Robustness active road noise control method based on coherence constraint

Abstract

The invention discloses a robust active road noise control method based on coherence constraint. The method comprises the following steps: (1) performing hardware configuration on the active road noise control system; (2) calculating a multi-coherence coefficient of the error signal and the reference signal in real time, and using the multi-coherence coefficient as a step length constraint factor for controlling updating of a filter; (3) performing control filter updating through secondary path decomposition and reference signal whitening in combination with a frequency domain filtering error minimum mean square algorithm; and (4) converting the control filter to a time domain, outputting a control signal in real time to drive a cancellation loudspeaker to produce sound, propagating the sound through a secondary path, and coherently superposing the sound with a road noise signal at an error microphone to produce a quiet zone near a human ear. The method can be used in an active road noise control system in a multi-channel complex scene, a higher convergence speed and a lower steady-state error are obtained in a steady-state scene based on a coherent constraint stability protection method, and the robustness of the system is improved in a noisy scene.

Description

Technical Field

[0001] The present invention belongs to the technical field of active noise control, and in particular relates to a robust active road noise control method based on coherence constraints. Background Art

[0002] Active Road Noise Control (ARNC) technology can significantly reduce road noise in the cabin and improve the passenger experience through the combination of electronic devices and multi-channel algorithms. The ARNC system needs to install an acceleration sensor on the vehicle chassis to collect road vibration information as a reference signal, which is sent to the multi-channel digital signal processor (DSP). After filtering by the control filter, the control signal is output in real time to drive the cancellation speaker to sound. After propagation through the secondary path, it is coherently superimposed with the road noise signal near the ears to achieve the noise control function. Based on the reference signal and the error signal picked up by the error microphone, the DSP combines the multi-channel algorithm to adaptively update the control filter weights to optimize the noise reduction effect of the system. The use of the Decomposing and Whitening Frequency Domain Filtered-error Least MeanSquare (DWFDFeLMS) algorithm based on secondary path decomposition and reference signal whitening for road noise control has been verified as a reliable technical solution (Lian S, Li T, Gu J, et al. An online decoupling-whitening frequency domain filtered-error least meansquare algorithm for active road noise control [J]. The Journal of the Acoustical Society of America, 2024, 156 (2): 1413-1424.). However, in actual applications, interference noises such as human voice communication and road bumps are picked up by the error microphone and fed into the algorithm update, which will affect the convergence process of the control filter and even cause the algorithm to diverge. Therefore, it is necessary to select a suitable protection algorithm to improve the robustness of the system.

[0003] Using a state transition strategy in noisy scenarios and stopping filter updates when interference noise is detected can avoid system divergence, but the effectiveness of this method is highly dependent on the selection of state transition parameters (Shen X, Gan WS, Shi D. Alternative switching hybrid ANC [J]. Applied Acoustics, 2021, 173: 107712.). Using a weight constraint algorithm to limit the maximum value of the control filter update for impulse noise improves system robustness, but affects the steady-state performance of the algorithm (Lan H, Zhang M, Ser W. A weight-constrained FxLMS algorithm for feedforward active noise control systems [J]. IEEE Signal Processing Letters, 2002, 9 (1): 1-4.). Summary of the invention

[0004] In order to solve the above technical problems, the present invention proposes a robust active road noise control method based on coherence constraints. The method is based on a stable protection method of coherence constraints, which can obtain faster convergence speed and lower steady-state error in steady-state scenarios and has stronger robustness in noisy scenarios.

[0005] The technical solution adopted by the present invention is:

[0006] A robust active road noise control method based on coherence constraints, the method comprising the following steps:

[0007] Step 1, hardware configuration of the active road noise control system, including placing an acceleration sensor on the vehicle chassis to collect road vibration information as a reference signal; installing an active noise reduction headrest including an error microphone and a cancellation speaker at the head of the passenger, wherein the error microphone is used to collect error signals at both ears;

[0008] Step 2, calculate the multiple coherence coefficients of the error signal and the reference signal in real time, and use the multiple coherence coefficients as the step size constraint factor for controlling the filter update; the coherence coefficients of the error signal and the reference signal in the early stage of control are large, and a large step size is used to ensure that the method converges quickly; the coherence coefficients in the later stage of control are small, and a small step size is used to reduce the steady-state error of the system; the introduction of interference noise in noisy scenes reduces the signal coherence, and a small step size is used to improve the robustness of the system;

