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

By using a frequency domain block adaptive sound field partitioning method, and employing gradient descent and FFT fast cyclic convolution, the computational complexity of sound field partitioning in dynamic environments is solved, achieving stable sound contrast control and bright area signal distortion suppression, thus meeting real-time processing requirements.

CN121645076APending Publication Date: 2026-03-10CHONGQING UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511928190.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-19
Publication Date
2026-03-10

AI Technical Summary

Technical Problem

Existing sound field zoning techniques struggle to achieve real-time, low-computational-complexity stable sound contrast control and bright-area signal distortion suppression in dynamic environments.

Method used

A frequency-domain block adaptive sound field partitioning method based on acoustic contrast control is adopted. By constructing a time-domain block-by-block updated signal model, the gradient descent method is used to update the loudspeaker control filter coefficients, and the frequency-domain fast cyclic convolution of FFT is combined to reduce computational complexity.

Benefits of technology

Stable target acoustic contrast control and suppression of bright area signal distortion are achieved in dynamic acoustic environments, reducing computational complexity and meeting real-time processing requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121645076A_ABST
    Figure CN121645076A_ABST
Patent Text Reader

Abstract

A frequency domain block adaptive sound field partitioning method based on sound contrast control comprises the steps that 1) L loudspeakers are arranged in a region to be partitioned, control filter coefficients of the L loudspeakers at the initial moment are defined, and L is a positive integer; 2) on the basis of the control filter coefficient at the current moment, calculating to obtain playback sound fields at different control points; 3) constructing a time domain block-by-block updating signal model based on the replayed sound fields at different control points; 4) solving the time domain block-by-block updating signal model by using a gradient descent method based on frequency domain fast cyclic convolution of FFT (Fast Fourier Transform) to obtain an updated control filter coefficient; and 5) based on the updated control filter coefficient, calculating to obtain output signals of the L loudspeakers, and further realizing self-adaptive sound field partitioning in the to-be-partitioned region. The method can realize tracking of the target sound contrast and suppression of distortion of a replay sound field in a bright region under a dynamically changing acoustic environment and an input signal.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of sound field zoning technology, and more specifically to a frequency domain block adaptive sound field zoning method based on sound contrast control. Background Technology

[0002] Sound field zoning systems aim to provide independent sound zones for different listeners in a shared space using loudspeaker arrays. In recent years, their application in spaces such as car cabins, home theaters, and open offices has garnered widespread attention. This technology manages the amplitude and phase of the output signals of each array element by designing control filters within the loudspeaker array, creating a bright zone in the target sound field reproduction area and a dark zone in the target sound field suppression area. The ratio of acoustic energy between the bright and dark sound zones is called acoustic contrast, and the reproduction error in the bright zone is called signal distortion; both are closely related to the transfer function from the loudspeaker to the bright and dark sound zones.

[0003] For offline control using fixed filters, the control filter for the loudspeaker array is designed once using a pre-measured sound field transfer function. However, the sound field transfer function in real-world environments often changes over time, making it difficult for the fixed filter method to guarantee sound field zoning control performance in dynamic environments. Therefore, updating the control filter in real time based on changes in the sound field transfer function has become an important research direction in the field of sound field zoning. Currently, one of the main challenges facing adaptive sound field zoning control algorithms is their high computational complexity, making real-time deployment difficult. Summary of the Invention

[0004] The purpose of this invention is to provide a frequency domain block adaptive sound field partitioning method based on acoustic contrast control, comprising the following steps:

[0005] 1) Arrange L loudspeakers in the area to be partitioned, and define the control filter coefficients of the L loudspeakers at the initial time, where L is a positive integer.

[0006] 2) Based on the control filter coefficients at the current moment, the reproduced sound field at different control points is calculated.

[0007] 3) Based on the reproduced sound field at different control points, construct a time-domain block-by-block updated signal model.

[0008] 4) Based on FFT, the frequency domain fast cyclic convolution is used to solve the time domain block-by-block update signal model using the gradient descent method to obtain the updated control filter coefficients.

[0009] 5) Based on the updated control filter coefficients, the output signals of L loudspeakers are calculated, and then adaptive sound field zoning is achieved in the area to be zoned.

[0010] Furthermore, the control filter coefficients at the initial time are as follows:

[0011] (1)

[0012] In the formula, Indicates the speaker index. Represents the discrete time index. Indicates the first The speaker in the A vector of control filter coefficients at discrete time points. Indicates the filter length. These represent the control filter coefficients for different filter lengths. This indicates transpose.

[0013] Furthermore, the reproduced sound field at the different control points is shown below:

[0014] (2)

[0015] In the formula, Indicates the control point index. Indicates the first The reproduced sound field at each control point. Indicates the speaker index. Indicates the first At the discrete time... The speaker to the Transpose of the impulse response function estimate for each control point. Indicates the first The speaker in the A vector of control filter coefficients at discrete time points. This represents the control filter coefficient vector for all loudspeakers.

[0016] Wherein, the input signal matrix Input signal diagonally repeated stacked matrix , No. Transpose of the impulse response function estimated at each control point As shown below:

[0017] (3)

[0018] (4)

[0019] (5)

[0020] In the formula, Represents the identity matrix. This represents the Kronecker product. This represents the length of the estimated impulse response function. They represent the first From the discrete time point to the... The input signal at discrete time points. Each represents a different speaker up to the [number]. Transpose of the impulse response function estimate for each control point.

