Intelligent calibration method and system for system parameters applied to in-vehicle road noise control
Through the partition control based on face recognition and multiple coherence analysis method to filter reference signals, combined with the variable step-free delay subband filtering algorithm of symbol error, the high computational complexity, slow convergence speed and unbalanced noise reduction effect are solved. Problem, effective medium and low frequency noise control in electric vehicles is achieved.
Patent Information
- Application Number
- CN202211041732.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-29
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-08-29
AI Technical Summary
The existing active road noise control (ARNC) system has shortcomings in the problems of high computational complexity, slow convergence speed, poor stability and unbalanced noise reduction effects, especially in electric vehicles with poor low-frequency noise control.
Using the partition control mode based on face recognition, noise data is collected through in-vehicle sensors, the optimal reference signal is filtered by multiple coherence analysis method, and secondary noise signals are generated by combining the variable step-free delayed subband filtering algorithm of symbol error, and the secondary noise signal is generated by performing Interference phase cancellation and noise reduction are used to optimize algorithm parameters to improve noise reduction performance.
It has achieved the reduction of calculation complexity, improved convergence speed and stability, and ensured that people sitting everywhere could obtain better noise reduction effects, especially in electric vehicles to effectively control medium and low frequency noise.
Smart Images

Figure CN115482803B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of automobile technology, and in particular to a method and system for intelligently calibrating system parameters for active road noise control in a vehicle. Background Art
[0002] With the rapid development of new energy vehicles represented by electric vehicles, the automotive power system is developing from internal combustion engine drive to electric motor drive. Without the sound masking effect of internal combustion engine noise, road noise has become the main source of noise inside electric vehicles. Effectively reducing road noise is an important way for new energy vehicles to enhance their brand image and improve comfort. Passive noise control (PNC) methods such as body structure design or adding shock absorbers will bring about problems such as difficulty in decoupling vehicle performance, poor low-frequency noise control effect, complex optimization solution structure, and high production costs. It usually also increases the weight of the vehicle, which is contrary to the development trend of lightweight vehicles.
[0003] The active road noise control (ARNC) method uses the principle of interference destructiveness to introduce a speaker (secondary sound source) into the sound field, and controls the speaker to emit a canceling noise with the same amplitude and opposite phase as the noise to be eliminated (primary noise), which can form a quiet zone in a specific area. The ARNC system can not only effectively reduce the low-frequency noise that is difficult to control with the PNC method, but also track the changes in the spectrum of road noise in the car for adaptive control, and can achieve the attenuation of broadband noise in the car without changing the vehicle structure and working performance. Therefore, ARNC technology has become a hot spot in NVH research in the automotive industry.
[0004] ARNC technology faces two key problems in real vehicle applications. One is the accurate acquisition of reference signals with good correlation with the road noise inside the vehicle, and the other is the implementation of an adaptive filtering algorithm with low computational complexity and high noise reduction. The reference signal provides prior information for the active control of the road noise inside the vehicle, and the correlation between the two is directly related to the noise reduction amount. At present, ARNC systems mostly use the traditional filter-x least mean square (FxLMS) algorithm. The FxLMS algorithm has the disadvantages of high computational complexity, slow convergence speed and poor stability in the application of ARNC. The present invention adopts an active noise control algorithm of delay-free subband filtering (DSAF), which can decompose the reference signal into subbands for downsampling, and complete the adaptive adjustment of the control filter coefficients in each subband, so that the computational complexity is reduced; since the signal is decomposed into subbands, the signal energy is also decomposed into subbands, and in theory the algorithm is more stable when dealing with impact interference.
[0005] In addition, studies have found that controlling the road noise in the car according to the number and position of the drivers and passengers can improve the noise reduction in specific areas and improve the sound quality in the car to a certain extent. However, there are not many actual industrial applications at present, mainly because the matching and calibration of the ARNC hardware in the car partitions under different driver and passenger scenarios are unstable and cannot be promoted. In particular, the low- and medium-frequency road noise of electric vehicles is very serious after losing the masking effect of engine order noise. Summary of the invention
[0006] The technical problem to be solved by the present invention is to provide a method and system for intelligent calibration of system parameters applied to in-vehicle road noise control, so as to solve the shortcomings of the active control method and system for in-vehicle road noise, such as high computational complexity, slow convergence speed and poor stability.
[0007] Secondly, the present invention aims to solve the problem that the noise reduction effect is currently uneven among the driver and passengers, so that people sitting in various places in the car can obtain a better noise reduction effect.
[0008] Furthermore, the present invention aims to solve the problem of the prominence of medium and low frequency road noise in electric vehicles after the sound masking effect of engine order noise is lost.
[0009] In order to solve the above technical problems, the present invention adopts the following technical solutions:
[0010] A method for intelligent calibration of system parameters applied to in-vehicle road noise control, characterized by comprising:
[0011] S1: Deploy sensors on the vehicle body and use them to collect vehicle noise data;
[0012] S2: Use the in-vehicle visual sensor to identify the control area and obtain the in-vehicle road noise partition control mode and control area according to the passenger's seating situation;
[0013] S3: Based on the determined partition control mode and in accordance with the principle of maximizing noise reduction in the control area, the reference signal is screened using a multiple coherence analysis method to screen out an optimal reference signal combination for the control area;
[0014] S4: Based on the above selected optimal reference signal combination and in combination with the in-vehicle road noise partition control mode, an active road noise control algorithm based on a variable step-size delay-free sub-band filtering algorithm of a sign error is used to generate a secondary noise signal to achieve noise reduction in a specific area through interference cancellation; the secondary noise has the same amplitude and opposite phase to the noise signal to be cancelled or the primary noise in the vehicle;
[0015] S5: Use the quadratic programming algorithm to optimize the parameters of the active road noise control algorithm based on the variable step-size delay-free sub-band filtering algorithm based on the symbol error.
