Self-adaptive filtering method based on full-range zero-attraction LMS (Least Mean Square) algorithm

Abstract

The invention relates to an adaptive filtering method based on a full-range zero-attraction LMS algorithm, and belongs to the field of digital signal processing. The method comprises the following steps: establishing an echo canceller system model; a plurality of groups of audio signals are collected through pickup equipment, a signal matrix is constructed, and channel noise is introduced; inputting the audio signal into an adaptive filter for processing and outputting to obtain an echo signal; inputting the audio signal into an echo channel, and superposing an output signal of the echo channel with channel noise to obtain a desired signal; calculating an error between the expected signal and the echo signal, and inputting the error into an adaptive filter to iteratively update a coefficient of the adaptive filter to obtain an optimal filtering parameter; and removing the echo signal generated by the adaptive filter from the expected signal of the obtained real-time voice signal under the optimal filtering parameter to obtain a voice signal after filtering processing. According to the method, the performance and the stability of the echo canceller can be improved, and meanwhile, the parameter selection of the algorithm is more robust.

Description

Technical Field

[0001] The invention belongs to the field of digital signal processing and relates to an adaptive filtering method based on a full-range zero-attraction LMS algorithm. Background Art

[0002] During audio communication, due to the relative position of the pickup device and the speaker, the ambient acoustic characteristics of the room, etc., the audio signal generates echoes through the complex echo channel, and the two parties on the call may hear their own voices reflected. Echoes not only affect the call quality and cause auditory discomfort, but also interfere with the normal communication between the two parties, resulting in confusion and reduced efficiency in information transmission. The adaptive filter can adjust its own parameters in real time according to the statistical characteristics of the input and output signals, so that the filter coefficients match the complex echo channel characteristics to minimize the residual echo.

[0003] In a typical sparse system, most coefficients are zero or close to zero, and only a small number of coefficients are very large. These large coefficients contain most of the key information of the system. Although the ZA-LMS and RZA-LMS algorithms proposed by Y.Chen have certain zero attraction weights for coefficients in the entire coefficient spectrum, ZA-LMS applies the same zero attraction weight to all coefficients, resulting in its performance being very dependent on the zero attraction strength ρ and measurement noise. Although RZA-LMS has the ability to change dynamically, it applies too large zero attraction weights to large coefficients, resulting in a decrease in algorithm performance. The l0-LMS algorithm proposed by Y.Gu and the OZ-LMS algorithm proposed by Luo, L limit the zero attraction range, and only the coefficients within the range will be attracted to zero, but the large coefficients outside the zero attraction range are completely abandoned, resulting in poor performance of the algorithm for dynamically changing systems and environments with large measurement noise, while increasing the algorithm's sensitivity to parameter selection. Summary of the invention

[0004] In view of this, an object of the present invention is to provide an adaptive filtering method based on a full-range zero-attraction LMS algorithm.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] An adaptive filtering method based on a full range zero attraction LMS algorithm, the method comprising the following steps:

[0007] S1. Establishing an echo canceller system model, which at least includes: a sound pickup device, an echo channel, and an adaptive filter;

[0008] S2. Collect several groups of audio signals through a sound pickup device, construct a signal matrix, and introduce channel noise;

[0009] S3, inputting the audio signal into an adaptive filter to process and output an echo signal; inputting the audio signal into an echo channel, and superimposing the output signal of the echo channel with the channel noise to obtain a desired signal;

[0010] S4, calculating the error between the expected signal and the echo signal, and inputting it into the adaptive filter to iteratively update the coefficients of the adaptive filter to obtain the optimal filtering parameters;

[0011] S5. Under the optimal filtering parameters, the echo signal generated by the adaptive filter is removed from the desired signal of the acquired real-time speech signal to obtain a speech signal after filtering.

[0012] Further, in step S1, the sound pickup device is used to obtain an audio signal and generate an echo signal, and the adaptive filter is used to cancel the echo; the audio signal obtained by the sound pickup device can be transmitted to the echo channel and the adaptive filter for processing respectively, and the length of the echo channel and the adaptive filter are both L.

[0013] Further, in step S2, the sound pickup device obtains the input signal power as The audio signal x(n) is combined into a signal matrix X n =[x(n)x(n-1)…x(n-L+1)] T , where n is the serial number of the audio data; the introduced channel noise n0(n) is zero-mean Gaussian white noise, and its power is

[0014] Further, in step S3, the echo signal is calculated as follows:

[0015] y(n)=W T (n)X(n)

[0016] Where W T (n) represents the transpose of the coefficient W(n) of the adaptive filter, and X(n) is the audio signal matrix.

