Active noise control method and system based on dimensionality reduction FxLMS

Abstract

The invention discloses an active noise control method and system based on dimensionality reduction FxLMS. The method comprises the following steps: acquiring a reference signal; obtaining a filtering reference signal based on the reference signal; performing dimension reduction on the filtering reference signal based on singular value decomposition; based on an ANC method of FxLMS, obtaining a dimensionality reduction adaptive filter by using the filtering reference signal after dimensionality reduction; and realizing active noise control based on the dimensionality reduction adaptive filter. The invention provides a dimensionality reduction FxLMS-based adaptive ANC filter, so as to improve the noise reduction level in a reverberation environment.

Description

Technical Field

[0001] The present invention belongs to the field of active noise control, and in particular relates to an active noise control (ANC) method and system based on dimension reduction FxLMS (filtered-x-LMS). Background Art

[0002] Traditional feedforward active noise control systems such as Figure 1 shown.

[0003] The reference microphone is placed near the noise source to receive the reference signal x(t), where t is the time index. p (t) is the transfer function from the noise source to the error microphone, and the secondary path h s (t) is the transfer function from the control speaker to the error microphone. The noise signal transmitted from the noise source signal to the reference microphone is d(t), where d(t) is the convolution of the noise signal and the primary path. The purpose of the adaptive active noise control method is to design the speaker control filter w(t), where x(t)*w(t) is the speaker drive signal. The control speaker plays the speaker drive signal and the noise signal at the error microphone, and the error signal can be expressed as

[0004] e(t)=d(t)-a(t)

[0005] =x(t)*h p (t)-w T (txt),

[0006] Where T represents the transpose operation, w(t)=[w1(t) w2(t) ... w L (t)] T is the speaker control filter vector, x′(t)=[x′(t)x′(t-1)...x′(t-L+1)] T is the filtering reference signal vector, and the filtering reference signal is expressed as * indicates convolution operation, is the secondary path room impulse response h s (t) is estimated.

[0007] Traditional FxLMS adaptive methods will degrade in highly reverberant environments. Summary of the invention

[0008] In view of the deficiencies of the prior art, the present invention proposes an active noise control method and system based on dimension reduction FxLMS, ​​and proposes an adaptive ANC filter based on dimension reduction to improve the noise reduction level in a reverberant environment.

[0009] To achieve the above object, the present invention provides the following solutions:

[0010] An active noise control method based on dimension reduction FxLMS comprises the following steps:

[0011] Acquire a reference signal, wherein the reference signal is a noise source signal received by a reference microphone and a self-noise of the reference microphone;

[0012] Based on the reference signal, obtaining a filtered reference signal;

[0013] Based on singular value decomposition, reducing the dimension of the filtered reference signal;

[0014] Based on the FxLMS ANC method, the dimension-reduced adaptive filter is obtained by using the dimension-reduced filtered reference signal;

[0015] Based on the dimension reduction adaptive filter, active noise control is achieved.

[0016] Preferably, the method for obtaining the reference signal is:

[0017]

[0018] Where x(t) is the reference signal and v(t) is the reference microphone self-noise.

[0019] Preferably, based on the reference signal, a method for obtaining a filtering reference signal is:

[0020]

[0021] in, is the secondary path room impulse response h s (t) is estimated.

[0022] Preferably, the method for reducing the dimension of the filtered reference signal based on singular value decomposition includes:

[0023]

[0024] in, vector is the room impulse response of the secondary path of length L, U T is the transpose of a semi-orthogonal matrix of size L×Q, where Q is the number of training microphones.

[0025] Preferably, based on the FxLMS ANC method, the method of obtaining a reduced-dimensionality adaptive filter using the reduced-dimensionality filtering reference signal includes:

[0026]

[0027] in, The length of is Q, μ>0 is the adaptive step size, and e(t) is the error signal.

[0028] The present invention also provides an active noise control system based on dimension reduction FxLMS, ​​which is used to implement the active noise control method based on dimension reduction FxLMS, ​​including: a signal pickup device, a signal processing device and a signal noise reduction device;

[0029] The signal pickup device comprises a reference microphone and an error microphone, wherein the reference microphone is located near the noise source and is used to pick up a reference signal, and the error microphone is located at a control point and is used to pick up an error signal; based on the reference signal, a filtered reference signal is obtained;

[0030] The signal processing device is used to reduce the dimension of the filter reference signal based on singular value decomposition; based on the ANC method of FxLMS, ​​the dimension-reduced filter reference signal and the error signal are used to obtain a dimension-reduced adaptive filter;

[0031] The signal noise reduction device is used to achieve active noise control based on the dimension reduction adaptive filter.