[0016] In the above technical solution, step S1 collects noise data under various steady-state test conditions of uniform speed driving and full throttle acceleration as a time-varying non-steady-state test condition, and the noise data includes the spectrum noise signal of the vibration acceleration sensor at the connection point between the vehicle suspension and the body or the side of the body as a reference signal, and the spectrum signal of the noise to be offset (primary) emitted by the microphone on the ear side of the headrest in a total of four positions in the front and rear rows of the vehicle.
[0017] In the above technical solution, step S2 is based on the arrangement and combination of different numbers and seating positions of the vehicle occupants. The in-vehicle road noise zoning control modes are divided into: a centralized mode when the vehicle is fully occupied, and four distributed modes in which the passengers are seated in a single zone, two zones, and three zones respectively. In each control mode, the passenger seating positions are all in the road noise control area.
[0018] In the above technical solution, step S1 collects noise data according to the partition control mode:
[0019] When the face recognition system identifies that there is only one passenger in the car, that is, the driver's seat, that needs to implement noise control, the microphone at that position is powered on to collect the noise spectrum signal in the car, and the microphones at other positions are powered off and do not need to collect signals;
[0020] When the face recognition system identifies area two, area three and when the car is full, the microphones at the corresponding positions are powered on and collect the primary noise spectrum signal in the car in that area. The vibration acceleration sensor is always in working state when collecting the spectrum noise signal. In the subsequent control, the secondary sound sources (speakers) or microphones in the non-control area are all in the off state, while the microphones in the road noise control area are all in the on state to facilitate the generation of secondary sound sources for interference cancellation.
[0021] In the above technical solution, step S3 imports the collected noise spectrum signal into MATLAB simulation analysis software to use the multiple coherence analysis method to select the optimal reference signal:
[0022] The input signal is the vibration noise spectrum signal collected by the acceleration sensor at the passive end of the connection between the suspension and the body, and the output signal is the sound pressure signal collected by the microphone placed at the passenger headrest in the car;
[0023] Calculate the multiple coherence coefficients of the reference signal and the signal to be offset (primary), obtain the target noise reduction amount or the theoretical maximum noise reduction amount based on the coherence coefficient, and sort and screen the different combination signals according to the principle of the theoretical maximum noise reduction amount or the maximum coherence of the reference signal, and then re-combine the signals based on the screened reference points.
[0024] In the above technical solution, step S3 is to sort and screen different combination signals according to the principle of maximum theoretical noise reduction or maximum reference signal coherence as follows:
[0025] First, the multi-correlation coefficients of all reference signals and the primary noise spectrum signal to be offset are calculated, and the reference signal corresponding to the maximum multi-correlation coefficient is taken, sorted as 1 and stored in a new reference signal selection set;
[0026] In the second iteration loop, the reference signal ranked 1 is removed, the multi-correlation coefficients of the remaining N-1 reference signals and the primary noise spectrum signal to be offset are calculated, the reference signal corresponding to the maximum multi-correlation coefficient is taken, ranked 2 and stored in a new reference signal selection set;
[0027] This cycle is repeated until the last reference signal is stored and recorded in a new reference signal selection set; the position of the reference signal with the highest ranking in the reference signal selection set is selected as the layout position of the vibration acceleration sensor and used as the reference signal.
[0028] In the above technical solution, step S4 adopts the variable step-size delay-free sub-band filtering algorithm VSS-DSAF based on symbol error to decompose the reference signal into sub-bands, independently set the iteration step size for each sub-band, and generate a secondary noise signal in the full frequency band.
[0029] In the above technical solution, step S4 adopts the variable step-size delay-free sub-band filtering algorithm VSS-DSAF based on symbol error, which specifically includes the following steps:
[0030] Step S41, design analysis filter:
[0031] h n (i)=h0(i)e -j2πi / N , i=0,2...,L p -1, n=0, 2, ... N-1;
[0032] Among them, L p is the length of the analysis filter, which determines the computational complexity; N is the number of subbands, h0(i) is obtained by the Matlab function h0(i) = fir1(L p -1, 1 / N) implementation;
[0033] Step S42: sub-band decomposition of reference signal and error signal:
[0034] Subband reference signal:
[0035]
[0036] Where: x n (t) is the reference signal sequence, h n (i) is the analysis filter bank;
[0037] Subband error signal:
[0038]
[0039] Where: e n (t) is the error signal sequence, h0(i) is obtained by Matlab function h0(i)=fir1(L p -1, 1 / N) implementation;
[0040] Step S43, calculating the sub-band filtering reference signal and the sub-band error signal:
[0041]
[0042] Where: D is the extraction factor, r is n (t) is the self-filtered reference signal sequence, hs n (k) is the sub-band secondary path estimation, K is the number of reference signals;
[0043] Step S44, sub-band control filter coefficient iteration:
[0044]
[0045]
[0046]
[0047] μ i (n) = λμ i (nD)+(1-λ)min{μ i,opt (n), μ i (nD)};
[0048]
[0049] Where: r n (t) = [r n (t), r n (tD),......,r n (t-M+D)], w n (t+D) is the subband control filter coefficient sequence; M is the full-band control filter length, which determines the convergence speed; H represents the conjugate transpose, * represents complex conjugate; μ i is the iterative step sequence, 0≤i≤N / 2, which determines the ability to resist shock interference; ∈ is a constant coefficient to prevent the denominator from being zero, sgn(e n (t)) is the sign function;
[0050] Step S45, convert the sub-band control filter coefficients into full-band coefficients, perform a 2M-point Fourier transform on the control filter coefficients of each sub-band, and then stack them according to the following (a)-(c) rules to obtain the Fourier transform of the full-band weight coefficients, and then perform a 2M-point discrete Fourier transform to obtain the full-band coefficients:
[0051]
[0052] Determine the optimal parameters of the analysis filter: the length L of the analysis filter p , full-band control filter length M, iterative step sequence μ i .
