CN108600894A - A kind of earphone adaptive active noise control system and method - Google Patents

Abstract

The invention discloses an earphone self-adaptive active noise control system and method. The system at least includes a reference microphone for collecting and collecting noise signals, and an error microphone for collecting error noise signals. The reference microphone is arranged outside the earphone. , the error microphone is set inside the earphone; a speaker unit is used to play the secondary sound signal; two analog-to-digital converters AD are used to convert the reference noise signal and the error noise signal into corresponding digital signals; a digital-to-analog converter A device DA is used to convert the filtered output signal into an analog signal and feed it to the secondary speaker; a processor is used to update the filter coefficient and generate the input signal of the secondary speaker. The invention can solve the problems of insufficient real-time tracking ability and excessive calculation delay of ordinary adaptive algorithms, and has good adaptive performance for different environments and different earphone wearing methods, and improves the performance of adaptive active noise control of earphones. Convergence performance and tracking performance.

Description

A headphone adaptive active noise control system and method

technical field

The invention relates to the field of noise elimination, in particular to an earphone adaptive active noise control system and method.

Background technique

Noise affects people's life all the time. For example, in the cabin, subway and office environment, people are disturbed by noise all the time. The existence of noise makes people feel restless and restless. Harm has always been one of the important goals of scientific research. Scientific research shows that long-term exposure to strong noise can easily lead to permanent loss of all or part of the hearing frequency sensitivity, and even cause deafness. There are two main ways to reduce noise: active noise control and passive noise control. Passive noise control mainly uses the reflection of special materials to isolate sound transmission or uses the porosity and viscosity of materials to attenuate the passing sound energy. Passive noise control methods are more effective for high-frequency noise. Active noise-canceling earphones use active noise control technology to generate an acoustic signal inside the earphone that has the same amplitude as the noise but has an opposite phase, so that the secondary sound waves at the target and the sound waves in the original sound field cancel each other out, thereby achieving the purpose of noise control. Noise control technology works better for low frequency noise.

A key issue in active noise control is how to update the filter coefficients. The FxLMS (Filtered-xleast mean square) algorithm is the most popular adaptive algorithm for active noise control. However, due to its implementation in the time domain, the algorithm has high computational complexity and is limited in resource-limited systems. In addition, This algorithm belongs to the LMS algorithm in essence, and its convergence speed depends on the condition number of the autocorrelation matrix of the input signal. Frequency-domain adaptive filtering algorithm has been widely used in echo cancellation, beamforming and howling suppression due to its low computational complexity and good convergence performance. There are also literatures that have attempted to use frequency-domain adaptive algorithms for active noise control. For example: the document "Q.Shen and ASSpanias, "Time and frequency domain xblock LMS algorithms for single channel active noise control," in Proceedings of the 2nd International Congress of Recent Developments in Air-and Structure-Borne SoundVibration, 1992, pp.353–360" first proposed the Frequency domain algorithm applied to active noise control, literature "Das,DP,G.Panda,andS.M.Kuo."New block filtered-X LMS algorithms for active noisecontrolsystems."IET Signal Processing 1.2(2007):73-81 》A more thorough frequency-domain adaptive active control technique is further discussed. However, because these methods are based on block processing, data caching and processing introduce delays, which will not cause performance degradation for system identification applications, but only bring algorithm delays, but the impact on active noise control systems But it is very large, because the delay may directly cause the active system to be non-causal, especially for the application of headphones, the distance between the feedforward microphone and the secondary speaker is very short, so the delay allowed by the algorithm is very small. For example, in most earphones, the typical value of the distance between the feed-forward microphone and the loudspeaker is d = 1 cm, assuming the sampling rate f _S = 16000 Hz, then the maximum delay of the algorithm allowed to ensure the causality of the system is 4.7 sampling periods, and the directly implemented frequency domain algorithm usually introduces a delay of, for example, 128 sampling periods, so it is not feasible to directly apply the frequency domain algorithm to the earphone active noise reduction system. In order to solve this problem, the document "X.Qiu and CHHansen, "Multidelay adaptive filters for active noise control," in Proceedings of the 14th International Congress on Sound and Vibration (ICSV), Cairns, Australia, 2007, pp.724–732" adopts no delay algorithm, but the peak complexity of this method is very high, because the algorithm requires a series of complex operations such as Fourier transform to be completed within one sampling period, which is a great challenge for most DSP chips .