[0032] Preferably, the system further comprises: an audio playback device;

[0033] The audio playback device is used to convolve the dimension reduction adaptive filter with the reduced-dimensional filtering reference signal to generate a speaker driving signal; the speaker driving signal and the noise signal are superimposed at the error microphone to obtain an error signal.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention proposes an adaptive ANC filter based on dimensionality reduction to improve the noise reduction level in a reverberant environment. The present invention requires placing training microphones around the error microphone. The goal is to reduce the dimensionality of the filtered reference signal. Subsequently, the FxLMS-based ANC method uses the reduced-dimensional filtered reference signal to generate a reduced-dimensional adaptive filter. The reduced-dimensional control filter is then convolved with the reduced-dimensional filtered reference signal to generate a speaker drive signal. Simulation results show that the reduced-dimensional FxLMS method achieves a lower noise reduction level in a reverberant environment compared to the traditional FxLMS-based method. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0037] Figure 1 The block diagram of a single-channel feed-forward ANC system based on the traditional FxLMS method is shown below;

[0038] Figure 2 1 is a block diagram of an ANC system based on a dimension reduction FxLMS method according to an embodiment of the present invention;

[0039] Figure 3 A schematic diagram of a simulation setting of an embodiment of the present invention;

[0040] Figure 4 It is a schematic diagram of the noise reduction performance after the algorithm converges using different numbers of training microphones according to an embodiment of the present invention.

[0041] Figure 5 Schematic diagram of noise reduction performance after algorithm convergence under different signal-to-noise ratios (SNR) of an embodiment of the present invention;

[0042] Figure 6 It is a schematic diagram of the noise reduction performance after the algorithm converges under different reverberation times according to an embodiment of the present invention;

[0043] Figure 7 4 is a block diagram of an active noise control system based on dimension reduction FxLMS according to an embodiment of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] Embodiment 1

[0047] The present invention provides an active noise control method based on dimension reduction FxLMS, ​​comprising the following steps:

[0048] Acquire a reference signal, wherein the reference signal is a noise source signal received by a reference microphone and a self-noise of the reference microphone;

[0049] Based on the reference signal, obtaining a filtered reference signal;

[0050] Based on singular value decomposition, reducing the dimension of the filtered reference signal;

[0051] Based on the FxLMS ANC method, the dimension-reduced adaptive filter is obtained by using the dimension-reduced filtered reference signal;

[0052] Based on the dimension reduction adaptive filter, active noise control is achieved.

[0053] In this embodiment, the present invention considers that the signal received by the reference microphone may have microphone self-noise in practice, and assumes that the noise is additive noise. In this case, the noisy signal received by the reference microphone is Where x(t) is the reference signal and v(t) is the reference microphone self-noise. The filtered reference signal becomes in, is the secondary path room impulse response h s In order to achieve better noise reduction effect at the control microphone, the secondary path room impulse response information is introduced here to implement FxLMS to achieve adaptive active noise control. The filter update formula is:

[0054]

[0055] Where μ>0 is the adaptive step size. In an environment with a long reverberation time, if you want to get better noise control effect, the secondary path room impulse response estimation The length of must be long enough, and the length of the filter w(t) must also be long enough.

[0056] In this embodiment, a signal dimension reduction method based on singular value decomposition is applied to reduce the dimension of the FxLMS-based active noise control filter.

[0057] Consider placing Q training microphones around the error microphone, where Q << L. The room impulse response vector of length L from the control speaker to the qth training microphone is

[0058] h q =[h q,1 h q,2 ... h q,L ] T ,q=1,2,...,Q

[0059] The following matrix can be formed

[0060] H=[h1 h2 ... h Q ]

[0061] The size of matrix H is L×Q. Assume that the column rank of H is Q. By performing singular value decomposition on matrix H, we get

[0062] H=UΣV T

[0063] Where U is a semi-orthogonal matrix of size L×Q, V is a matrix of dimension R×R, whose column vectors are the right singular vectors of H, and Σ is a diagonal matrix of size Q×Q, whose diagonal elements are the singular values ​​of H. Using singular value decomposition, the reduced-dimensional reference signal is obtained as follows:

[0064]

[0065] in

[0066]

[0067] vector is the room impulse response of the secondary path of length L, U T is the transpose of a semi-orthogonal matrix of size L×Q, where Q is the number of training microphones. It can be seen that the filtered reference signal after dimensionality reduction The length Q is significantly lower than the length L of the original filtered reference signal.