[0021] Furthermore, the optimization function for the time-domain block-by-block update signal model is as follows:

[0022] (6)

[0023] In the formula, This represents the optimization function for updating the signal model block by block in the time domain. This indicates the length of the time-domain block. k represents the time block index. Indicates the first The control filter coefficient vector of all loudspeakers at each discrete time. Indicates the control point index. , These represent the number of control points in the bright sound zone and the dark sound zone, respectively. Indicates the first The first discrete time of the bright sound region The replay sound field vector of each control point. Indicates the first The first discrete moment of the dark region The replay sound field vector of each control point. Indicates the first The first discrete time of the bright sound region The desired sound field at each control point. Represents the norm. This represents the cost function corresponding to sound contrast control. This indicates the distortion of the sound field reproduced in the bright area. A linear representation of the target acoustic contrast. The regularization parameter represents the distortion of the reproduced sound field.

[0024] Among them, in the first The first discrete moment in the bright sound region and the dark sound region The replay sound field vectors for each control point are shown below:

[0025] (7)

[0026] (8)

[0027] In the formula, Indicates the speaker index. , They represent the first time. At the nth discrete time, the th The first speaker in the bright sound zone and the dark sound zone Filtered input signals for each control point. , They represent the first time. The first discrete moment in the bright sound region and the dark sound region The Toeplitz matrix is ​​composed of the filtered input signals of each control point. Indicates the first The speaker in the The control filter coefficient vector at discrete time points. Indicates the first The control filter coefficient vector of all loudspeakers at each discrete time.

[0028] Furthermore, the aforementioned in the first At the nth discrete time, the th The first speaker in the bright sound zone Filtered input signal at each control point The first of the dark sound zone Filtered input signal at each control point Use parameters uniformly This is represented as follows:

[0029] (9)

[0030] (10)

[0031] (11)

[0032] In the formula, Indicates the filter length. They represent the first time step at different discrete times. The speaker in the The filtered input signal vector of each control point. They represent the first time step at different discrete times. The speaker in the Filtered input signals for each control point. This represents the vector used to calculate the filtered input signal. The Toeplitz matrix is ​​composed of the input signals. Indicates the first At the nth discrete time, the th The speaker in the Impulse response function estimation for each control point. This represents the length of the estimated impulse response function. These represent the input signals at different discrete times.

[0033] Furthermore, the steps for solving the time-domain block-by-block update signal model using the gradient descent method to obtain the updated control filter coefficients are as follows:

[0034] S1 uses the gradient descent method to minimize the optimization function of the time-domain block-by-block update signal model, obtaining the... The control filter coefficients at discrete time points are shown below:

[0035] (12)

[0036] In the formula, This indicates the length of the time-domain block. k represents the time block index. , They represent the first time. , The control filter coefficient vector of all loudspeakers at each discrete time. This represents the step size parameter. This represents the optimization function for updating the signal model block by block in the time domain.

[0037] Wherein, gradient vector As shown below:

[0038] (13)

[0039] In the formula, Indicates the control point index. , These represent the number of control points in the bright sound zone and the dark sound zone, respectively. , They represent the first time. The first discrete moment in the bright sound region and the dark sound region The transpose of the Toeplitz matrix, which consists of the filtered input signals of control points. , They represent the first time. The first discrete moment in the bright sound region and the dark sound region The replay sound field vector of each control point. Indicates the first The first discrete time of the bright sound region The desired sound field at each control point. A linear representation of the target acoustic contrast. The regularization parameter represents the distortion of the reproduced sound field.

[0040] Among them, the block average acoustic contrast As shown below:

[0041] (14)

[0042] In the formula, Represents the norm.

[0043] S2 uses a sound field modeling method to model the reproduced sound field.

[0044] S3 substitutes the replay sound field constructed in step S2 into the... The gradient vectors of the control filter coefficients at discrete time points are used to obtain the updated control filter coefficients.

[0045] Furthermore, the reproduced sound field constructed in step S2 is shown below:

[0046] (15)

[0047] (16)

[0048] (17)

[0049] In the formula, Indicates the speaker index. , They represent the first time. At the nth discrete time, the th The first speaker in the bright sound zone and the dark sound zone Filtered input signals for each control point. Indicates the first At the nth discrete time, the th The control filter coefficient vector of each speaker. Indicates the first The vector of speaker control filter coefficients corresponding to the desired sound field at each discrete time point.

[0050] Furthermore, the output signals of the L loudspeakers are as follows:

[0051] (18)

[0052] (19)

[0053] In the formula, This indicates the length of the time-domain block. k represents the time block index. Indicates the speaker index. Indicates the first At the nth discrete time, the th The output signal vector of each speaker. Indicates the first The Toeplitz matrix, composed of the input signals used to calculate the loudspeaker output signal at discrete time points. Indicates the first At the nth discrete time, the th The control filter coefficient vector of each speaker. Indicates the filter length. These represent the input signals at different discrete times.

[0054] Furthermore, the frequency-domain fast circular convolution of the FFT is used to filter the input signal vector. gradient vector Reproduced sound field vector Output signal vector Convert it to a circular convolution form to speed up the solution.

