A vehicle noise reduction method for broadband noise

Abstract

The present invention discloses a vehicle noise reduction method for broadband noise. The vehicle noise reduction method: collect a reference signal of the vehicle road noise at each sampling moment, and record the residual noise signal at each sampling moment in the area where noise reduction is required as an error signal. If the data accumulations of the reference signal and the error signal reach N respectively, then execute the frequency domain calculation step, and update the filter parameters obtained in the frequency domain to the time domain. In the frequency domain calculation step, perform a pseudo-inverse calculation on the two-dimensional matrix constructed by the reference signal at each frequency point, and then obtain the gradient vector in the frequency domain at each frequency point; or, calculate the pseudo-inverse of the secondary channel function matrix offline, and then calculate the pseudo-inverse of the reference signal. The present invention can control the broadband noise caused by the friction between the vehicle tire and the road surface, reduce the in-vehicle noise pollution and has a relatively fast convergence speed.

Description

Technical Field

[0001] The present invention belongs to the field of vehicle noise control, relates to the field of vehicle active noise reduction, and specifically is a vehicle noise reduction method for broadband noise. Background Art

[0002] With the development of modern industry, the problem of noise pollution has attracted more and more attention, and high-intensity noise signals also affect the comfort of listeners. Due to the effect of acoustic masking, it is necessary to increase the volume to obtain a higher signal-to-noise ratio for a clear listening effect. The long-term continuous high sound pressure brought about by this will cause irreversible damage to hearing. With the improvement of vehicle intelligence, the requirements of passengers and drivers for the in-vehicle acoustic environment are becoming increasingly strict. In-vehicle noise will reduce the comfort of passengers and drivers, cause irritability and fatigue of in-vehicle occupants; it will also affect the clarity of communication calls, and even affect the driver's perception of external signal sounds of the vehicle, increasing traffic hazards. Automotive NVH (Noise, Vibration, Harshness) is an important issue that vehicle manufacturers are concerned about. Modifying the structural design, adding damping materials or using shock-absorbing springs and other devices to reduce noise are collectively referred to as passive noise control; this method has a relatively good noise reduction effect on medium and high-frequency noises. However, this method has a relatively poor effect on low frequencies, especially the road noise caused by the collision and friction between the road surface and the tires, which is often concentrated in the low frequencies. In addition, passive noise control requires a long tuning time and is difficult to control costs. The active noise reduction solution uses the in-vehicle audio system to construct an anti-signal of the noise signal, form a secondary sound wave, cancel the noise in the target area, reduce noise pollution, and improve the subjective listening comfort. However, it hardly adds extra weight to the vehicle, which helps to reduce exhaust emissions and is a green and energy-saving solution.

[0003] The FxLMS (Filtered-x Least Mean Square) algorithm is a commonly used algorithm in active noise control. Because its consumption of computing resources is small and the algorithm robustness is good, it is widely applied. FxLMS is a special application of the gradient descent method in active noise control, mainly searching through the negative gradient direction of the objective function combined with the step size, and its convergence speed is slow. The Newton method has a second-order convergence speed and only needs very few iteration steps to approach convergence. For example, Patent CN106990800B proposes a narrowband adaptive method based on a pre-stage cross-band filter bank structure, which introduces a multi-channel FxNewton algorithm for narrowband situations, but this method cannot be applied to broadband noise control and cannot be used in the active control of vehicle road noise. Summary of the Invention

[0004] The object of the present invention is to provide a vehicle noise reduction method for broadband noise, which can control the broadband noise caused by the friction between vehicle tires and the road surface, reduce the noise pollution in the vehicle and has a relatively fast convergence speed.

[0005] According to the first aspect of the present invention, a vehicle noise reduction method for broadband noise includes the following steps:

[0006] Collect the J-channel reference signals of the vehicle road noise at each sampling moment, denoted as x j (n), j = 1, 2,..., J, where J is the number of channels of the reference signal and n represents the sampling moment;

[0007] Generate a control signal according to the filter control coefficient at the current moment and the multi-channel reference signal, and feed it to the sound reproduction device in the corresponding noise reduction area of the vehicle;

[0008] Collect the residual noise signals at each sampling moment in the noise reduction area, denoted as the error signal e m (n), m = 1, 2,..., M;

[0009] The vehicle road noise control method further includes: if the data accumulations of the reference signal and the error signal reach N respectively, then execute the frequency domain calculation step, and the frequency domain calculation step includes:

[0010] S101, generate the reference signal X in the frequency domain j , where X j = [X j (0), X j (1),... X j (2N - 1)] T , X j (0), X j (1), X j (2N - 1) respectively represent the characteristics of the corresponding reference signals at the 1st, 2nd, and 2Nth analysis frequencies;

[0011] S102, generate the error signal E in the frequency domain m , where E m = [E m (0), E m (1),... E m (2N - 1)] T , E m (0), E m (1), E m (2N - 1) respectively represent the characteristics of the corresponding error signals at the 1st, 2nd, and 2Nth analysis frequencies;

[0012] S103, generate the filtered reference signal as shown in the following formula

[0013]

[0014] wherein, S l，m =[S l，m (0), S l，m (1), … S l，m (2N - 1)] T , l = 1, 2, …, L; m = 1, 2, …, M, S l，m represents the transfer function between the l-th loudspeaker and the m-th residual noise signal acquisition device, and M is the number of residual noise signal acquisition devices;

[0015] S104. Transform the frequency-domain filtered reference signal into the time domain, intercept the first N points, and denote them as

