Frequency domain adaptive sound field control method based on sound contrast control

Abstract

The invention belongs to the technical field of sound field control, and relates to a frequency domain adaptive sound field control method based on sound contrast control, which comprises the following steps of: constructing an update model MDH of a sound transfer function by adopting a frequency domain adaptive algorithm based on a data frame; constructing an update model MDW for controlling a filter coefficient by using a sound contrast control algorithm; acquisition of an acoustic transfer function of a current frame

Description

Technical Field

[0001] The present invention belongs to the technical field of sound field control and relates to a frequency domain adaptive sound field control method based on sound contrast control. Background Art

[0002] Sound field control technology mainly uses speakers or specific sound sources to generate different sound fields in the target area. In the same space, different listening experiences can be met without physical isolation. On smartphones, this method can prevent sound leakage and improve the privacy and security of calls; in smart cockpits, this method can create different listening experiences for drivers and passengers without affecting each other; in smart home systems, this method can allow people in the same living room to entertain and work without disturbing each other. In short, with the development of the economy and society, sound field control technology has been applied in many fields.

[0003] Mainstream sound field control algorithms include acoustic contrast control, pressure matching, and the fusion of the two algorithms. Most studies are based on a stable acoustic environment. Once the acoustic environment changes, such as a change in the control area, the performance of the algorithm will drop sharply, or even fail to achieve sound field control. For this application scenario, it is imperative to study adaptive sound field control algorithms.

[0004] At present, some adaptive sound field control algorithms have been proposed, but the time domain adaptive sound field control algorithm still faces the problem of high computational complexity, and does not realize the real-time estimation of the transfer function, and still cannot solve the sound field control problem when the control area changes. Summary of the invention

[0005] In order to solve the above-mentioned prior art problems, the present invention adopts a frequency domain adaptive sound field control method based on sound contrast control, comprising:

[0006] S1. Determine a sound field control area, where the sound field control area includes: a bright area and a dark area; set a speaker array including a plurality of speakers; each speaker corresponds to a control filter;

[0007] S2, using the frequency domain adaptive algorithm based on the data frame to construct the updated model MD of the sound transfer function from the speaker array to the sound field control area H ; Using the acoustic contrast control algorithm to build an update model MD for controlling filter coefficients W ;

[0008] S3. Get the acoustic transfer function in the current frame H i , according to the acoustic transfer function H i Using the updated model MD WUpdate the control filter coefficients to obtain the control filter coefficients in the current frame W i ;

[0009] S4, obtain the current frame input signal, load the current frame input signal to the speaker, and the speaker W i Output signal;

[0010] S5, obtain the input signal of the previous frame, according to the input signal of the previous frame and the current frame and the coefficient W i Adopting the updated model MD H Acoustic Transfer Function H i Update and get the acoustic transfer function in the next frame H i+1 .

[0011] Update Model MD H ={MD H,B ,MD H,D}; Among them, MD H,B ,MD H,D The updated models of the sound transfer function from the loudspeaker array to the bright area and the dark area are constructed respectively; the updated model MD of the sound transfer function from the loudspeaker array to the bright area is constructed H,B include:

[0012] S11, construct the input signal of the i-1th frame and the ith frame and control the filter coefficient W i , according to the input signal of the i-1th frame and the i-th frame and the control filter coefficient W i Construct a diagonal matrix of the loudspeaker output signal in the frequency domain U i ;

[0013] S12. Constructing the acoustic transfer function H i,B , according to the diagonal matrix U i Harmonic Transfer Function H i,B Construct the total sound pressure generated by the bright area in the frequency domain P i,B ;

[0014] S13, total sound pressure P i,B Perform inverse discrete Fourier transform to obtain the total sound pressure P in the bright area in the time domain i,B ;

[0015] S14, constructing the expected signal, according to the expected signal and the total sound pressure P in the time domaini,B Construct the signal error, convert the signal error to the frequency domain, and obtain the frequency domain error e i,B ;

[0016] S15, according to the frequency domain error e i,B and the diagonal matrix U i Building Block Gradient Estimation According to the block gradient estimate Update the acoustic transfer function H i,B , get the updated model MD H,B .

