A fast convergence method for decoupled whitening suitable for active road noise control systems

Abstract

This invention discloses a decoupling whitening fast convergence method suitable for active road noise control systems. The steps include: (1) hardware configuration in an ARNC system based on a feedforward strategy; (2) decomposing the secondary path into a minimum-phase part and an all-pass part based on internal and external integral solutions, then filtering the output of the control filter through the minimum-phase part and filtering the error signal through the all-pass part; (3) decomposing the spectral density matrix of the reference signal to obtain the spectral factor function; (4) using the filtered error signal and the whitened reference signal, combined with the minimum mean square algorithm for frequency domain filtering error, updating the control filter, and outputting the control signal to drive the cancelling speaker to emit sound, which propagates through the secondary path and coherently superimposes with the road noise signal near the ear. This invention can be used in ARNC systems in multi-channel complex scenarios, accelerating the convergence speed of the FeLMS algorithm through secondary path decoupling and reference signal whitening operations.

Description

Technical Field

[0001] This invention belongs to the technical field of active noise control, specifically relating to a decoupled whitening fast convergence method suitable for active road noise control systems. Background Technology

[0002] The lightweighting of electric vehicles has significantly increased cabin noise. Road noise, a major source of cabin noise, exhibits low-frequency random noise characteristics that vary with road conditions. Traditional passive noise control is only effective for mid-to-high frequency noise. Therefore, Active Road Noise Control (ARNC), which combines low-frequency noise reduction with road noise control systems, has become an important research direction for cabin noise reduction. The feedforward multi-channel ARNC system uses multiple accelerometers deployed on the vehicle chassis to collect road vibration information as reference signals. This information is then fed into a digital signal processor (DSP), filtered by a control filter, and output as a control signal to drive a noise cancellation speaker. After propagation through a secondary path, the noise signal coherently superimposes with the road noise signal near the listener's ear, thus achieving noise control. Therefore, selecting a suitable adaptive algorithm to update the control filter is a prerequisite for achieving noise reduction.

[0003] The Filtered-reference Least Mean Square (FxLMS) algorithm for road noise control has been proven to be a reliable technical solution (Sutton TJ, Elliott SJ, McDonald AM, et al. Active control of road noise inside vehicles[J]. Noise Control Engineering Journal, 1994, 42(4):137-147.). However, in actual ARNC systems, due to the complexity of vibration sources and vehicle structure, the coupling between different secondary paths and the cross-correlation between reference signals can slow down the convergence speed of the FxLMS algorithm based on the gradient descent principle. Therefore, decoupling of secondary paths and whitening of reference signals are key to improving the convergence performance of multi-channel adaptive algorithms. In addition, the computational burden introduced by using a large number of reference signals is not conducive to the implementation of the FxLMS algorithm in practical applications. Considering that the number of error signals in the ARNC system is much smaller than the number of reference signals, a filtered error structure can be used instead of a filtered reference structure for path noise control (DeBrunner VE, Zhou D. Hybrid filtered error LMS algorithm: Another alternative to filtered-x LMS[J]. IEEE Transactions on Circuits and Systems I: Regular Papers, 2006, 53(3): 653-661.). Simultaneously, utilizing the efficient operation of time-domain convolution in the frequency domain, transforming the FeLMS algorithm to the frequency domain can further reduce the computational load. Therefore, applying the secondary path decoupling and reference signal whitening strategies to the frequency-domain FeLMS algorithm for path noise control is of great significance for the rapid convergence and engineering implementation of the ARNC system. Summary of the Invention

[0004] To address the aforementioned technical issues, this invention proposes a decoupled whitening fast convergence method suitable for active road noise control systems. This method can be used in real-time road noise control systems in multi-channel complex scenarios, and has the ability to converge quickly with low computational complexity.

[0005] The technical solution adopted in this invention is as follows:

[0006] A fast convergence method for decoupled whitening suitable for active road noise control systems, comprising the following steps:

[0007] Step 1: Configure the hardware in the active road noise control system based on the feedforward strategy, including placing an acceleration sensor on the vehicle chassis, installing an active noise-canceling headrest including an error microphone and a cancelling speaker at the head of the driver, and connecting the acceleration sensor to a multi-channel digital signal processor.