[0055] Furthermore, the filtered input signal vector gradient vector Reproduced sound field vector Output signal vector The steps to convert it to a circular convolution form are as follows:

[0056] Q1 introduces the Overlap-Save method to filter the input signal vector. gradient vector Reproduced sound field vector Output signal vector Equivalent to circular convolution, as shown below:

[0057] (20)

[0058] (twenty one)

[0059] (twenty two)

[0060] (twenty three)

[0061] In the formula, This indicates the length of the time-domain block. k represents the time block index. Indicates the speaker index. Indicates the control point index. This represents the vector used to calculate the filtered input signal. The Toeplitz matrix is ​​composed of the input signals. Indicates the first At the nth discrete time, the th The speaker in the Impulse response function estimation for each control point. This represents the length of the estimated impulse response function. Indicates the filter length. , , , All represent zero matrices. , , , Both represent identity matrices. , Indicates the first , At the nth discrete time, the th The control filter coefficient vector of each speaker. Indicates the first At the nth discrete time, the th The first speaker in the bright sound zone Filtered input signals for each control point. Indicates the first At the nth discrete time, the th The first speaker in the bright sound zone The replay sound field vector of each control point. Indicates the first The Toeplitz matrix, composed of the input signals used to calculate the loudspeaker output signal at discrete time points. Refers to the gradient vector Part of the gradient in. Indicates the first The first discrete time of the bright sound region The transpose of the Toeplitz matrix, which consists of the filtered input signals of control points. Indicates the first The first discrete time of the bright sound region The replay sound field vector of each control point. This represents the time inversion matrix. Represents the replay sound field vector The vectors arranged in reverse order. , They represent the first time. , Input signal at discrete time points. , They represent the first time. , At the nth discrete time, the th The first speaker in the bright sound zone Filtered input signals for each control point.

[0062] Q2 uses FFT to accelerate the recurrent convolution obtained in step Q1, as shown below:

[0063] (twenty four)

[0064] (25)

[0065] (26)

[0066] (27)

[0067] (28)

[0068] (29)

[0069] (30)

[0070] (31)

[0071] (32)

[0072] In the formula, This represents the discrete inverse Fourier transform matrix. This represents the discrete Fourier transform matrix. , These represent the frequency domain components of the input signal and the estimated impulse response function, respectively. This represents the Hadamard product. , These represent the frequency domain components of the filtered input signal and the control filter coefficients, respectively. Indicates the Ming Sheng District No. Frequency domain components of the reproduced sound field at each control point. Indicates conjugate.

[0073] The technical effectiveness of this invention is undeniable. This invention proposes a low-computational-complexity frequency-domain block adaptive sound field partitioning algorithm for stable control of acoustic contrast and suppression of bright-area signal distortion. First, a comprehensive optimization objective is constructed, using acoustic contrast control deviation as the loss function and bright-area signal distortion as the regularization term. Gradient descent is used to minimize the objective function to update the loudspeaker control filter, enabling stable target acoustic contrast control in dynamically changing acoustic environments while ensuring a certain level of sound field reproduction accuracy. Second, time-domain block processing is combined to reduce the filter update frequency, and Fast Fourier Transform (FFT) fast cyclic convolution is used instead of the original linear convolution, further reducing the algorithm's computational complexity.

[0074] The FBFxLAD algorithm proposed in this invention can track the target sound contrast and suppress sound field distortion in bright-area playback under dynamically changing acoustic environments and input signals. This algorithm combines temporal block processing and FFT fast circular convolution, resulting in low computational complexity and enabling efficient deployment on hardware systems to meet real-time processing requirements. Attached Figure Description

[0075] Figure 1 Here is a block diagram of the FBFxLAD algorithm;

[0076] Figure 2 A block diagram illustrating the principle of a personal vocal zone system;

[0077] Figure 3 This is a schematic diagram of the array layout and measured RIR. Figure 3 (a) is a schematic diagram of the layout of the arc-shaped loudspeaker array; Figure 3 (b) is a schematic diagram showing the RIR from the reference loudspeaker to a control point;

[0078] Figure 4 For different audio inputs and Adaptive result diagram; Figure 4 (a) White noise input Adaptive result diagram; Figure 4 (b) Music signal input Adaptive result diagram; Figure 4 (c) White noise input Adaptive result diagram; Figure 4 (d) Music signal input Adaptive result diagram;

[0079] Figure 5 Schematic diagram of the sound field imaging in areas A and B under different audio inputs; Figure 5 (a) is a schematic diagram of the reproduced sound field imaging in region A with white noise input; Figure 5 (b) is a schematic diagram of the sound field imaging of the music signal input in region A; Figure 5 (c) is a schematic diagram of the reproduced sound field imaging in region B with white noise input; Figure 5 (d) is a schematic diagram of the sound field imaging of the music signal input in region B. Detailed Implementation

[0080] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0081] Example 1:

[0082] See Figures 1 to 5 The frequency domain block adaptive sound field partitioning method based on acoustic contrast control includes the following steps:

[0083] 1) Arrange L loudspeakers in the area to be partitioned, and define the control filter coefficients of the L loudspeakers at the initial time, where L is a positive integer.

[0084] 2) Based on the control filter coefficients at the current moment, the reproduced sound field at different control points is calculated.

[0085] 3) Based on the reproduced sound field at different control points, construct a time-domain block-by-block updated signal model.

[0086] 4) Based on FFT, the frequency domain fast cyclic convolution is used to solve the time domain block-by-block update signal model using the gradient descent method to obtain the updated control filter coefficients.

[0087] 5) Based on the updated control filter coefficients, the output signals of L loudspeakers are calculated, and then adaptive sound field zoning is achieved in the area to be zoned.

[0088] Example 2:

[0089] The frequency-domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical content of which is described in Example 1, further wherein the control filter coefficients at the initial moment are as follows:

[0090] (1)

[0091] In the formula, Indicates the speaker index. Represents the discrete time index. Indicates the first The speaker in the A vector of control filter coefficients at discrete time points. Indicates the filter length. These represent the control filter coefficients for different filter lengths. This indicates transpose.

[0092] Example 3:

[0093] The frequency-domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical contents of which are described in any one of Embodiments 1 and 2, further, the reproduced sound field at different control points is as follows:

[0094] (2)

[0095] In the formula, Indicates the control point index. Indicates the first The reproduced sound field at each control point. Indicates the speaker index. Indicates the first At the discrete time... The speaker to the Transpose of the impulse response function estimate for each control point. Indicates the first The speaker in the A vector of control filter coefficients at discrete time points. This represents the control filter coefficient vector for all loudspeakers.