Another problem with frequency-domain adaptive algorithms is that the choice of step size must be a compromise between convergence speed and steady-state misalignment. A larger step size can ensure that the algorithm has a faster initial convergence speed, and has the ability to quickly re-track when the system changes, but a large step size can easily make the algorithm have a larger misalignment in the steady state, which is is an undesirable property. In the actual environment, the transfer functions of the main channel and the secondary channel are time-varying. For example, the size of the ears of different people is different, so the secondary channel is different after wearing the earphones. In practice, the wearer will often adjust The tightness of the earphones causes changes in the above two transfer functions. Therefore, the adaptive algorithm must have good real-time tracking performance.

Contents of the invention

In view of the above-mentioned deficiencies existing in the prior art, the purpose of the present invention is to provide an earphone adaptive active noise control system and method, which can solve the problems of insufficient real-time tracking ability and excessive calculation delay of ordinary adaptive algorithms, and is suitable for different The environment and different earphone wearing styles have good adaptive performance, which improves the convergence performance and tracking performance of the adaptive active noise control of the earphone.

An earphone adaptive active noise control system, the system at least includes:

A first audio collector and a second audio collector, the first audio collector is installed outside the earphone for collecting reference noise signals; the second audio collector is installed inside the earphone for collecting error noise signals after noise reduction;

The first analog-to-digital converter AD and the second analog-to-digital converter AD, the first analog-to-digital converter AD is electrically connected to the first audio collector, and is used to convert the reference noise signal collected by the first audio collector into a corresponding digital signal, the second analog-to-digital converter AD is electrically connected to the second audio collector, and is used to convert the error noise signal collected by the second audio collector into a corresponding digital signal;

The processor unit, the first analog-to-digital converter AD is connected to the first information input end of the processor unit, and the second analog-to-digital converter AD is connected to the second information input end of the processor unit;

The digital-to-analog converter DA and the secondary speaker, the information output end of the processor unit is connected to the secondary speaker after the digital-to-analog converter DA, the secondary speaker is installed in the earphone, and the digital-to-analog converter DA is used to convert the output of the processor unit The filtered signal is converted to a corresponding analog signal and fed to a secondary speaker, which is used to play the secondary sound signal.

Further, the control system also includes a memory unit, and the memory unit and the processor unit are connected in bidirectional communication.

Further, the control system also includes a power module, which is connected to a power supply terminal of the processor unit.

Further, the control system also includes a control logic unit, which is used to control the switch of the active noise reduction function.

A kind of earphone adaptive active noise control method, comprises the following steps:

The reference noise signal is collected in real time by the first audio collector, and the reference noise signal is converted into a corresponding digital signal x(n) by the first analog-to-digital converter and then transmitted to the processor unit; the error is collected in real time by the second audio collector noise signal, and the error noise signal is converted into a corresponding digital signal e(n) by the second analog-to-digital converter and then transmitted to the processor unit;

Calculation of the filter coefficient vector by the processor unit and calculate the output filtered signal y(n), and use the Kalman filter to calculate the filter coefficient vector in the frequency domain In the time domain for the digital signal x(n) and the time domain filter coefficients Perform convolution to obtain the output filtered signal y(n); where, the time domain filter coefficient is the filter coefficient vector The inverse Fourier transform of ;

The output filtered signal y(n) is converted into a corresponding analog signal by a digital-to-analog converter and fed to the secondary speaker, which is used to play the secondary sound signal.