[0009] Step 3, through secondary path decomposition and reference signal whitening, combined with the frequency domain filtering error minimum mean square algorithm, the control filter is updated;

[0010] Step 4: Convert the control filter to the time domain, and output the control signal in real time to drive the cancellation speaker to emit sound. After propagating through the secondary path, it is coherently superimposed with the road noise signal at the error microphone, generating a quiet zone near the human ear to achieve the noise reduction function.

[0011] The present invention can predict the maximum noise reduction amount of the system by calculating the multiple coherence coefficients of the error signal and the reference signal in real time, and use it as a step size constraint factor for controlling the filter update. The coherence coefficient is large in the early stage of control, and a large step size is used to ensure rapid convergence of the algorithm; the coherence coefficient is small in the later stage of control, and a small step size is used to reduce the steady-state error of the system. When there is interference noise in the system, the coherence of the error signal and the reference signal deteriorates, and a smaller step size constrains the weight update amount of the control filter, thereby improving the robustness of the ARNC system. Therefore, compared with the prior art, the present invention, in the real-time control system of road noise, can obtain a faster convergence speed and a lower steady-state error in a steady-state scenario based on a stable protection method of coherence constraints, and can improve the robustness of the system in noisy scenarios, and has extremely high practical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 It 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 flow chart of a multi-channel frequency domain FeLMS (Coherence-based DWFDFeLMS, C-DWFDFeLMS) algorithm based on secondary path decomposition and reference signal whitening after adding stable protection of coherence constraints.

[0013] Figure 2 is an updated block diagram of a medium filter based on secondary path decomposition and reference signal whitening.

[0014] Figure 3 2 is a hardware configuration diagram in this embodiment, (a) is a layout diagram of the vehicle chassis acceleration sensor, and (b) is an installation diagram of the active noise reduction headrest.

[0015] Figure 4 1 is a noise reduction curve at the error microphone in a steady-state scenario in this embodiment, (a) is a noise reduction curve at the left ear microphone, and (b) is a noise reduction curve at the right ear microphone.

[0016] Figure 5 It is the step size variation curve of the C-DWFDFeLMS algorithm in the steady-state scenario in this embodiment.

[0017] Figure 6 1 is a noise reduction curve at the error microphone in a noisy scene in this embodiment, (a) is a noise reduction curve at the left ear microphone, and (b) is a noise reduction curve at the right ear microphone.

[0018] Figure 7is the step size variation curve of the C-DWFDFeLMS algorithm in the noisy scene in this embodiment. DETAILED DESCRIPTION

[0019] This embodiment provides a robust active road noise control method based on coherence constraints, and the main steps include the following parts:

[0020] 1. Hardware configuration of ARNC system

[0021] 1) Acceleration sensors are placed on the vehicle chassis

[0022] In the feedforward ARNC system, acceleration sensors can be placed on the vehicle subframe, steering shaft, shock absorber, trailing arm and other locations to pick up road vibration information and obtain a reference signal that is highly correlated with the road noise signal at the vehicle's ears (Zhang JA, Murata N, Maeno Y, et al. Coherence-based performance analysis on noise reduction in multichannel active noise control systems[J]. The Journal of the Acoustical Society of America, 2020, 148(3): 1519-1528.), which is used to update the control filter weights in the multi-channel control algorithm.

[0023] 2) Install active noise reduction headrests in the car seat

[0024] Installing an active noise reduction headrest at the head of the driver, using a cancellation speaker to control road noise, and using an error microphone to pick up the error signal to control the update of the filter is a common solution for ARNC systems (Jung W, Elliott SJ, Cheer J. Local active control of road noise inside a vehicle [J]. Mechanical Systems and Signal Processing, 2019, 121: 144-157.). On the one hand, the secondary path between the speaker and the microphone should have a relatively flat frequency domain response; on the other hand, the headrest should be as comfortable and beautiful as possible to provide passengers with a better riding experience.