[0096] Wherein, the input signal matrix Input signal diagonally repeated stacked matrix , No. Transpose of the impulse response function estimated at each control point As shown below:

[0097] (3)

[0098] (4)

[0099] (5)

[0100] In the formula, Represents the identity matrix. This represents the Kronecker product. This represents the length of the estimated impulse response function. They represent the first From the discrete time point to the... The input signal at discrete time points. Each represents a different speaker up to the [number]. Transpose of the impulse response function estimate for each control point.

[0101] Input signal diagonally repeated stacked matrix Essentially, it involves repeating the input matrix multiple times and arranging it diagonally. The number of repetitions is related to the number of loudspeakers, and the purpose is to construct the reproducible sound pressure level. and filter long vector The relationship between them is shown in formula (2).

[0102] Example 4:

[0103] The frequency-domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical contents of which are described in any one of Examples 1 to 3, further wherein the optimization function for the time-domain block-by-block update signal model is as follows:

[0104] (6)

[0105] In the formula, This represents the optimization function for updating the signal model block by block in the time domain. This indicates the length of the time-domain block. k represents the time block index. Indicates the first The control filter coefficient vector of all loudspeakers at each discrete time. Indicates the control point index. , These represent the number of control points in the bright sound zone and the dark sound zone, respectively. Indicates the first The first discrete time of the bright sound region The replay sound field vector of each control point. Indicates the first The first discrete moment of the dark region The replay sound field vector of each control point. Indicates the first The first discrete time of the bright sound region The desired sound field at each control point. Represents the norm. This represents the cost function corresponding to sound contrast ratio (the ratio of sound energy in bright and dark areas) control. This indicates the distortion of the sound field reproduced in the bright area. This represents a linear representation of the target sound contrast; for example, if the target sound contrast is 20 dB, then... . The regularization parameter represents the distortion of the reproduced sound field. The absolute value of the first part of Formula (6) is used to control the contrast of the target sound, and the square of the second norm of the second part is used to penalize the distortion of the reproduced sound field.

[0106] Among them, in the first The first discrete moment in the bright sound region and the dark sound region The replay sound field vectors for each control point are shown below:

[0107] (7)

[0108] (8)

[0109] In the formula, Indicates the speaker index. , They represent the first time. At the nth discrete time, the th The first speaker in the bright sound zone and the dark sound zone Filtered input signals for each control point. , They represent the first time. The first discrete moment in the bright sound region and the dark sound region The Toeplitz matrix is ​​composed of the filtered input signals of each control point. Indicates the first The speaker in the The control filter coefficient vector at discrete time points. Indicates the first The control filter coefficient vector of all loudspeakers at each discrete time.

[0110] Example 5:

[0111] The frequency domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical contents of which are described in any one of embodiments 1 to 4, further, in the... At the nth discrete time, the th The first speaker in the bright sound zone Filtered input signal at each control point The first of the dark sound zone Filtered input signal at each control point Use parameters uniformly This is represented as follows:

[0112] (9)

[0113] (10)

[0114] (11)

[0115] In the formula, Indicates the filter length. They represent the first time step at different discrete times. The speaker in the The filtered input signal vector of each control point. They represent the first time step at different discrete times. The speaker in the Filtered input signals for each control point. This represents the vector used to calculate the filtered input signal. The Toeplitz matrix is ​​composed of the input signals. Indicates the first At the nth discrete time, the th The speaker in the Impulse response function estimation for each control point. This represents the length of the estimated impulse response function. These represent the input signals at different discrete times.

[0116] Example 6:

[0117] The frequency-domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical contents of which are described in any one of Examples 1 to 5, further includes the following step of using the gradient descent method to solve the time-domain block-by-block updated signal model to obtain the updated control filter coefficients:

[0118] S1 uses the gradient descent method to minimize the optimization function of the time-domain block-by-block update signal model, obtaining the... The control filter coefficients at discrete time points are shown below:

[0119] (12)

[0120] In the formula, This indicates the length of the time-domain block. k represents the time block index. , They represent the first time. , The control filter coefficient vector of all loudspeakers at each discrete time. This represents the step size parameter. This represents the optimization function for updating the signal model block by block in the time domain.

[0121] Wherein, gradient vector As shown below:

[0122] (13)

[0123] In the formula, Indicates the control point index. , These represent the number of control points in the bright sound zone and the dark sound zone, respectively. , They represent the first time. The first discrete moment in the bright sound region and the dark sound region The transpose of the Toeplitz matrix, which consists of the filtered input signals of control points. , They represent the first time. The first discrete moment in the bright sound region and the dark sound region The replay sound field vector of each control point. Indicates the first The first discrete time of the bright sound region The desired sound field at each control point. A linear representation of the target acoustic contrast. The regularization parameter represents the distortion of the reproduced sound field.

[0124] Among them, the block average acoustic contrast As shown below:

[0125] (14)

[0126] In the formula, Represents the norm.

[0127] S2 uses a sound field modeling method to model the reproduced sound field.

[0128] S3 substitutes the replay sound field constructed in step S2 into the... The gradient vectors of the control filter coefficients at discrete time points are used to obtain the updated control filter coefficients.

[0129] Example 7:

[0130] The frequency domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical contents of which are described in any one of Embodiments 1 to 6, further, the playback sound field constructed in step S2 is as follows:

[0131] (15)

[0132] (16)

[0133] (17)

[0134] In the formula, Indicates the speaker index. , They represent the first time. At the nth discrete time, the th The first speaker in the bright sound zone and the dark sound zone Filtered input signals for each control point. Indicates the first At the nth discrete time, the th The control filter coefficient vector of each speaker. Indicates the first The vector of speaker control filter coefficients corresponding to the desired sound field at each discrete time point.

