阅读说明：本技术 基于fpga的主动噪声控制系统模块化设计方法 (Active noise control system modular design method based on FPGA ) 是由 袁军 吕韦喜 王巍 张涛 李军 于 2019-11-11 设计创作，主要内容包括：本发明请求保护基于现场可编辑门阵列(FPGA)的主动噪声控制系统模块化设计方法。主要包括6个模块：白噪声信号发生器、卷积模块3、权值更新模块2、卷积模块2、权值更新模块1以及卷积模块1。本发明目的在于提高ANC系统的运算速度。创新点在于为了解决在实际的噪声环境下,最小均方(LMS)算法的收敛速度较慢,提出了基于动量最小均方(MLMS)算法主动噪声控制系统。使用1个权值更新模块和2个卷积模块实现ANC系统中动量LMS算法,为了对次级路径进行实时的建模使用1个权值更新模块和1个卷积模块实现动量LMS算法。这种模块化的ANC设计方法实现了一个快速的设计,具有快速收敛,为了提高系统的运算速度以及系统对数据的吞吐量,采用并行的处理方式进行主动噪声控制系统的设计。(The invention discloses a modular design method of an active noise control system based on a Field Programmable Gate Array (FPGA). Mainly include 6 modules: the white noise signal generator comprises a white noise signal generator, a convolution module 3, a weight updating module 2, a convolution module 2, a weight updating module 1 and a convolution module 1. The invention aims to improve the operation speed of an ANC system. The innovation point is that in order to solve the problem that the convergence speed of a Least Mean Square (LMS) algorithm is low in an actual noise environment, an active noise control system based on a Momentum Least Mean Square (MLMS) algorithm is provided. And the momentum LMS algorithm in the ANC system is realized by using 1 weight updating module and 2 convolution modules, and the momentum LMS algorithm is realized by using 1 weight updating module and 1 convolution module for carrying out real-time modeling on a secondary path. The modularized ANC design method realizes a rapid design, has rapid convergence, and adopts a parallel processing mode to design an active noise control system in order to improve the operation speed of the system and the throughput of the system to data.)

1. An active noise control system modular design method based on FPGA is used for an FPGA development board and is characterized by comprising the following steps: the white noise generator comprises a white noise signal generator, a first convolution module, a first weight updating module, a second convolution module, a second weight updating module and a third convolution module; the white noise signal generator is respectively connected with the first convolution module and the first weight updating module, the first weight updating module is connected with the second convolution module, the second convolution module and the first convolution module are also respectively connected with the second weight updating module, the second weight updating module is connected with the third convolution module, and the input signal of the noise source is respectively connected with the third convolution module and the second convolution module;

a white noise generator module: for generating white gaussian noise v_n(n)，v_n(n) injecting an input signal and a noise source into the secondary path to generate a signal that is not wanted to be correlated, and using a white noise generator to generate a white noise signal for the secondary path to solve the problem;

a first volume module: for Gaussian white noise v of input signal_n(N) FIR filter S with real weight N_n(n) filtering to obtain an output signal of the modeling filter, and generating v' (n) ═ s (n) × v (n);

the first weight value updating module: for inputting Gaussian white noise signal v_n(n) inputting the reconstructed error signal f (n) ═ e (n) — v' (n), multiplying the error signal f (n) — e (n) ((n)) to obtain f_n(n)v_n(n) is obtained after shift calculation

a second convolution module: for filtering reference noise signal

The second weight value updating module: updating the weight of the control filter w (z), and outputting the updated weight of w (z), specifically including: inputting Gaussian white noise signal x'_n(n) inputting the error signal f_n(n) is multiplied to obtain f_n(n)x′_n(n)，x′_n(n) represents a noise source x_n(n) Signal pass modeling Filter

a third convolution module: filtering the reference noise signal y (n) w (n) x (n), and superimposing the filtered signal y (n) with the white noise signal v (n) to obtain the anti-noise signal y (n) v (n).

2. The modular design method for the active noise control system based on the FPGA of claim 1, wherein the active noise control system of the momentum lms (mlms) algorithm is divided into five modular designs, three identical convolution modules, and two weight updating modules, and this design method improves the operation speed of the whole system, reduces the period of system design, and improves the throughput of the system.

3. The active noise control system modular design method based on the FPGA of claim 1, wherein the momentum LMS algorithm adds momentum terms to the original LMS algorithm:

W(n+1)＝W(n)-2μe(n)x′(n)+α[W(n)-W(n-1)]

α is a momentum factor, | α | <1, and the introduced momentum item can make the weight coefficient converge faster and more smoothly.