[0025] 3) Connect to multi-channel DSP

[0026] The reference signal collected by the acceleration sensor is sent to the multi-channel DSP. After being filtered by the control filter, the control signal is output in real time to drive the cancellation speaker to sound. The signal is transmitted to the ears through the secondary path and coherently superimposed with the road noise signal. The error signal after control is picked up by the error microphone and sent to the DSP. The DSP calculates the multiple coherence coefficients of the error signal and the reference signal to obtain the step size constraint factor for the filter update, and then combines the C-DWFDFeLMS algorithm to update the control filter weight. During the convergence process, the noise in the binaural noise reduction area is gradually reduced, thereby realizing the road noise control function.

[0027] 2. Derivation of DWFDFeLMS algorithm

[0028] Example A multi-channel ARNC system including R reference signals, C control signals and M error signals is as follows: Figure 1 As shown in (a). Define 2L discrete frequency points k = 0, 1, ..., 2L-1, and obtain the autospectral density matrix of the reference signal and the cross-spectral density matrix of the road noise signal and the reference signal respectively.

[0029]

[0030] In formula (1), Indicates mathematical expectation, superscript H represents conjugate transpose, x(k) and d(k) are the Fast Fourier Transform (FFT) of 2L discrete frequency points of the reference signal and the road noise signal, respectively.

[0031] 1) Secondary path decomposition and reference signal whitening

[0032] Define S(k) as the frequency domain response of the secondary path and decompose it using inner and outer integrals to obtain the all-pass part S all (k) and the minimum phase part S min (k)(Wesselink JM, Berkhoff AP. Fast affine projections and theregularized modified filtered-error algorithm in multichannel active noisecontrol[J]. The Journal of the Acoustical Society of America, 2008, 124(2):949-960.)

[0033] S(k)=S all (k)S min (k) (2)

[0034] The error signal is filtered through the full-pass part, and the filtered error signal is obtained as

[0035]

[0036] Where e(k) is the FFT of 2L discrete frequency points of the error signal.

[0037] The autospectral density matrix P of the reference signal xx (z) Perform spectral factorization

[0038]

[0039] Among them, the spectral density function F min (k) can be regarded as a transfer function containing the characteristics of the reference signal (Elliott SJ. Optimal controllers and adaptive controllers for multichannel feedforward control of stochastic disturbances [J]. IEEE Transactions on signal Processing, 2000, 48 (4): 1053-1060.). Therefore, F min The inverse matrix of (k) preprocesses the reference signal to realize the whitening function and obtains a set of uncorrelated whitened reference signals:

[0040]

[0041] Among them, the superscript -1 Indicates the inverse operation.

[0042] 2) Implementation of multi-channel frequency domain FeLMS algorithm

[0043] The paper (Lian S, Li T, Gu J, et al. An online decoupling-whitening frequencydomain filtered-error least mean square algorithm for active road noise control [J]. The Journal of the Acoustical Society of America, 2024, 156 (2): 1413-1424.) points out that the optimal solution of the frequency domain control filter is

[0044]

[0045] In formula (6), {·}+ By defining the medium filter Ψ(k), the control filter can be calculated as follows:

[0046]

[0047] Among them, the optimal solution of Ψ(k) is Based on Newton's algorithm (Widrow B, Stearns SD. Adaptive signal processing prentice-hall [J]. Englewood Cliffs, NJ, 1985: 52.), the update equation of Ψ(k) can be written as

[0048] Ψ new (k) = (1-β)Ψ old (k)+βΨ opt (k) (8)

[0049] In the above formula, β is the update step size of the medium filter.

[0050] For non-steady-state road noise signals, the present invention uses the mean of T frame data based on the block update strategy to obtain a more accurate spectral density estimate. At this time, equation (8) can be rewritten as P iterations to approximate the optimal solution:

[0051]

[0052] In the above formula, p is the current number of iterations. is the filtered error signal of the tth frame, v t (k) is the whitened reference signal of the tth frame. The update block diagram of the medium filter is as follows: Figure 2 As shown. The residual signal passes through The reference signal is filtered to obtain the filtered residual signal; The whitened reference signal is filtered to obtain the whitened reference signal. The whitened reference signal is conjugate transposed and multiplied with the filtered residual signal in the frequency domain. After completing the T frame data processing, the causal constraint is added and the medium filter is updated. When p = 0, there is Ψ 0 (k) = Ψ old (k) When p>0, is the filtered residual signal of the t-th frame data at the p-th iteration:

[0053]

[0054] After completing P iterations, there is Ψ P+1 (k) = Ψ new (k). At this time, according to Ψ new (k) The update formula of the time domain control filter is obtained as follows:

[0055]

[0056] Among them, IFFT{·} + Indicates the causal part of the Inverse Fast Fourier Transform (IFFT). During the control filter update process, the multi-channel digital signal processor outputs a control signal in real time to drive the cancellation speaker to sound, and the noise in the binaural noise reduction area is gradually reduced, thereby realizing the road noise control function.

[0057] 3. Calculation of multiple coherence coefficients and step-size constraint factors

[0058] Without considering causality, the maximum noise reduction of the ARNC system can be predicted using the multiple coherence coefficients of the road noise signal and the reference signal:

[0059]

[0060] represents the multi-correlation coefficient of the mth road noise signal and the reference signal, and satisfies for all frequency points k P xdm is the cross-spectral density vector of the mth road noise signal and the reference signal, P xx (k) is the spectral density matrix of the reference signal, P dmdm is the spectral density of the mth road noise signal. However, the error microphone cannot directly obtain the road noise signal during the control process, so the multi-correlation coefficient of the error signal and the reference signal can be calculated in real time

[0061]

[0062] In formula (13), P xem is the cross-spectral density vector of the mth error signal and the reference signal, P emem is the spectral density of the mth error signal. In the initial stage of control, e m (k) = d m (k) The value of is the largest. As the control process gradually converges, the coherence between the error signal and the reference signal gradually weakens. The value decreases gradually.

[0063] In the actual ARNC system, the signal picked up by the error microphone contains not only the road noise signal related to the reference signal, but also the irrelevant in-vehicle interference noise a(n) such as passenger voice. In this case, d'(n) = d(n) + a(n) is defined in the noisy scene. The optimal solution of the control filter in equation (6) can be rewritten as

[0064]

[0065] In formula (14), It can be expressed as

[0066]

[0067] Where a(k) is the FFT of 2L discrete frequency points of the interference noise. Substituting equation (15) into equation (14) and comparing it with equation (6), it is found that the introduction of incoherent noise will not affect the optimal solution of the algorithm. However, when incoherent noise is fed into the algorithm iteration, it will cause the filter weights to deviate from the previous converged solution, which not only reduces the convergence speed of the algorithm, but may also cause the algorithm to diverge.

[0068] The multi-coherence coefficient in the noise-added scenario is defined as

[0069]

[0070] where e′ m (n) is the mth noisy error signal. Considering that a(n) and x(n) are uncorrelated, the noisy error vector e′(n) can be written as

[0071] e′(n)=e(n)+a(n) (17)

[0072] At this time, according to the approximation of formula (15), formula (16) can be written as

[0073]

[0074] When the system is far from the optimal solution, the controllable part of the error signal (that is, the part related to the reference signal) is large. The value of is large; when the system approaches convergence, The introduction of interference noise will increase the noisy error signal, but it is not related to the reference signal and can still ensure The value is at a lower level.

[0075] In a multi-channel system, the step size constraint factor for controlling the filter update is obtained by calculating the average value of M multi-coherence coefficients.

[0076]

[0077] At this time, the update formula of the medium filter based on the coherence constraint is:

[0078]

[0079] The step size μ(k) is modified by the step size constraint factor to

[0080]

