CN103915091A - Active noise control method based on adaptive algorithm free of secondary channel modeling - Google Patents

Abstract

The invention provides an active noise control method based on an adaptive algorithm free of secondary channel modeling. Due to dependency on a secondary channel model, an existing self-adaption active noise control system is lack of adaptability on a wide system. Based on the active noise control algorithm free of secondary channel modeling, a self-adaption model-free control method based on a synchronous disturbance stochastic approximation algorithm is provided, and the problem of the uncertainty of a secondary channel in the active noise control process is solved.

Description

Active Noise Control Method Based on Adaptive Algorithm Without Secondary Channel Modeling

technical field

The invention belongs to the field of vibration, impact and noise, and is an active noise control method based on an adaptive ANC algorithm without secondary channel modeling.

Background technique

Because it does not involve the complex spatial sound field environment, the noise reduction effect of the LMS algorithm in the simulation environment is remarkable. In the actual sound field environment, due to the existence of the secondary channel, the output signal of the adaptive filter can only be sent to the error microphone through the secondary channel, so the traditional adaptive algorithm cannot be directly applied to the active noise control.

Contents of the invention

In order to overcome the deficiencies of the prior art, the object of the present invention is to provide an active noise control method based on an adaptive algorithm without secondary channel modeling.

An active noise control method based on an adaptive algorithm without secondary channel modeling. The noise signal generated by the engine is collected through a microphone, and the secondary sound is generated through an adaptive ANC algorithm without secondary channel modeling, combined with an error microphone. The source signal is played through the indoor noise reduction speakers, and the active noise control is realized in the limited space of the cabin through the superposition of the secondary noise and the primary noise in the sound field space.

Described adaptive ANC algorithm comprises the steps:

A. Determine the update direction of the weight coefficient

①Do not update the weight coefficient of the adaptive filter first, select N sets of sample data, and calculate the error signal power Maximum error signal amplitude e _max =max(|e(i)|), reference signal power Among them, N is a natural number, e(i) is an error signal, and x(i) is a reference signal;

②After updating the weight coefficients of the adaptive filter, select N sets of sample data by generating the synchronous disturbance vector _Δk , and calculate the error signal power and reference signal power like

|e(i)|＞(1+δ ₂ )e _max will stop updating;

③If Or |e(i)|>(1+δ ₂ )e _max , then the sign before the step size μ is changed.

B. Adaptive update weight coefficient

④Update the weight coefficient of the adaptive filter;

C. Performance monitoring

⑤ When n=1, initialize χ(0)=χ ₁ , ξ(0)=ξ ₁ ;

⑥Use ξ(n)=λξ(n-1)+e ² (n) to calculate the average power of the error signal, and use χ(n)=λχ(n-1)+x ² (n) to calculate the average power of the reference signal, where λ∈[0.5,1), is the forgetting factor, take

⑦ if $\frac{\mathrm{\ξ}\left(\mathrm{no}\right)}{\mathrm{\χ}\left(\mathrm{no}\right)}>(1+{\mathrm{\δ}}_1)c\′\frac{\mathrm{\ξ}(\mathrm{no}-N)}{\mathrm{\χ}(\mathrm{no}-N)}$ or $\frac{\mathrm{\ξ}\left(\mathrm{no}\right)}{\mathrm{\χ}\left(\mathrm{no}\right)}>(1+{\mathrm{\δ}}_1)c\′\frac{{\mathrm{\ξ}}_1}{{\mathrm{\χ}}_1},$ Then return to step ① to re-search the direction of weight coefficient update, otherwise return to step ④ to continue to update the weight coefficient.

Beneficial effects of the present invention: by comparing in real time the reference signal energy value and error signal energy value collected into the system each time with the reference signal energy value and error signal energy value collected into the system in the previous time unit, the direction of the step factor is dynamically updated , so that the weight coefficient of the adaptive filter is adjusted in real time, and then the corresponding secondary sound source control amount is generated to achieve a stable noise reduction effect. This algorithm is especially suitable for closed spaces where the space is extremely irregular and the sound field environment is extremely complex due to the influence of structural design and space layout. Description of drawings

The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

Figure 1 is a schematic diagram of the principle of an active noise control system based on an adaptive ANC algorithm without secondary channel modeling.