[0008] Step 2: Based on the internal and external integral solution operation, the secondary path is decomposed into a minimum phase part and an all-pass part. Then, the output of the control filter is filtered through the minimum phase part, and the error signal is filtered through the all-pass part to achieve the decoupling function of the secondary path.

[0009] Step 3: Based on the spectral factor decomposition operation, the spectral density matrix of the reference signal is decomposed to obtain the spectral factor function. This function is then used to preprocess the reference signal to achieve the whitening function of the reference signal.

[0010] Step 4: In the multi-channel digital signal processor, the filter error signal obtained in step 2 and the whitening reference signal obtained in step 3 are used together with the minimum mean square algorithm of frequency domain filter error to update the control filter. The output control signal drives the cancellation speaker to emit sound. After propagating through the secondary path, it is coherently superimposed with the road noise signal near the ear to achieve the noise reduction function.

[0011] Furthermore, in step 1, the hardware configuration of the active road noise control system based on the feedforward strategy specifically includes: deploying an acceleration sensor on the vehicle chassis to pick up road vibration information, which is used to obtain a reference signal highly correlated with the road noise signal; installing an active noise-canceling headrest including an error microphone and a cancellation speaker at the passenger's head, where the error microphone is used to collect the error signal at the ear, and the cancellation speaker is used to emit control sound waves, which are coherently superimposed with the road noise signal after propagation through a secondary path, creating a quiet zone near the ear; sending the collected reference signal and error signal into a multi-channel digital signal processor for processing, updating the control filter in combination with an adaptive algorithm, and simultaneously outputting a control signal to drive the cancellation speaker to emit sound, thereby realizing the real-time road noise control function.

[0012] Compared with the prior art, the beneficial effects of the present invention are as follows: In the real-time road noise control system, the use of secondary path decoupling and reference signal whitening strategies accelerates the convergence speed of the FeLMS algorithm, enabling passengers to obtain a short-term and noticeable noise reduction experience; in addition, the method of the present invention has low computational complexity, making it effective in DSP and having extremely high practical value. Attached Figure Description

[0013] Figure 1 Here is an overall schematic block diagram of the method of the present invention, (a) is a system configuration diagram of a multi-channel ARNC system, and (b) is a flowchart of the decoupled whitening fast convergence algorithm.

[0014] Figure 2This is a block diagram of the signal processing of a multi-channel ARNC system.

[0015] Figure 3 This is a block diagram of the FeLMS algorithm based on secondary path decoupling and reference signal whitening strategies.

[0016] Figure 4 Here are the hardware configuration diagrams in this embodiment: (a) is a layout diagram of the vehicle chassis acceleration sensor, and (b) is an installation diagram of the active noise-canceling headrest.

[0017] Figure 5 These are the secondary path impulse response and amplitude-frequency response curves of the active noise-canceling headrest in this embodiment. (a) and (b) are the secondary path impulse response and amplitude-frequency response curves from the left speaker to the binaural error microphone; (c) and (d) are the secondary path impulse response and amplitude-frequency response curves from the right speaker to the binaural microphone.

[0018] Figure 6 These are the noise reduction curves at the error microphone in this embodiment. (a) is the noise reduction curve at the left ear microphone, and (b) is the noise reduction curve at the right ear microphone. Detailed Implementation

[0019] This embodiment provides a decoupled whitening fast convergence method suitable for active road noise control systems, the main steps of which include the following parts:

[0020] 1. ARNC System Hardware Configuration

[0021] 1) Place acceleration sensors on the vehicle chassis

[0022] In a feedforward ARNC system, by placing accelerometers on the vehicle chassis to pick up information about vehicle vibration and undulation, a reference signal highly correlated with the road noise signal at the human ear can be obtained, thereby achieving the desired noise reduction. After optimization, accelerometers can be placed in locations such as the subframe, steering shaft, shock absorber, and trailing arm to obtain a reference signal with strong coherence.