[0017] Construct a diagonal matrix of the loudspeaker output signal in the frequency domain U i Including: performing discrete Fourier transform on the input signal of the i-1th frame and the ith frame, converting the signal after discrete Fourier transform into a diagonal matrix X i ; According to the coefficient of the control filter W i and the diagonal matrix X i Construct the output signal y for each speaker l i,l ; For the output signal y i,l Perform inverse discrete Fourier transform, take the last N data of the signal after inverse discrete Fourier transform, and get signal y′ i，l ; for signal y′ i，l Perform discrete Fourier transform and convert the discrete Fourier transformed signal into a diagonal matrix U i，l , the diagonal matrix of each speaker l is U i，l Combined, we get the diagonal matrix U i ; Where N is the length of the input signal.

[0018] Block Gradient Estimation It is expressed as:

[0019]

[0020] Among them, E i，B =diag{ e i，B , ..., e i，B} is a diagonal matrix, G=[I N×N 0 N×N ] is a matrix, F -1 represents the inverse discrete Fourier transform, I N×Nis the identity matrix, 0 N×N is a zero matrix and N is the length of the input signal.

[0021] Update the acoustic transfer function based on the block gradient estimate H i，B include: Among them, μ i =diag{μ, ..., μ} is a 2N×2N diagonal matrix, μ represents the step size, E i，B =diag{ e i，B , ..., e i，B} is a diagonal matrix, G=[I N×N 0 N×N ] is a matrix, I N×N is the identity matrix, 0 N×N is a zero matrix, N is the length of the input signal, F -1 represents inverse discrete Fourier transform, and F represents discrete Fourier transform.

[0022] Adopting the updated model MD H Acoustic Transfer Function H i The update includes: measuring the current real sound transfer function, obtaining the signal currently output by the speaker, convolving the signal currently output by the speaker with the real sound transfer function to obtain the expected signal, and convolving the expected signal, the input signal of the previous frame and the current frame, and the coefficient of the control filter in the current frame. W i Input update model MD H , and obtain the acoustic transfer function H in the next frame i+1 .

[0023] Construct an update model MD that controls the coefficients of the filter W Including: constructing the sound transfer function H from the loudspeaker to the bright area and the dark area at different frequencies i，B [ω] and H i，D [ω]; construct the coefficients of the control filter W i , according to the acoustic transfer function H i，B [ω] H i，D [ω] and the coefficients of the control filter W i Construct the total sound pressure of the bright and dark areas p i，B [ω] and p i，D [ω]; based on the total sound pressure in the bright area and the dark area p i，B[ω] and p i，D [ω] Construct the ratio of the acoustic potential energy of the bright and dark areas, construct the objective function based on the ratio of the acoustic potential energy of the bright and dark areas, and construct the update model MD of the control filter coefficient based on the objective function W .

[0024] The objective function is:

[0025]

[0026] Among them, M D is the number of microphones in the dark area, M B is the number of microphones in the bright area.

[0027] Construct an update model MD of the coefficients of the control filter according to the objective function W include:

[0028] Constructing the Matrix Among them, the eigenvector corresponding to the maximum eigenvalue of the matrix V is proportional to the optimal solution vector of the objective function;

[0029] Construct an update model of the characteristic matrix of matrix V; the update model of the characteristic matrix includes: bright area update model MD V,B Update model MD in dark zone V,D :

[0030]

[0031]

[0032] Among them, α and β are attenuation factors, A i [ω] B i [ω] is the feature matrix of the i-th frame;

[0033] An improved eigenvector tracking method based on power iteration is used to solve the bright area update model MD V,B Update model MD with dark zone V,D , get the updated model MD of the control filter coefficients W .