Further, the Kalman filter is used to calculate the filter coefficient vector in the frequency domain The steps are as follows:

In the frequency domain, the digital signal x(n) corresponding to the reference noise signal is estimated by the secondary channel Filter to obtain the Kalman filter input signal v(n);

Fourier transform the sequence composed of the nearest 2L-point Kalman filter input signal v(n) to obtain the frequency domain vector v(k), and place the elements of the frequency domain vector v(k) on the diagonal to form the diagonal matrix V(k);

Perform Fourier transform on the sequence composed of the nearest L point error noise signal e(n) to obtain the frequency domain vector E(k);

Update the Kalman gain matrix K(k)=P(k)V ^H (k)[V(k)P(k)V ^H (k)+2ψ _ee (k)] ^-1 , where P(k) is The state error matrix, ψ _ee (k) is the diagonal covariance matrix of the observation noise error, and the superscript H represents the conjugate transpose operation;

Update frequency-domain filter coefficient vector in is a constraint matrix, I _L is an identity matrix whose dimension is L×L, 0 _L is a zero matrix whose dimension is L×L, F is a Fourier transform matrix, and K(k) is a Kalman gain matrix;

Update state error matrix P(k): Among them, ψ _ΔΔ (k) is the process noise diagonal matrix, and I _2L is the identity matrix whose dimension is 2L×2L.

Further, in order to calculate the Kalman gain matrix K (k), we need to know ψ _ee (k), the observation noise error matrix ψ _ee (k) in the present invention is calculated by using the smooth power spectrum of the error noise signal e (n) get.

Further, in order to calculate the state error matrix P(k), we also need to calculate the process noise covariance matrix ψ _ΔΔ (k), the diagonal element ψ of the process noise diagonal matrix ψ _ΔΔ (k) described in the present invention _ΔΔ,i (k) is calculated by in, is the filter coefficient vector The i-th element of .

Compared with the prior art, the present invention has the following advantages:

An earphone self-adaptive active noise control system and method disclosed in the present invention adopts a Kalman filter algorithm to update the filter coefficients in the frequency domain. Since the Kalman filter algorithm has good tracking performance, it solves the problem of common self-adaptive algorithms. The problem of insufficient tracking ability makes it possible to meet different application environments and different wearing styles of headphones. At the same time, the filter coefficients in the frequency domain are converted to the time domain, and the output filter is directly calculated by convolution in the time domain, thereby avoiding the block delay introduced by the frequency domain calculation. Through the control system and method, the convergence performance and tracking performance of calculation are improved.

Description of drawings

Fig. 1 is a system block diagram of an earphone adaptive active noise control system in an embodiment of the present invention;

FIG. 2 is a schematic diagram of control of earphone adaptive active noise in an embodiment of the present invention;

Fig. 3 is the filtering flowchart of processor unit in the embodiment of the present invention;

Fig. 4 is the flowchart of adopting Kalman filter in the frequency domain in the embodiment of the present invention;

Fig. 5 is a flow chart of calculating the power spectrum of the observation noise in the embodiment of the present invention.

Reference signs:

102. A first audio collector; 104. A second audio collector; 106. A secondary speaker; 108. A first analog-to-digital converter AD; 110. A digital-to-analog converter DA; 112. A second analog-to-digital converter AD; 114, estimated secondary channel; 116, filter time domain coefficient; 118, Kalman filter; 120, controller unit; 150, memory unit; 160, power module; 170, control logic unit; 180, earphone; 202, Shift unit; 204, shift unit; 206, shift unit; 208, multiplier; 210, multiplier; 212, multiplier; 214, multiplier; 216, adder; 218, adder; 220, adder ; 302, Fourier transform unit; 304, Kalman gain matrix; 306, Fourier transform unit; 308, multiplier; 310, inverse Fourier transform operation unit; 312, the second half of the sequence is set to zero; 314, Fourier transform; 316, adder; 318, delay unit; 320, inverse Fourier transform; 402, conjugate module; 404, multiplication unit; 406, multiplication unit; 408, addition unit; 410, multiplication unit ; 412, delay unit.