[0016]

[0017] wherein, respectively represent the filtered reference signals at the first, second, …, N-th sampling time points;

[0018] S105. Re-transform to obtain the frequency-domain filtered reference signal

[0019] S106. For each frequency point, construct a two-dimensional matrix of the filtered reference signal

[0020]

[0021] wherein, ∈R J·L×M represents a matrix with JL rows and M columns;

[0022] S107. For each frequency point, generate a frequency-domain gradient vector,

[0023] Φ(k) = μ · [conj(U(k))] -1 · Err(k) ∈R J·L×1 k = 0, 1, …, 2N - 1;

[0024] wherein, Err(k) = [E 1 (k), E 2 (k), … E M (k)] T , E 1 (k), E 2 (k), E M (k) are the frequency-domain error signals of the first, second, and M-th noise signal channels respectively; conj(·) represents taking its conjugate; μ is the convergence factor;

[0025] Among them, the elements of Φ(k) are defined as Φ(k) = [Ψ 1，1 (k), Ψ 1，2 (k), … Ψ J，L (k)] T ; The gradient increment of each channel, denoted as

[0026] Ψ j，l = [Ψ j，l (0), Ψ j，l (1), … Ψ j，l (k) … Ψ j，l (2N - 1)] T ;

[0027] S108. Generate a gradient vector in the time domain according to the gradient vector Ψ j，l in the frequency domain. After transforming the numerical values in the frequency domain to the time domain, intercept the first N points, which is

[0028] S109. Regenerate the gradient vector in the frequency domain according to

[0029] S110. Update the filter control parameter in the frequency domain according to the following formula

[0030] W j，l，new = W j，l，old + Ψ j，l

[0031] Among them, W j，l，old , W j，l，new are the filter control parameters in the frequency domain before and after update respectively;

[0032] S111. Transform to obtain the filter control coefficient w j，l，new in the time domain according to the filter control parameter W j，l .

[0033] In a preferred embodiment, in step S101, the reference signal in the frequency domain is generated according to the following formula

[0034]

[0035] Among them, FFT represents the Fourier transform, x j (n) = [x j (n - N + 1), …, x j (n - 1), x j (n)] T , x j (n + N) represents the numerical value of the reference signal at the (n + N) sampling moment and the previous N time sampling moments.

[0036] ​In a preferred embodiment, in step S102, an error signal in the frequency domain is generated by the following formula:

[0037]

[0038] where e m (n)=[e m (n - N + 1),..., e m (n - 1), e m (n)] T .

[0039] In a preferred embodiment, in step S104, the frequency-filtered reference signal is transformed into the time domain by the following formula:

[0040]

[0041] In a preferred embodiment, in step S105, the filtered reference signal in the frequency domain is re-transformed according to the following formula as shown in the following formula

[0042]

[0043] where represents the vector composed of the sampling points of the filtered reference signal transformed into the time domain, its length is N, and the initial sampling time is n.

[0044] In a preferred embodiment, in step S108, a gradient vector in the time domain is generated according to the following formula:

[0045]

[0046] In a preferred embodiment, in step S109, the gradient vector in the frequency domain is re-generated according to the following formula:

[0047]

[0048] In a preferred embodiment, in step S111, the filter control coefficient w j，l is as shown in the following formula

[0049] w j，l =G·IFFT[W j，l，new , j = 1, 2,..., J; l = 1, 2,..., L;

[0050] where IN represents an N×N identity matrix, its diagonal elements are 1, and the rest of the elements are 0; 0 N represents an N×N matrix, all of whose elements are 0; IFFT represents the inverse Fourier transform;

[0051] Take the first N points, wj，l = [w j，l (1), w j，l (2), … w j，l (N)] T 。

[0052] In a preferred embodiment, the control signal is generated according to the following formula:

[0053]

[0054] where w j，l (i) is the filter coefficient at the current moment.

[0055] In a preferred embodiment, the frequency-domain calculation step is performed within a time period of N sampling time points. If the data of the reference signal and the error signal reach N, the counter is cleared, and then the filter parameters obtained in the frequency domain are updated to the time domain through the frequency-domain calculation step, and a new round of frequency-domain calculation is started; if not, the vehicle road noise and the error signal are continuously collected, and the counter continues to accumulate.

[0056] In a preferred embodiment, the vibration signal generated by the friction between the wheel and the road surface is collected by a vibration sensor as the reference signal.

[0057] More preferably, the vibration sensor is disposed on the vehicle floor.

[0058] In a preferred embodiment, the noise signal generated by the friction between the wheel and the road surface is collected by a first microphone as the reference signal.

[0059] More preferably, the first microphone is disposed at a position adjacent to the wheel of the vehicle.

[0060] In a preferred embodiment, the sound reproduction device includes an in-vehicle speaker disposed in the vehicle compartment. The in-vehicle speaker is disposed in the vehicle compartment or at least radiates sound to the vehicle compartment, including but not limited to: headrest speakers, ceiling speakers, door panel speakers, etc.

[0061] In a preferred embodiment, the error signal acquisition device includes a plurality of second microphones. The sound signal in the vehicle compartment is collected by the plurality of second microphones, and the plurality of second microphones are disposed at a plurality of sampling positions in the vehicle compartment.