[0068] In this embodiment, by applying the reduced-dimensional filtered reference signal, the length of the FxLMS-based ANC filter can be significantly reduced. The error signal can be written as:

[0069]

[0070] in is the control filter after dimensionality reduction. By solving the following optimization problem:

[0071]

[0072] The optimal dimension reduction control filter can be obtained as

[0073]

[0074] in yes The covariance matrix of for and the cross-correlation of d(t).

[0075] The matrix inversion in the above formula is computationally intensive, which increases the ANC delay and prevents the control filter from adapting to rapidly changing noise in practice. Therefore, an adaptive method is used to reduce the computational complexity of the algorithm. The adaptive method is implemented by minimizing the instantaneous square error, and the optimization method problem is simplified to:

[0076]

[0077] Apply the steepest descent method to solve the above equation, and get the update formula of the adaptive filter:

[0078]

[0079] in The length of the filter is Q, which is significantly shorter than the filter length of the traditional FxLMS. Figure 2 Figure 3 is a schematic diagram of the dimensionality reduction FxLMS method.

[0080] Embodiment 2

[0081] This embodiment demonstrates the technical effect of the solution through an application example. Consider a room with a size of 6m×5m×3m. Figure 3 The simulation setup is shown. There is a sound source in the room that generates a noise signal, which is a band-limited white Gaussian noise between 600Hz and 800Hz. A control speaker generates an anti-noise signal to cancel the noise signal propagating to the error microphone. In order to implement the dimension reduction FxLMS method, multiple training microphones are placed in a square area around the error microphone position. Specifically, the training microphones are evenly distributed in a square area with vertices at (4.5m, 2.25m, 1.5m), (4.5m, 2.75m, 1.5m), (4m, 2.25m, 1.5m) and (4.5m, 2.25m, 1.5m). The error microphone is not included in the training microphones. The speaker and the microphone are located in the same plane and are 1.5m above the ground. The room impulse responses in the example are all generated using the mirror model method. We set the speed of sound in the air to 343m / s and the sampling frequency to 8000Hz. Assume that the reference signal is contaminated by the microphone self-noise, that is Here, the self-noise is assumed to be Gaussian. We use the signal-to-noise ratio (SNR) to control the self-noise level. Here, the noise reduction level at the control point after the adaptive method converges is used as the performance metric of the active noise control. This indicator quantifies the effectiveness of the active noise controller and is defined as the ratio of the signal power observed at the control point when the active noise controller is turned on to the noise signal power recorded at the control point when the active noise controller is turned off. We abbreviate the dimension reduction FxLMS method as DR-FxLMS.

[0082] Example 1

[0083] This example studies the performance of the proposed reduced-dimensional ANC method when using different numbers of training microphones. 60is set to about 220ms. We set the signal-to-noise ratio (SNR) at the reference microphone to 20dB. For comparison, we also evaluate the performance of the conventional FxLMS-based ANC method. The filter length of the conventional FxLMS filter is set to L=1600. The step size of the conventional FxLMS method is set to μ=1.28. The step sizes of the reduced-dimensional ANC method are shown in Table 1. All step sizes are selected to obtain the lowest steady-state noise reduction level under different simulation conditions. The steady-state performance of the adaptive ANC method is shown in Table 1. Figure 4 As shown in Figure 2, the steady-state performance of the proposed dimensionality reduction ANC method becomes better with the increase of Q. When Q is greater than 441, the steady-state performance of the proposed method exceeds that of the FxLMS-based method. After reaching Q = 676, the increasing trend of the noise reduction level becomes less obvious.

[0084] Table 1 Step size of dimensionality reduction FxLMS method under different Q values

[0085]

[0086] Example 2

[0087] In the second set of examples, we evaluate the performance of the proposed dimensionality reduction method under different levels of added noise at the reference microphone. We set T 60 ≈220ms. The filter length of the traditional FxLMS filter is set to L=1600. For the dimension reduction method, Q=676 is set. The results are as follows Figure 5 As shown in Table 2, the noise reduction level of both methods decreases with the increase of SNR. Under all simulation conditions, the dimensionality reduction-based method has better noise control performance than the traditional FxLMS method. The step size under different SNR conditions is shown in Table 2.

