Echo Canceller Based on Proportional Population Adaptive Method

Abstract

The present invention discloses an echo canceller based on a proportional overall adaptive method, and belongs to the field of acoustic signal processing. The echo canceller is established by minimizing the overall mean square error cost function constrained by a condition. By introducing a proportional matrix, the echo canceller uses the sparsity of the system to accelerate convergence. The overall cost function enables the echo canceller to obtain a lower steady-state misalignment when both the input and output signals are interfered by noise, thereby having a higher accuracy. The echo canceller disclosed by the present invention can be applied to communication systems such as video conferencing and hands-free phones.

Description

Technical Field

[0001] The invention discloses an echo canceller based on a proportional overall adaptive method, and belongs to the field of acoustic signal processing. Background Art

[0002] In video conferencing, hands-free phones and other communication systems, an echo canceller is required to reduce echoes and improve call quality. The echo canceller generates an approximate signal of the echo by estimating the impulse response of the echo path, and then reduces the echo by subtracting the echo signal from the constructed approximate signal of the echo.

[0003] The impulse response of the echo path is often sparse, that is, most of the elements of the impulse response are close to zero, and a few elements have large values. Traditional echo cancellers based on the least mean square and normalized least mean square methods are easy to implement, but the convergence performance is poor when both the input speech and the echo are interfered by noise. To address this situation, echo cancellers based on the overall adaptive method have been proposed in the literature, but these echo cancellers do not fully utilize the sparse characteristics of the echo path, so the convergence speed is slow. Summary of the invention

[0004] In order to solve the above problems, the present invention proposes an echo canceller based on proportional total adaptive method (hereinafter referred to as PTNLMS). The echo canceller has better performance and faster convergence when both input and output signals are interfered by noise.

[0005] The echo canceller updates the weight vector including the following steps:

[0006] 1) The noisy input signal vector and the weight vector w of the echo canceller n Perform inner product to generate the output signal of adaptive echo canceller in: The noisy input signal sampled at time n and the previous M-1 consecutive time points Composition, w n =[w n,0 ,w n,1 ,...,w n,M-1 ] T is the weight vector composed of the M tap coefficients of the echo canceller at time n, and the superscript T represents the transposition operation;

[0007] 2) According to Calculate the error signal where: represents the noisy expected response of the echo canceller;

[0008] 3) Calculate the intermediate variable p(w n,i )=min(γ|wn,i |,1), where the parameter γ is a positive constant used to control the strength of the estimated sparse system;

[0009] 4) Construct vector p(w) from intermediate variables n )=[p(w n,0 ),p(w n,1 ),...,p(w n,M-1 )] T ;

[0010] 5) According to Calculate the proportional coefficient, where: i is an integer and satisfies 0≤i<M, α is the proportional coefficient control parameter, and its value range is -1≤α≤1, ε is a small positive constant to avoid the denominator being zero, and ||||1 represents the l1 norm of the vector.

[0011] 6) Construct the scale matrix G according to the scale coefficient n =diag(g n,0 ,g n,1 ,...,g n,M-1 ), where: diag means to change the element g n,0 ,g n,1 ,...,g n,M-1 As diagonal elements construct diagonal elements;

[0012] 7) According to the expression Calculate the vector f n ,in: is the augmented weight vector of the echo canceller, is the variance of the noise contained in the desired signal, is the variance of the noise contained in the input signal of the echo canceller;

[0013] 8) Using expressions Update the weight vector w n , where: μ is a positive step size parameter, and δ is a small positive constant to avoid zero denominator.

[0014] Beneficial Effects

[0015] Compared with the existing technical solutions, the echo canceller based on the proportional overall adaptive method proposed in the present application can not only reduce the impact of noise on the input speech signal, but also make full use of the sparse characteristics of the echo path impulse response to accelerate the convergence speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described below in conjunction with the accompanying drawings and embodiments:

[0017] Figure 11 is a schematic diagram of the structure of an echo canceller based on a proportional population adaptive method in an embodiment of the present invention;

[0018] Figure 2 is a speech signal used in the embodiment of the present invention;

[0019] Figure 3 is the impulse response of the echo path used in the embodiment of the present invention;

[0020] Figure 4 Comparison of normalized mean square deviation curves of GD-TLS, RZA-TLS and PTNLMS in the embodiments of the present invention; DETAILED DESCRIPTION

[0021] Example

[0022] The present invention proposes an echo canceller based on a proportional overall adaptive method. The structure principle of the echo canceller is as follows: Figure 1 As shown in Figure 2, the echo canceller has better performance and converges faster when both the input and output signals are interfered by noise.

