CN-122024687-A - Active noise reduction method for automobile based on iterative APA algorithm

CN122024687ACN 122024687 ACN122024687 ACN 122024687ACN-122024687-A

Abstract

The invention discloses an active noise reduction method of an automobile based on an iterative APA algorithm, which comprises the following steps of arranging an analog error microphone near the ear of the automobile (at a headrest and other places), wherein the microphone is required to have a good frequency response curve at a low-frequency (< 500 HZ) part. The original heavy bass, bass and midrange speakers of the automobile are used as secondary sound sources. APA dimension setting-the APA dimension is set to L, which must be less than or equal to the dimension N of the filter coefficients. The iterative process of iterative APA algorithm includes setting the transfer function vector of primary channel as Let the filter vector be First, the reference vector is read And Error signal And , In Find out above Projection of (a) . Followed by reading And , In Determination of Projection of (a) And so on until reading And , In Determination of Projection of (a) . The acquisition of prior filter vectors, namely, in the algorithm development stage, testing various speeds on various types of road surfaces, collecting reference signals and error signals, off-line estimating corresponding filter vectors, averaging the vectors to obtain prior vectors . Updating of the filter vector according to the formula Determination of If (3) If it is too large, discard Otherwise utilize Updating 。

Inventors

FU CHI
LIU WEIQIANG
ZOU XUEPING
ZHANG YOUSHENG

Assignees

永丰航盛电子有限公司

Dates

Publication Date: 20260512
Application Date: 20260204

Claims (3)

1. An active noise reduction method for an automobile based on an iterative APA algorithm is characterized by comprising the following steps: S1, hardware deployment, namely, arranging an analog error microphone near the ear of an automobile (a headrest and the like), wherein the microphone is required to have a good frequency response curve in a low-frequency (< 500 HZ) part, and the PCM6240 is used in an ADC scheme. The original heavy bass, bass and midrange speakers of the automobile are used as secondary sound sources, and the special speakers for noise reduction can be arranged at the positions of doors, trunk, control desk, ceilings and headrests of the speakers, even at the positions of the tops of passengers. Note that speakers with poor low frequency performance such as tweeters should not be used. The speaker is driven using class D power amplifier FDA901. The noise source CAN be constructed through the engine rotating speed acquired by the CAN bus for noise reduction of the engine, the triaxial acceleration sensor of the A2B interface is arranged at the chassis for noise reduction of the road, and the microphone is used for acquiring noise sources such as wind noise, air conditioner noise and the like. The processor of the noise reduction platform is ADSP21565; And S2. Selecting APA parameters, namely selecting the linear space dimension of the work of the APA algorithm as N and selecting the dimension processed by the APA algorithm as L. The higher the parameter N, the higher the calculation amount required by the algorithm and the higher the memory required, the parameter needs to be determined according to the actual vehicle test result, and is generally selected to be 128, 256 or 512, and the parameter L should be as high as possible, but must be less than or equal to N, because Must be regarded as Is determined by the nature of the subspace of (a); S3, collecting sensor data, namely sampling noise source signals and error microphone signals at a sampling rate of 48KHZ, and performing 48 times downsampling to obtain L reference vectors and L corresponding error signals finally, wherein the working frequency Fs of an algorithm is 1 KHZ. Thus, the update period of the algorithm is L/Fs; s4, self-adaptive filtering calculation based on iterative APA algorithm, namely obtaining the coefficient of the FIR filter to be updated by adopting the iterative APA algorithm Because of the real-time requirement of the system, L-dimension calculation is required to be completed within one working cycle (1/Fs) of the algorithm, if the L parameter is too high, the calculated amount is too large, and the system cannot work normally. One improvement of the iterative algorithm is that a group of parameters are acquired, namely, calculation is carried out once until L groups of parameters are acquired, so that the calculation of the algorithm is carried out after the L groups of parameters are acquired, and the calculation of the algorithm is carried out once every acquisition of a group of parameters, namely, the work in the duration of 1/Fs is distributed into the duration of L/Fs; s5, obtaining a priori value of the filter coefficient, namely performing a large number of tests at various speeds (within the allowable speed range of an automobile) under various road conditions (rough asphalt pavement, smooth cement pavement, grooved cement pavement and the like) in an algorithm development stage to obtain a series of reference signals and error signals, and using an LMS method (or an APA method and an LS method) to estimate the corresponding filter coefficient offline according to the reference model and the error signals Average the coefficients As a priori value for the filter coefficients. S6, iteration of the filter coefficients, wherein in the algorithm operation stage, L reference vectors are linearly related due to extremely small probability (almost impossible to happen), so that the calculated reference vectors There are two cases 1 where the deviation from the expected value is too great, affecting the convergence of the system, which needs to be discarded . A normal state Must fall to Inside the N-dimensional sphere which is the center of the sphere, thus when Oversized, discard This iteration is abandoned. 2, the deviation from the expected value is not large, and although the deviation is abnormal, the influence on the convergence of the system is small, and the deviation can be directly ignored and is not processed; s7, there is also a possibility that the calculated result is caused by the extra vibration and noise of the speed reducing belt, the thunder and the like Excessive deviation from the expected value, i.e Too large, this case also gives up the iteration.
2. The method for actively reducing noise of an automobile based on an iterative APA algorithm as claimed in claim 1, wherein the iterative APA algorithm in the step S2 is implemented by iterative projection to avoid matrix inversion operation, wherein L samples are collected to form L reference vectors, and projection iteration is sequentially performed on subspaces formed by two or more reference vectors, so as to find an optimal projection vector to update a filter coefficient.
3. The method of claim 1, wherein the iterative APA algorithm in step S2 estimates the filter coefficients that deviate from the prior filter coefficients too much, and discards the iteration.