[0135] Example 8:

[0136] The frequency domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical contents of which are described in any one of Embodiments 1 to 7, further, the output signals of the L loudspeakers are as follows:

[0137] (18)

[0138] (19)

[0139] In the formula, This indicates the length of the time-domain block. k represents the time block index. Indicates the speaker index. Indicates the first At the nth discrete time, the th The output signal vector of each speaker. Indicates the first The Toeplitz matrix, composed of the input signals used to calculate the loudspeaker output signal at discrete time points. Indicates the first At the nth discrete time, the th The control filter coefficient vector of each speaker. Indicates the filter length. These represent the input signals at different discrete times.

[0140] Example 9:

[0141] The frequency-domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical contents of which are described in any one of Examples 1 to 8, further wherein the frequency-domain fast circular convolution of the FFT is used to filter the input signal vector gradient vector Reproduced sound field vector Output signal vector Convert it to a circular convolution form to speed up the solution.

[0142] Example 10:

[0143] The frequency-domain block adaptive sound field partitioning method based on acoustic contrast control, the main technical contents of which are described in any one of Examples 1 to 9, further, the filtered input signal vector gradient vector Reproduced sound field vector Output signal vector The steps to convert it to a circular convolution form are as follows:

[0144] Q1 introduces the Overlap-Save method to filter the input signal vector. gradient vector Reproduced sound field vector Output signal vector Equivalent to circular convolution, as shown below:

[0145] (20)

[0146] (twenty one)

[0147] (twenty two)

[0148] (twenty three)

[0149] In the formula, This indicates the length of the time-domain block. k represents the time block index. Indicates the speaker index. Indicates the control point index. This represents the vector used to calculate the filtered input signal. The Toeplitz matrix is ​​composed of the input signals. Indicates the first At the nth discrete time, the th The speaker in the Impulse response function estimation for each control point. This represents the length of the estimated impulse response function. Indicates the filter length. , , , All represent zero matrices. , , , Both represent identity matrices. , Indicates the first , At the nth discrete time, the th The control filter coefficient vector of each speaker. Indicates the first At the nth discrete time, the th The first speaker in the bright sound zone Filtered input signals for each control point. Indicates the first At the nth discrete time, the th The first speaker in the bright sound zone The replay sound field vector of each control point. Indicates the first The Toeplitz matrix, composed of the input signals used to calculate the loudspeaker output signal at discrete time points. Refers to the gradient vector The gradient in the middle (only includes the sound energy part of the sound field reproduced in the bright area; the rest can be obtained in the same way). Indicates the first The first discrete time of the bright sound region The transpose of the Toeplitz matrix, which consists of the filtered input signals of control points. Indicates the first The first discrete time of the bright sound region The replay sound field vector of each control point. This represents the time inversion matrix. Represents the replay sound field vector The vectors arranged in reverse order. , They represent the first time. , Input signal at discrete time points. , They represent the first time. , At the nth discrete time, the th The first speaker in the bright sound zone Filtered input signals for each control point.

[0150] Q2 uses FFT to accelerate the recurrent convolution obtained in step Q1, as shown below:

[0151] (twenty four)

[0152] (25)

[0153] (26)

[0154] (27)

[0155] (28)

[0156] (29)

[0157] (30)

[0158] (31)

[0159] (32)

[0160] In the formula, This represents the discrete inverse Fourier transform matrix. This represents the discrete Fourier transform matrix. , These represent the frequency domain components of the input signal and the estimated impulse response function, respectively. This represents the Hadamard product. , These represent the frequency domain components of the filtered input signal and the control filter coefficients, respectively. Indicates the Ming Sheng District No. Frequency domain components of the reproduced sound field at each control point. Indicates conjugate.

[0161] Example 11:

[0162] See Figures 1 to 5 The frequency domain block adaptive sound field partitioning method based on acoustic contrast control includes the following steps:

[0163] Step 1: Construct a time-domain block-by-block update signal model

[0164] Figure 2 This is a schematic diagram of a personal vocal zone system. Each speaker enables dual-zone control of both bright and dark sound, with the number of control points in each zone being [number missing]. and Unless otherwise specified, subsequent variables will use the subscripts "B" and "D" to distinguish between bright and dark sound zones. Define the... Each speaker at discrete time The control filter coefficients at that time are The filter length is The superscript "T" indicates transpose. The audio input signal to be played in the bright area. Convolved with the speaker adaptive control filter, we obtain the first... The output of each speaker is as follows:

[0165] (1)

[0166] in, The input signal vector is used. The reproduced sound field at the control point is obtained by convolving the loudspeaker output signal with the room impulse response function in the time domain, such as the control point. The corresponding number The reproduced sound pressure at each discrete moment can be expressed as:

[0167] (2)

[0168] in, Discrete time Time The speaker to the The room impulse response (RIR) function is estimated for each control point, with a length of [length missing]. And it is assumed that its change over time is known (such as the RIRs before and after the change of the offline predicted sound field or the real-time measurement by arranging several feedback microphones). This is the corresponding loudspeaker output signal vector. Combining equations (1) and (2), we can see that:

[0169] (3)

[0170] in, Let be the long vector formed by stacking all the speaker control filters. , and They are represented as follows:

[0171] (4)

[0172] (5)

[0173] (6)

[0174] in, The Hankel matrix constructed for the input signal, represent identity matrix This represents the Kronecker product.