4. The modular design method for an FPGA-based active noise control system of claim 1, wherein the anti-noise signal data reflects the magnitude of its noise reduction performance and the accuracy of the secondary channel modeling according to the following formulas:

wherein R: the noise reduction performance of the ANC system is good or bad; e (n): the ANC system is used for controlling an error function of the adaptive filter; d (n): the ANC system is used for controlling a desired signal of the adaptive filter; Δ S: the accuracy of modeling of a secondary channel in the ANC system; s_i(n): a path function of an actual secondary channel in the ANC system;a path function of the secondary channel is simulated in the ANC system.

5. The modular design method for active noise control system based on FPGA of claim 1 or 2, characterized in that the secondary path is onlineNew adaptive filter

Technical Field

The invention belongs to the technical field of noise elimination, and particularly relates to a research on an active noise control system modular design method based on an FPGA.

Background

Noise pollution gradually develops into a problem to be solved urgently in large and medium-sized cities, particularly in areas where residential buildings and main roads are very close to each other. The long-term exposure to the noise environment can cause serious harm to the physiology and the psychology of people, and the normal work and life of people can be influenced by general noise interference. The main approach for solving the noise pollution of modern cities is to install a full-sealed sound insulation window, a double-layer glass window and the like, the cost of the measure in the high-rise buildings of the cities increases exponentially along with the increase of floors, and the measure only can block the noise of middle and high frequency bands, so that the control effect on the noise of low frequency bands is poor. However, in hot summer or in tropical high-temperature areas, the window cannot be ventilated and ventilated after being closed for a long time, so that the indoor air quality is affected, and the indoor temperature cannot be reduced, so that the indoor living comfort is reduced. Therefore, it is a real urgent need to develop an effective application for Active Noise Control (ANC) in high-rise houses while maintaining sufficient ventilation.

The traditional ventilation sound insulation window has obvious noise effect aiming at the middle and high frequency bands, has poor sound effect aiming at the low frequency bands, and has high manufacturing cost and heavy volume. Active noise control systems are widely used for noise in low frequency bands.

Active noise control systems typically employ a transversal adaptive filter for digital signal processing, which is traditionally implemented using a Digital Signal Processor (DSP) due to the need for fast floating-point arithmetic. With the integration level of the FPGA chip becoming higher and the continuous integration and development of the embedded digital signal processing module, the FPGA device has become a powerful competitor in the field of active noise control.

In a hardware design using a DSP for active noise control, an inherent software program needs to be executed in an early stage, which reduces the operating efficiency of the system. In contrast, designing an active noise control system using an FPGA yields higher efficiency than a DSP approach because the design is directly hardware-oriented. The text introduces an implementation method of an active noise control system modular design method based on an FPGA.

Disclosure of Invention

The present invention is directed to solving the above problems of the prior art. An active noise control system modular design method based on FPGA is provided. The technical method of the invention is as follows:

an active noise control system modular design method based on FPGA is used for an FPGA development board and comprises the following steps: the white noise generator comprises a white noise signal generator, a first convolution module, a first weight updating module, a second convolution module, a second weight updating module and a third convolution module; the white noise signal generator is respectively connected with the first convolution module and the first weight updating module, the first weight updating module is connected with the second convolution module, the second convolution module and the first convolution module are also respectively connected with the second weight updating module, the second weight updating module is connected with the third convolution module, and the input signal of the noise source is respectively connected with the third convolution module and the second convolution module;

a white noise generator module: for generating white gaussian noise v_n(n)，v_n(n) injecting the input signal in the secondary path with a noise source to generate a signal that is not wanted to be correlated, and to solve this problem, generating a white noise signal for the secondary path using a white noise generator.

A first volume module: for Gaussian white noise v of input signal_n(N) FIR filter S with real weight N_n(n) filtering to obtain an output signal of the modeling filter, and generating v' (n) ═ s (n) × v (n);

the first weight value updating module: for inputting Gaussian white noise signal v_n(n) inputting the reconstructed error signal f (n) ═ e (n) — v' (n), multiplying the error signal f (n) — e (n) ((n)) to obtain f_n(n)v_n(n) is obtained after shift calculationAnd S_n(n) adding the sum, and α (S) adding the result of the addition_n(n)-S_n-1(n)) obtaining the updated weight coefficient S_n(n +1), α represents momentum factor, | α<1; the first weight value updating module estimates a secondary path S (z) to obtain a filter