[0053] In the above technical solution, step S5 uses a quadratic programming algorithm to calculate the parameter L p , M, μ i The adjustable parameters are optimized so that the parameter L p , M, μ i The optimum value is reached so that the resulting synthetic noise immunity tracks broadband noise for ideal ARC performance:
[0054] Taking the minimization of the negative average noise reduction as the performance indicator, the quadratic programming optimization goal is:
[0055]
[0056] Where: f U is the upper limit of the frequency of interest, f L The lower limit of the frequency is of concern. The road noise is a medium-low frequency broadband random noise, so the frequency range of concern is 50H z ~500H z , MNR is the minimum average noise reduction; the noise reduction target NR(f i )=10log 10 {e off (f i ) / e on (f i )}; where e off (f i ) is the spectrum of the steady-state error signal when the variable step-size delay-free subband filtering algorithm based on symbol error is not applied, e on (f i ) is the steady-state error signal spectrum when applying the variable step-size delay-free sub-band filtering algorithm based on symbol error;
[0057] The constraint condition is: 1≤K M ≤5, 1≤K Lp ≤5,μ L ≤μ i ≤μ U ;
[0058] Where: μ L , μ U They are the upper and lower limits of the iteration step, generally ranging from 0 to 1;
[0059] Call the quadratic programming algorithm function in the MATLAB optimization toolbox:
[0060] X=quadprog(H, f, A, b, Aeq, beq, lb, ub);
[0061] The optimization objective of the quadratic programming is:
[0062] The constraints are:
[0063]
[0064] Where: min[f(K M , K Lp , μ i )]-objective function, H-symmetric matrix of quadratic objective term; f-real vector of linear objective term, A, b-linear inequality constraints, Aeq, beq-linear equality constraints, lb, ub-upper and lower limits of constraints.
[0065] In the above technical solution, hs generated by S4 as a secondary noise signal n (k) is the sub-band secondary path estimation using an offline identification method, which is mainly the electro-acoustic path of the process in which the electrical signal output by the controller generates secondary noise through the secondary sound source and is transmitted to the error microphone, including a D / A converter, a reconstruction filter, a power amplifier, a speaker, a propagation path from the speaker to the error microphone, an error microphone, a preamplifier, an anti-aliasing filter and an A / D converter.
[0066] A system parameter intelligent calibration system for in-vehicle road noise control, characterized in that a program for implementing any of the above methods is stored.
[0067] Therefore, the method and system of the present invention adopt a partition control mode based on face recognition, and complete the development of an intelligent calibration system for algorithm parameters according to the matching of in-vehicle partition ARNC hardware and the adjustment method of algorithm parameters under different driver and passenger scenarios.
[0068] At the same time, partition control and global control have different requirements in the selection of the optimal reference signal and the determination of the parameters of the adaptive filtering algorithm. Through partition control, the present invention only needs to use the quadratic programming function in terms of the algorithm's computational complexity, convergence speed, and anti-impact interference ability to complete the optimization of the parameters of the variable step-size delay-free subband filtering algorithm VSS-DSAF based on the symbol error, achieve the best noise reduction performance, and avoid the shortcomings of the existing algorithm that relies on trial and error decoupling, high computational complexity, slow convergence speed, and poor stability.
[0069] Secondly, the present invention generates secondary signals according to the number and positions of drivers and passengers, and the noise reduction effect at each position can be achieved according to the design requirements, and the noise reduction effect is more targeted.
[0070] Finally, the present invention utilizes the control to generate a secondary signal with low efficiency of the primary noise signal, which can truly solve the problem of the prominent medium and low frequency road noise of electric vehicles after losing the sound masking effect of the engine order noise. BRIEF DESCRIPTION OF THE DRAWINGS
[0071] Figure 1 This is a schematic diagram of the active road noise control principle on which the method of the present invention is based.
[0072] Figure 2 Flow chart of the intelligent calibration method of system parameters for active road noise control in a vehicle according to the present invention
[0073] Figure 3 FIG. 4 is a flow chart of the screening procedure for the optimal reference signal combination of the present invention.
[0074] Figure 4 It is a principle diagram of a variable step-size delay-free sub-band filtering algorithm (VSS-DSAF) based on symbol error of the present invention. DETAILED DESCRIPTION
[0075] The present invention mainly solves the problem of the prominent mid- and low-frequency road noise of electric vehicles after the sound masking effect of engine order noise is lost. It mainly uses active noise control technology to perform zone control on the mid- and low-frequency noise of the vehicle and intelligently optimize and calibrate the control parameters of the controller to achieve the purpose of noise reduction in the car.