[0175] Assuming each filter coefficient The update is only updated once for each sampling point, so the first sampling point... Control points under each time block The replay sound field vector at that location It can be represented as:

[0176] (7)

[0177] in, Indicates the length of the time-domain block. The Toeplitz matrix, composed of the filtered input signals, is expressed as follows (the variables for bright and dark sound regions are expressed the same way, and subscripts are omitted here):

[0178] (8)

[0179] (9)

[0180] (10)

[0181] (11)

[0182] in, Discrete time The corresponding impulse response function estimation, The Toeplitz matrix, composed of the input signals, is used to calculate the filtered input signal vectors arranged in ascending order. The filtered input is essentially a convolution of the speaker input signal and the estimated RIRs of the bright and dark sound zones.

[0183] Step 2: Update the filter coefficients in the time domain block by block.

[0184] Based on the reproducible sound field calculation in equation (7), the first... The optimization objective function for each time block is as follows:

[0185] (12)

[0186] in, and They represent the first The sound field reproduced in the bright and dark areas of each time block. Then it means the first The expected sound field in the bright area under each time block This represents the 2-norm.

[0187] Minimize using gradient descent , obtained the The time-domain filter coefficients are updated as follows for each time block:

[0188] (13)

[0189] in, The step size parameter under block update. The gradient vector is as follows:

[0190] (14)

[0191] also, The block average acoustic contrast ratio is calculated as follows:

[0192] (15)

[0193] For the reproduced sound field of the bright and dark sound areas in equation (14) and In practice, considering the difficulty of deploying a sufficient number of monitoring microphones near the human ear, this embodiment uses a sound field modeling method to obtain... and Assuming the room impulse response function can be estimated using virtual sensing techniques, the reproduced sound field is modeled using RIR estimation, input signal, and control filters as follows:

[0194] (16)

[0195] (17)

[0196] (18)

[0197] in, Indicates the first The first time block The control filter vector corresponding to the speaker is used to model the . The reproduced sound field within each time block, i.e., the filter update, has a delay. Furthermore, This represents the loudspeaker control filter corresponding to the desired sound field.

[0198] Step 3: Time-domain loudspeaker array output calculation

[0199] Speaker output The calculation is as follows:

[0200] (19)

[0201] (20)

[0202] in, The Toeplitz matrix, composed of the input signals, is used to calculate the speaker output signals arranged in ascending order.

[0203] Step 4: Frequency Domain Fast Recurrent Convolution Based on FFT

[0204] To ensure continuity between blocks, the Overlap-Save method is introduced. The default overlap rate is... corresponding block length This is both the block input length for a single iteration and the overlap length. At this point, the temporal linear convolutions in equations (10), (16), and (19) and the temporal linear correlation in equation (14) can be equivalently represented by the following circular convolutions (the remaining equations can be obtained similarly):

[0205] (twenty one)

[0206] (twenty two)

[0207] (twenty three)

[0208] (twenty four)

[0209] in, and Represent Identity matrix and zero matrix and Represent Identity matrix and zero matrix This represents the circular convolution of two vectors. The gradient component used in equation (14) (which only includes the sound energy part of the sound field reproduced in the bright area; the rest can be obtained similarly) This is the time inversion matrix (all values ​​on the second diagonal are 1, and all other positions are 0). superscript " "" indicates that the vectors are arranged in reverse order.

[0210] To further improve computational efficiency, FFT is used to accelerate the calculation of the above circular convolution, defining... For length The discrete Fourier transform matrix, whose elements are... ( Then the circular convolution in equation (21) can be expressed as:

[0211] (25)

[0212] (26)

[0213] (27)

[0214] in, and These represent the frequency domain components of the input signal and the impulse response function, respectively. Represents the Hadamard product. Let represent the discrete Fourier inverse transform matrix. Similarly, equation (22) can be expressed as:

[0215] (28)

[0216] (29)

[0217] (30)

[0218] in, and Let represent the frequency domain components of the filtered input signal and the control filter, respectively. Equation (23) is then expressed as:

[0219] (31)

[0220] Finally, equation (24) can also be implemented using FFT, as expressed below:

[0221] (32)

[0222] (33)

[0223] in, For Ming District No. The frequency domain components of the reproduced sound field at each control point are marked with " " indicates finding conjugate, The conjugation in the simplified time inversion matrix is ​​due to... What it produces.

[0224] The proposed frequency domain block adaptive method is named FBFxLAD (Frequency-domain BlockFilter-x Least Absolute Deviation), and its algorithm flow is as follows: Figure 1 As shown, it mainly includes four parts: filter input calculation, replay sound field modeling, filter update, and output signal calculation.

[0225] Performance verification based on measured impulse response function: To verify the correctness and effectiveness of the proposed method, multi-region acoustic reproduction experiments were conducted using measured RIR functions. For example... Figure 3 (a) shows the array layout, with 10 loudspeakers spaced 6° apart to form an arc-shaped loudspeaker array. The green ☆ represents the reference loudspeaker corresponding to the desired sound location in the bright area. 35 RIRs sampling points are arranged in each of regions A and B. (5 cm intervals), of which 12 control points are used to design the control filters for the loudspeaker array, and 23 monitoring points are used to evaluate the sound field zoning control effect. Figure 3(b) shows the RIR display from the reference loudspeaker to a certain control point, with reverberation time... The sampling frequency is 16 kHz, and the sampling length is In adaptive algorithms, the modeling length of RIRs is limited by computational cost. Typically, this setting can only cover a portion of the RIRs sampling length. This is sufficient to cover the direct sound and major reflections in RIRs. Additionally, a modeling delay is set. Filter length FFT length is Regularization parameters Taking into account the speaker array and control point spacing, and balancing low-frequency performance with high-frequency upper limits, the sound field zoning control frequency range is set as follows: .