[0023] This embodiment uses a computer experiment method to verify the performance of the PTNLMS echo canceller. In the experiment, the PTNLMS echo canceller disclosed in the present invention is used to cancel the echo in an environment where both the input and output are interfered by white noise, and its performance is compared with the performance of the echo canceller based on GD-TLS [Analysis of the Gradient-Descent Total Least-Squares Adaptive Filtering Algorithm, 2014, 62 (5): 1256-1264] and RZA-TLS [Reweighted l1regularized TLS linear neuron for the sparse system identification, 2016, 173, 1972-1975]. The PTNLMS echo canceller disclosed in the embodiment of the present application performs echo cancellation including the following steps:

[0024] 1) The noisy input signal vector and the weight vector w of the echo canceller n Perform inner product to generate the output signal of adaptive echo canceller in: The noisy input signal sampled at time n and the previous M-1 consecutive time points Composition, w n =[w n,0 ,w n,1 ,...,w n,M-1 ]T is the weight vector composed of the M tap coefficients of the echo canceller at time n, and the superscript T represents the transposition operation;

[0025] 2) According to Calculate the error signal, where represents the noisy expected response of the echo canceller;

[0026] 3) Calculate the intermediate variable p(w n,i )=min(γw| n,i |,1), where the parameter γ is a positive constant used to control the strength of the estimated sparse system;

[0027] 4) Construct vector p(w) from intermediate variables n )=[p(w n,0 ),p(w n,1 ),...,p(w n,M-1 )] T ;

[0028] 5) According to Calculate the proportional coefficient, where: i is an integer and satisfies 0≤i<M, α is the proportional coefficient control parameter, and its value range is -1≤α≤1, ε is a small positive constant to avoid the denominator being zero, and ||||1 represents the l1 norm of the vector.

[0029] 6) Construct the scale matrix G according to the scale coefficient n =diag(g n,0 ,g n,1 ,...,g n,M-1 ), where: diag means to change the element g n,0 ,g n,1 ,...,g n,M-1 As diagonal elements construct diagonal elements;

[0030] 7) According to the expression Calculate the vector f n ,in: is the augmented weight vector of the echo canceller, is the variance of the noise contained in the desired signal, is the variance of the noise contained in the input signal of the echo canceller;

[0031] 8) Using expressions Update the weight vector w n , where: μ is a positive step size parameter, and δ is a small positive constant to avoid zero denominator.

[0032] In the experiment, the input signal is a speech signal, such as Figure 2 shown.

[0033] The impulse response of the echo path w o like Figure 3 As shown, its length is 512. The normalized mean square deviation (NMSD) is used as the performance measure, defined as NMSD = 10log 10 [||w n -w o || 2 / ||w o || 2 ], in dB. Input noise u n and output noise v n are all zero-mean Gaussian white noise, u n The variance of v n The variance of

[0034] Depend on Figure 4 It can be seen that the echo canceller designed by the present invention (ie, the PTNLMS echo canceller) can obtain the lowest NMSD while achieving the best convergence speed, that is, estimating w o The highest accuracy.

[0035] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with the technology to understand the content of the present invention and implement it accordingly, and they cannot be used to limit the protection scope of the present invention. Any equivalent transformation or modification made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. An echo canceller based on a proportional population adaptive method, characterized in that: The echo canceller is established by using proportional update and Lagrange multiplier method to solve the overall cost function that minimizes the constraints. The echo canceller updates the weight vector including the following steps: 1) The noisy input signal vector and the weight vector w of the echo canceller n Perform inner product to generate the output signal of adaptive echo canceller in: The noisy input signal sampled at time n and the previous M-1 consecutive time points Composition, w n =[w n,0 ,w n,1 ,...,w n,M-1 ] T is the weight vector composed of the M tap coefficients of the echo canceller at time n, and the superscript T represents the transposition operation; 2) According to Calculate the error signal, where represents the noisy expected response of the echo canceller; 3) Calculate the intermediate variable p(w n,i )=min(γ|w n,i |,1), where: the parameter γ is a positive constant used to control the strength of the estimated sparse system; 4) Construct vector p(w) from intermediate variables n )=[p(w n,0 ),p(w n,1 ),...,p(w n,M-1 )] T ; 5) According to calculate the proportionality coefficient, where: i is an integer and satisfies 0 ≤ i < M, α is the proportionality coefficient control parameter, whose value ranges from -1 ≤ α ≤ 1, ε is a small positive constant to avoid a zero denominator, and ||||1 represents the l1 norm of the vector; 6) Construct the scale matrix G according to the scale coefficient n =diag(g n,0 ,g n,1 ,...,g n,M-1 ), where: diag means to change the element g n,0 ,g n,1 ,...,g n,M-1 As diagonal elements construct diagonal elements; 7) According to the expression Calculate the vector f n ,in: is the augmented weight vector of the echo canceller, 8) The echo canceller uses the expression Update the weight vector w n , where μ is a positive step size parameter and δ is a small positive constant that avoids the denominator being zero.

2. The echo canceller based on the proportional population adaptive method according to claim 1, characterized in that: In step 7), in: is the variance of the noise contained in the desired signal, σ u 2 is the variance of the noise contained in the input signal of the echo canceller.

Images