[0076] The intelligent optimization and calibration method and system for the technical parameters of the active control of road noise in the vehicle zone proposed in the present invention mainly collects noise data as a reference signal through the on-board sensor of the vehicle chassis, identifies the control area by using the sensor in the vehicle, and obtains the control area; maximizes the noise reduction of the control area based on the partition control mode, and uses the multiple coherence analysis method to screen the reference signal to screen out the optimal reference signal combination of the control area; and based on the multi-channel active road noise control algorithm with low computational complexity, generates secondary noise with the same amplitude and opposite phase as the noise signal to be offset (primary noise in the vehicle) to achieve interference cancellation of the noise in the control area, and uses the optimization algorithm to optimize the parameters of the active road noise control algorithm. The active road noise control principle based on which the method of the present invention is based is as follows Figure 1 shown.
[0077] Vehicle driving conditions and working conditions: Two new energy test vehicles with multi-link independent suspension structure and torsion beam non-independent suspension structure were used as test data collection objects, and vibration and noise tests were carried out on rough asphalt roads and smooth asphalt roads. The vehicle speed adjustment was set to cruise control mode, and each increase of 10km / h from 30 to 120km / h was used as a steady-state test condition; the acceleration method used full throttle acceleration and used this as a time-varying non-steady-state test condition. The test content includes: the vibration reference signal at the connection point between the vehicle suspension and the body (body side) and the sound pressure signal next to the headrest at a total of four positions in the front and rear rows of the vehicle, thereby obtaining a data analysis sample library, and the test conditions are shown in Table 1.
[0078]
[0079] The required instruments and equipment are shown in Table 2:
[0080] Serial number Device Name quantity 1 LMS Test.Lab test equipment 1 set 2 PCB three-axis acceleration sensor several 3 Microphone (sound pressure sensor) several 4 Triaxial BNC Plug Cable several 5 Laptop 1 unit 6 Magnetic base several
[0081] The specific positions of the sensor arrangements are as follows: one vibration acceleration sensor is arranged at each passive end of the connection point between the front end of the left and right triangular arms of the front suspension and the shock absorber, one vibration acceleration sensor is arranged at each passive end of the connection point between the rear end of the left and right triangular arms of the front suspension and the shock absorber, one is arranged at each passive end of the upper end of the left and right shock absorbers of the front suspension, one is arranged at each passive end at the center of each left and right front wheel, one is arranged at the passive end in the middle of the front axle subframe, one is arranged at each passive end in the middle of the left and right connecting beams between the body and the chassis, one is arranged at each passive end at the front end of the left and right torsion beams of the rear suspension, one is arranged at each passive end at the rear end of the left and right torsion beams of the rear suspension, one is arranged at each passive end of the upper end of the left and right shock absorbers of the rear suspension, one is arranged at each passive end at the center of each left and right rear wheel, and one is arranged at the passive end in the middle of the torsion beam of the rear suspension. A total of 20 vibration acceleration sensors are arranged at 20 collection points, and microphones (sound pressure sensors) are installed on the right side of the driver (DR), the left ear of the co-driver (PL), the left right ear of the rear row (RLR), and the right left ear of the rear row (RRL).
[0082] Design of active control of road noise in the car: face recognition technology is used to identify the seating situation of passengers in the car according to the arrangement and combination of different numbers and seating positions of passengers in the car. It should be pointed out that face recognition technology is very mature. This invention is just borrowing the existing technology, mainly used to identify the seating situation of passengers in each seat in the car, that is, to distinguish how many passengers are in the car and their respective seating positions; thus forming active control architecture forms such as centralized and distributed, weight ratio between error channels, etc., to establish a partitioned active road noise control rule base. The centralized type means that the car is full of passengers, and the distributed type means that the passengers are seated separately in the car, that is, single area, two areas, and three areas, and the facial recognition system can clearly identify the seating position of the passengers, that is, the seating position of the passengers is the road noise control area.
[0083] Reference signal screening method: Use the LMS Test.Lab test equipment data acquisition module to collect the spectral noise signal of the accelerometer and the spectral noise signal of the microphone next to the ear. The spectral signal obtained by the accelerometer is the reference signal, and the spectral noise signal of the microphone next to the ear is the signal to be offset, that is, the primary signal. Import the collected signals into the MATLAB simulation software and use the multiple coherence analysis method to screen the optimal reference signal combination. The principle of the multiple coherence analysis method is to calculate the coherence relationship between the frequency domain signals of multiple input and multiple output signals, and sort and screen different combination signals according to the theoretical maximum noise reduction amount (the reference signal has the largest coherence), and then re-combine the signals based on the screened reference points until the target effect is achieved.
[0084] Adoption of noise reduction algorithm: Based on the best reference signal combination screened out above, and combined with the in-vehicle road noise zoning control, the active road noise control algorithm based on symbol error variable step size delay-free sub-band filtering algorithm (VSS-DSAF) is adopted to generate a secondary cancellation signal to achieve noise reduction in specific areas through secondary sound sources.
[0085] Intelligent optimization of algorithm parameters: The cancellation signal (secondary signal) emitted by the secondary sound source affects the noise control performance, which is not only related to the strong coherence of the reference signal but also to the algorithm processing capability. The algorithm parameters mainly affect its computational complexity and convergence speed. The parameter optimization of the variable step-size delay-free subband filtering algorithm (VSS-DSAF) based on the symbol error is mainly optimized using the quadratic programming algorithm.