[0226] To verify the adaptive sound contrast tracking capability and bright-area signal distortion suppression capability of FBFxLAD, a bright-and-dark sound zone swapping experiment was conducted. The target sound contrast was set to 15 dB, and the input signals were 30 s white noise and music signals. Initially, region A was designated as the bright zone and region B as the dark zone. At a midpoint, the positions of the bright and dark sound zones were swapped. The sound contrast of regions A and B was observed. Signal distortion in the bright area The convergence performance over time is as follows: Figure 4 As shown. Figure 4 (a) and (c) show the convergence results under white noise input, from which the FBFxLAD's convergence can be seen. The FBFxLAD exhibits good convergence to 15 dB in the initial stage and converges rapidly to -15 dB after switching between bright and dark sound zones, indicating that it has excellent performance. Adaptive tracking capability. Furthermore, from the bright area... It can be seen that FBFxLAD ensures After converging to the expected value, it will be reduced as much as possible. . Figure 4(b) and (d) show the convergence results with a music signal input. Their overall convergence trend is consistent with the white noise results, but the convergence speed is significantly slower after the exchange between bright and dark sound regions. This is because the music signal is a correlated input, which leads to a further reduction in the maximum step size. Furthermore, after convergence... and The fluctuations are also greater because the frequency domain components and time domain amplitude of the music signal change rapidly over time. Figure 5 This supplements the aforementioned imaging of the reproduced sound field in regions A / B of white noise and music signal input.

Claims

1. A frequency-domain block-adaptive soundfield partitioning method based on sound contrast control, characterized in that, The method comprises the following steps: 1) arranging L loudspeakers in a region to be partitioned, and defining control filter coefficients of the L loudspeakers at an initial time, L being a positive integer; 2) calculating a playback sound field at different control points based on the control filter coefficients at the current time; 3) constructing a time-domain block-by-block updating signal model based on the playback sound field at the different control points; 4) solving the time-domain block-by-block updating signal model by using a gradient descent method based on FFT frequency-domain fast convolution, to obtain updated control filter coefficients; 5) calculating output signals of the L loudspeakers based on the updated control filter coefficients, and then implementing adaptive sound field partitioning in the region to be partitioned.

2. The sound contrast control based frequency domain block adaptive sound field partitioning method according to claim 1, characterized in that, The control filter coefficients at the initial time are as follows: (1) wherein denotes a loudspeaker index, denotes a discrete time index; denotes a control filter coefficient vector of the th loudspeaker at the th discrete time; denotes a filter length; denotes a control filter coefficient of the denotes a transpose.

3. The sound contrast control based frequency domain block adaptive sound field partitioning method according to claim 1, wherein, The playback sound field at the different control points is as follows: (2) In the formula, Indicates the control point index; Indicates the first The reproduced sound field at each control point; Indicates the speaker index. Indicates the first At the discrete time... The speaker to the Transpose of the impulse response function estimate for each control point; Indicates the first The speaker in the A vector of control filter coefficients at discrete time points; This represents the control filter coefficient vector for all loudspeakers; where the input signal matrix , the input signal diagonal-repeated stacked matrix , the impulse response function estimate of the th control point is given by (3) (4) (5) wherein denotes the identity matrix; denotes the Kronecker product; denotes the length of the impulse response function estimate; denotes the input signal from the first discrete time instant to the first discrete time instant, respectively; denotes the transpose of the impulse response function estimate from the different loudspeaker to the first control point, respectively.

4. The sound contrast control based frequency domain block adaptive sound field partitioning method of claim 1, wherein, The optimization function of the time-domain block-by-block updating signal model is as follows: (6) wherein represents an optimization function for time-domain block-wise updating signal model; represents a time-domain block length; k represents a time block index; represents a control filter coefficient vector of all loudspeakers at the discrete time instant; represents a control point index; , respectively represent the number of control points in the sound- on region and the sound-off region; represents a playback soundfield vector of the control point in the sound-on region at the discrete time instant; represents a playback soundfield vector of the control point in the sound-off region at the discrete time instant; represents an expected soundfield of the control point in the sound-on region at the discrete time instant; represents a norm; represents a cost function corresponding to the sound contrast control, represents a sound-on region playback soundfield distortion; represents a linear representation of the target sound contrast; represents a regularization parameter for the playback soundfield distortion; wherein at the first discrete time instant the playback sound field vectors of the first control points of the bright sound zone and the dark sound zone are as follows: (7) (8) wherein, represents a loudspeaker index, , respectively represent the filtered input signal of the th control point of the near sound zone, far sound zone of the th loudspeaker at the th discrete time instant; , respectively represent a Toeplitz matrix composed of the filtered input signal of the th control point of the near sound zone, far sound zone at the th discrete time instant; represents the control filter coefficient vector of the th loudspeaker at the th discrete time instant; represents the control filter coefficient vector of all loudspeakers at the th discrete time instant.

5. The sound contrast control based frequency domain block adaptive sound field partitioning method according to claim 4, characterized in that, The filtered input signal at the control point of the loudspeaker in the loud area at the discrete time instant The filtered input signal at the control point of the loudspeaker in the loud area at the discrete time instant The filtered input signal at the control point of the loudspeaker in the loud area at the discrete time instant The filtered input signal at the control point of the loudspeaker in the loud area at the discrete time instant The filtered input signal at the control point of the loudspeaker in the loud area at the discrete time instant The filtered input signal at the control point of the loudspeaker in the loud area at the discrete time instant The filtered input signal at the control point of the loudspeaker in the loud area at the discrete time instant are represented by a parameter (9) (10) (11) wherein denotes the filter length; denote the filtered input signal vector of the th loudspeaker at the th control point at different discrete time instants; denote the filtered input signal of the th loudspeaker at the th control point at different discrete time instants; denotes a Toeplitz matrix of input signal components used to compute the filtered input signal vector denotes a Toeplitz matrix of input signal components used to compute the filtered input signal vector denote the impulse response function estimate of the th loudspeaker at the th control point at a discrete time instant; denote the impulse response function estimate of the denotes the length of the impulse response function estimate; denote the input signal at different discrete time instants.