[0062] According to the second aspect of the present invention, a vehicle noise reduction method for broadband noise includes the following steps:

[0063] Collect the J-channel reference signals of the vehicle road noise at each sampling moment, denoted as x j (n), j = 1, 2, …, J, where J is the number of channels of the reference signal, and n represents the sampling moment;

[0064] Generate a control signal based on the filter control coefficient at the current moment and the multi-channel reference signal, and feed it to the sound reproduction device in the corresponding noise reduction area of the vehicle;

[0065] Collect the residual noise signal at each sampling moment in the area where noise reduction is required, denoted as the error signal e m (n), m = 1, 2,..., M;

[0066] The vehicle road noise control method further includes: if the data accumulations of the reference signal and the error signal reach N respectively, then perform a frequency domain calculation step, and the frequency domain calculation step includes:

[0067] S101. Generate a reference signal X in the frequency domain j , where X j = [X j (0), X j (1),... X j (2N - 1)] T , X j (0), X j (1), X j (2N - 1) respectively represent the characteristics of the corresponding reference signals at the 1st, 2nd, and 2Nth analysis frequencies;

[0068] S102. Generate the pseudo-inverse signal X -1 (k) of the reference signal, where

[0069] S103. Generate an error signal E in the frequency domain m , where E m = [E m (0), E m (1),... E m (2N - 1)] T , E m (0), E m (1), E m (2N - 1) respectively represent the characteristics of the corresponding error signals at the 1st, 2nd, and 2Nth analysis frequencies;

[0070] S104. Generate the reference signal after inverse filtering as shown in the following formula

[0071]

[0072] where represents the frequency domain pseudo-inverse model of the transfer function between the lth speaker and the mth residual noise signal acquisition device, and M is the number of residual noise signal acquisition devices;

[0073] S105. Generate the gradient vector Ψ in the frequency domain for each frequency point, j，l , as shown in the following formula,

[0074]

[0075] where μ is the convergence factor;

[0076] The gradient increment Ψ of each channel j，l is as follows,

[0077] Ψ j，l = [Ψ j，l (0), Ψ j，l (1), … Ψ j，l (k) … Ψ j，l (2N - 1)] T ;

[0078] S106. Generate the gradient vector in the time domain according to the gradient vector Ψ in the frequency domain. After transforming the numerical values in the frequency domain to the time domain, the first N points are intercepted, which is j，l S107. Regenerate the gradient vector in the frequency domain according to

[0079] S108. Update the filter control parameter in the frequency domain according to the following formula,

[0080]

[0081] W j，l，new = W j，l，old + Ψ j，l

[0082]

[0083] where W j，l，old , W j，l，new are the filter control parameters in the frequency domain before and after update respectively;

[0084] S109. Transform the filter control parameter W in the frequency domain to obtain the filter control coefficient w in the time domain j，l，ne w j，l .

[0084] In a preferred embodiment, in step S101, the reference signal in the frequency domain is generated by the following formula,

[0085]

[0086] where FFT represents the Fourier transform, x j (n) = [x j (n - N + 1), …, x j (n - 1), x j (n)]T , x j (n + N) represents the values of the reference signal at the (n + N)-th sampling moment and the previous N time sampling moments.

[0087] In a preferred embodiment, in step S102, the pseudo-inverse signal of the reference signal is calculated according to the following formula

[0088]

[0089] In a preferred embodiment, in step S103, the error signal in the frequency domain is generated by the following formula

[0090]

[0091] where, e m (n) = [e m (n - N + 1), …, e m (n - 1), e m (n)] T .

[0092] In a preferred embodiment, in step S106, the gradient vector in the time domain is generated according to the following formula

[0093]

[0094] In a preferred embodiment, in step S107, the gradient vector in the frequency domain is regenerated according to the following formula

[0095]

[0096] In a preferred embodiment, in step S109, the filter control coefficient w j，l is as shown in the following formula

[0097] w j，l = G · IFFT[W j，l，new , j = 1, 2, …, J; l = 1, 2, …, L;

[0098] where, I N represents an N×N identity matrix with diagonal elements being 1 and the remaining elements being 0; 0 N represents an N×N matrix with all elements being 0; IFFT represents the inverse Fourier transform;

[0099] The first N points are intercepted, W j，l = [w j，l (1), w j，l (2), … w j，l (N)] T .

[0100] In a preferred embodiment, the control signal is generated according to the following formula:

[0101]

[0102] where w j，l (i) is the filter coefficient at the current moment.

[0103] In a preferred embodiment, the frequency-domain calculation step is performed within a time period of N sampling time points. If the data of the reference signal and the error signal reach N, the counter is cleared, and then the filter parameters obtained in the frequency domain are updated to the time domain through the frequency-domain calculation step, and a new round of frequency-domain calculation is started; if not, the vehicle road noise and the error signal are continuously collected, and the counter continues to accumulate.

[0104] In a preferred embodiment, the vibration signal generated by the friction between the wheel and the road surface is collected by a vibration sensor as the reference signal.

[0105] More preferably, the vibration sensor is disposed on the vehicle floor.

[0106] In a preferred embodiment, the noise signal generated by the friction between the wheel and the road surface is collected by a first microphone as the reference signal.

[0107] More preferably, the first microphone is disposed at a position adjacent to the wheel of the vehicle.

[0108] In a preferred embodiment, the sound reproduction device includes an in-vehicle speaker disposed in the vehicle compartment. The in-vehicle speaker is arranged in the vehicle compartment or at least radiates sound to the vehicle compartment, including but not limited to: headrest speakers, ceiling speakers, door panel speakers, etc.