[0017] Further, in step S3, the expected signal is calculated as follows:

[0018] d(n)=H(n) T X(n))+n0(n)

[0019] In the formula, H(n) T represents the transpose of the echo channel coefficient n(n), and n0(n) is the introduced channel noise.

[0020] Further, in step S4, the error between the desired signal d(n) and the echo signal y(n) is calculated:

[0021] e(n)=d(n)-y(n)

[0022] The error e(n) is input into the adaptive filter to iteratively update the coefficients in the full-range zero-attraction LMS algorithm of the adaptive filter. When the coefficients W(n) of the adaptive filter match the echo channel H(n), the iteration process stops. The coefficient iteration formula of the full-range zero-attraction LMS algorithm is expressed as:

[0023] W(n+1)=W(n)+μe(n)X(n)-f(W(n))

[0024]

[0025] Where W(n) is the filter coefficient; parameter μ is the step size of the filter coefficient iteration, and its range is The parameter ρ is the zero attraction factor, which is used to adjust the zero attraction strength of the zero attraction function; f(W(n)) is the zero attraction function, which applies zero attraction weight to the coefficient; the parameter β is the zero attraction range, which is used to specify the coefficient boundary attracted by the zero attraction function; the parameter α is a constant, which is used to adjust the zero attraction strength imposed by the zero attraction function on the coefficient outside the zero attraction range; sgn(W(n)) is the sign function, and the corresponding expression is:

[0026]

[0027] The sign function sgn(W(n)) is used to confirm the range of the current coefficient and adjust the zero attraction strength of the coefficient W(n).

[0028] Furthermore, before iteratively updating the coefficients of the full-range zero-attraction LMS algorithm in the adaptive filter, the parameters in the algorithm are initialized according to the system parameters and the target steady-state MSD. First, the step size μ is set according to the system parameters and the target steady-state MSD initialization algorithm parameters to satisfy Parameter a and zero attraction strength ρ, and then continuously adjust the step size μ and the value of zero attraction strength ρ according to the actual effect;

[0029] Among them, the target steady-state MSD calculation method is expressed as:

[0030]

[0031] In the formula, P is the number of coefficients outside the zero attraction range.

[0032] The beneficial effects of the present invention are:

[0033] The present invention extends the zero attraction weight to the entire coefficient spectrum, and adopts dynamically scaled zero attraction weights for smaller coefficients within the zero attraction range to promote rapid convergence without causing overfitting; for larger coefficients beyond the range, the present invention implements decreasing zero attraction weights, thereby improving the convergence speed and the robustness of the algorithm to parameter adjustment. The full-range zero attraction LMS algorithm proposed by the present invention is conducive to improving the performance and stability of the echo canceller, and is more robust to the parameter selection of the algorithm.

[0034] The zero attraction function applies a larger zero attraction weight to small coefficients within the zero attraction range, and applies a zero attraction weight that gradually decreases as the coefficient increases to large coefficients outside the zero attraction range, thereby preserving the characteristics of the system. Compared with other algorithms, the proposed algorithm not only has a lower MSD and faster convergence speed in simulation, but also shows good robustness in parameter selection and measurement noise power. In the application scenarios of dynamic system tracking and echo cancellation, the algorithm also shows better performance.

[0035] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:

[0037] Figure 1 It is a structural diagram of the basic system identification model based on adaptive filter;

[0038] Figure 2 It is a structural schematic diagram of the echo canceller system model of the present invention;

[0039] Figure 3 It is the zero attraction function image of the full range zero attraction LMS algorithm of the present invention;

[0040] Figure 4 It is a schematic diagram of transient MSD of different adaptive filtering algorithms in a changing system under the embodiment;

[0041] Figure 5 It is a schematic diagram of the simulation result of 8kHz real speech signal input under the embodiment;

[0042] Figure 6 is an echo signal generated by the echo canceller system model of the present invention under the embodiment;

[0043] Figure 7 The echo signal generated by a typical echo channel;

[0044] Figure 8 is the relationship between the echo signal and the estimated signal;

[0045] Fig. 9 is the relationship between the difference between the estimated signal and the echo signal and the measurement noise. DETAILED DESCRIPTION

[0046] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0047] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0048] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0049] See also Figures 1 to 9 , which is an adaptive filtering method based on the full-range zero-attraction LMS algorithm.