[0086] The flow chart of the intelligent calibration method of system parameters for active road noise control in a vehicle according to the present invention is as follows: Figure 2 As shown. It includes the following steps:
[0087] Two new energy test vehicles with multi-link independent suspension and torsion beam non-independent suspension chassis structures are used as test data collection objects. Due to the different chassis structures of multi-link independent suspension and torsion beam non-independent suspension, the positions of sensor layout are also different, but they should all follow the arrangement of sensors at the passive end of the attachment point between the vehicle body and the chassis structure, so as to more comprehensively collect the vehicle road noise signal. The present invention takes the arrangement position of sensors in the torsion beam non-independent suspension chassis structure as an example: a vibration acceleration sensor is arranged at the passive end of the connection point between the front end of the left and right triangular arms of the front suspension and the shock absorber, a vibration acceleration sensor is arranged at the passive end of the connection point between the rear end of the left and right triangular arms of the front suspension and the shock absorber, one is arranged at the passive end of the upper end of the left and right shock absorbers of the front suspension, one is arranged at the passive end of the left and right front wheel centers, one is arranged at the passive end of the middle of the front axle subframe, one is arranged at the passive end of the middle of the left and right connecting beams between the vehicle body and the chassis, and one is arranged at the passive end of the rear suspension. There are 20 vibration acceleration sensors arranged at the front end of the left and right torsion beams of the front suspension, one at the rear end of the left and right torsion beams of the rear suspension, one at the upper passive end of the left and right shock absorbers of the rear suspension, one at the passive end of the left and right rear wheel centers, and one at the middle passive end of the rear suspension torsion beam. A total of 20 collection points are arranged with 20 vibration acceleration sensors, and microphones (sound pressure sensors) are installed on the right side of the driver's seat (DR), the left ear of the co-driver (PL), the right ear on the left side of the rear row (RLR), and the left ear on the right side of the rear row (RRL). The required instruments and equipment are shown in Table 2.
[0088] The road conditions for the car to drive are rough asphalt road and smooth asphalt road. The vehicle speed adjustment is set to cruise control mode, and each increase of 10km / h from 30 to 120km / h is used as a steady-state test condition; the acceleration method uses full throttle acceleration and uses this as a time-varying non-steady-state test condition. The data acquisition method is to use the LMSTest.Lab test equipment data acquisition module to collect the spectral noise signal of the acceleration sensor and the spectral noise signal of the microphone on the side of the ear at each road surface and each vehicle speed. The noise spectrum signal obtained by the acceleration sensor is the reference signal, and the noise spectrum signal of the microphone on the side of the ear is the signal to be offset, that is, the primary signal. The test conditions are shown in Table 1.
[0089] Since a car has multiple passenger seats that all need noise reduction, and the noise transmission path of each seat is different, if they are controlled as a whole, the noise reduction performance will be affected compared to zone control. If control is performed based on the seating situation of the passengers, it will not only improve the noise reduction effect of the local area but also make efficient use of the resources in the car. Therefore, the road noise control in the car should be zoned according to the seating situation of the passengers.
[0090] In view of the different numbers of passengers and the arrangement and combination of different seating positions in the car, face recognition technology is used to identify the seating situation of passengers in the car. It should be pointed out that face recognition technology is very mature. The present invention is just borrowing the existing technology, mainly used to identify the seating situation of passengers in each seat in the car, that is, to distinguish how many passengers are in the car and their respective seating positions; thus forming active control architecture forms such as centralized and distributed, weight ratio between error channels, etc., to establish a partitioned active road noise control rule base. The centralized type means that the car is full of passengers, and the distributed type means that the passengers are seated separately in the car, that is, single area, two areas, and three areas, and the facial recognition system can clearly identify the passenger seat position, that is, the passenger seat position is the road noise control area.
[0091] The principle of noise signal collection is to use the LMS Test.Lab test equipment data acquisition module to collect the spectrum noise signal of the acceleration sensor and the spectrum noise signal of the microphone beside the ear. The spectrum noise signal should also be collected according to the partition control requirements. When the face recognition system recognizes the seat status of the occupant, for example, when the face recognition system recognizes that there is only one occupant in the car, that is, the driver's seat, and noise control needs to be implemented, the microphone at this position is powered on to realize the collection of the noise spectrum signal in the car. The microphones at other positions are in the power-off state and do not need to collect signals. Similarly, when the face recognition system recognizes the second area, the third area, and the full seat, the microphones at the corresponding positions are powered on and collect the noise spectrum signal in the car in the area. The vibration acceleration sensor is always in working state when collecting road noise. In the subsequent control, the secondary sound sources (speakers) in the non-control area are all in the off state, and the control area is in the on state to facilitate the generation of secondary sound sources for interference cancellation.