[0109] In a preferred embodiment, the error signal acquisition device includes a plurality of second microphones. The sound signals in the vehicle compartment are collected by the plurality of second microphones, and the plurality of second microphones are arranged at a plurality of sampling positions in the vehicle compartment.

[0110] The present invention adopts the above scheme and has the following advantages compared with the prior art:

[0111] For the vehicle noise reduction method of the present invention, aiming at broadband noises such as road noise caused by the friction between the tire and the road surface, the multi-channel FxNewton algorithm in the time-frequency domain is adopted, which has a faster convergence speed than the traditional FxLMS algorithm in the time-frequency domain, and the noise reduction performance is equivalent to or even better than that of the traditional algorithm, with a larger noise reduction amount, can effectively control the noise in the vehicle caused by the friction between the tire and the road surface, and has a good noise reduction effect. Description of the Drawings

[0112] To more clearly illustrate the technical solution of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0113] Figure 1 It is a flowchart of the time-domain part of the multi-channel FxNewton algorithm according to Embodiment 1 of the present invention.

[0114] Figure 2 It is a flowchart of the frequency-domain part of the multi-channel FxNewton algorithm according to Embodiment 1 of the present invention.

[0115] Figure 3 It is a block diagram of the multi-channel FxNewton algorithm according to Embodiment 1 of the present invention.

[0116] Figure 4 It is a comparison chart of the noise reduction performance of the traditional multi-channel FxLMS and the multi-channel FxNewton algorithm according to Embodiment 1 at position 1.

[0117] Figure 5 It is a comparison chart of the noise reduction performance of the traditional multi-channel FxLMS and the multi-channel FxNewton algorithm according to Embodiment 1 at position 4.

[0118] Figure 6 It is a flowchart of the frequency-domain part of the multi-channel FxNewton algorithm according to Embodiment 2 of the present invention.

[0119] Figure 7 It is a block diagram of the multi-channel FxNewton algorithm according to Embodiment 2 of the present invention. Detailed implementation manners

[0120] The following will elaborate on the preferred embodiments of the present invention in conjunction with the drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art. It should be noted here that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation to the present invention.

[0121] This embodiment provides a vehicle noise reduction method for broadband noise, which is mainly road noise caused by the friction between vehicle tires and the road surface. This method uses a multi-channel algorithm (MIMO, Multiple Inputs Multiple Outputs) for controlling road noise. Specifically, it is the FxNewton algorithm in the case of MIMO. Especially considering the computational complexity of the algorithm, the execution of the algorithm in the frequency domain can effectively reduce the consumption of computing resources. Also considering the delay introduced by frequency domain calculation, a multi-channel normalization algorithm combining time and frequency domains, namely the FxNLMS algorithm, is proposed. This algorithm includes two parts, one is time domain calculation and the other is frequency domain calculation. Combining Figures 1 to 3 as shown, this method is specifically described as follows.

[0122] Figure 1 is the time domain calculation step of the algorithm. The calculation is performed at each sampling time point.

[0123] A. Reference signal acquisition: At each sampling moment n, a reference signal is acquired from a sensor. For example, the vibration signal of the vehicle floor is acquired from a vibration sensor, or the acoustic signal is acquired using the first microphone near the wheel, etc. The vibration signal acquired by the vibration sensor or the noise signal acquired by the first microphone is used as the reference signal for the road noise to be reduced. There are a total of J channels of reference signals, denoted as x j (n), j = 1, 2,..., J.

[0124] B. Error signal acquisition: At each sampling moment n, the residual noise signal is acquired from a microphone. There are a total of M microphone signals, denoted as e m (n), m = 1, 2,..., M.

[0125] C. Control signal generation: As shown in the following formula, according to the parameter w j，l (n) at the current moment and the reference signal obtained in the previous step, a control signal y l (n) is generated and fed to an acoustic playback unit such as a speaker

[0126]

[0127] where the number of speaker channels is L and the order of the adaptive filter is N. w j，l (i) represents the control coefficient at the current moment of the filter whose input is the jth reference signal and the output corresponds to the lth control sound source (in-vehicle speaker).

[0128] Judge whether the data accumulation of the reference signal and the error signal has N. If so, clear the counter, update the filter coefficients obtained in the frequency domain to the time domain, and start a new round of frequency domain calculation; if not, continue to accumulate the counter and continue the above steps A to C.

[0129] Figure 2 These are the frequency-domain calculation steps of the algorithm. The calculation is performed over a time period of N sampling time points.

[0130] S101. Generate a reference signal in the frequency domain. According to the overlap-save method, perform an FFT operation on 2N data of the reference signal for each channel, specifically expressed as

[0131]

[0132] where x j (n) = [x j (n - N + 1), …, x j (n - 1), x j (n)] T ; X j = [X j (0), X j (1), … X j (2N - 1)] T .

[0133] S102. Generate an error signal in the frequency domain. According to the overlap-save method, perform an FFT calculation on the zero-padded error signal for each channel, specifically expressed as

[0134]

[0135] where e m (n) = [e m (n - N + 1), …, e m (n - 1), e m (n)] T ; E m = [E m (0), E m (1), … E m (2N - 1)] T .