Detailed ways

The principle of the active noise control system based on the adaptive ANC algorithm without secondary channel modeling in the present invention is shown in FIG. 1 . When the secondary channel is fixed in the control process, the FX-LMS algorithm controlled by the secondary channel model based on off-line modeling technology can achieve a satisfactory noise reduction effect. However, in practical applications, the secondary channel will continue to change with the environmental parameters of the sound field and noise characteristics. If the secondary channel model based on off-line modeling technology is continued to be used for system control, the noise reduction performance of the active noise control system will be reduced. deterioration, and even system instability. An effective way to solve this problem is to use a control method that does not require a secondary channel model for noise control. In the present invention, the need for secondary channel modeling is not to ignore the impact of the secondary channel on the noise reduction performance of the actual system, but to reflect the dynamic change of the secondary channel in the adaptive update process of the system in other ways. By comparing the reference signal energy value and error signal energy value collected into the system each time in real time with the reference signal energy value and error signal energy value collected into the system in the previous time unit, the direction of the step factor is dynamically updated, thereby adjusting the self-adaptation in real time The weight coefficient of the filter, and then generate the corresponding secondary sound source control amount to achieve a stable noise reduction effect.

Using the weight updating formula of FX-LMS algorithm

w(n)=w(n-1)+μe(n)r _s (n) (1)

Assuming that the input signal is a pure sinusoidal component r _ω (n) with frequency ω, use and W _ω (n) denote the transfer functions of the primary channel, secondary channel, secondary channel estimation and adaptive controller at frequency ω. but

$\begin{array}{c}{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&mu;W</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; S</span>}}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; [</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; r</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; ]</span>\\ <span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; \&mu;P</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&omega;</span>}<span\; class="notranslate">\; )</span>{\stackrel{<span\; class="notranslate">\; ^</span>}{\mathrm{<span\; class="notranslate">\; S</span>}}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; [</span>\frac{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}}{{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; W</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ]</span>\end{array}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

where P _r (ω) represents the reference noise power at frequency ω. After the algorithm converges, there are W _ω (n)=W _ω (n-1) and W _ω (∞)=P _ω /S _ω . When there is no error between the secondary channel model and the real secondary channel, formula (2) becomes

W _ω (n)＝W _ω (n-1)+μP _r (ω)|S _ω | ² [P _ω /S _ω -W _ω (n-1)] (3)

Obviously, W _ω (n) updates the weight coefficients of the adaptive filter along the direction of P _ω /S _ω starting from W _ω (n-1), and the distance of each iteration of W _ω (n) is _μ P _r (ω )|S _ω | ² [P _ω /S _ω -W _ω (n-1)]. In order to ensure that the system can converge, _μ P _r (ω)|S _ω | ² <2 is required, then

$\mathrm{<span\; class="notranslate">\; \&mu;</span>}<span\; class="notranslate">\; <</span>\frac{\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; \&angle;</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; P</span>}}_{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; \&omega;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; |</span>{\mathrm{<span\; class="notranslate">\; S</span>}}_{\mathrm{<span\; class="notranslate">\; \&omega;</span>}}<span\; class="notranslate">\; |</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

Among them, ∠S _ω represents the phase of S _ω , and |S _ω | represents the magnitude of S _ω . Only when ∠S _ω ∈[-90°,90 _o °, the algorithm converges. Otherwise, the algorithm convergence can be ensured by changing the sign before the step size μ, that is

w(n)=w(n-1)+μe(n)r(n),∠S _ω ∈[-90°,90°] (5)

w(n)=w(n-1)-μe(n)r(n), otherwise (6)

A. Determine the update direction of the weight coefficient

①Do not update the weight coefficient of the adaptive filter first, select N sets of sample data, and calculate the error signal power Maximum error signal amplitude e _max =max(|e(i)|), reference signal power Among them, N is a natural number, e(i) is an error signal, and x(i) is a reference signal;

②After updating the weight coefficients of the adaptive filter, select N sets of sample data by generating the synchronous disturbance vector _Δk , and calculate the error signal power and reference signal power If |e(i)|＞(1+δ ₂ )e _max , stop updating;

③If Or |e(i)|>(1+δ ₂ )e _max , then the sign before the step size μ is changed.