Detailed ways

Embodiments of the technical solutions of the present invention will be described in detail below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solution of the present invention more clearly, so they are only examples, and should not be used to limit the protection scope of the present invention.

Example:

Referring to Figure 1 and Figure 2, a headphone adaptive active noise control system, the system includes:

The first audio collector 102 and the second audio collector 104, the first audio collector is installed outside the earphone 180 for collecting reference noise signals; the second audio collector is installed inside the earphone for collecting errors after noise reduction noise signal;

The first analog-to-digital converter AD108 and the second analog-to-digital converter AD112, the first analog-to-digital converter AD is electrically connected to the first audio collector, for converting the reference noise signal collected by the first audio collector into corresponding digital Signal x(n), the second analog-to-digital converter AD is electrically connected to the second audio collector, and is used to convert the error noise signal collected by the second audio collector into a corresponding digital signal e(n);

Processor unit 140, the first analog-to-digital converter AD is connected to the first information input end of the processor unit, and the second analog-to-digital converter AD is connected to the second information input end of the processor unit; digital signal x(n) and digital signal e(n) is input to the processor unit, and the processor unit analyzes and obtains the time domain coefficient of the filter 116 and the output filter signal y (n); the processor unit 140 has two functions, one is to utilize the digital signal x (n) corresponding to the reference noise signal and the digital signal e (n) corresponding to the error noise signal to update the filter coefficient vector The second is to calculate the filtered output signal y(n). where the filter coefficient vector is updated frame by frame in the frequency domain, and the filtered output signal y(n) is calculated point by point in the time domain.

Digital-to-analog converter DA110 and secondary loudspeaker 106, the information output end of the processor unit is connected with the secondary loudspeaker after the digital-to-analog converter DA, the secondary loudspeaker is installed in the earphone, and the digital-to-analog converter DA is used to convert the processor unit The output filtered signal is converted into a corresponding analog signal and fed to the secondary speaker for playing the secondary sound signal; the output filtered signal y(n) is sent to the secondary speaker 106 for playback after passing through the digital-to-analog converter DA110 . During specific implementation, both the first audio collector and the second audio collector may use microphones.

In this embodiment, the control system further includes a memory unit 150, and the memory unit is connected to the processor unit in bidirectional communication. The control system also includes a power module 160, which is connected to the power supply terminal of the processor unit. The control system also includes a control logic unit 170 for controlling the switch of the active noise reduction function. During specific implementation, the processor unit may adopt DSP, ARM or other special-purpose processor chips. The memory unit is used to store programs and variables. The control system also includes an installation housing and a circuit board, the installation housing is provided with a circuit board installation cavity, the second audio collector, the first analog-to-digital converter, the second analog-to-digital converter, the processor unit, the digital-to-analog Both the converter and the secondary speakers are mounted on the circuit board. An installation base is provided on the outer surface of the installation housing, and the first audio collector is installed on the installation base.

Referring to Figures 2 to 5, an adaptive active noise control method for earphones includes the following steps:

The reference noise signal is collected in real time by the first audio collector, and the reference noise signal is converted into a corresponding digital signal x(n) by the first analog-to-digital converter and then transmitted to the processor unit; the error is collected in real time by the second audio collector noise signal, and the error noise signal is converted into a corresponding digital signal e(n) by the second analog-to-digital converter and then transmitted to the processor unit;

Calculation of the filter coefficient vector by the processor unit and calculate the output filtered signal y(n), and use the Kalman filter 118 to calculate the filter coefficient vector in the frequency domain In the time domain for the digital signal x(n) and the time domain filter coefficients Perform convolution to obtain the output filtered signal y(n); where, the time domain filter coefficient is the filter coefficient vector The inverse Fourier transform of ;

The output filtered signal y(n) is converted into a corresponding analog signal by a digital-to-analog converter and fed to the secondary speaker, which is used to play the secondary sound signal.