[0136] S103. Generate a filtered reference signal. An important step in the FxNewton algorithm is to filter the reference signal. It is generally considered that the transfer function of the secondary channel includes the transfer path of the digital control signal y(n) through the DAC module, analog filter, power amplifier module, speaker, spatial propagation of sound waves, microphone, analog filter, and ADC module. The transfer function S of the secondary channel is obtained through online and offline system identification methods and is a digital filter of length N. Here, considering its frequency-domain form is expressed as S l，m = [S l，m (0), S l，m (1), … S l，m(2N - 1)] T , l = 1, 2, …, L; m = 1, 2, …, M, S l，m represents the transfer function between the l-th speaker and the m-th microphone. M is the number of microphones. The filtered reference signal is calculated as

[0137]

[0138] where k represents the k-th frequency point. Since the length of the FFT is 2N, there are 2N frequency points;

[0139] S104. Transform the frequency-domain filtered reference signal into the time domain and intercept the first N points

[0140]

[0141] where

[0142] S105. Re-transform to obtain the filtered reference signal in the frequency domain

[0143]

[0144] S106. For each frequency point, construct a two-dimensional matrix from the filtered reference signal

[0145]

[0146] where ∈R J·L×M represents a matrix with JL rows and M columns.

[0147] S107. For each frequency point, generate a gradient vector in the frequency domain

[0148] Φ(k) = μ · [conj(U(k))] -1 · Err(k) ∈R J·L×1 k = 0, 1, …, 2N - 1;

[0149] where Err(k) = [E 1 (k), E 2 (k), … E M (k)] T ; conj(·) represents taking its conjugate; μ is the convergence factor, which is a constant and its value is usually between 0 and 2.

[0150] It should also be noted as follows: (1) [·] -1 represents the Moore-Penrose pseudoinverse here; there are many conventional algorithms for solving this inverse, which will not be elaborated here; (2) The elements of Φ(k) are defined as Φ(k) = [Ψ1，1 (k), Ψ 1，2 (k), …Ψ J，L (k)] T ; Meanwhile, the gradient increment from each channel, denoted as Ψ j，l = [Ψ j，l (0), Ψ j，l (1), …Ψ j，l (k) …Ψ j，l (2N - 1)] T ; The elements are reorganized from the perspective of the frequency domain and the number of channels.

[0151] S108. Generate the gradient vector in the time domain. After transforming the numerical values in the frequency domain to the time domain, intercept the first N points

[0152]

[0153] Among them,

[0154] S109. Regenerate the gradient vector in the frequency domain

[0155]

[0156] S110. Update the filter control parameters in the frequency domain

[0157] W j，l，new = W j，l，old + Ψ j，l j = 1, 2, …, J; l = 1, 2, …, L;

[0158] Among them, W j，l = [W j，l (0), W j，l (1), …W j，l (2N - 1)] T .

[0159] S111. Transform to obtain the filter control coefficients in the time domain and intercept the first N points

[0160] w j，l = G·IFFT[W j，l，new j = 1, 2, …, J; l = 1, 2, …, L;

[0161] Among them, w j，l = [w j，l (1), w j，l (2), …w j，l (N)] T .

[0162] Simulation Example 1

[0163] The convergence performance of the algorithm was simulated. In the simulation experiment, the target noise adopted was a broadband signal with a frequency band covering 30 Hz - 400 Hz, which was the typical frequency band distribution of road noise. The noise signal was generated by passing white noise through a band - pass filter. The number of channels of the reference signal was set to K = 3, the number of speakers was set to L = 4, and the number of error microphones was set to M = 4. The transfer function between the speakers and the microphones, which was also the transfer function of the above - mentioned secondary channel, was obtained by in - vehicle acquisition. In the simulation experiment, the variation relationships of the noise energy before and after active noise control with the number of iterations (which also corresponded to time) were compared respectively. More importantly, the traditional multi - channel FxLMS algorithm and the multi - channel FxNewton algorithm of this embodiment were compared.

[0164] Figure 4 The variation relationship of the residual noise signal amplitude at the first position with the number of iterations was given. As can be seen from Figure 4 it, the traditional FxLMS algorithm had a certain noise reduction effect; for the multi - channel FxNewton algorithm of this embodiment, the algorithm converged faster and the final noise reduction amount was almost the same.

[0165] Figure 5 The variation relationship of the residual noise signal amplitude at the fourth position with the number of iterations was given. As can be seen from Figure 5 it, the traditional FxLMS algorithm had almost no noise reduction effect; for the multi - channel FxNewton algorithm of this embodiment, it had an obvious noise reduction effect, a fast convergence speed, and a large noise reduction amount.

[0166] Example 2

[0167] In step S107 of Embodiment 1, it was necessary to calculate the Moore - Penrose pseudoinverse of a matrix of size J·L×M in real time. If the number of channels was relatively large, this was a relatively complex calculation because it involved matrix inversion. In active noise control, the transfer function of the secondary channel was generally measured offline. Therefore, in this embodiment, the Moore - Penrose pseudoinverse of the secondary channel function matrix could be calculated offline, and then the pseudoinverse of the reference signal was calculated, and the calculation was relatively simple. Especially when the size of the reference signal was a vector of J×1, Embodiment 2 calculated its pseudoinverse with a simpler calculation formula, which will be introduced in detail below.

[0168] The time - domain calculation steps of the algorithm in this embodiment were the same as those in Embodiment 1. Referring to Figure 1 shown, the time - domain calculation was performed at each sampling time point.

[0169] A. Reference signal acquisition: At each sampling moment n, the reference signal was collected from the sensor. For example, the vibration signal was collected from the vibration sensor, or the sound signal was collected from the microphone, etc. There were a total of J channels of reference signals, denoted as x j(n), j = 1, 2, …, J.