[0023] 2) Install an active noise-canceling headrest

[0024] Active noise-canceling headrests are typically installed at the head of the passenger in a vehicle and include a cancelling speaker and two error microphones simulating the binaural quiet zone. On the one hand, the speaker's low-frequency response should be relatively flat to ensure good low-frequency response in the secondary path between the speaker and the microphone; on the other hand, the headrest should be as comfortable and aesthetically pleasing as possible to provide passengers with a better riding experience.

[0025] 3) Connect to a multi-channel DSP

[0026] The reference signal acquired by the accelerometer is sent to the DSP, filtered by a control filter, and outputs a real-time control signal to drive the cancellation speaker to emit sound. This sound propagates through a secondary path to the ear and coherently superimposes with the road noise signal. An error microphone picks up the controlled error signal and sends it to the DSP. The control filter updates its coefficients based on the reference and error signals using an adaptive algorithm. During convergence, the noise in the binaural noise reduction area gradually decreases, thus achieving road noise control.

[0027] 2. Multi-channel FeLMS algorithm

[0028] Example: A multi-channel system containing R reference signals, C control signals, and M error signals, such as... Figure 2 As shown. W i S is the i-th order C×R dimension coefficient matrix of a control filter of length I. j Let J be the j-th order M×C dimension coefficient matrix of the secondary path impulse response. Then, the error signal vector e(n) can be written as...

[0029]

[0030] In equation (1), the superscript T Let d(n) represent the transpose operation, d(n) be the desired signal vector at time n, and x(·) be the reference signal vector. At this point, the system cost function is:

[0031]

[0032] In the above formula, trace{·} represents the trace of the matrix. This represents finding the expected value, and

[0033]

[0034] Define the autocorrelation matrices of the desired signal and the reference signal, and the cross-correlation matrices of the desired signal and the reference signal, respectively.

[0035]

[0036] Substituting equation (4) into equation (3) yields

[0037]

[0038] According to the definition in equation (1), the cross-correlation matrix between the error signal and the reference signal can be written as follows:

[0039]

[0040] According to the rule of differentiation of matrix trace, equation (2) is applied to the coefficient matrix W. i Differentiation yields

[0041]

[0042] Based on the steady-state signal assumption, the filter error signal is defined. Equation (7) can be rewritten as

[0043]

[0044] At this point, equation (8) uses a time-advanced filtered error signal. It cannot be implemented in a real physical system. Therefore, it is possible to... By simultaneously introducing a J-point delay into x(n), a causal version of the FeLMS algorithm is obtained:

[0045]

[0046] In equation (9), α is the convergence step size.

[0047] In multi-channel ARNC systems, coupling between multiple secondary paths and cross-correlation between multiple reference signals can slow down the convergence speed of the FxLMS algorithm based on gradient descent. Therefore, these problems can be addressed by decoupling secondary paths and whitening reference signals.

[0048] 3. Decoupled whitening fast convergence algorithm

[0049] For ease of discussion, this embodiment performs algorithm derivation in the Z domain, and the relevant variable definitions are shown in Table 1.

[0050] Table 1. Time-domain representation of variables and their corresponding Z-domain representation.

[0051] Time domain representation <![CDATA[W i ]]> <![CDATA[S j ]]> x(n) e(n) <![CDATA[R xx (n′)]]> <![CDATA[R xd (n′)]]> <![CDATA[R xe (n′)]]> Z-domain represents W(z) S(z) x(z) e(z) <![CDATA[P xx (from)]]> <![CDATA[P xd (from)]]> <![CDATA[P xe (from)]]>

[0052] 1) Secondary path decoupling and reference signal whitening

[0053] The secondary path S(z) is decomposed into a fully passable part S using the inner and outer integral solutions. all (z) and the minimum phase part S min (z):

[0054] S(z)=S all (z)S min (z) (10)

[0055] And the full passage part satisfies

[0056]

[0057] superscript -1 This represents the inverse operation. For S all The adjoint matrix of (z), I CLet be a C×C dimensional identity matrix. By setting the desired signal to zero, we obtain the filter error signal from the output signal y(z) of the control filter to update the control filter.