In this embodiment, the Kalman filter is used to calculate the filter coefficient vector in the frequency domain The steps are as follows:

In the frequency domain, the digital signal x(n) corresponding to the reference noise signal is estimated by the secondary channel Filter to obtain the Kalman filter input signal v(n);

Fourier transform the sequence composed of the nearest 2L-point Kalman filter input signal v(n) to obtain the frequency domain vector v(k), and place the elements of the frequency domain vector v(k) on the diagonal to form the diagonal matrix V(k);

Perform Fourier transform on the sequence composed of the nearest L point error noise signal e(n) to obtain the frequency domain vector E(k);

Update the Kalman gain matrix K(k)=P(k)V ^H (k)[V(k)P(k)V ^H (k)+2ψ _ee (k)] ^-1 , where P(k) is The state error matrix, ψ _ee (k) is the diagonal covariance matrix of the observation noise error, and the superscript H represents the conjugate transpose operation;

Update frequency-domain filter coefficient vector in is a constraint matrix, I _L is an identity matrix whose dimension is L×L, 0 _L is a zero matrix whose dimension is L×L, F is a Fourier transform matrix, and K(k) is a Kalman gain matrix;

Update state error matrix P(k): Among them, ψ _ΔΔ (k) is the process noise diagonal matrix, and I _2L is the identity matrix whose dimension is 2L×2L.

During specific implementation, in order to calculate the Kalman gain matrix K (k), we need to know ψ _ee (k), the observation noise error matrix ψ _ee (k) in the present invention adopts the smooth power spectrum calculation of the error noise signal e (n) get. In order to calculate the state error matrix P(k), we also need to calculate the process noise covariance matrix ψ _ΔΔ (k), the diagonal elements of the process noise diagonal matrix ψ _ΔΔ (k) in the present invention ψ _ΔΔ,i (k ) is calculated as:

in, is the filter coefficient vector The i-th element of . The controller unit 120 consists of a first analog-to-digital converter AD, a digital-to-analog converter DA, a second analog-to-digital converter AD, estimated secondary channels, filtered time-domain coefficients and Kalman filtering.

Below, we will describe the specific implementation of the 116 module in Figure 2: with reference to Figure 3, the calculation process of the output filter signal y(n), wherein the length of the filter vector of L The relationship between the digital signal x(n) and the output filtered signal y(n) is:

in, Represents a vector of time-domain filter coefficients The i-th element of . When each new sampling data arrives, we need to perform L times of multiplication and L-1 times of addition according to the expression (1) to obtain the convolution to obtain the output filter signal y(n). Specifically, the system maintains a buffer or shift register to store current and past sampling data. With reference to Fig. 3, whenever new sampling data arrives, old data obtains x(n-1), x(n-2) through a series of shift units 202, 204, 206 and other units, until x(n -L+1) has a total of L elements. Then these elements are multiplied by the multiplier unit and the filter weight, that is, the digital signal x(n) and the weight Multiplied by multiplier 208, digital signal x(n-1) and weight Multiplied by the multiplier 210, the digital signal x(n-2) and the weight Multiplied by multiplier 212, and so on, finally x(n-L+1) and weight are multiplied by the multiplier 214, and then all the multiplication results are added by the adders 216, 218 and 220 to obtain the output filtered signal y(n). The signal is output to the digital-to-analog converter DA110 when the next sampling moment arrives, that is, the delay of the filtering module is one sampling period T _s . In practice, we can make the sampling period T _s sufficiently small by increasing the sampling frequency f _s of the digital system, so that the causal impact of the delay on the entire system can be ignored. Of course, the sampling frequency f _s cannot be increased indefinitely, because the larger f _s is, the larger the order L of the FIR filter required by the modeling controller is, so the amount of calculation required by the expression (1) also increases sharply. A balanced choice needs to be made in practical applications. The operation speed of the processor must be fast enough to ensure that the time required to execute expression (1) is less than one sampling period, otherwise this is not a real-time system, resulting in performance degradation or even failure of the active control system.