[0034] Solving the bright area update model MD V,B Update model MD with dark zone V,D include:

[0035] Update model MD according to dark zone V,D Constructing the feature matrix First update the model to obtain the intermediate variables of the transfer function

[0036] The first update model is transformed according to the relationship between the inverse and diagonalization of the Hermitian matrix to obtain the characteristic matrix The second updated model of;

[0037] For the feature matrix The second update model is approximated to obtain the feature matrix The approximate matrix K i Update model of

[0038] Update model MD according to the bright area V,B , Dark Zone Update Model MD V,D , intermediate variables of the transfer function And the approximate matrix K i The updated model builds the feature matrix A i [ω] Normalized eigenvector matrix C i Update model of

[0039] According to the eigenvector matrix C i The updated model building matrix V is the eigenvector matrix Update model of

[0040] According to the approximate matrix K i The eigenvector matrix of the matrix V Convert to the solution vector space of the objective function and obtain the control filter coefficients W i Update model

[0041] Beneficial effects:

[0042] 1. The present invention adopts a frequency domain adaptive algorithm based on data frames to dynamically track the changes of the sound transfer function and realize accurate online modeling of the sound transfer function. Compared with the time domain adaptive algorithm, the frequency domain adaptive algorithm can greatly improve the calculation efficiency and better meet the real-time requirements; 2. The traditional sound contrast control algorithm involves matrix square root and inverse operations when solving the optimal filter coefficient, which requires that the condition number of the autocorrelation matrix of the transfer function cannot be too large, otherwise it will cause robustness problems, and the inverse operation of the matrix will increase the calculation overhead. The present invention proposes an improved eigenvector tracking method based on power iteration to calculate the control filter coefficient, which can avoid the matrix square root and inverse operations, and improve the robustness and efficiency of the control filter coefficient calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 A schematic diagram of a sound range provided by an embodiment of the present invention;

[0044] Figure 2 A frequency domain adaptive sound field control flow chart based on sound contrast control provided by an embodiment of the present invention;

[0045] Figure 3 A diagram of a sound field simulation setting provided by an embodiment of the present invention;

[0046] Figure 4 A schematic diagram of the acoustic contrast between light and dark areas provided by an embodiment of the present invention;

[0047] Figure 5 A schematic diagram of the online modeling result of the acoustic transfer function provided by an embodiment of the present invention;

[0048] Figure 6 A sound pressure spectrum at a microphone provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] like Figure 2 As shown, the present invention adopts a frequency domain adaptive sound field control method based on sound contrast control, which specifically includes:

[0051] S1. Determine a sound field control area, where the sound field control area includes: a bright area and a dark area; set a speaker array including a plurality of speakers; each speaker corresponds to a control filter;

[0052] like Figure 1 As shown, in the time domain, the input signal x(n) is filtered and radiated by the speaker to the control area, and the output of the lth speaker is:

[0053] y l (n) = x T (n)w1 (1)

[0054] Where x(n) = [x(n), x(n-1), ..., x(n-J+1)] T represents the input signal, J represents the length of the input signal, w l (n) = [w(1), w(2), ..., w(J)] represents the coefficient of the control filter of the lth speaker, l = 1, 2, ..., L, L represents the number of speakers in the system.

[0055] There are M microphones evenly distributed in the light and dark control areas. The transfer function between the lth loudspeaker and the mth microphone in the bright area can be expressed as h B，ml , then the sound pressure p at the mth microphone in the bright area B，m (n) can be expressed as:

[0056]

[0057] Among them, y l (n)=[y1(n),y2(n-1),...,y l (n-I+1)] T represents the output signal of the lth speaker, h B，ml =[h B，ml (1), h B，ml (2), ..., h B，ml (I)] represents a transfer function with a length of I. It is not difficult to see that the above calculation process includes multiple convolutions, the processing speed of a single sample is slow, the calculation complexity is high, and it is difficult to meet the real-time requirements.