[0081] The update block diagram of the C-DWFDFeLMS algorithm after adding the stable protection of the coherent constraint of the present invention is as follows: Figure 1 As shown in (b). The dotted box above is the time domain processing part, including real-time filtering of the control filter and point-by-point output of the control signal by DSP; the dotted box below is the frequency domain processing part, including FFT of the signal, calculation of the coherence coefficient, and update of the medium filter and the control filter. First, the two reference signal data blocks in the time domain are connected in parallel, and the all-0 data block of point L is inserted before the error signal data block, and the reference data and error data of the 2L points are converted to the frequency domain by FFT. Then, the multiple coherence coefficients are calculated according to formula (16), and substituted into formula (20) to update the medium filter. After the medium filter is updated, it is substituted into formula (11) to calculate the time domain control filter weights, thereby forming a time-frequency domain hybrid control structure without delay. In the early stage of control, the coherence coefficients of the error signal and the reference signal are large, and a large step size is used to ensure rapid convergence of the algorithm; in the later stage of control, the coherence coefficients of the error signal and the reference signal are small, and a small step size is used to reduce the steady-state error of the system. When there is interference noise in the system, the coherence between the noisy error signal and the reference signal deteriorates, and a smaller step size constrains the weight update amount of the control filter, thereby improving the robustness of the ARNC system.

[0082] Example

[0083] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. It should be understood that these examples are only used to illustrate the present invention and are not used to limit the scope of the present invention. After reading the present invention, various equivalent forms of modifications to the present invention by those skilled in the art all fall within the scope defined by the claims attached to this application.

[0084] 1. Hardware configuration and algorithm parameter selection of ARNC system

[0085] The vehicle model used in this experiment is Changan UNI-K. A total of 4 3-axis acceleration sensors are placed on the vehicle chassis to pick up 12 reference signals. The acceleration sensors are placed as follows: Figure 3 As shown in (a); the active noise reduction headrest consists of a left ear speaker and a right ear speaker, a left ear microphone and a right ear microphone that simulate the noise reduction point of the human ear, as shown in Figure 3As shown in (b); Texas Instruments' TMS320C6678 digital signal processor is used for noise control in the area to be reduced in the cabin. The sampling frequency is 2500Hz, the lengths of the control filter and the secondary path are both 256 orders, and the number of data frames T and the number of iterations P are both set to 10. Based on the 50% overlap retention method, the FFT size is set to 512, and the A-weighted (dBA) noise reduction amount of 20-1250Hz is statistically used as an evaluation indicator of the algorithm performance. The algorithm performances of the existing DWFDFeLMS algorithm that does not use coherent constraint stability protection and the C-DWFDFeLMS algorithm of the present invention that uses coherent constraint stability protection are tested respectively, and the results are compared with the Wiener solution (Wiener) under the current driving conditions.

[0086] 2. Real vehicle road noise control performance test in steady-state scenario

[0087] Under the steady-state condition of 50 km / h, the update step size β of the DWFDFeLMS algorithm and the C-DWFDFeLMS algorithm is set to 0.3. The 30-second data of the error microphone with ARNC turned off and on are recorded as the road noise signal before control (Primarynoise) and the error signal after control. The noise reduction curves of different algorithms are statistically shown as follows: Figure 4 The results show that due to the use of the coherent constraint method, the proposed method can quickly converge to the Wiener solution in about 11 seconds, with a noise reduction of 5.3dBA for the left ear and 5.0dBA for the right ear; however, the DWFDFeLMS algorithm cannot converge to the Wiener solution in the later stage due to its large step size. The step size change curve of the C-DWFDFeLMS algorithm is shown in Figure 5 As shown in Figure 2, the step size gradually decreases as the system converges. The coherent constraint method ensures that a large step size is used in the early stage to speed up the convergence of the system, and a small step size is used in the later stage to reduce the steady-state error of the system. In addition, Figure 4 It can also be seen from the bumpy road conditions marked by the arrows that the C-DWFDFeLMS algorithm has better stability than the DWFDFeLMS algorithm.

[0088] 3. Real vehicle road noise control performance test in noisy scenes

[0089] To construct a real noisy driving scenario, a 10-second speech signal was added at 20 seconds of driving at a speed of 50 km / h to evaluate the stability of different algorithms. The update step size β of the DWFDFeLMS algorithm and the C-DWFDFeLMS algorithm was set to 0.2. 50 seconds of data with ARNC turned off and on at the error microphone were recorded as the road noise signal before control and the error signal after control, and the noise reduction curves of different algorithms were statistically analyzed as follows: Figure 6As shown in (a) and (b) of the figure, it can be seen that the control performance of the DWFDFeLMS algorithm deteriorates after the noise is removed, while the C-DWFDFeLMS algorithm can quickly recover to a better control level and has better stability. The step size change curve of the C-DWFDFeLMS algorithm is shown in Figure 7 The introduction of incoherent noise deteriorates the coherence between the noisy error signal and the reference signal. The C-DWFDFeLMS algorithm uses a small step size to limit the impact of noise on the control filter update. When the noise is removed, the coherence is restored, and the C-DWFDFeLMS algorithm uses a larger step size to quickly converge to the Wiener solution.