[0050] Example

[0051] like Figure 1 The basic model of the application of adaptive filter in system identification is given. Figure 2The model of the echo canceller in practical application is shown. In this model, when the input signal x(n) is captured by the pickup device, a noise signal n(n) is introduced, thereby forming the expected output signal d(n). The expected signal is offset by the output y(n) of the adaptive filter to generate an error signal e(n). This signal is used as the far-end audio to generate an echo through the echo channel H(n). The echo will be captured by the pickup device and cause interference. In addition, the error signal e(n) will also be used as the input of the adaptive filter to iteratively update the filter coefficient W(n). When the coefficient W(n) of the adaptive filter matches the echo channel H(n), the iteration process stops. Under this condition, the output expected signal e(n) will be the near-end speech signal after echo cancellation.

[0052] Figure 3 It is the zero attraction function image of the proposed algorithm.

[0053] In the current conventional scheme, the coefficient iteration formula of the ZA-LMS algorithm is:

[0054] W(n+1)=W(n)+μe(n)X(n)+ρf ZA-LMS (W(n))

[0055] f ZA-LMS (W(n)) = -sgn[W(n)]

[0056] in

[0057]

[0058] The coefficient iteration formula of the l0-LMS algorithm is:

[0059]

[0060]

[0061] The coefficient iteration formula of the OZ-LMS algorithm is:

[0062] W(n+1)=W(n)+μe(n)X(n)+ρf OZ-LMS (W(n))

[0063]

[0064] Where ε is a constant, which is set to 1000 as required.

[0065] The coefficient iteration formula of the present invention is:

[0066] W(n+1)=W(n)+μe(n)X(n)+f(W(n))

[0067]

[0068] Where a is a constant used to adjust the zero attraction strength of large coefficients outside zero attraction, and is generally set to 0.02.

[0069] The prior art mainly deals with zero coefficients and near-zero coefficients within the ZA range, and pays little attention to large coefficients outside the zero attraction range. The present invention proposes a full-range zero attraction LMS algorithm, which uses a simple zero attraction function for small coefficients within the zero attraction range. The function increases as the coefficient increases, and is continuous at zero, which can avoid excessive attraction. For large coefficients outside the zero attraction range, a modified sieve function is used, which gradually decreases as the coefficient increases, and can also have a certain attraction for larger coefficients. This treatment reduces the sensitivity of parameter selection and measurement noise while improving performance and accelerating convergence speed.

[0070] In order to better illustrate the advantages of the present invention, a simulation experiment was carried out on the proposed algorithm under a changing system, and the mean square error (MSD) was used as the performance evaluation criterion, namely

[0071] MSD(n)=tr{G(n)}

[0072] in,

[0073]

[0074] The average power of the simulated input signal is 1, the signal-to-noise ratio SNR is 20dB, the length of the echo channel L is set to 200, and the sparsity of the channel is That is, the number of large coefficients is 10, and the rest are zero coefficients. The step size μ of all algorithms is set to 0.005, and the zero attraction factor ρ is set to: ρ ZA-LMS =4×10 -5 , ρ OZ-LM =4×10 -5 ,ρ FZM-L7MS =4×10 -5 The steady-state MSD of different algorithms under different zero attraction ranges β is shown in Figure 4 As shown in (a), it can be seen from the figure that the proposed algorithm (FZA-LMS) has a larger tolerance range for parameter β and has the best performance. Figure 4 (b) is the steady-state MSD of different algorithms under different power measurement noise. The zero attraction range of the simulation experiment is β, It is set to 300. From the figure, it can be seen that the proposed algorithm has a larger noise tolerance and also has better performance under large noise conditions.

[0075] The average power of the simulated input signal is 1, the signal-to-noise ratio (SNR) is 20 dB, and the length of the echo channel L is set to 200. In the changing system, the sparsity of the channel will change. The sparsity of the channel in the three stages is set to The step size μ of all algorithms is set to 0.005, the zero attraction range β, is set to 300, and the zero attraction factor ρ is set to: ZA- =4×10 -5 , ρ OZ-L =4×10 -5 ,ρ FZA-LNS =4×10 -5 The simulation results are as follows: Figure 5 As shown in the figure, it can be seen that the transient MSD of the proposed algorithm can show faster convergence speed and better performance when the channel is sparse, and has better tracking ability under changing systems.

[0076] The input signal x(n) with a sampling frequency of 8kHz and the audio signal pass through the echo channel H(n) to generate the echo signal as follows: Figure 6 As shown. The channel length is L = 300, and the sparsity A typical echo channel, such as Figure 7 shown. Figure 8 is the echo signal and the estimated echo signal y(n). Fig. 9 Then is the relationship between the estimated echo signal and the difference between the echo signal and the measurement noise. Figure 6-9 It can be seen that the proposed algorithm performs well in matching the echo channel.