Description

Active noise reduction method for automobile based on iterative APA algorithm Technical Field The invention belongs to the technical field of active noise reduction of automobiles, and particularly relates to an active noise reduction method of an automobile based on an iterative APA algorithm. Background Currently, active noise reduction technology for automobiles is widely applied to improving driving comfort, and the core of the active noise reduction technology is to inhibit noise in a carriage by an acoustic cancellation principle. In the prior art, the mainstream scheme is mostly based on an adaptive filtering algorithm, wherein a Least Mean Square (LMS) algorithm and variants thereof (such as FxLMS, block LMS, frequency domain LMS) are one of common technical paths due to relatively simple implementation. Such schemes typically deploy microphones in the car to collect error signals and combine the speaker output of the inverted sound waves to cancel. However, the LMS algorithm is basically a random gradient descent method, and inherent contradiction exists between the convergence speed and the steady-state error, and the selection of step parameters is particularly critical, namely, the step is large, the convergence speed is high, the steady-state error is large, the steady-state error is easy to destabilize, the step is small, the steady-state precision is high, the convergence speed is slow, and the time-varying characteristic of the automobile noise is difficult to track rapidly. Part of the technology can also acquire working condition signals such as vehicle speed, engine speed and the like from an automobile ECU (electronic control unit) as auxiliary basis for noise characteristic judgment so as to adapt to noise changes (such as engine noise at low speed, wind noise and road noise at high speed) under different running states. Meanwhile, the prior art has realized multichannel signal processing, and through the cooperation of a plurality of microphones and loudspeakers, the coverage area of noise reduction is enlarged, and the basic requirement of noise suppression of the whole vehicle is primarily met. Although the prior art can realize a certain degree of noise reduction, various limitations still exist. From the algorithm level, a traditional Affine Projection Algorithm (APA) was proposed in 1984, which theoretically has a faster convergence speed and better steady-state performance. The APA algorithm is popularized in a high-dimensional space as an LMS algorithm, and many improvements of the LMS algorithm (aiming at specific vehicle types and specific noise types) can be applied to the APA algorithm, such as step-variable, order-variable and the like. However, the classical implementation of the APA algorithm needs to judge the rank of the matrix, calculate the inverse of the matrix, and the calculation complexity and memory occupation are significantly higher than those of the LMS algorithm, so that the method is difficult to directly apply under the limited calculation power and real-time requirements of the vehicle-mounted embedded system. Disclosure of Invention In order to overcome the defects of the prior art, the embodiment of the invention provides an active noise reduction method for an automobile based on an iterative APA algorithm, which solves the problems of large calculated amount, high memory occupation and poor instantaneity of a classical APA algorithm in the traditional active noise reduction of the automobile in the prior art. In order to achieve the above purpose, the present invention provides the following technical solutions: An active noise reduction method for an automobile based on an iterative APA algorithm comprises the following steps: S1, hardware deployment, namely, arranging an analog error microphone near the ear of an automobile (a headrest and the like), wherein the microphone is required to have a good frequency response curve in a low-frequency (< 500 HZ) part, and the PCM6240 is used in an ADC scheme. The original heavy bass, bass and midrange speakers of the automobile are used as secondary sound sources, and the special speakers for noise reduction can be arranged at the positions of doors, trunk, control desk, ceilings and headrests of the speakers, even at the positions of the tops of passengers. Note that speakers with poor low frequency performance such as tweeters should not be used. The speaker is driven using class D power amplifier FDA901. The noise source CAN be constructed through the engine rotating speed acquired by the CAN bus for noise reduction of the engine, the triaxial acceleration sensor of the A2B interface is arranged at the chassis for noise reduction of the road, and the microphone is used for acquiring noise sources such as wind noise, air conditioner noise and the like. The processor of the noise reduction platform is ADSP21565; And S2. Selecting APA parameters, namely selecting the linear space dimension of the work of the