Claims

1. A robust active road noise control method based on coherence constraints, characterized in that: The method comprises the following steps: Step 1, hardware configuration of the active road noise control system, including placing an acceleration sensor on the vehicle chassis to collect road vibration information as a reference signal; installing an active noise reduction headrest including an error microphone and a cancellation speaker at the head of the passenger, wherein the error microphone is used to collect error signals at both ears; Step 2, calculate the multiple coherence coefficients of the error signal and the reference signal in real time, and use the multiple coherence coefficients as the step size constraint factor for controlling the filter update; the coherence coefficients of the error signal and the reference signal in the early stage of control are large, and a large step size is used to ensure that the method converges quickly; the coherence coefficients in the later stage of control are small, and a small step size is used to reduce the steady-state error of the system; the introduction of interference noise in noisy scenes reduces the signal coherence, and a small step size is used to improve the robustness of the system; Step 3, updating the control filter by combining the frequency domain filtering error minimum mean square algorithm through secondary path decomposition and reference signal whitening; Step 4: Convert the control filter to the time domain, and output the control signal in real time to drive the cancellation speaker to emit sound. After propagating through the secondary path, it is coherently superimposed with the road noise signal at the error microphone, generating a quiet zone near the human ear to achieve the noise reduction function.

2. The method for robust active road noise control based on coherence constraints according to claim 1, characterized in that: In step 2, the multiple coherence coefficients of the error signal and the reference signal are calculated in real time, specifically including: Define 2L discrete frequency points k=0, 1, ..., 2L-1, and the calculation formula of the multi-coherence coefficient is: is the cross-spectral density vector of the mth error signal and the reference signal, P xx (k) is the spectral density matrix of the reference signal, is the spectral density of the mth error signal; superscript -1 and H They represent the inversion operation and conjugate transpose operation respectively.

3. The method for robust active road noise control based on coherence constraints as claimed in claim 2, characterized in that: In step 2, using the multiple coherence coefficients as a step size constraint factor for controlling filter update specifically includes: Calculate the mean of M multi-coherence coefficients to obtain the step size constraint factor for controlling filter update: At this time, the update step size of the control filter can be written as Among them, β is a constant.

4. The method for robust active road noise control based on coherence constraints according to claim 1, characterized in that: In step 3, the secondary path decomposition and reference signal whitening are performed, specifically including: Decompose the secondary path into all-pass parts S all (k) and the minimum phase part S min (k), and use the all-pass part to filter the error signal to obtain the filtered error signal Where e(k) is the fast Fourier transform of the error signal at 2L discrete frequency points; The spectral density matrix P of the reference signal xx (k) Perform spectral factor decomposition to obtain the spectral density function F containing the reference signal characteristics min (k), and use F min The inverse matrix of (k) is used to filter the reference signal to obtain the whitened reference signal Where x(k) is the fast Fourier transform of 2L discrete frequency points of the reference signal.

5. The method for robust active road noise control based on coherence constraints as claimed in claim 4, characterized in that: In step 3, the control filter is updated by combining the frequency domain filtering error minimum mean square algorithm, which specifically includes: Combined with the frequency domain filtering error minimum mean square algorithm, the update formula of the frequency domain medium filter is obtained as follows: In the above formula, p is the number of iterations, T is the number of data frames used for each update, is the filtered error signal of the tth frame, v t (k) is the whitening reference signal of the tth frame; when p = 0, Ψ 0 (k) = Ψ old (k) When p>0, is the filtered residual signal of the t-th frame data at the p-th iteration: After completing P iterations, there is Ψ P+1 (k) = Ψ new (k); At this time, according to Ψ new (k) The update formula of the time domain control filter is obtained as follows: Among them, IFFT{·} + represents the causal part of the inverse fast Fourier transform.

Images