The weight coefficient of (a);

a second convolution module: for filtering reference noise signal

The second weight value updating module: updating the weight of the control filter w (z), and outputting the updated weight of w (z), specifically including: inputting Gaussian white noise signal x'_n(n) inputting the error signal f_n(n) is multiplied to obtain f_n(n)x′_n(n)，x′_n(n) represents a noise source x_n(n) Signal pass modeling Filter

The obtained filtering signal is obtained after shift calculation

And w_n(n) the integrated result is summed with momentum term α (w)_n(n)-w_n-1(n)) adding to obtain updated weight coefficient S_n(n +1), α represents momentum factor, | α<1；

A third convolution module: filtering the reference noise signal y (n) w (n) x (n), and superimposing the filtered signal y (n) with the white noise signal v (n) to obtain the anti-noise signal y (n) v (n).

The active noise control system is divided into five modularized designs, three identical convolution modules and two weight updating modules, and the design method improves the operation speed of the whole system and reduces the period of system design.

The momentum LMS algorithm adds momentum terms to the original LMS algorithm:

W(n+1)＝W(n)-2μe(n)x′(n)+α[W(n)-W(n-1)]

where α is the momentum factor, | α | < 1. the introduced momentum term may make the faster, smoother action of the weight coefficients convergence.

Further, the anti-noise signal data may reflect the magnitude of its noise reduction performance and the accuracy of the secondary channel modeling according to the following equations:

wherein R: the noise reduction performance of the ANC system is good or bad; e (n): the ANC system is used for controlling an error function of the adaptive filter; d (n): the ANC system is used for controlling a desired signal of the adaptive filter; Δ S: the accuracy of modeling of a secondary channel in the ANC system; s_i(n): a path function of an actual secondary channel in the ANC system;

a path function of the secondary channel is simulated in the ANC system.

Further, the secondary path updates the adaptive filter on line

For modeling the channel S (z), whereinUpdating the weight coefficients using the momentum LMS algorithm, u_sIs composed of

Step size parameter of, error signal of modeling filter

The main noise d (n) ═ p (n) × (n), the anti-noise signal y '(n) ═ s (n) × (n), y (n) ═ w (n) × (n) is the output signal of the filter w (z), v' (n) ═ s (n) (n) is the white noise signal v (n) passing through the modeling filter

V (n) is white additive noise of the input signal for modeling, the unit impulse responses of linear convolution, P (z), S (z) and W (z) are respectively expressed as p (n), s (n) and w (n), wherein the active noise control filter W (z) updates the weight by using the momentum LMS algorithm, u (n)_wFor the iteration step size of w (z),

is the reference signal x (n) is passedFiltered to obtain (g), (n), (e), (n), (d), (n) -y '(n) + v' (n)]To control the error signal of filter w (z).

The invention has the following advantages and beneficial effects:

the invention provides the realization of the active noise control system modular design method based on the FPGA by combining the problems in the ANC system, the convergence time of a control filter is greatly reduced due to the sensitivity of the dispersion degree of the characteristic value of the autocorrelation matrix of the reference signal in the LMS algorithm, and the convergence time of the whole ANC system is greatly reduced due to the variable step length algorithm adopted by the secondary path.

Active Noise Control (ANC) is a method of suppressing acoustic noise signals using electromechanical combination, mainly based on the principle of acoustic superposition. Compared with the traditional Passive Noise Control (PNC) method, the traditional noise control method can only reduce low-frequency signals with narrow frequency bands, and the required devices are large and heavy, and the application scenes are limited. The ANC system has good effects in the aspects of noise reduction of low-frequency noise, convenience in installation, stability of working performance and the like, and can offset noises with different characteristics by controlling parameters.

The active noise control system of the Momentum LMS (MLMS) algorithm is divided into five modularized designs, three identical convolution modules and two weight updating modules, and the design method improves the operation speed of the whole system, reduces the period of system design and improves the throughput of the system.

The technical key difficulty lies in that in order to solve the problem that in the traditional hardware design for active noise control by using a DSP, the operation efficiency of a system can be reduced because an inherent software program needs to be executed in the early stage, the active noise is designed by using an FPGA, and the operation efficiency higher than that of the DSP can be obtained because the active noise is directly designed facing hardware.