[0058] S2, using the frequency domain adaptive algorithm based on the data frame to construct the updated model MD of the sound transfer function from the speaker array to the sound field control area H ; Using the acoustic contrast control algorithm to construct the update model MD of the control filter coefficients W ;

[0059] The present invention considers adopting frequency domain adaptive filter to model the transfer function, and the processing method based on data frame can significantly improve the efficiency of the algorithm.

[0060] Constructing the updated model MD of the sound transfer function from the loudspeaker array to the bright area H，B include:

[0061] Specifically, assuming that a frame of data contains N sample points and a signal overlap (overlap-save) scheme is used between frames, the output y of the th speaker is i，l It can be expressed as:

[0062] y i，l =X i w i，l (3)

[0063] Where i represents the frame index value, X i =diag{F[x(iN-N),...,x(iN-1),x(iN),...,x(iN+N-1)] T} is a diagonal matrix consisting of the i-1th frame and the i-th frame, representing the input signal in the frequency domain; represents the filter coefficient of the control filter in the frequency domain, F represents the discrete Fourier transform matrix, whose dimension is 2N×2N, and the output of the lth speaker in the time domain only takes the last N samples to obtain the signal y′ i，l , expressed as:

[0064] y′ i，l =ZF -1 y i，l (4)

[0065] Among them, y′ i，l =[y l (iN), ..., y l (iN+N-1)] T , F -1 represents the inverse discrete Fourier transform, z = [0 N I N ] is a matrix, 0 N is a zero matrix, I N is the identity matrix.

[0066] The sound pressure generated by the lth loudspeaker at the mth microphone in the bright area p i，ml can be re-expressed as:

[0067] p i，ml = U i，l h B，ml (5)

[0068] Among them, U i，l =diag{F[y l (iN), ..., y l (iN+N-1)] T} is a diagonal matrix, is the transfer function.

[0069] In one embodiment, the transfer function is modeled as a finite impulse response FIR filter, and the input data of the FIR filter is composed of the output of the speaker, including y i-1，l and i，l , therefore, U i，l =diag{F[y l (iN-N), ..., y l (iN-1), y l (iN), ..., y l (iN+N-1)] T}.

[0070] Total sound pressure at the mth microphone in the bright areaP i，B，m It can be expressed as:

[0071]

[0072] in, U i =[ U i，l , ..., U i，L ] T is the diagonal matrix of the loudspeaker output signal in the frequency domain, H i，B，m =[ h i，B，m1 ..., h i，B，mL ] T is the transfer function; similar to equation (4), the sound pressure at the mth microphone in the bright area can be expressed in the time domain as P i，B，m :

[0073] P i，B，m =ZF -1 P i，B，m (7)

[0074] Construct the expected signal d i，B，m , we can get the signal error e i，B，m for:

[0075] e i，B，m =d i，B，m -P i，B，m (8)

[0076] To convert the error vector to the frequency domain, it is necessary to perform zero padding after the vector to obtain the frequency domain error. e i，B，m :

[0077] e i，B，m =FZ T e i,B,m (9)

[0078] By similar derivation, the block gradient estimate It can be expressed as:

[0079]

[0080] in, E i，B，m =diag{ e i，B，m , ..., e i，B，m} is a diagonal matrix, G = [I N×N 0N×N ] is a matrix used to extract the first N data of each column, I N×N is the identity matrix, 0 N×N is a zero matrix.

[0081] After converting the gradient estimate to the frequency domain, it is added to the current transfer function to obtain the updated model of the acoustic transfer function in the bright area:

[0082]

[0083] Among them, μ i =diag{μ, ..., μ} is a 2N×2N diagonal matrix, μ represents the step size. Generally speaking, each step size varies according to the signal power of the frequency point. If the input signal is statistically stationary, the step size matrix can be fixed.