Next we discuss another key issue, the update of the processor filter coefficient vector, the most commonly used adaptive algorithm is the FxLMS algorithm. The processor is represented by a transfer function W(z), the transfer function from the secondary speaker to the second audio collector is written as S(z), and the transfer function from the first audio collector to the second audio collector is written as P( z). Then when the adaptive algorithm converges to a steady state, we get W(z)=-P(z)/S(z), then the sound pressure at the second audio collector is zero, thus achieving the goal of perfect noise control .

In practical applications, the primary channel transfer function P(z) and the secondary channel transfer function S(z) may be time-varying. Specifically in the application of earphones, each person's head size is individualized and the degree of tightness that individuals prefer to wear earphones is also different. During the wearing process, the wearer will also adjust the position and tightness of the earphones from time to time, which leads to The secondary channel transfer function S(z) varies, and thus the processor transfer function W(z) is also time-varying. This requires the adaptive filter used to have good tracking performance. In the patent of this invention, we use the frequency-domain Kalman filter algorithm to obtain the filter coefficients of the 116 modules in Figure 2 That is, the function to be realized by the module 118 in FIG. 2 . In order to make technical personnel better understand the main idea of the present invention, we specifically show the implementation steps of module 118 in FIG. 2 in FIG. 4 .

However, the variation of our controller transfer function is difficult to describe with an accurate mathematical model. In order to facilitate the description of the problem and meet the actual needs, we use a simplified first-order Markov model to describe the controller, that is,

W(k+1)=W(k)+Δ(k) (2)

In the above formula, W(k) represents the filter coefficient vector in the frequency domain, and Δ(k) represents the change of the filter coefficient in the frequency domain from the kth frame to the k+1 frame, which is called process noise. When Δ(k) is close to the zero vector, it means that there is almost no change in the system, and when Δ(k) is large, it means that the system has a large change. In the language of Kalman filtering, expression (2) is called the equation of state. Now let's give a frequency-domain adaptive filtering algorithm based on Kalman filtering.

The input signal v(n) of the Kalman filter is the estimated secondary channel of the input digital signal x(n) Obtained by filtering, that is, the expression realized by module 114 in Fig. 2 is:

This operation can be done in time-domain convolution or frequency-domain Fourier transform. It should also be noted that in order to successfully implement the method of the invention, it is necessary to estimate the transfer function of the secondary channel The estimation of the transfer function can be obtained in advance by using the classical system identification method. For example, Simon Hekin’s classic treatise "S. Haykin, Adaptive Filter Theory, 5th Edition, Prentice Hall, 2013" describes that using various adaptive algorithms can be used to estimate the transfer function.

Referring to FIG. 4 , the specific implementation of the Kalman filter is described. First, the Fourier transform unit 302 performs Fourier transform on the sequence composed of the nearest 2L points v(n) to obtain the frequency domain vector v(k)=F[v(kL-2L+1),...,v(kL )] ^T , where F represents the Fourier transform matrix, and the elements of v(k) are placed on the diagonal in turn to form a diagonal matrix V(k).

The Fourier transform unit 306 performs Fourier transform on the sequence composed of the nearest L points e(n) to obtain the frequency domain vector

Then according to expression (2), the update equation of frequency domain Kalman filter is:

in, is a vector of frequency-domain filter coefficients, in fact it is a vector of time-domain filter coefficients The 2L point Fourier transform of , K(k) is the Kalman gain matrix 304, is a constraint matrix, I _L is an identity matrix with a dimension of L×L, and 0 _L is a zero matrix with a dimension of L×L.

Now, we illustrate the specific implementation of expression (4) according to FIG. 4 . Unit 304 calculates the Kalman gain matrix K(k) according to the input V(k), and the multiplier 308 completes the multiplication operation of the input matrix V(k) and the gain matrix K(k) to obtain a diagonal matrix C(k). Take the diagonal elements of the diagonal matrix C(k) to obtain the vector c(k), and then the inverse Fourier transform operation unit 310 completes the inverse Fourier transform of the vector c(k) to obtain a real number whose dimension is 2L×1 vector a(k). Next, the zero-setting module 312 in the second half of the sequence implements the constraint function, that is, the L elements behind the sequence a(k) are set to zero while the previous L elements remain unchanged to obtain a new sequence b(k). Next, the Fourier transform module 314 performs Fourier transform on the sequence b(k), and at the adder 316 and the output of the delay unit 318 Add up to get At the same time will Perform inverse Fourier transform in unit 320 to obtain Used to assign values to the module 116 of FIG. 2 in the next frame, the expression of the Kalman gain matrix K (k) is:

K(k)＝P(k)V ^H (k)[V(k)P(k)V ^H (k)+2ψ _ee (k)] ^-1 (5)

Among them, P(k) is the state error matrix, ψ _ee (k)=diag{[ψ _ee,0 (k),ψ _ee,1 (k),…,ψ _ee,2L-1 (k)] ^T } is the system noise error diagonal matrix, and the superscript H stands for the conjugate transpose operation. The calculation method of P(k) is:

Among them, ψ _ee (k)=diag{[ψ _ee,0 (k),ψ _ee,1 (k),…,ψ _ee,2L-1 (k)] ^T } is the process noise diagonal matrix, I is An identity matrix with a dimension of 2L×2L, that is, the implementation of module 304 in FIG. 4 is described by Expression (5) and Expression (6).

Estimation of the power spectrum ψ _ee,i (k) of the system observation noise signal is an important issue. In practice, we do not have a means to measure the system noise signal superimposed on the error microphone. However, when the filter converges to a certain degree, the digital signal e(n) can better approximate the observed noise signal of the system. Based on this fact, in the patent of the present invention, we use the power spectrum of the error signal to replace the power spectrum of the observation noise. That is, we use the instantaneous power of the error signal to smooth through a low-pass filter, as follows:

ψ _ee,i (k)=αψ _ee,i (k-1)+(1-α)|E _i (k)| ² (7)

Among them, α is a smoothing factor, it is recommended to take α=0.8. According to this formula, Fig. 5 shows the specific block diagram for calculating the power spectrum ψ _ee,i (k) of the system observation noise signal. First, the conjugate operation is performed on the input E _i (k) in module 402, and then E _i (k ) and its conjugate are multiplied at unit 404, and then the output of multiplication unit 404 is multiplied at 406 by a factor of 1-α. Delay unit 412 carries out delay operation to ψ _ee,i (k) and obtains ψ _ee,i (k-1), then multiplies with factor α in unit 410, and the output of multiplication unit 406 and the output of multiplication unit 410 are in The addition unit 408 sums to obtain ψ _ee,i (k).

In order to make the above algorithm run smoothly, we also need to estimate the covariance matrix ψ _ΔΔ (k) of the process noise Δ(k). We know that the second term on the right in the expression (4) formula reflects the fluctuation situation of the filter coefficient vector to some extent, so the present invention utilizes this term to estimate the actual process noise matrix:

Then in the initial convergence stage of the system, ψ _ΔΔ,i (k) takes a relatively large value, which can speed up the convergence of the filter; when the algorithm reaches a steady state and the actual system fluctuation is small, ψ _ΔΔ,i (k) takes A small value is beneficial for the algorithm to achieve a smaller misalignment in the steady state; and once the system changes and needs to be tracked quickly, ψ _ΔΔ,i (k) can take a relatively large value to speed up the tracking performance of the algorithm. This explains in principle that the method proposed by the present invention has better performance than the traditional frequency domain algorithm.