6. The sound contrast control based frequency domain block adaptive sound field partitioning method according to claim 5, characterized in that, The step of solving the time-domain block-by-block updating signal model by using the gradient descent method to obtain the updated control filter coefficients is as follows: S1minimizes the optimization function of the time-domain block-wise updated signal model using the gradient descent method to obtain the control filter coefficients at the first discrete time instant, as follows: (12) wherein denotes the time-domain block length; k denotes the time block index; , denote the control filter coefficient vector of all loudspeakers at the , discrete time instances, respectively; denotes a step size parameter; denotes an optimization function for time-domain block-wise updating the signal model; where the gradient vector As shown below: (13) In the formula, Indicates the control point index; , These represent the number of control points in the bright sound zone and the dark sound zone, respectively. , They represent the first time. The first discrete moment in the bright sound region and the dark sound region Transpose of the Toeplitz matrix composed of the filtered input signals of control points; , They represent the first time. The first discrete moment in the bright sound region and the dark sound region The replay sound field vector of each control point; Indicates the first The first discrete time of the bright sound region The desired sound field at each control point; A linear representation of the target acoustic contrast; Regularization parameter representing the distortion of the reproduced sound field; where the block average acoustic contrast as follows: (14) wherein denotes the norm; S2 models the playback sound field by using a sound field modeling method; S3 substitutes the playback sound field constructed in step S2 into the gradient vector of the control filter coefficient at the i-th discrete time point to obtain the updated control filter coefficient. S3 substitutes the playback sound field constructed in step S2 into the gradient vector of the control filter coefficient at the i-th discrete time point to obtain the updated control filter coefficient.

7. The sound contrast control based frequency domain block adaptive sound field partitioning method according to claim 6, characterized in that, The playback sound field constructed in the step S2 is as follows: (15) (16) (17) In the formula, Indicates the speaker index. , They represent the first time. At the nth discrete time, the th The first speaker in the bright sound zone and the dark sound zone Filtered input signals for each control point; Indicates the first At the nth discrete time, the th The control filter coefficient vector of each loudspeaker; Indicates the first The vector of speaker control filter coefficients corresponding to the desired sound field at each discrete time point.

8. The sound contrast control based frequency domain block adaptive sound field partitioning method according to claim 7, characterized in that, The output signals of the L loudspeakers are as follows: (18) (19) wherein denotes the time-domain block length; k denotes the time block index; denotes the loudspeaker index, denotes the output signal vector of the th loudspeaker at the th discrete time instant; denotes the output signal vector of the th loudspeaker at the th discrete time instant; denotes the control filter coefficient vector of the th loudspeaker at the denotes the filter length; denotes the input signal at different discrete time instants, respectively.

9. The sound contrast control based frequency domain block adaptive sound field partitioning method according to claim 8, characterized in that, The frequency domain fast convolution of the FFT is used to convert the filtering input signal vector , gradient vector , playback soundfield vector , output signal vector into a convolution form, which is then solved faster.

10. The sound contrast control based frequency domain block adaptive sound field partitioning method of claim 9, wherein, the filtered input signal vector , gradient vector , playback soundfield vector , output signal vector The step of converting to a circular convolution form is as follows: Q1 introduces the Overlap-Save method, which filters the input signal vector , the gradient vector , the playback soundfield vector , the output signal vector is equivalent to a circular convolution, as follows: (20) (21) (22) (23) In the formula, Indicates the time-domain block length; k represents the time block index; Indicates the speaker index. Indicates the control point index; This represents the vector used to calculate the filtered input signal. The Toeplitz matrix composed of the input signals; Indicates the first At the nth discrete time, the th The speaker in the Estimation of the impulse response function at each control point; This represents the length of the estimated impulse response function; Indicates the filter length; , , , Both represent zero matrices; , , , Both represent identity matrices; , Indicates the first , At the nth discrete time, the th The control filter coefficient vector of each loudspeaker; Indicates the first At the nth discrete time, the th The first speaker in the bright sound zone Filtered input signals for each control point; Indicates the first At the nth discrete time, the th The first speaker in the bright sound zone The replay sound field vector of each control point; Indicates the first The Toeplitz matrix, composed of the input signals used to calculate the loudspeaker output signal at discrete time points; Refers to the gradient vector Partial gradient in; Indicates the first The first discrete time of the bright sound region Transpose of the Toeplitz matrix composed of the filtered input signals of control points; Indicates the first The first discrete time of the bright sound region The replay sound field vector of each control point; Represents the time inversion matrix; Represents the replay sound field vector The vectors arranged in reverse order; , represent input signals at the 、 discrete time instants, respectively; 、 represent filtered input signals at the 、 discrete time instants, respectively, for the th loudspeaker at the th control point in the near-field region; Q2 accelerates the convolution obtained in the step Q1 by using FFT, as follows: (24) (25) (26) (27) (28) (29) (30) (31) (32) wherein denotes the inverse discrete Fourier transform matrix; denotes the discrete Fourier transform matrix; , denote the frequency domain components of the input signal and the impulse response function estimate, respectively; denotes the Hadamard product; , denote the frequency domain components of the filtered input signal and the control filter coefficients, respectively; denotes the frequency domain component of the playback sound field at the m-th control point in the listening area; denotes the frequency domain component of the playback sound field at the m-th control point in the listening area; denotes the conjugate.