[0084] Similarly, the updated model of the acoustic transfer function in the dark zone is

[0085] To avoid the first iteration update W i When the value is stable, all elements of the initial value of the sound transfer function are set to a minimum value, such as e -6 .

[0086] Using the acoustic contrast control algorithm to construct an update model MD of the coefficients of the control filter W include:

[0087] In this embodiment, the sound field control method adopts acoustic contrast control (ACC), which aims to maximize the ratio of the sound potential energy in the bright area to the sound potential energy in the dark area.

[0088] Specifically, construct the transfer function from the loudspeaker to the bright area and the dark area at different frequencies. H i，B [ω] and H i，D [ω]; According to the transfer function H i，B [ω] H i，D [ω] and the coefficients of the control filter W i Construct the total sound pressure of the bright and dark areas p i，B [ω] and p i，D [ω]; At a given frequency ω, the sound pressure at the mth microphone in the bright region p i，B，m [ω] can be expressed as:

[0089]

[0090] in, h i，B，m [ω]=[ h i，B，m1 [ω], ..., h i，B，mL [ω]] T Contains the transfer function from all speakers to the microphone at the current frequency point, W ii [ω]=[ W i，1 [ω], ..., W i，L [ω]] T Indicates the filter coefficients of different speakers at the current frequency point.

[0091] Bright area sound pressure p i，B [ω] can be expressed as:

[0092] p i，B [ω]= H i，B [ω] W i [ω] (13)

[0093] in, is a matrix containing all transfer functions in the bright region.

[0094] Similarly, the dark zone sound pressure p i，D [ω] can be expressed as:

[0095] p i，D [ω]=H i，D [ω] W i [ω] (14)

[0096] in, is a matrix containing all transfer functions of the dark region.

[0097] According to the total sound pressure of bright area and dark area p i，B [ω] and p i，D [ω] calculates and maximizes the ratio of the acoustic potential energy of the light and dark areas. The maximum ratio of the acoustic potential energy of the light and dark areas can be expressed as:

[0098]

[0099] Among them, M D is the number of microphones in the dark area, M B is the number of microphones in the bright area.

[0100] The optimal solution vector and matrix of equation (15) is proportional to the eigenvector corresponding to the maximum eigenvalue, so the problem can be converted into a problem of solving the eigenvector.

[0101] The update model of the feature matrix of the update matrix V is modeled as:

[0102]

[0103]

[0104] Among them, α and β are attenuation factors. A i [ω] B i [ω] is the feature matrix of the i-th frame, which is used to track the main eigenvectors of the signal.

[0105] Directly adopting the solution method based on power iteration requires matrix square root operation and solving the matrix inverse, both of which require very high computational complexity. The present invention will improve the algorithm and enhance the computational efficiency.

[0106] Specifically, the steps include:

[0107] Constructing the feature matrix The first updated model:

[0108]

[0109] in,

[0110]

[0111] The first update model is transformed according to the relationship between the inverse and diagonalization of the Hermitian matrix to obtain the characteristic matrix The second updated model of;

[0112] Specifically, the Hermitian matrix C satisfies the equation (DcDH) -1 / 2 =C -1 / 2 D -1 , then the second update model can be obtained from equation (18):

[0113]

[0114] For the feature matrix The second update model is approximated to obtain the feature matrix The approximate matrix K i Update model of

[0115] Specifically, for any vector c, it satisfies the following equation:

[0116]

[0117] Then, through equation (21), we can get the recursive equation of equation (20), that is, the approximate matrix K i The updated model:

[0118]

[0119] Among them, the approximate matrix And the intermediate variable γ i for:

[0120]

[0121] in, The matrix is ​​calculated The square of the 2-norm of the elements in each column, where 1 is a vector whose elements are all 1.