[0092] Import the collected signals into the MATLAB simulation software and use the multiple coherence analysis method to screen the optimal reference signal combination. The principle of the multiple coherence analysis method is to calculate the coherence relationship between the frequency domain signals of multiple input and multiple output signals, and sort and screen different combination signals according to the theoretical maximum noise reduction (reference signal coherence is maximum), and then re-combine the signals based on the screened reference points until the target effect is achieved. The specific process of multiple coherence analysis method to screen the optimal reference signal combination is as follows: Figure 3 As shown:
[0093] In the present invention, the input signal is the vibration signal collected by 20 acceleration sensors at the passive end of the connection between the suspension and the vehicle body, and the output signal is the sound pressure signal collected by the microphone placed at the passenger headrest in the vehicle. The target noise reduction amount is obtained using the following formula:
[0094]
[0095] In formula (1), the multi-coherence coefficient The calculation formula is as follows:
[0096]
[0097] Where: Δ(f) is the theoretical maximum noise reduction, is the multi-correlation coefficient, is the input signal x n The autopower spectrum of the output y(n) caused by dd (f) is the auto-power spectrum of the primary noise signal d(n). xd For x n The cross spectrum with d(n), the superscript "T" indicates the transpose, S xx =X*X T For x n Autopower spectrum. The superscript “-1” indicates that the matrix is generalized inverse.
[0098] First, the multi-coherence coefficients of all N reference signals and the signal to be cancelled are calculated, and the reference signal corresponding to the maximum multi-coherence coefficient is taken, sorted as 1 and stored.
[0099] In the second iteration loop, the reference signal ranked 1 is removed, the multi-coherence coefficients of the remaining N-1 reference signals and the signal to be eliminated are calculated, and the reference signal corresponding to the maximum multi-coherence coefficient is taken and ranked 2.
[0100] This cycle continues until the last reference signal is obtained. The position of the reference signal with the highest ranking can be selected as the layout position of the vibration acceleration sensor of the road noise control system and used as the reference signal for system control.
[0101] The calculation formula for screening the reference signal using the multiple coherence coefficient method is:
[0102]
[0103] Where: n is the number of input signals, ranging from 1 to N, N is the total number of input signals or reference signals, f u is the highest frequency point, f l is the lowest frequency point.
[0104] In order to solve the problem of heavy computational burden and slow convergence speed of FxLMS algorithm in broadband noise control, the present invention will adopt variable step-size delay-free subband filtering algorithm (VSS-DSAF) based on symbol error. The reference signal is decomposed into subbands, and each subband is independently set with iteration step size, so that the optimal noise reduction effect can be achieved in the whole frequency band. The algorithm principle diagram is shown in Figure 4 shown.
[0105] The algorithm implementation process is as follows:
[0106] Step 1: Design and analyze the filter
[0107] h n (i)=h0(i)e -j2πi / N , i=0,2...,L p -1, n=0, 2, ... N-1 (4)
[0108] L p is the length of the analysis filter, N is the number of subbands, and h0(i) is obtained by the Matlab function h0(i)=fir1(L p -1, 1 / N) to achieve.
[0109] Step 2: Sub-band decomposition of reference signal and error signal:
[0110] 1. Subband reference signal:
[0111]
[0112] Where: x n (t) is the reference signal sequence, h n (i) is the analysis filter bank
[0113] 2. Sub-band error signal:
[0114]
[0115] Where: e n (t) is the error signal sequence, h0(i) is obtained by Matlab function h0(i)=fir1(L p -1, 1 / N) to achieve.
[0116] Step 3: Calculate sub-band filtering reference signal and sub-band error signal
[0117]
[0118] Where: D is the extraction factor, r is n (t) is the self-filtered reference signal sequence, hs n (k) is the sub-band secondary path estimation, and K is the number of reference signals.
[0119] Step 4: Subband control filter coefficient iteration
[0120]
[0121]
[0122]
[0123] μ i (n) = λμi (nD)+(1-λ)min{μ i,opt (n), μ i (nD)} (11)
[0124]
[0125] Where: r n (t) = [r n (t), r n (tD),......,r n (t-M+D)], w n (t+D) is the sub-band control filter coefficient sequence, M is the full-band control filter length, H represents the conjugate transpose, * represents the complex conjugate, μ i is the iterative step sequence, ∈ is a constant coefficient to prevent the denominator from being zero, sgn(e n (t)) is a sign function.
[0126] Step 5: Convert sub-band control filter coefficients to full-band coefficients (FFT-2 stacking method). Perform a 2M-point Fourier transform (FFT) on the control filter coefficients of each sub-band, and then stack them according to the following rules to obtain the Fourier transform (FFT) of the full-band weight coefficients, and then perform a 2M-point discrete Fourier transform (IFFT) to obtain the full-band coefficients.
[0127]
[0128] Since the algorithm parameter L p , M, μ i The computational complexity, convergence speed and anti-impact interference capability of the algorithm are determined. Therefore, simply using empirical methods to try out the algorithm parameters obviously cannot meet its optimal performance design. Therefore, the present invention uses quadratic programming to optimize the algorithm parameters.
[0129] Effectively eliminating broadband noise requires adjusting some of the most influential parameters L p , M, μ i(0≤i≤N / 2) is associated with a delay-free subband filtering algorithm based on a delay-free architecture. The optimal length (M) of the adaptive filter is easily determined by theoretical calculation or by trial and error for a single-channel active road noise control system, but this is not the case for a multi-channel active road noise control system. For simplicity, it is assumed that the frequency resolution is determined as follows: N is a fixed value before each optimization instance, so there are 1+1+(N / 2+1)=N / 2+3 design variables. In the present invention, the adjustable parameters are optimized using a quadratic programming algorithm. The parameters are optimized so that the resulting synthetic anti-noise tracks the broadband noise to achieve the desired active road noise control performance.