[0088] Table 2 Step size of FxLMS method and reduced dimension FxLMS method under different SNRs

[0089]

[0090] Example 3

[0091] The last set of examples evaluates the performance under different reverberation times. For the dimensionality reduction method, we set Q = 676. The filter lengths of the traditional FxLMS method under different reverberation conditions are shown in Table 3. The performance is shown in Figure 6 As shown in Table 4, the performance of both the traditional FxLMS and the proposed method decreases with the increase of reverberation time. The proposed dimensionality reduction method outperforms the traditional FxLMS method under all reverberation conditions. The adaptive filter step size under different reverberation times is experimentally determined to minimize the steady-state noise reduction level, as shown in Table 4.

[0092] Table 3 Different T 60Step size for the lower FxLMS method and the reduced dimension FxLMS method

[0093]

[0094] Table 4 Different T 60 Filter lengths for the lower FxLMS method and the reduced dimension FxLMS method

[0095]

[0096] Embodiment 3

[0097] like Figure 7 As shown, the present invention also provides an active noise control system based on dimension reduction FxLMS, ​​which is used to implement the active noise control method based on dimension reduction FxLMS, ​​including: a signal pickup device, a signal processing device and a signal noise reduction device;

[0098] The signal pickup device comprises a reference microphone and an error microphone, wherein the reference microphone is located near the noise source and is used to pick up a reference signal, and the error microphone is located at a control point and is used to pick up an error signal; based on the reference signal, a filtered reference signal is obtained;

[0099] The signal processing device is used to reduce the dimension of the filter reference signal based on singular value decomposition; based on the ANC method of FxLMS, ​​the dimension-reduced filter reference signal and the error signal are used to obtain a dimension-reduced adaptive filter;

[0100] The signal noise reduction device is used to realize active noise control based on the dimension reduction adaptive filter.

[0101] In this embodiment, the system further includes: an audio playback device;

[0102] The audio playback device is used to convolve the shortened adaptive filter with the reduced-dimensional filtering reference signal to generate a loudspeaker drive signal; the loudspeaker drive signal and the noise signal are superimposed at the error microphone to obtain an error signal.

[0103] In this embodiment, the signal pickup device further includes: a training microphone and an analog-to-digital converter; the signal processing device includes: an adaptive dimension reduction FxLMS filter; and the audio playback device includes: a digital-to-analog converter and a control speaker.

[0104] Specifically, consider the case where the signal received by the reference microphone in practice has microphone self-noise, and assume that the noise is additive noise. In this case, the noisy signal received by the reference microphone is Where x(t) is the reference signal and t is the time index. p (t) is the transfer function from the noise source to the error microphone, and the secondary path hs (t) is the transfer function from the control speaker to the error microphone. The filtered reference signal becomes in, is the secondary path room impulse response h s In order to achieve better noise reduction effect at the control microphone, the secondary path room impulse response information is introduced to implement FxLMS to achieve adaptive active noise control, and its filter update formula is:

[0105]

[0106] Where μ>0 is the adaptive step size. In an environment with a long reverberation time, if you want to get a better noise control effect, the length of the secondary path room impulse response estimate must be long enough, and the length of the filter w(t) must also be long enough.

[0107] A signal dimensionality reduction method based on singular value decomposition is applied to reduce the dimensionality of the FxLMS-based active noise control filter.

[0108] Consider placing Q training microphones around the error microphone, where Q << L. The room impulse response vector of length L from the control speaker to the qth training microphone is

[0109] h q =[h q,1 h q,2 ... h q,L ] T ,q=1,2,...,Q

[0110] The following matrix can be formed

[0111] H=[h1 h2 ... h Q ]

[0112] The size of matrix H is L×Q. Assume that the column rank of H is Q. By performing singular value decomposition on matrix H, we get

[0113] H=UΣV T

[0114] Where U is a semi-orthogonal matrix of size L×Q, V is a matrix of dimension R×R, whose column vectors are the right singular vectors of H, and Σ is a diagonal matrix of size Q×Q, whose diagonal elements are the singular values ​​of H. Using singular value decomposition, the reduced-dimensional reference signal is obtained as follows:

[0115]

[0116] in

[0117]

[0118] vector is the room impulse response of the secondary path with length L. It can be seen that the filtered reference signal after dimensionality reduction The length Q is significantly lower than the length L of the original filtered reference signal.