[0077] Therefore, it can be understood that the adaptive filter implemented by the method of the present invention improves performance and accelerates convergence speed while reducing the sensitivity of parameter selection and measurement noise, and also performs well in the application of echo canceller.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. An adaptive filtering method based on a full range zero-attraction LMS algorithm, characterized in that: The method comprises the following steps: S1. Establishing an echo canceller system model, which at least includes: a sound pickup device, an echo channel, and an adaptive filter; S2. Collect several groups of audio signals through the sound pickup device, construct a signal matrix, and introduce channel noise; S3, input the audio signal into an adaptive filter to process and output an echo signal; input the audio signal into an echo channel, and superimpose the output signal of the echo channel with the channel noise to obtain a desired signal; S4, calculating the error between the expected signal and the echo signal, and inputting it into the adaptive filter to iteratively update the coefficients of the adaptive filter to obtain the optimal filtering parameters; S5. Under the optimal filtering parameters, the echo signal generated by the adaptive filter is removed from the desired signal of the acquired real-time speech signal to obtain a speech signal after filtering.

2. The adaptive filtering method based on the full range zero attraction LMS algorithm according to claim 1 is characterized in that: In step S1, the sound pickup device is used to obtain audio signals, the echo channel is used to generate echo signals, and the adaptive filter is used to cancel echoes; the audio signals obtained by the sound pickup device can be transmitted to the echo channel and the adaptive filter for processing respectively, and the lengths of the echo channel and the adaptive filter are both L.

3. The adaptive filtering method based on the full range zero attraction LMS algorithm according to claim 2 is characterized in that: In step S2, the sound pickup device obtains the input signal power as The audio signal x(n) is combined into a signal matrix X n =[x(n) x(n-1)…x(n-L+1)] T , where n is the serial number of the audio data; the introduced channel noise n0(n) is zero-mean Gaussian white noise, and its power is 4. The adaptive filtering method based on the full range zero attraction LMS algorithm according to claim 3 is characterized in that: In step S3, the echo signal is calculated as follows: y(n)=W T (n)X(n) Where W T (n) represents the transpose of the coefficient W(n) of the adaptive filter, and X(n) is the audio signal matrix.

5. The adaptive filtering method based on the full range zero attraction LMS algorithm according to claim 4 is characterized in that: In step S3, the expected signal is calculated as follows: d(n)=H(n) T X(n)+n0(n) In the formula, H(n) T represents the transpose of the echo channel coefficient H(n), and n0(n) is the introduced channel noise.

6. The adaptive filtering method based on the full range zero attraction LMS algorithm according to claim 5 is characterized in that: In step S4, the error between the desired signal d(n) and the echo signal y(n) is calculated: e(n)=d(n)-y(n) The error e(n) is input into the adaptive filter to iteratively update the coefficients in the full-range zero-attraction LMS algorithm of the adaptive filter. When the coefficients W(n) of the adaptive filter match the echo channel H(n), the iteration process stops. The coefficient iteration formula of the full-range zero-attraction LMS algorithm is expressed as: W(n+1)=W(n)+μe(n)X(n)-f(W(n)) Where W(n) is the filter coefficient; parameter μ is the step size of the filter coefficient iteration, and its range is The parameter ρ is the zero attraction factor, which is used to adjust the zero attraction strength of the zero attraction function; f(W(n)) is the zero attraction function, which applies zero attraction weight to the coefficient; the parameter β is the zero attraction range, which is used to specify the coefficient boundary attracted by the zero attraction function; the parameter α is a constant, which is used to adjust the zero attraction strength imposed by the zero attraction function on the coefficient outside the zero attraction range; sgn(W(n)) is the sign function, and the corresponding expression is: The sign function sgn(W(n)) is used to confirm the range of the current coefficient and adjust the zero attraction strength of the coefficient W(n).

7. The adaptive filtering method based on the full range zero attraction LMS algorithm according to claim 6 is characterized in that: Before iteratively updating the coefficients of the full-range zero-attraction LMS algorithm in the adaptive filter, the step size μ is set according to the system parameters and the target steady-state MSD initialization algorithm parameters to meet Parameter a and zero attraction strength ρ, and then continuously adjust the step size μ and the value of zero attraction strength ρ according to the actual effect; Among them, the target steady-state MSD calculation method is expressed as: In the formula, P is the number of coefficients outside the zero attraction range.

Images