[0170] B. Error signal acquisition: At each sampling time n, the residual noise signal is collected from the microphone. There are M microphone signals, denoted as e m (n), m = 1, 2, …, M.

[0171] C. Control signal generation: As shown in the following formula, according to the filter control coefficient w j，l (n) at the current moment and the reference signal obtained in the previous step, a control signal y l (n) is generated and fed to a sound reproduction unit such as a speaker

[0172]

[0173] where the number of speaker channels is L and the order of the adaptive filter is N. w j，l (i) indicates that the input of the filter is the jth reference signal, and the output corresponds to the control coefficient at the current moment of the filter for the lth control sound source.

[0174] Judge whether the data accumulation of the reference signal and the error signal has N. If so, clear the counter, update the filter coefficients obtained in the frequency domain to the time domain, and start a new round of frequency domain calculation; if not, continue to accumulate the counter and continue the above steps A to C.

[0175] Figure 6 This is the frequency domain calculation step of the algorithm in this embodiment. The calculation is performed within a time period of N sampling time points.

[0176] S101. Generate a reference signal in the frequency domain. According to the overlap - save method, perform FFT operation on 2N data of the reference signal for each channel, specifically expressed as

[0177]

[0178] where x j (n)=[x j (n - N + 1), …, x j (n - 1), x j (n)] T ; X j =[X j (0), X j (1), … X j (2N - 1)] T .

[0179] S102. Generate the pseudo - inverse signal of the reference signal: For each frequency point k, the reference signal is expressed as X(k)=[X 1 (k), X 2(k), …X J (k)] T , calculate its Moore-Penrose pseudo-inverse, and this calculation needs to be performed in real time. Specifically, the pseudo-inverse calculation of the reference signal is expressed as

[0180]

[0181] wherein, the pseudo-inverse of the reference signal is also denoted as

[0182] S103. Generate an error signal in the frequency domain. According to the overlap-save method, zero-padding is performed on the error signal of each channel and then FFT calculation is carried out, which is specifically expressed as

[0183]

[0184] wherein, e m (n)=[e m (n - N + 1), …, e m (n - 1), e m (n)] T ; E m =[E m (0), E m (1), …E m (2N - 1)] T .

[0185] S104. Generate a reference signal after inverse filtering. An important step in the FxNewton algorithm is to filter the reference signal. Generally, it is considered that the transfer function of the secondary channel includes the transfer path of the digital control signal y(n) passing through the DAC module, analog filter, power amplifier module, speaker, spatial propagation of sound waves, microphone, analog filter, and ADC module. The transfer function S of the secondary channel is obtained through on-line and off-line system identification methods and is a digital filter with a length of N. Here, instead of directly using this transfer function to generate the filtered reference signal, its Moore-Penrose pseudo-inverse is used, which can be calculated offline and does not require real-time computing resources. In many computing tools such as matlab, the Moore-Penrose pseudo-inverse of a signal can be directly calculated, and no more introduction will be made here. The frequency-domain pseudo-inverse model of the transfer function of the channel is denoted as M is the number of microphones. The reference signal after inverse filtering is calculated as

[0186]

[0187] wherein, k represents the kth frequency point. Since the length of the FFT is 2N, there are 2N frequency points;

[0188] S105. Generate a gradient vector in the frequency domain for each frequency point.

[0189]

[0190] where μ is the convergence factor, a constant, and its value usually ranges from 0 to 2. The gradient increment for each channel is denoted as Ψ. j，l = [Ψ j，l (0), Ψ j，l (1), … Ψ j，l (k) … Ψ j，l (2N - 1)] T .

[0191] S106. Generate a gradient vector in the time domain. After transforming the numerical values in the frequency domain to the time domain, intercept the first N points.

[0192]

[0193] where,

[0194] S107. Regenerate the gradient vector in the frequency domain.

[0195]

[0196] S108. Update the filter control parameters in the frequency domain.

[0197] W j，l，new = W j，l，old + Ψ j，l j = 1, 2, …, J; l = 1, 2, …, L;

[0198] where, W j，l = [W j，l (0), W j，l (1), … W j，l (2N - 1)] T .

[0199] S109. Transform to obtain the filter control coefficients in the time domain and intercept the first N points.

[0200] w j，l = G · IFFT[W j，l，new j = 1, 2, …, J; l = 1, 2, …, L;

[0201] where, w j，l = [w j，l (1), w j，l (2), … w j，l (N)] T .

[0202] Figure 7 A block diagram of the algorithm of this embodiment is shown. The differences from the traditional algorithm include at least the inverse processing of the reference signal and the inverse model of the signal passing through the secondary channel function.

[0203] Those skilled in the art of this technology can understand that unless specifically stated, the singular forms "a", "an", "the", and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of this application means the presence of features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or their groups.