[0130] The set of parameters that need to be adjusted are:
[0131] X=[K M , K Lp , μ1, μ2, … μ N / 2 ] (14)
[0132] Where: K M =log2M-4, its confidence interval is {1, 2, 3, 4, 5}, and the confidence interval of M is {32, 64, 128, 256, 512}; K Lp =log2(L p / N)+2, and its confidence interval is {1, 2, 3, 4, 5}.
[0133] Since the goal of the present invention is to achieve noise reduction, this index value is related to the noise reduction performance of active road noise control within a specific frequency range. In order to obtain the optimal solution, the noise reduction target is:
[0134] NR(f i )=10log 10 {e off (f i ) / e on (f i )} (15)
[0135] where e off (f i ) is the steady-state error signal spectrum when the variable step-size delay-free subband filtering algorithm based on symbol error (VSS-DSAF) is not applied, e on (f i ) is the steady-state error signal spectrum when applying the variable step-size delay-free subband filtering algorithm based on symbol error (VSS-DSAF).
[0136] Taking the minimization of the negative average noise reduction as the performance indicator, the optimization problem is expressed as:
[0137] Optimization goal:
[0138]
[0139] Where: f U is the upper limit of the frequency of interest, f L The lower limit of the frequency is of concern. The road noise is a medium-low frequency broadband random noise, so the frequency range of concern is 50H z ~500H z , MNR is the minimum average noise reduction.
[0140] The constraints are;
[0141] 1≤K M ≤5, 1≤K Lp ≤5,μ L ≤μ i ≤μ U (17)
[0142] Where: μ L , μ U They are the upper and lower limits of the iteration step, and generally range from 0 to 1.
[0143] Combine the above optimization objectives with the constraints, and according to the algorithm calculation results, call the quadratic programming algorithm function X=quadprog(H, f, A, b, Aeq, beq, lb, ub) in the MATLAB optimization toolbox, and rewrite the above formula into a quadratic programming form:
[0144]
[0145] The constraints are:
[0146]
[0147] Where: min[f(K M , K Lp , μ i )]-objective function, H-quadratic objective term symmetric matrix; f-linear objective term real vector, A, b-linear inequality constraints, Aeq, beq-linear equality constraints, lb, ub-upper and lower limits of constraints. The present invention only uses the quadratic programming function to optimize the algorithm parameters and achieve the best noise reduction performance.
[0148] Secondary channel hs n (k) An offline identification method is adopted, which mainly includes the electro-acoustic path of the process in which the electrical signal output by the controller generates secondary noise through the secondary sound source and is transmitted to the error microphone, including a D / A converter, a reconstruction filter, a power amplifier, a speaker, a transmission path from the speaker to the error microphone, an error microphone, a preamplifier, an anti-aliasing filter and an A / D converter.
[0149] It should be understood that those skilled in the art can make improvements or changes based on the above description, but the specific implementation of the present invention is not limited to this example. All alternative solutions with similar effects proposed according to the principles and ideas of the present invention should be considered as the protection scope of the present invention.
Claims
1. A method for intelligent calibration of system parameters for in-vehicle road noise control, characterized in that include: S1: Deploy sensors on the vehicle body and use them to collect vehicle noise data; S2: Use the in-vehicle visual sensor to identify the control area and obtain the in-vehicle road noise partition control mode according to the passenger's seating situation; S3: Based on the determined partition control mode and in accordance with the principle of maximizing noise reduction in the control area, the reference signal is screened using a multiple coherence analysis method to screen out an optimal reference signal combination for the control area; S4: Based on the above-screened optimal reference signal and in combination with the in-vehicle road noise partition control mode, an active road noise control algorithm based on a variable step-size delay-free sub-band filtering algorithm of a symbol error is used to generate a secondary noise signal to achieve noise reduction in a specific area through interference cancellation; the secondary noise has the same amplitude and opposite phase to the noise signal to be cancelled or the primary noise in the vehicle; S5: Use the quadratic programming algorithm to optimize the parameters of the active road noise control algorithm based on the variable step-size delay-free sub-band filtering algorithm based on the symbol error; Step S4 uses a variable step-size delay-free sub-band filtering algorithm VSS-DSAF based on symbol error to decompose the reference signal into sub-bands, independently set the iteration step size for each sub-band, and generate a secondary noise signal in the full frequency band; Step S4 adopts the variable step-size delay-free sub-band filtering algorithm VSS-DSAF based on symbol error, which specifically includes the following steps: Step S41, design analysis filter: Among them, L p is the length of the analysis filter, which determines the computational complexity; N is the number of subbands, h0(i) is obtained by the Matlab function h0(i) = fir1(L p -1, 1 / N) implementation; Step S42: sub-band decomposition of reference signal and error signal: Subband reference signal: Where: x n (t) is the reference signal sequence, h n (i) is the analysis filter bank; Subband error signal: Where: e n (t) is the error signal sequence, h0(i) is obtained by Matlab function h0(i)=fir1(L p -1, 1 / N) implementation; Step S43, calculating the sub-band filtering reference signal and the sub-band error signal: Where: D is the extraction factor, r is n (t) is the self-filtered reference signal sequence, hs n (k) is the sub-band secondary path estimation; Step S44, sub-band control filter coefficient iteration: m i (n)=λμ i (nD)+(1-λ)min{μ i,opt (n),μ i (nD)}; Where: r n (t) = [r n (t), r n (tD),......,r n (t-M+D)], w n (t+D) is the subband control filter coefficient sequence; M is the full-band control filter length, which determines the convergence speed; H represents the conjugate transpose, represents complex conjugate; μ i is the iterative step sequence, 0≤i≤N / 2, which determines the ability to resist shock interference; ∈ is a constant coefficient to prevent the denominator from being zero, sgn(e n (t)) is the sign function; Step S45, convert the sub-band control filter coefficients into full-band coefficients, perform a 2M-point Fourier transform on the control filter coefficients of each sub-band, and then stack them according to the following (a)-(c) rules to obtain the Fourier transform of the full-band weight coefficients, and then perform a 2M-point discrete Fourier transform to obtain the full-band coefficients: Determine the optimal parameters of the analysis filter: the length L of the analysis filter p , full-band control filter length M, iterative step sequence μ i .