[0122] Update the model based on the feature matrix and transfer the intermediate variables of the model and approximate values K i The update model transforms the problem from a generalized eigenvector problem to a standard eigenvector problem, and obtains the standard eigenmatrix C i :

[0123]

[0124] in, C i is a matrix A i-1 The standardized feature matrix;

[0125] Constructing intermediate variables According to the above intermediate variables, the standard feature matrix is ​​obtained C i The updated model:

[0126]

[0127] According to the eigenvector matrix C i The updated model building matrix V is the eigenvector matrix The updated model:

[0128]

[0129]

[0130]

[0131] in, is the intermediate variable that controls the filter coefficients, Represents the eigenvector matrix Perform orthogonal decomposition, Q i , R i For the matrix Decomposed matrix, First r columns of Q i Indicates Q i The first r columns, r usually refers to The number of columns.

[0132] By K i The eigenvector matrix Convert to the solution vector space of the objective function and obtain the required filter coefficient matrix W i The updated model:

[0133]

[0134] Controls the initial coefficients of the filter W 0 is an all-zero matrix;

[0135] S3, each time the input signal of the current frame is input for adaptive sound field control, the acoustic transfer function of the current frame obtained in the previous iteration is obtained H i , according to the acoustic transfer function H i Using the updated model MD W Update the control filter coefficients to obtain the control filter coefficients in the current frame W i ;

[0136] S4, obtain the current frame input signal, load the current frame input signal to the speaker, and the speaker is based on the coefficient of the current frame W i Output signal and generate the desired sound in the bright area and zero signal in the dark area;

[0137] S5, obtain the input signal of the previous frame, according to the input signal of the previous frame and the current frame and the coefficient W i Adopting the updated model MD H Acoustic Transfer Function H i Update and get the acoustic transfer function in the next frame H i+1; When the input signal of the next frame is input for adaptive sound field control, the acoustic transfer function of the next frame H i+1 Used to update the control filter coefficients to obtain the control filter coefficients in the next frame.

[0138] Adopting the updated model MD H Acoustic Transfer Function H i The update includes: measuring the current real sound transfer function, obtaining the signal currently output by the speaker, convolving the signal currently output by the speaker with the real sound transfer function to obtain the expected signal, and convolving the expected signal, the input signal of the previous frame and the current frame, and the coefficient of the control filter in the current frame W i Input update model MD H , get the acoustic transfer function in the next frame H i+1 .

[0139] like Figure 3 As shown, the loudspeaker and the microphone are placed on a plane with a height of 1.5m in a room with a size of 7.22m (length) × 4.90m (width) × 5.29m (height); among them, six loudspeakers represented by black triangles are placed along an arc with a radius of 1.5m to form two listening areas 0.8m apart, and the interval between the loudspeakers is 0.15m; 4 microphones are used in both the bright area and the dark area, with an interval of 0.04m; the real sound transfer function is measured in the anechoic chamber of the UTS Technology Laboratory with a sampling rate of 48kHz, and the sampling rate is reduced to 4kHz in this simulation; the length of the sound transfer function is set to 3640, the length of the filter is set to 128, and the number of Fourier transform points is 256; the positions of the bright and dark areas are exchanged in the simulation to test the performance of the method of the present invention; for comparison, the simulation results when the adaptive sound field control algorithm is not used are given.

[0140] Take white noise as input, step size μ = 5 × 10 -7 , in order to smooth the forgetting factor λ=0.99, α and β are both 0.001. Initially, area A is used as the bright area and area B is used as the dark area. After 15 seconds, the two areas are swapped. Figure 4The acoustic contrast curves of the light and dark areas are shown. The blue curve represents the case where the adaptive control algorithm is not used. When the transfer function is given, the contrast can be maintained at 20dB in the first 15 seconds. However, when the positions of the light and dark areas suddenly change, the contrast between the light and dark areas is only 0dB when the transfer function cannot be correctly modeled in real time. The red curve represents the case where the method of the present invention is used. The acoustic transfer function is initialized to 0, which leads to a decrease in the convergence speed in the first 15 seconds, but it can still converge in a short time. When the positions of the light and dark areas are suddenly exchanged, the algorithm can also quickly model the transfer function, so that the acoustic contrast returns to a stable 20dB.