The reference signal x (n) generated by the noise source generates an interference signal d (n) through a main channel, the reference signal passes through a control filter w (z) to generate an anti-noise signal y' (n), and the output signal y (n) is generated by y (n) through a secondary path, and in order to ensure the stability of the momentum LMS algorithm on the update of the control filter weight, the reference signal x (n) must pass through a secondary modeling filter

In order to solve the problem that the secondary path changes along with time, the secondary modeling filter needs to be estimated on line, random noise irrelevant to a reference signal is injected into the secondary path, a white noise generator generates a group of random signals v (n), v (n) generates a modeling signal v' (n) through the secondary path, and the other end v (n) passes through the modeling filter

Generating a modeling signal v' (n), and generating an error signal f (n) by subtracting the participation error signal e (n) from the modeling signal. f (n) as an error signal for the momentum LMS algorithm and the momentum LMS algorithm.

The invention improves the performance of the whole ANC system in the noise reduction of low-frequency noise to a certain extent, and has the following outstanding advantages:

1. the convergence rate is high, the momentum LMS only adds a momentum term introduced due to the correlation of the weight coefficients compared with the LMS algorithm, and under the condition that the weight coefficients change greatly, the current weight coefficients are increased, so that the acceleration gradient reduction can be realized, and the convergence of the weight coefficient mean value is faster and more stable. After the momentum LMS algorithm is adopted, the value of the convergence coefficient is increased compared with the LMS algorithm, so that the sensitivity of the step length to the dispersion degree of the characteristic value of the reference signal autocorrelation matrix is reduced, and the convergence speed of the control filter is accelerated.

2. The method is characterized in that the operating frequency is high, the system is easy to migrate, the active noise control system of the whole active noise control system is designed in a modularized mode, the active noise control system of the Momentum LMS (MLMS) algorithm is divided into five modularized designs, three identical convolution modules and two weight updating modules, the operating speed of the whole system is improved by the aid of the design method, the period of system design is shortened, the throughput of the system is improved, the frequency obtained by the FPGA is 120MHz, and the whole system can obtain higher convergence speed by the aid of the modularized implementation of the ANC system.

3. According to the active noise control system modular design method based on the FPGA, the weight value is updated by using the following algorithm for a control filter.

In the equation, α is momentum factor, and is taken as | α<1,u_wRepresenting the step size parameter that controls the filter.

Drawings

FIG. 1 is a block diagram of a secondary path online modeling system proposed by a preferred embodiment of the present invention;

FIG. 2 is a block diagram of a hardware implementation structure of a secondary path online identification ANC system;

FIG. 3 is a diagram of the comprehensive simulation results of the ANC system, wherein FIG. 3(a) shows the comparison of the output residual error with the residual error after simulation on MATLAB; fig. 3(b) shows a diagram of the theoretical residual error of the model sim simulation and algorithm of the hardware code.

Detailed Description

The technical method in the embodiment of the invention will be described in detail in the following with reference to the accompanying drawings in the embodiment of the invention. The described embodiments are only some of the embodiments of the present invention.

The technical method for solving the technical problems comprises the following steps:

the ANC system provided by the invention adopts Quartus II for simulation.

As shown in fig. 1, the reference signal is first filtered through the secondary path s (z), and then the filtered signal x' (n) is updated with the weight coefficients of the adaptive filter w (z). Secondary path online update adaptive filter

For modeling the channel S (z), whereinUpdating the weight coefficients using the momentum LMS algorithm, u_sIs composed of

Step size parameter of, error signal of modeling filter

The main noise d (n) ═ p (n) × (n), the anti-noise signal y '(n) ═ s (n) × (n), y (n) ═ w (n) × (n) is the output signal of the filter w (z), v' (n) ═ s (n) (n) is the white noise signal v (n) passing through the modeling filter

V (n) is white additive noise for the modeled input signal, the unit impulse responses of which represent linear convolution, p (z), s (z) and w (z) are denoted as p (n), s (n), w (n), respectively. Wherein the active noise control filter w (z) uses a momentum LMS algorithm to update the weights. u. of_wFor the iteration step size of w (z),

is the reference signal x (n) is passed

Filtered to obtain (g), (n), (e), (n), (d), (n) -y '(n) + v' (n)]To control the error signal of filter w (z). Test forIn the experiment, a reference noise pickup, an error pickup, a cancellation loudspeaker and an Artix7 series FPGA development board are adopted, two WM8731 audio codecs are used in the development board, one WM8731 chip carries out audio acquisition on a reference noise signal, the other WM8731 chip carries out audio acquisition on the error pickup, and the cancellation noise signal is transmitted to the cancellation loudspeaker after being decoded.

