios, the iterative forgetting factor value, nonlinear function, input vector and expected signal sampling value are substituted into formula (7) to obtain the array block element after iteration. The sparse intensity penalty strength factor is iterated based on formula (8) to obtain the sparse intensity penalty strength factor .
[0088] S800: Update a weight vector of an echo canceller according to the rear array block elements, the iterative sparse penalty strength factor value and the sparse penalty function.
[0089] In step S800, the formula for updating the weight vector of the echo canceller is:
[0090] (9);
[0091] in, is the iterative sparse penalty strength factor value at time n, is a square-root lower triangular matrix defined by the robust weighted autocorrelation matrix of the input vector, , is the cross-correlation vector between the input vector and the expected signal observation value, is the sparse penalty function, is a symbolic function in vector form.
[0092] In actual application scenarios, the rear array block elements, the iterative sparse penalty strength factor value and the sparse penalty function are used to update the weight vector of the echo canceller through equation (9) to eliminate the echo.
[0093] The input signal in this application is a speech signal with a sampling rate of 8KHz. The weight vector of the unknown echo channel is The length of is 512. The noise signal used contains a zero mean and a variance of Gaussian white noise and an impulse noise ,Right now . Impulse noise Produced by a Bernoulli Gaussian process, that is ,in, is a Bernoulli process, and the probability of taking 1 is Pr, and the probability of taking 0 is 1-Pr, is a zero-mean Gaussian white noise with a variance of 100. In order to make the result more general, the impulse noise in this embodiment is set to Pr=0.01 and Pr=0.1 respectively.
[0094] The normalized mean square deviation (NMSD) is used as the performance measure, i.e. , the unit is ,in The NMSD curves obtained by simulation are obtained by taking the average value of 200 independent iterations. Figure 2 and Figure 3 As shown, it is the simulation result of this application. Figure 2 For Pr = 0.01, Figure 3 Pr = 0.1. The VSS-PAPTA echo generator in the embodiment of the present application has good anti-pulse performance, and has the fastest convergence speed and the lowest steady-state imbalance in echo cancellation of telephone communication.
[0095] Second, refer to Figure 4 As shown, the present application provides a numerically stable echo cancellation system for telephone communication, the system comprising: a vector construction module 100, an estimation value calculation module 200, a calculation module 300, and an update module 400;
[0096] The vector construction module 100 is configured to: obtain input signal sampling values and expected signal sampling values; construct an input vector according to the input signal sampling values;
[0097] The estimated value calculation module 200 is configured to: generate an error signal according to the input vector and the expected signal sampling value; generate a nonlinear function and a variance estimate of the error signal according to the error signal; generate a measurement noise variance estimate according to the expected signal sampling value and the variance estimate;
[0098] The calculation module 300 is configured to: generate an iterative forgetting factor value according to the measurement noise variance estimate and the variance estimate; iteratively generate an iterative sparse penalty strength factor value for the rear array block elements according to the iterative forgetting factor value, the nonlinear function, the input vector and the expected signal sample value;
[0099] The updating module 400 is configured to update the weight vector of the echo canceller according to the rear array block elements, the iterative sparse penalty strength factor value and the sparse penalty function.
[0100] The effects of the above-mentioned system when applying the above-mentioned method can be found in the description of the above-mentioned method embodiment, which will not be repeated here.
[0101] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
[0102] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0103] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0104] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0105] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.
Claims
1. A numerically stable echo cancellation method for telephone communications, It is characterized in that The method comprises: Obtaining input signal sampling values and expected signal sampling values; Constructing an input vector according to the input signal sampling values; generating an error signal according to the input vector and the expected signal sample value; generating a nonlinear function and a variance estimate of the error signal based on the error signal; Generate a measurement noise variance estimate value according to the expected signal sampling value and the variance estimate value; generating an iterative forgetting factor value according to the measurement noise variance estimate and the variance estimate; Iteratively generate an iterative sparse penalty strength factor value for the rear array block elements according to the iterative forgetting factor value, the nonlinear function, the input vector and the expected signal sampling value; updating a weight vector of an echo canceller according to the rear array block elements, the iterative sparse penalty strength factor value and the sparse penalty function; Wherein, the input vector is expressed as: x n =[x n ,x n-1 ,...,x n-M+1 ] T ; Among them, x n ,x n-1 ,...,x n-M+1 is the input signal sampling value at time n and the M-1 consecutive times before it, M represents the weight vector length of the echo canceller, and the superscript T represents the transposition operation; The error signal is expressed as: Among them, w n-1 =[w n-1,0 ,w n-1,1 ,...,w n-1,M-1 ] T is the M tap weights w of the echo canceller at time n-1 n-1,0 ,w n-1,1 ,...,w n-1,M-1 The weight vector formed; The formula for updating the weight vector of the echo canceller is: Among them, ρ n is the iterative sparse penalty strength factor value at time n, is a square-root lower triangular matrix defined by the robust weighted autocorrelation matrix of the input vector, z n is the cross-correlation vector between the input vector and the expected signal observation value, γ′(w n-1 )=sgn(w n-1 ) is the sparse penalty function, sgn(·) is the sign function in vector form; Iterate over the array block elements and r n The formula is: Among them, λ n is the iterative forgetting factor value, Θ(n) represents the unitary rotation amount, d n is the expected signal sampling value at time n, is a nonlinear function f(e n ), f(e n ) value according to Calculation, parameter p is the shape parameter that controls the nonlinear function and satisfies -2≤p<2, τ is the regularization parameter used to overcome the difficulty of numerical calculation, b 22 (n) and b 32 (n) are the elements at the corresponding positions of the rear array, and 0 is the all-zero vector; The formula for iterative forgetting factor value is: Among them, λ max is the upper bound of the forgetting factor between [0.999, 0.9999], is the variable value k at time n n The variance estimate of The square root of calculate, β is a smoothing factor with a value between [0.9,1), is the error signal e at time n n The variance estimate of The square root of calculate, The noise v is measured at time n n Variance estimate The square root of , ∈ is the regularization parameter, and θ is the adjustment parameter with a value between [1,2]; Measurement noise v n The formula for estimating the variance of is: in, is the estimated variance of the expected signal sampling value at time n, and its value is based on calculate, is the estimated value of the output signal variance at time n, and its value is based on calculate; The formula for the iterative sparse penalty strength factor value is as follows: Among them, max means the maximum value operation.
2. A numerically stable echo cancellation system for telephone communication, configured to execute the numerically stable echo cancellation method for telephone communication as claimed in claim 1, It is characterized in that The system comprises: a vector construction module, an estimated value calculation module, a calculation module, and an update module; The vector construction module is configured to: obtain input signal sampling values and expected signal sampling values; construct an input vector according to the input signal sampling values; The estimated value calculation module is configured to: generate an error signal according to the input vector and the expected signal sampling value; generate a nonlinear function and a variance estimate of the error signal according to the error signal; generate a measurement noise variance estimate according to the expected signal sampling value and the variance estimate; The calculation module is configured to: generate an iterative forgetting factor value according to the measurement noise variance estimate and the variance estimate; iteratively generate an iterative sparse penalty strength factor value for the rear array block elements according to the iterative forgetting factor value, the nonlinear function, the input vector and the expected signal sampling value; The updating module is configured to update a weight vector of an echo canceller according to the rear array block elements, the iterative sparse penalty strength factor value and a sparse penalty function.