[0058]

[0059] The transmission path between them is

[0060]

[0061] Among them, z -J Let J represent the time delay at point J in the Z domain. Substituting equations (10) and (11) into equation (13), we obtain the propagation path at this time as z. -J That is, the pure time delay at point J, thus realizing the secondary path decoupling function.

[0062] Based on the secondary path decomposition approach in equation (10), the spectral density matrix P of the reference signal can be obtained. xx (z) Perform spectral factor decomposition

[0063]

[0064] Among them, F min (z) is P xx The minimum phase part of (z) can be viewed as a transfer function that incorporates characteristics of the reference signal. For F min The adjoint matrix of (z). Therefore, F can be used. min (z) The inverse matrix The whitening function is implemented by preprocessing the reference signal to obtain a set of uncorrelated whitened reference signals:

[0065]

[0066] Where v(z) is the Z-domain representation of the whitened reference signal, and x(z) is the Z-domain representation of the reference signal.

[0067] Figure 3 A flowchart of the FeLMS algorithm based on secondary path decoupling and reference signal whitening operations is given. The update formula for the control filter can then be rewritten as:

[0068]

[0069] 2) Frequency domain implementation of the decoupled whitening FeLMS algorithm

[0070] When the control filter corresponding to equation (7) converges to the optimal solution, we have:

[0071]

[0072] The Z-transform of the above equation can be written as

[0073] {S T (z -1 )P xe (z)} + =0 (18)

[0074] Among them, S T (z -1 Let be the adjoint matrix of S(z). Let {·} be the cross-spectral density matrix of the error signal and the reference signal. + To extract the causal part within the parentheses (Elliott S J. Optimal controllers and adaptive controllers for multichannel feedforward control of stochastic densities[J]. IEEE Transactions on Signal Processing, 2000, 48(4):1053-1060.). Furthermore, the cross-spectral density matrix P of the desired signal and the reference signal is defined. xd (z) Spectral density matrix P of the reference signal xx (z) are respectively

[0075]

[0076] Taking the Z-transform of both sides of equation (6), we have:

[0077] P xe (z)=P xd (z)+S(z)W(z)P xx (z) (20)

[0078] Substituting equation (20) into equation (18) yields

[0079] {S T (z -1 )P xd (z)+S T (z -1 )S(z)W opt (z)P xx (z)} + =0 (21)

[0080] Among them, W opt (z) represents the optimal solution for the control filter.

[0081] Substituting equations (10), (11), and (14) into equation (21) yields...

[0082]

[0083] in, For S min The adjoint matrix of (z), for The inverse matrix. and Since they are all minimum phase, equation (22) can be written as follows:

[0084]

[0085] Because S min (z) and F min (z) are all causal sequences, therefore

[0086]

[0087] Now, define the medium filter Ψ(z), and the control filter can be calculated as follows:

[0088]

[0089] From equation (24), the optimal solution for Ψ(z) is: According to Newton's algorithm (Widrow B, Stearns SD. Adaptive signal processing prentice-hall[J]. Englewood Cliffs, NJ, 1985: 52.), the update equation for Ψ(z) can be written as:

[0090] Ψ new (z)=Ψ old (z)+βΔΨ(z) (26)

[0091] In the above equation, β is the update step size of the media filter, and ΔΨ(z) is the update amount in the p-th iteration:

[0092]

[0093] Based on the definitions of the filtering error signal in equation (12) and the whitening reference signal in equation (15), equation (27) can be rewritten as follows:

[0094]

[0095] By defining 2L discrete frequency points k = 0, 1, ..., 2L-1, when transforming equation (26) to the frequency domain, it is impossible to obtain an accurate spectrum estimate using finite frame data. Therefore, the frequency domain medium filter Ψ(k) can be rewritten in the form of multiple iterations to approximate the optimal solution:

[0096]

[0097] Where p is the iteration number, and the superscript... H Representing the conjugate transpose, v(k) is the Fast Fourier Transform (FFT) of the whitened reference signal at 2L points. H (k) is the conjugate transpose of v(k). This is the filtered residual signal from the p-th iteration:

[0098]

[0099] In equation (30), S all (k) and S min (k) is the frequency domain representation of the all-pass portion and the minimum-phase portion of the secondary path S(k). and S represents all The conjugate transpose of (k) and S min The inverse matrix of (k), e(k) is the fast Fourier transform of the error signal at 2L points; it can be found that when p=0, Ψ 0 (k)=Ψ old (k), after completing P iterations, we have Ψ P+1 (k)=Ψ new (k). At this point, according to Ψ new (k) The update formula for the time-domain controlled filter is obtained as follows:

[0100]

[0101] in, It is the inverse matrix of the reference signal spectral factor matrix Fmin(k) in the frequency domain, IFFT{·} + This represents the causal component of the inverse fast fourier transform (IFFT).