Finally, it is noted that the above embodiments are only used to illustrate the technical solutions of the present invention without limitation. Although the present invention has been described in detail with reference to the embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or Equivalent replacements without departing from the spirit and scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims (8)

1. An earphone adaptive active noise control system, characterized in that the system at least includes: A first audio collector and a second audio collector, the first audio collector is installed outside the earphone for collecting reference noise signals; the second audio collector is installed inside the earphone for collecting error noise signals after noise reduction; The first analog-to-digital converter AD and the second analog-to-digital converter AD, the first analog-to-digital converter AD is electrically connected to the first audio collector, and is used to convert the reference noise signal collected by the first audio collector into a corresponding digital signal, the second analog-to-digital converter AD is electrically connected to the second audio collector, and is used to convert the error noise signal collected by the second audio collector into a corresponding digital signal; The processor unit, the first analog-to-digital converter AD is connected to the first information input end of the processor unit, and the second analog-to-digital converter AD is connected to the second information input end of the processor unit; The digital-to-analog converter DA and the secondary speaker, the information output end of the processor unit is connected to the secondary speaker after the digital-to-analog converter DA, the secondary speaker is installed in the earphone, and the digital-to-analog converter DA is used to convert the output of the processor unit The filtered signal is converted to a corresponding analog signal and fed to a secondary speaker, which is used to play the secondary sound signal. 2. The earphone adaptive active noise control system according to claim 1, characterized in that the control system further comprises a memory unit, and the memory unit and the processor unit are connected in bidirectional communication. 3. The earphone adaptive active noise control system according to claim 1, characterized in that the control system further comprises a power supply module connected to a power supply terminal of the processor unit. 4. The headphone adaptive active noise control system according to claim 1, characterized in that the control system further comprises a control logic unit for controlling the switch of the active noise reduction function. 5. an earphone self-adaptive active noise control method, is characterized in that, comprises the following steps: The reference noise signal is collected in real time by the first audio collector, and the reference noise signal is converted into a corresponding digital signal x(n) by the first analog-to-digital converter and then transmitted to the processor unit; the error is collected in real time by the second audio collector noise signal, and the error noise signal is converted into a corresponding digital signal e(n) by the second analog-to-digital converter and then transmitted to the processor unit; Calculation of the filter coefficient vector by the processor unit and calculate the output filtered signal y(n), and use the Kalman filter to calculate the filter coefficient vector in the frequency domain In the time domain for the digital signal x(n) and the time domain filter coefficients Perform convolution to obtain the output filtered signal y(n); where, the time domain filter coefficient is the filter coefficient vector The inverse Fourier transform of ; The output filtered signal y(n) is converted into a corresponding analog signal by a digital-to-analog converter and fed to the secondary speaker, which is used to play the secondary sound signal. 6. earphone adaptive active noise control method according to claim 5, is characterized in that, adopts Kalman filtering to calculate filter coefficient vector in frequency domain The steps are as follows: In the frequency domain, the digital signal x(n) corresponding to the reference noise signal is estimated by the secondary channel Filter to obtain the Kalman filter input signal v(n); Fourier transform the sequence composed of the nearest 2L-point Kalman filter input signal v(n) to obtain the frequency domain vector v(k), and place the elements of the frequency domain vector v(k) on the diagonal to form the diagonal matrix V(k); Perform Fourier transform on the sequence composed of the nearest L point error noise signal e(n) to obtain the frequency domain vector E(k); Update the Kalman gain matrix K(k)=P(k)V ^H (k)[V(k)P(k)V ^H (k)+2ψ _ee (k)] ^-1 , where P(k) is The state error matrix, ψ _ee (k) is the diagonal covariance matrix of the observation noise error, and the superscript H represents the conjugate transpose operation; Update frequency-domain filter coefficient vector in is a constraint matrix, I _L is an identity matrix whose dimension is L×L, 0 _L is a zero matrix whose dimension is L×L, F is a Fourier transform matrix, and K(k) is a Kalman gain matrix; Update state error matrix P(k): Among them, ψ _ΔΔ (k) is the process noise diagonal matrix, and I _2L is the identity matrix whose dimension is 2L×2L. 7. The earphone adaptive active noise control method according to claim 6, wherein the observed noise error matrix ψ _ee (k) is calculated by using the smooth power spectrum of the error noise signal e(n). 8. The earphone adaptive active noise control method according to claim 6, characterized in that, the diagonal element ψ _ΔΔ of the process noise diagonal matrix ψ _ΔΔ (k), the calculation method of i(k) is in, is the filter coefficient vector The i-th element of .