[0204] The above embodiments are only for illustrating the technical concept and features of the present invention. They are a preferred embodiment, and their purpose is to enable those who are familiar with this technology to understand the content of the present invention and implement it accordingly. However, the protection scope of the present invention cannot be limited thereby. Any equivalent transformation or modification made according to the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A vehicle noise reduction method for broadband noise, comprising the following steps: Collect the J-channel reference signal of the vehicle road noise at each sampling moment, denoted as x j (n), j = 1, 2, …, J, where J is the number of channels of the reference signal, and n represents the sampling moment; Generate a control signal according to the filter control coefficient at the current moment and the reference signal, and feed it to the sound reproduction device in the corresponding noise reduction area of the vehicle; Collect the residual noise signals at each sampling moment in the area that needs noise reduction, denoted as the error signal e m (n), m = 1, 2, …, M; Characterized in that, The vehicle noise reduction method further includes: if the data accumulations of the reference signal and the error signal reach N respectively, perform a frequency domain calculation step, and the frequency domain calculation step includes: S101. Generate a reference signal X in the frequency domain j , where X j = [X j (0), X j (1), … X j (2N - 1)] T , and X j (0), X j (1), and X j (2N - 1) respectively represent the characteristics of the reference signals corresponding to the 1st, 2nd, …, 2Nth analysis frequencies; S102. Generate the error signal E in the frequency domain m , where E m = [E m (0), E m (1), … E m (2N - 1)] T , and E m (0), E m (1), E m (2N - 1) respectively represent the characteristics of the corresponding error signals at the 1st, 2nd, …, 2Nth analysis frequencies; S103. Generate a filtered reference signal As expressed by the following formula where s l,m = [s l,m (0), S l,m (1), … S l,m (2N - 1)] T , l = 1, 2, …, L; m = 1, 2, …, M, S l,m represents the transfer function between the l-th loudspeaker and the m-th residual noise signal acquisition device, and M is the number of residual noise signal acquisition devices; S104. Transform the frequency-domain filtered reference signal into the time domain, intercept the first N points, and denote them as Among them, respectively represent the filtered reference signals at the first, second, and Nth sampling time points; S105. Re-transform to obtain a filtered reference signal in the frequency domain S106. For each frequency point, construct a two-dimensional matrix of the filtered reference signal k = 0, 1, …, 2N - 1; where ∈R J·L×M represents a matrix with JL rows and M columns; S107. For each frequency point, generate a gradient vector in the frequency domain, Φ(k) = μ · [conj(U(k))] -1 · Err(k) ∈ R J·L×1 k = 0, 1, …, 2N - 1; where Err(k) = [E 1 (k), E 2 (k), … E M (k)] T , E 1 (k), E 2 (k), E M (k) are the frequency-domain error signals of the first, second, and Mth noise signal channels, respectively; conj(·) denotes taking its conjugate; μ is the convergence factor; Among them, the elements of Φ(k) are defined as Φ(k) = [Ψ 1,1 (k), Ψ 1,2 (k), … Ψ J,L (k)] T ; The gradient increment of each channel is expressed as Ψ j,l = [Ψ j,l (0), Ψ j,l (1), … Ψ j,l (k) … Ψ j,l (2N - 1)] T ; S108. Generate a gradient vector in the time domain according to the gradient vector Ψ in the frequency domain. After transforming the numerical values in the frequency domain to the time domain, intercept the first N points, which are j,l ​ S109. According to Regenerate the gradient vector in the frequency domain S110. Update the filter control parameters in the frequency domain according to the following formula, W j,l,new = W j,l,old + Ψ j,l Among them, W j,l,old and W j,l,new are the filter control parameters in the frequency domain before and after the update, respectively; S111. Obtain the filter control coefficient w in the time domain by transforming according to the filter control parameter W in the frequency domain j,l,new j,l .​ 2. The vehicle noise reduction method according to claim 1, Characterized in that, In step S101, generate a reference signal in the frequency domain through the following formula, where FFT represents the Fourier transform, x j (n) = [x j (n - N + 1), …, x j (n - 1), x j (n)] T , x j (n + N) represents the value of the reference signal at the sampling time of n + N and the previous N time sampling times; and / or, In step S102, generate an error signal in the frequency domain through the following formula, where, e m (n) = [e m (n - N + 1), …, e m (n - 1), e m (n)] T ; and / or, In step S104, transform the reference signal after frequency filtering into the time domain through the following formula, and / or, In step S105, the filtering reference signal in the frequency domain is re-transformed according to the following formula as shown in the following formula Among them, represents a vector composed of time-domain sampling points obtained by transforming the filtered reference signal, with a length of N and an initial sampling time of n; and / or, In step S108, generate a gradient vector in the time domain according to the following formula, and / or, In step S109, regenerate a gradient vector in the frequency domain according to the following formula, and / or, In step S111, the filter control coefficient w j,l is as shown in the following formula w j,l = G·IFFT[w j,l,new , j = 1, 2, …, J; l = 1, 2, …, L; Among them, I N represents an N×N identity matrix with diagonal elements equal to 1 and the remaining elements equal to 0; 0 N represents an N×N matrix with all elements equal to 0; IFFT represents the inverse Fourier transform; Intercept the first N points, w j,l = [w j,l (1), w j,l (2), … w j,l (N)] T .

3. The vehicle noise reduction method according to claim 1, Characterized in that, Generate the control signal according to the following formula, where w j,l (i) is the filter control coefficient at the current moment.

4. The vehicle noise reduction method according to claim 1, Characterized in that, The frequency domain calculation step is performed within a time period of N sampling time points. If the data of the reference signal and the error signal reach N, the counter is cleared, and then the filter parameters obtained in the frequency domain will be updated to the time domain through the frequency domain calculation step, and a new round of frequency domain calculation will start; if not, continue to collect the vehicle road noise and error signals, and the counter will continue to accumulate.

5. The vehicle noise reduction method according to claim 1, Characterized in that, Collect the vibration signal generated by the friction between the wheel and the road surface as the reference signal through a vibration sensor or a first microphone; and / or, the sound reproduction device includes an in-vehicle speaker arranged in the vehicle compartment; and / or, collect the residual noise signal in the vehicle compartment through a plurality of second microphones, and the plurality of second microphones are arranged at a plurality of sampling positions in the vehicle compartment.