[0102] Example

[0103] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that these examples are for illustrative purposes only and are not intended to limit the scope of the invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0104] 1. Install hardware facilities

[0105] The vehicle used in this experiment is a Changan UNI-K. A total of four 3-axis accelerometers were deployed on the vehicle chassis to collect 12 reference signals. The placement of the accelerometers is as follows: Figure 4As shown in (a); the active noise-canceling headrest consists of a left-ear speaker (Speaker1) and a right-ear speaker (Speaker2), and a left-ear microphone (Microphone1) and a right-ear microphone (Microphone2) simulating the noise reduction points of the human ear, as follows. Figure 4 (b) shows that the Texas Instruments TMS320C6678 digital signal processor was used for noise control in the cabin area to be noise-reduced.

[0106] 2. Obtaining secondary paths

[0107] This experimental example uses, for example Figure 4 (b) shows the secondary paths obtained from the active noise-canceling headrest. The sound emitted by Speaker 1 is collected by Microphone 1 and Microphone 2, yielding secondary paths S11 and S12. The sound emitted by Speaker 1 is collected by Microphone 1 and Microphone 2, yielding secondary paths S21 and S22. To ensure the secondary path modeling is as accurate as possible, a quiet road section without noise interference is selected during measurement, and the car engine is turned off. To simulate real-world driving conditions, a tester should be present in the passenger seat. The 512th-order impulse response and amplitude-frequency response curves of secondary paths S11 and S12 are shown below. Figure 5 As shown in (a) and (b); the 512th-order impulse response and amplitude-frequency response curves of secondary paths S21 and S22 are as follows. Figure 5 As shown in (c) and (d).

[0108] 3. Real-world road noise control performance test based on A-weighting

[0109] This embodiment compares the multi-channel time-domain FxLMS algorithm (Approach 1) widely used in ARNC systems with the method of this invention (Approach 2). All algorithms were optimized for performance during the experiment. Under steady-state conditions at 50 km / h, 20 seconds of data were recorded at the error microphone with ARNC off and on at a sampling rate of 3125 Hz as the desired signal and the error signal after control. The differences in noise reduction of different algorithms under A-weighting were statistically analyzed. The Wiener Solution under the current driving conditions was also presented for comparison (Elliott S. Signal processing for active control [M]. Elsevier, 2000.). The noise reduction curves for the left and right ears are shown below. Figure 6As shown in (a) and (b), the results show that, due to the use of secondary path decoupling and reference signal whitening strategies, the method of this invention can quickly converge to the Wiener solution in about 15 seconds, with a noise reduction of 5.1 dBA for the left ear and 5.0 dBA for the right ear. The convergence performance is significantly better than the time-domain FxLMS algorithm.