[0119] By applying the reduced-dimensional filtered reference signal, the length of the FxLMS-based ANC filter can be significantly reduced. The error signal can be written as:

[0120]

[0121] in is the control filter after dimensionality reduction. By solving the following optimization problem:

[0122]

[0123] The optimal dimension reduction control filter can be obtained as

[0124]

[0125] in yes The covariance matrix of for and the cross-correlation of d(t).

[0126] The matrix inversion in the above formula is computationally intensive, which increases the ANC delay and prevents the control filter from adapting to rapidly changing noise in practice. Therefore, an adaptive method is used to reduce the computational complexity of the algorithm. The adaptive method is implemented by minimizing the instantaneous square error, and the optimization method problem is simplified to:

[0127]

[0128] Apply the steepest descent method to solve the above equation, and get the update formula of the adaptive filter:

[0129]

[0130] in The length of the filter is Q, which is significantly shorter than the filter length of the traditional FxLMS.

[0131] The speaker is controlled to play a speaker driving signal and a noise signal which are superimposed at the error microphone to obtain an error signal.

[0132] Active noise control in reverberant environments is a challenging task due to the need for long secondary path room impulse response information to achieve effective noise reduction performance. Therefore, reducing the signal dimensionality during the ANC process is essential for developing more efficient noise reduction control filters. In this proposal, a dimension-reduced FxLMS method is proposed to achieve ANC in reverberant environments. The method first applies a signal dimension reduction method based on singular value decomposition to the filtered reference signal, and then uses the dimension-reduced filtered reference signal in the FxLMS-based ANC method. Simulation results show that under various simulation conditions, by appropriately selecting the number of training microphones, the proposed dimension-reduced FxLMS method achieves lower noise reduction levels than the traditional FxLMS method.

[0133] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. An active noise control method based on dimension reduction FxLMS, ​​characterized in that: The following steps are involved: Acquire a reference signal, wherein the reference signal is a noise source signal received by a reference microphone and a self-noise of the reference microphone; Based on the reference signal, obtaining a filtered reference signal; Based on singular value decomposition, reducing the dimension of the filtered reference signal; Based on the FxLMS ANC method, the dimension-reduced adaptive filter is obtained by using the dimension-reduced filtered reference signal; Based on the dimension reduction adaptive filter, active noise control is achieved.

2. The active noise control method based on dimension reduction FxLMS according to claim 1, characterized in that: The method to obtain the reference signal is: Where x(t) is the reference signal and v(t) is the reference microphone self-noise.

3. The active noise control method based on dimension reduction FxLMS according to claim 2 is characterized in that: Based on the reference signal, a method for obtaining a filtering reference signal is: in, is the secondary path room impulse response h s (t) is estimated.

4. The active noise control method based on dimension reduction FxLMS according to claim 3 is characterized in that: The method of reducing the dimension of the filtered reference signal based on singular value decomposition includes: in, vector is the room impulse response of the secondary path of length L, U T is the transpose of a semi-orthogonal matrix of size L×Q, where Q is the number of training microphones.

5. The active noise control method based on dimension reduction FxLMS according to claim 4 is characterized in that: Based on the FxLMS ANC method, the method of obtaining the dimension-reduced adaptive filter by using the dimension-reduced filtered reference signal includes: in, The length of is Q, μ>0 is the adaptive step size, and e(t) is the error signal.

6. An active noise control system based on dimension reduction FxLMS, ​​characterized in that: The active noise control method based on dimension reduction FxLMS for implementing any one of claims 1 to 5 comprises: a signal pickup device, a signal processing device and a signal noise reduction device; The signal pickup device comprises a reference microphone and an error microphone, wherein the reference microphone is located near the noise source and is used to pick up a reference signal, and the error microphone is located at a control point and is used to pick up an error signal; based on the reference signal, a filtered reference signal is obtained; The signal processing device is used to reduce the dimension of the filter reference signal based on singular value decomposition; based on the ANC method of FxLMS, ​​the dimension-reduced filter reference signal and the error signal are used to obtain a dimension-reduced adaptive filter; The signal noise reduction device is used to achieve active noise control based on the dimension reduction adaptive filter.

7. The active noise control system based on dimension reduction FxLMS according to claim 6, characterized in that: The system further comprises: an audio playback device; The audio playback device is used to convolve the dimension reduction adaptive filter with the reduced-dimensional filtering reference signal to generate a speaker driving signal; the speaker driving signal and the noise signal are superimposed at the error microphone to obtain an error signal.

Images