As shown in fig. 2, the present invention provides a method for modular design of an active noise control system based on an FPGA, which mainly comprises 6 modules: (1) the white noise signal generator comprises a white noise signal generator, (2) a convolution module 1, (3) a weight updating module 1, (4) a convolution module 2, (5) a weight updating module 2 and (6) a convolution module 3.

White noise generator module (1) for generating white gaussian noise v_n(n)，v_n(n) injecting the input signal in the secondary path with a noise source to generate a signal that is not wanted to be correlated, and to solve this problem, generating a white noise signal for the secondary path using a white noise generator.

Convolution module 1: input signal gaussian white noise v_n(N) by FIR filter S with real weight N_nAnd (n) filtering to obtain an output signal of the modeling filter.

Weight updating module 1: for inputting Gaussian white noise signal v_n(n) inputting the reconstructed error signal f (n) ═ e (n) — v' (n), multiplying the error signal f (n) — e (n) ((n)) to obtain f_n(n)v_n(n) is obtained after shift calculation

And S_n(n) adding the sum, and α (S) adding the result of the addition_n(n)-S_n-1(n)) obtaining the updated weight coefficient S_n(n +1), α represents momentum factor, | α<1; the first weight value updating module estimates a secondary path S (z) to obtain a filter

The weight coefficient of (a);

and a convolution module 2: input signal gaussian white noise x_n(N) by FIR filter S with real weight N_n(n) Filtering x_n(n) obtaining the output signal of the filter.

A weight value updating module 2: updating the weight of the control filter w (z), and outputting the updated weight of w (z), specifically including: inputting Gaussian white noise signal x'_n(n) inputting the error signal f_n(n) is multiplied to obtain f_n(n)x′_n(n)，x′_n(n) represents a noise source x_n(n) Signal pass modeling Filter

The obtained filtering signal is obtained after shift calculation

And w_n(n) the integrated result is summed with momentum term α (w)_n(n)-w_n-1(n)) adding to obtain updated weight coefficient S_n(n +1), α represents momentum factor, | α<1；

And a convolution module 3: input signal gaussian white noise x_n(N) passing through FIR filter W with real weight N_n(n) Filtering x_n(n) obtaining the output signal y of the filter_n(n)。

An adder 2: input signal y_n(n) and additive white Gaussian noise v_n(n) the subtraction produces an anti-noise signal output.

The modular design of the ANC system is divided into 5 modules, 3 convolution modules and 2 LMS algorithm weight updating modules. As shown in fig. 2, the 1 st convolution module generates v '(n) ═ s (n) × v (n), outputs the reconstructed error signal f (n) ═ e (n) — v' (n) by the adder, and the 1 st LMS weight update module estimates the secondary path s (z) to obtain the filter

The 2 nd convolution module filters the reference noise signal to obtainThe 2 nd LMS weight value updating module updates the weight value of the control filter W (z) and outputsAnd the 3 rd convolution module filters the reference noise signal y (n) (w) (n) × (n), and the filtered signal y (n) is superposed with the white noise signal v (n) to obtain the anti-noise signal y (n) -v (n).

The resulting data may reflect the magnitude of its noise reduction performance and the accuracy of the secondary channel modeling according to the following equations:

wherein R: the noise reduction performance of the ANC system is good or bad;

e (n): the ANC system is used for controlling an error function of the adaptive filter;

d (n): the ANC system is used for controlling a desired signal of the adaptive filter;

Δ S: the accuracy of modeling of a secondary channel in the ANC system;

S_i(n): a path function of an actual secondary channel in the ANC system;

simulating path functions of secondary channels in an ANC system

As shown in fig. 3(a), the residual error e (n) of the ANC system is simulated and compared, the algorithm is simulated by MATLAB, the design method is realized by joint simulation of QUARTUS ii and modelmin, and the simulated output residual error is compared with the residual error simulated on MATLAB, so that the simulation result of the modular active noise-based design on QUARTUS ii is the same as the theoretical value, and the accuracy of the algorithm is ensured.

As shown in fig. 3(b), it can be seen that the model sim simulation of the hardware code is substantially consistent with the theoretical residual error of the algorithm. It can be seen from the figure that the FPGA design of the active noise control system is identified on line by the secondary path to perform adaptive noise reduction, and after a period of stable operation, the expected noise reduction effect is achieved, which illustrates the correctness of the design idea.

The above examples are to be construed as merely illustrative and not limitative of the remainder of the disclosure. After reading the description of the invention, the skilled person can make various changes or modifications to the invention, and these equivalent changes and modifications also fall into the scope of the invention defined by the claims.