[0141] Figure 5 The online modeling results of the transfer function are shown. It is obvious that the online modeled transfer function and the real transfer function are basically consistent at the main peak.

[0142] Figure 6 The sound pressure spectra measured at the specified microphones (0.42m, 0.2m) are shown. From the upper sub-figure, it can be seen that when the positions of the light and dark areas change, not applying the adaptive sound field control method will result in the sound pressure at the microphone not being suppressed. The lower sub-figure shows that after using the method of the present invention, when the microphone moves from the light area to the dark area, the measured sound pressure is also suppressed accordingly.

[0143] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation modes of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A frequency domain adaptive sound field control method based on acoustic contrast control, characterized in that: include: S1. Determine a sound field control area, where the sound field control area includes: a bright area and a dark area; set a speaker array including a plurality of speakers; each speaker corresponds to a control filter; S2, using the frequency domain adaptive algorithm based on the data frame to construct the updated model MD of the sound transfer function from the speaker array to the sound field control area H ; Using the acoustic contrast control algorithm to build an update model MD for controlling filter coefficients W ; S3. Get the acoustic transfer function in the current frame H i , according to the acoustic transfer function H i Using the updated model MD W Update the control filter coefficients to obtain the control filter coefficients in the current frame W i ; S4, obtain the current frame input signal, load the current frame input signal to the speaker, and the speaker W i Output signal; S5, obtain the input signal of the previous frame, according to the input signal of the previous frame and the current frame and the coefficient W i Adopting the updated model MD H Acoustic Transfer Function H i Update and get the acoustic transfer function in the next frame H i+1 .

2. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 1, characterized in that: Update Model MD H ={MD H,B ,MD H,D }; Among them, MD H,B ,MD H,D The updated models of the sound transfer function from the loudspeaker array to the bright area and the dark area are constructed respectively; the updated model MD of the sound transfer function from the loudspeaker array to the bright area is constructed H,B include: S11, construct the input signal of the i-1th frame and the ith frame and control the filter coefficient W i , according to the input signal of the i-1th frame and the i-th frame and the control filter coefficient W i Construct the diagonal matrix U of the speaker's output signal in the frequency domain i ; S12. Constructing the acoustic transfer function H i,B , according to the diagonal matrix U i Harmonic Transfer Function H i,B Construct the total sound pressure generated by the bright area in the frequency domain P i,B ; S13, total sound pressure P i,B Perform inverse discrete Fourier transform to obtain the total sound pressure P in the bright area in the time domain i,B ; S14, constructing the expected signal, according to the expected signal and the total sound pressure P in the time domain i,B Construct the signal error, convert the signal error to the frequency domain, and obtain the frequency domain error e i,B ; S15, according to the frequency domain error e i,B and the diagonal matrix U i Building Block Gradient Estimation According to the block gradient estimate Update the acoustic transfer function H i,B , get the updated model MD H,B .

3. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 2, characterized in that: Construct a diagonal matrix of the loudspeaker output signal in the frequency domain U i Including: performing discrete Fourier transform on the input signal of the i-1th frame and the ith frame, converting the signal after discrete Fourier transform into a diagonal matrix X i ; According to the coefficient of the control filter W i and the diagonal matrix X i Construct the output signal y for each speaker l i,l ; For the output signal y i,l Perform inverse discrete Fourier transform, take the last N data of the signal after inverse discrete Fourier transform, and get signal y′ i,l ; for signal y′ i,l Perform discrete Fourier transform and convert the discrete Fourier transformed signal into a diagonal matrix U i,l , the diagonal matrix of each speaker l is U i,l Combined, we get the diagonal matrix U i ; Where N is the length of the input signal.

4. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 2, characterized in that: Block Gradient Estimation It is expressed as: in, E i,B =diag{ e i,B ,..., e i,B } is a diagonal matrix, G=[I N×N 0 N×N ] is a matrix, F -1 represents the inverse discrete Fourier transform, I N×N is the identity matrix, 0 N×N is a zero matrix and N is the length of the input signal.

5. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 2, characterized in that: Update the acoustic transfer function based on the block gradient estimate H i,B include: Among them, μ i =diag{μ,...,μ} is a 2N×2N diagonal matrix, μ represents the step size, E i,B =diag{ e i,B ,..., e i,B } is a diagonal matrix, G=[I N×N 0 N×N ] is a matrix, I N×N is the identity matrix, 0 N×N is a zero matrix, N is the length of the input signal, F -1 represents inverse discrete Fourier transform, and F represents discrete Fourier transform.

6. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 2, characterized in that: Adopting the updated model MD H Acoustic Transfer Function H i The update includes: measuring the current real sound transfer function, obtaining the signal currently output by the speaker, convolving the signal currently output by the speaker with the real sound transfer function to obtain the expected signal, and convolving the expected signal, the input signal of the previous frame and the current frame, and the coefficient of the control filter in the current frame. W i Input update model MD H , get the acoustic transfer function in the next frame H i+1 .

7. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 1, characterized in that: Construct an update model MD that controls the coefficients of the filter W Including: constructing the sound transfer function from the speaker to the bright area and the dark area at different frequency points ω H i,B [ω] and H i,D [ω]; construct the coefficients of the control filter W i , according to the acoustic transfer function H i,B [ω] H i,D [ω] and the coefficients of the control filter W i Construct the total sound pressure of the bright and dark areas p i,B [ω] and p i,D [ω]; based on the total sound pressure in the bright area and the dark area p i,B [ω] and p i,D [ω] Construct the ratio of the acoustic potential energy of the bright and dark areas, construct the objective function based on the ratio of the acoustic potential energy of the bright and dark areas, and construct the update model MD of the control filter coefficient based on the objective function W .

8. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 7, characterized in that: The objective function is: Among them, M D is the number of microphones in the dark area, M B is the number of microphones in the bright area.

9. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 8, characterized in that: Construct an update model MD of the coefficients of the control filter according to the objective function W include: Constructing the Matrix Among them, the eigenvector corresponding to the maximum eigenvalue of the matrix V is proportional to the optimal solution vector of the objective function; Construct an update model of the feature matrix of matrix V; the update model of the feature matrix includes: bright area update model MD V,B Update model MD with dark zone V,D : MD V,B : MD V,D : Among them, α and β are attenuation factors, A i [ω] B i [ω] is the feature matrix of the i-th frame; An improved eigenvector tracking method based on power iteration is used to solve the bright area update model MD V,B Update model MD with dark zone V,D , get the updated model MD of the control filter coefficients W .

10. The frequency domain adaptive sound field control method based on acoustic contrast control according to claim 9, characterized in that: An improved eigenvector tracking method based on power iteration is used to solve the bright area update model MD V,B Update model MD with dark zone V,D include: Construct feature matrix based on dark area update model First update the model to obtain the acoustic transfer function H i,D Intermediate variable of [ω] The first update model is transformed according to the relationship between the inverse and diagonalization of the Hermitian matrix to obtain the characteristic matrix The second updated model of; For the feature matrix The second update model is approximated to obtain the feature matrix The approximate matrix K i Update model of Update model MD according to the bright area V,B , Dark Zone Update Model MD V,D , intermediate variables of the transfer function And the approximate matrix K i The updated model builds the feature matrix A i [ω] Normalized eigenvector matrix C i Update model of According to the eigenvector matrix C i The updated model building matrix V is the eigenvector matrix Update model of According to the approximate matrix K i The eigenvector matrix of the matrix V Convert to the solution vector space of the objective function and obtain the control filter coefficients W i The updated model.

Images