[0110] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A decoupled whitening fast convergence method suitable for an active road noise control system, characterized in that, The method includes the following steps: Step 1: Configure the hardware in the active road noise control system based on the feedforward strategy, including placing an acceleration sensor on the vehicle chassis, installing an active noise-canceling headrest including an error microphone and a cancelling speaker at the head of the driver, and connecting the acceleration sensor to a multi-channel digital signal processor. Step 2: Based on the internal and external integral solution operation, the secondary path is decomposed into a minimum phase part and an all-pass part. Then, the output of the control filter is filtered through the minimum phase part, and the error signal is filtered through the all-pass part to achieve the decoupling function of the secondary path. Step 3: Based on the spectral factor decomposition operation, the spectral density matrix of the reference signal is decomposed to obtain the spectral factor function. This function is then used to preprocess the reference signal to achieve the whitening function of the reference signal. Step 4: In the multi-channel digital signal processor, the filter error signal obtained in Step 2 and the whitening reference signal obtained in Step 3 are used, combined with the minimum mean square algorithm for frequency domain filter error, to update the control filter. The output control signal drives the cancellation speaker to emit sound, which propagates through the secondary path and coherently superimposes with the road noise signal near the ear, thus achieving noise reduction. Specifically, based on Newton's algorithm, the iterative formula for the medium filter Ψ(z) is obtained as follows: wherein, is the iteration step size, is the difference between the optimal solution and the current solution of the intermediate filter, denotes the causal part of the bracketed expression, denotes the mathematical expectation, is the filtered error signal in the Z-domain, is the companion matrix of v(z); when transforming the above equation to the frequency domain, using finite frame data cannot obtain accurate spectral estimation, therefore the frequency domain intermediate filter Ψ(k) is rewritten as a form of multiple iterations to approximate the optimal solution: In the above formula, p represents the iteration number, and the superscript... H Representing the conjugate transpose, v(k) is the fast Fourier transform of the whitened reference signal at point 2L. H (k) is the conjugate transpose of v(k). This is the filtered residual signal from the p-th iteration: Among them, S all (k) and S min (k) is the frequency domain representation of the all-pass portion and the minimum-phase portion of the secondary path S(k). and S represents all The conjugate transpose of (k) and S min The inverse matrix of (k), e(k) is the fast Fourier transform of the error signal at 2L points; it can be found that when From time to time After completing P iterations, there is At this time, according to The update formula for the time-domain controlled filter is obtained as follows: Among them, W i Let I be the i-th coefficient matrix of a control filter of length I. It is the inverse matrix of the reference signal spectral factor matrix in the frequency domain. This represents the causal part of the inverse fast Fourier transform; during the update process of the control filter, the multi-channel digital signal processor outputs control signals in real time, and the noise in the binaural noise reduction area gradually decreases, thereby realizing the road noise control function.

2. The decoupling whitening fast convergence method for active road noise control systems as described in claim 1, characterized in that, In step 1, the hardware configuration of the active road noise control system based on the feedforward strategy specifically includes: placing an acceleration sensor on the vehicle chassis to pick up road vibration information to obtain a reference signal highly correlated with the road noise signal; installing an active noise-canceling headrest including an error microphone and a cancellation speaker at the passenger's head, where the error microphone is used to collect the error signal at the ear, and the cancellation speaker is used to emit control sound waves, which are coherently superimposed with the road noise signal after propagation through a secondary path, creating a quiet zone near the ear; sending the collected reference signal and error signal into a multi-channel digital signal processor for processing, updating the control filter in combination with an adaptive algorithm, and simultaneously outputting a control signal to drive the cancellation speaker to emit sound, thereby realizing the real-time road noise control function.

3. The decoupling whitening fast convergence method for active road noise control systems as described in claim 1, characterized in that, The step 2 is implemented as follows: based on an inner-outer integral decomposition operation, the secondary path S(z) is decomposed into an all-pass part S all (z) and a minimum phase part S min (z). Where z is a complex variable in the Z-transform; The filtered error signal in the Z-domain is obtained by filtering the error signal using the full-pass component. Among them, superscript T and −1 These represent the transpose and inverse operations, respectively. for The adjoint matrix is ​​given, and e(z) is the Z-domain representation of the error signal; at the same time, the output of the control filter is filtered through the minimum phase part to achieve the decoupling function of the secondary path.

4. The decoupling whitening fast convergence method for active road noise control systems as described in claim 1, characterized in that, The specific implementation of the step 3 is to calculate the spectral density matrix P of the reference signal xx (z) and obtain the spectral factor function F min (z) based on the spectral factor decomposition operation is: F min (z) is the transfer function that incorporates the characteristics of the reference signal. For F min The adjoint matrix of (z) can therefore be obtained using F. min (z) The inverse matrix The reference signal is filtered to obtain a whitened reference signal, thus achieving the decorrelation function of the reference signal: Where v(z) is the Z-domain representation of the whitened reference signal, and x(z) is the Z-domain representation of the reference signal.

Images