6. A vehicle noise reduction method for broadband noise, comprising the following steps: Collect the J-channel reference signal of the vehicle road noise at each sampling moment, denoted as x j (n), j = 1, 2, …, J, where J is the number of channels of the reference signal, and n represents the sampling moment; Generate a control signal according to the filter control coefficient at the current moment and the reference signal, and feed it to the sound reproduction device in the corresponding noise reduction area of the vehicle; Collect the residual noise signals at each sampling moment in the area that needs noise reduction, denoted as the error signal e m (n), where m = 1, 2, …, M; Characterized in that, The vehicle noise reduction method further includes: if the data accumulations of the reference signal and the error signal reach N respectively, perform a frequency domain calculation step, and the frequency domain calculation step includes: S101. Generate a reference signal X in the frequency domain j , where X j = [X j (0), X j (1), …, X j (2N - 1)] T , and X j (0), X j (1), X j (2N - 1) respectively represent the characteristics of the reference signals corresponding to the 1st, 2nd, …, 2Nth analysis frequencies; S102. Generate the pseudo-inverse signal X -1 (k), where S103. Generate an error signal E in the frequency domain m , where E m = [E m (0), E m (1), … E m (2N - 1)] T , and E m (0), E m (1), E m (2N - 1) respectively represent the characteristics of the corresponding error signals at the 1st, 2nd, …, 2Nth analysis frequencies; S104. Generate a reference signal after inverse filtering As shown in the following formula Among them, represents the frequency-domain pseudo-inverse model of the transfer function between the l-th loudspeaker and the m-th residual noise signal acquisition device, and M is the number of residual noise signal acquisition devices; S105. Generate a gradient vector Ψ in the frequency domain for each frequency point, as shown in the following formula j,l , as shown in the following formula where μ is a convergence factor; The gradient increment Ψ of each channel j,l is as follows Ψ j,l = [Ψ j,l (0), Ψ j,l (1), … Ψ j,l (k) … Ψ j,l (2N - 1)] T ; S106. Generate a gradient vector in the time domain according to the gradient vector Ψ in the frequency domain. After transforming the values in the frequency domain to the time domain, intercept the first N points, which are j,l the gradient vector in the time domain. After transforming the values in the frequency domain to the time domain, intercept the first N points, which are S107. According to Regenerate the gradient vector in the frequency domain S108. Update the filter control parameters in the frequency domain according to the following formula, W j,l,new = W j,l,old + Ψ j,l Among them, W j,l,old and W j,l,new are the filter control parameters in the frequency domain before and after update, respectively. S109. Obtain the filter control coefficient w in the time domain by transforming according to the filter control parameter W in the frequency domain j,l,new as shown below j,l ​ w j,l = G · IFFT[W j,l,new , j = 1, 2, …, J; l = 1, 2, …, L; Among them, I N represents an N×N identity matrix with diagonal elements being 1 and the remaining elements being 0; 0 N represents an N×N matrix with all elements being 0; IFFT represents the inverse Fourier transform.

7. The vehicle noise reduction method according to claim 6, Characterized in that, In step S101, generate a reference signal in the frequency domain through the following formula, Among them, FFT represents the Fourier transform, x j (n) = [x j (n - N + 1), …, x j (n - 1), x j (n)] T , x j (n + N) represents the value of the reference signal at the sampling time of n + N and the previous N time sampling times; and / or, In step S102, calculate the pseudo-inverse signal of the reference signal according to the following formula, and / or, In step S103, generate an error signal in the frequency domain through the following formula, where, e m (n) = [e m (n - N + 1), …, e m (n - 1), e m (n)] T ; and / or, In step S106, generate a gradient vector in the time domain according to the following formula, and / or, In step S107, the gradient vector in the frequency domain is regenerated according to the following formula and / or In step S109, the filter control coefficient w j,l is as shown in the following formula w j,l = G · IFFT[w j,l,new , j = 1, 2, …, J; l = 1, 2, …, L; Among them, I N represents an N×N identity matrix with diagonal elements equal to 1 and the remaining elements equal to 0; 0 N represents an N×N matrix with all elements equal to 0; IFFT represents the inverse Fourier transform; Intercept the first N points, w j,l = [w j,l (1), w j,l (2), … w j,l (N)] T 。 8. The vehicle noise reduction method according to claim 6, characterized in that the control signal is generated according to the following formula where w j,l (i) are the filter coefficients at the current moment.

9. The vehicle noise reduction method according to claim 6, characterized in that the frequency domain calculation step is performed within a time period of N sampling time points. If the data of the reference signal and the error signal reach N, the counter is cleared, and the filter parameters obtained in the frequency domain are updated to the time domain through the frequency domain calculation step, and a new round of frequency domain calculation is started; if not, the vehicle road noise and the error signal are continuously collected, and the counter continues to accumulate.

10. The vehicle noise reduction method according to claim 6, characterized in that the vibration signal generated by the friction between the wheel and the road surface is collected by a vibration sensor or a first microphone as the reference signal; and / or, the sound reproduction device includes an in-vehicle speaker disposed in the vehicle compartment; and / or, the residual noise signal in the vehicle compartment is collected by a plurality of second microphones, and the plurality of second microphones are arranged at a plurality of sampling positions in the vehicle compartment.

Images