2. The intelligent calibration method for system parameters applied to in-vehicle road noise control according to claim 1 is characterized in that Step S1 collects noise data under various steady-state test conditions of uniform speed driving and full throttle acceleration as a time-varying non-steady-state test condition, wherein the noise data includes a spectrum noise signal of a vibration acceleration sensor at a connection point between the vehicle suspension and the vehicle body or at the side of the vehicle body as a reference signal, and a primary noise spectrum signal emitted by a microphone on the ear side of the headrest at four positions in the front and rear rows of the vehicle.
3. The system parameter intelligent calibration method for in-vehicle road noise control according to claim 1 is characterized in that In step S2, according to the arrangement and combination of different numbers and seating positions of the vehicle occupants, the vehicle road noise zoning control modes are divided into: a centralized mode when the vehicle is fully occupied, and three distributed modes when the vehicle occupants are seated in a single zone, two zones, and three zones. In each control mode, the seating positions of the occupants are all in the road noise control zone.
4. The intelligent calibration method for system parameters applied to in-vehicle road noise control according to claim 1, characterized in that Step S1 collects noise data according to the partition control mode: When the face recognition system identifies that there is only one passenger in the car, that is, the driver's seat, that needs to implement noise control, the microphone at that position is powered on to collect the noise spectrum signal in the car, and the microphones at other positions are powered off and do not need to collect signals; When the face recognition system identifies area two, area three and full seats, the microphones at the corresponding positions are powered on and collect the primary noise spectrum signal in the car in that area. The vibration acceleration sensor is always in working state when collecting the spectrum noise signal. In the subsequent control, the secondary sound sources or microphones in the non-control area are all in the off state, while the microphones in the road noise control area are all in the on state to facilitate the generation of secondary sound sources for interference cancellation.
5. The intelligent calibration method for system parameters applied to in-vehicle road noise control according to claim 1 is characterized in that Step S3 imports the collected noise spectrum signal into MATLAB simulation analysis software and uses the multiple coherence function method to select the optimal reference signal: The input signal is the vibration noise spectrum signal collected by the acceleration sensor at the passive end of the connection between the suspension and the body, and the output signal is the sound pressure signal collected by the microphone placed at the passenger headrest in the car; The coherence coefficient of each combination is calculated, and the target noise reduction amount or the theoretical maximum noise reduction amount is obtained based on the coherence coefficient. Different combination signals are sorted and screened according to the principle of the theoretical maximum noise reduction amount or the maximum coherence of the reference signal, and the signals are recombined based on the screened reference points.
6. The intelligent calibration method for system parameters applied to in-vehicle road noise control according to claim 1, characterized in that Step S3 sorts and screens different combination signals according to the principle of maximum theoretical noise reduction or maximum reference signal coherence as follows: First, the multi-correlation coefficients of all reference signals and the primary noise spectrum signal to be offset are calculated, and the reference signal corresponding to the maximum multi-correlation coefficient is taken, sorted as 1 and stored in a new reference signal selection set; In the second iteration loop, the reference signal ranked 1 is removed, the multi-correlation coefficients of the remaining N-1 reference signals and the primary noise spectrum signal to be offset are calculated, the reference signal corresponding to the maximum multi-correlation coefficient is taken, ranked 2 and stored in a new reference signal selection set; This cycle is repeated until the last reference signal is stored and recorded in a new reference signal selection set; the position of the reference signal with the highest ranking in the reference signal selection set is selected as the layout position of the vibration acceleration sensor and used as the reference signal.
7. The intelligent calibration method for system parameters applied to in-vehicle road noise control according to claim 1, characterized in that Step S5 uses a quadratic programming algorithm to adjust the adjustable parameters L of the analysis filter of the variable step-size delay-free subband filtering algorithm VSS-DSAF based on symbol error. p , M, μ i Optimize the adjustable parameter L p , M, μ i An optimum value is reached, resulting in a synthetic secondary anti-noise tracking broadband noise for ideal active road noise control performance.
8. The intelligent calibration method for system parameters applied to in-vehicle road noise control according to claim 1, characterized in that The sub-band secondary path estimation for generating the secondary noise signal in S4 adopts an offline identification method, which is mainly the electrical-acoustic path of the process in which the electrical signal output by the controller generates secondary noise through the secondary sound source and is transmitted to the error microphone, including a D / A converter, a reconstruction filter, a power amplifier, a speaker, a propagation path from the speaker to the error microphone, an error microphone, a preamplifier, an anti-aliasing filter and an A / D converter.
9. An intelligent calibration system for system parameters used in vehicle road noise control, characterized in that A program for implementing the method described in any one of claims 1 to 8 is stored.