B. Adaptive update weight coefficient

④Update the weight coefficient of the adaptive filter;

C. Performance monitoring

⑤ When n=1, initialize χ(0)=χ ₁ , ξ(0)=ξ ₁ ;

⑥Use ξ(n)=λξ(n-1)+e ² (n) to calculate the average power of the error signal, and use χ(n)=λχ(n-1)+x ² (n) to calculate the average power of the reference signal, where λ∈[0.5,1), is the forgetting factor, take

⑦ if $\frac{\mathrm{\&xi;}\left(\mathrm{no}\right)}{\mathrm{\&chi;}\left(\mathrm{no}\right)}>(1+{\mathrm{\&delta;}}_1)c\&prime;\frac{\mathrm{\&xi;}(\mathrm{no}-N)}{\mathrm{\&chi;}(\mathrm{no}-N)}$ or $\frac{\mathrm{\&xi;}\left(\mathrm{no}\right)}{\mathrm{\&chi;}\left(\mathrm{no}\right)}>(1+{\mathrm{\&delta;}}_1)c\&prime;\frac{{\mathrm{\&xi;}}_1}{{\mathrm{\&chi;}}_1},$ Then return to step ① to re-search the direction of weight coefficient update, otherwise return to step ④ to continue to update the weight coefficient.

The above is the process of the self-adaptive ANC algorithm without secondary channel modeling adopted by the present invention, which can save the secondary channel estimation link compared with the FX-LMS algorithm. The adaptive ANC algorithm without secondary channel modeling can be summarized into three parts. The most important thing is the selection of the update direction μ of the adaptive filter. First, μ is initialized to a sufficiently small value that meets the convergence condition, and then the residual noise energy is calculated. If the noise energy increases, change the sign of μ.

Claims (2)

1. the active noise control method of the adaptive algorithm based on without secondary channel modeling, it is characterized in that: gather by microphone the noise signal that engine produces, through the self-adaptation ANC algorithm without secondary channel modeling, in conjunction with error microphone, produce secondary sound source signal, play through indoor noise reduction loudspeaker, the stack by secondary noise and elementary noise in sound field space, in cabin, the finite space realizes Active noise control using.

2. method according to claim 1, is characterized in that: described self-adaptation ANC algorithm comprises the steps:

A, determine weight coefficient upgrade direction

1. first do not upgrade the weight coefficient of sef-adapting filter, choose N group sample data, error signal power maximum error signal amplitude e _max=max (| e (i) |), reference signal power wherein N is natural number, and e (i) is error signal, and x (i) is reference signal;

2. upgrade after the weight coefficient of sef-adapting filter, by producing simultaneous perturbation vector Δ _kchoose N group sample data, error signal power and reference signal power if | e (i) | > (1+ δ ₂) e _maxstop upgrading;

If 3. or | e (i) | > (1+ δ ₂) e _max, by the sign modification before step size mu.

B, adaptive updates weight coefficient

4. upgrade the weight coefficient of sef-adapting filter;

C, performance monitoring

5. in the time of n=1, initialization χ (0)=χ ₁, ξ (0)=ξ ₁;

6. utilize ξ (n)=λ ξ (n-1)+e ²(n) error signal average power, utilizes χ (n)=λ χ (n-1)+x ²(n) computing reference average power signal, wherein λ ∈ [0.5,1), be forgetting factor, get

If 7. $\frac{\mathrm{\&xi;}\left(n\right)}{\mathrm{\&chi;}\left(n\right)}>(1+{\mathrm{\&delta;}}_1)c\&prime;\frac{\mathrm{\&xi;}(n-N)}{\mathrm{\&chi;}(n-N)}$ Or $\frac{\mathrm{\&xi;}\left(n\right)}{\mathrm{\&chi;}\left(n\right)}>(1+{\mathrm{\&delta;}}_1)c\&prime;\frac{{\mathrm{\&xi;}}_1}{{\mathrm{\&chi;}}_1},$ Get back to 1. right of search coefficient update direction again of step, 4. continue to upgrade weight coefficient otherwise return.