MVDR beam forming method and device

Abstract

The invention provides an MVDR beam forming method and device, and the method comprises the steps: obtaining an observation acoustic matrix formed after a plurality of acoustic sensors receive a plurality of target signal sources, and enabling the plurality of acoustic sensors to be arranged in an array manner; drawing a potential function numerical diagram based on the observation acoustic matrix to determine the number of target signal sources and the direction of arrival corresponding to each target signal source; calculating a covariance matrix based on the observation acoustic matrix; revising the covariance matrix based on the number of the target signal sources to generate an optimized covariance matrix; and obtaining a beam forming result based on the observation acoustic matrix, the optimization covariance matrix and the array response matrix. The method and the device are used for improving signal extraction quality and accuracy of MVDR beam forming in a multi-signal-source and noise interference environment.

Description

Technical Field

[0001] The present application relates to the field of sound processing technology, and in particular to an MVDR beamforming method and device. Background Art

[0002] The MVDR (Minimum Variance Distortionless Response) beamforming algorithm is a milestone in beamforming technology. It was proposed in the 1960s and is mainly used to solve the signal orientation problem in radar and sonar systems. However, the existing MVDR algorithm has poor signal extraction quality and low recognition accuracy in environments with multiple signal sources and noise interference. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide an MVDR beamforming method and apparatus to improve the signal extraction quality and accuracy of MVDR beamforming in an environment with multiple signal sources and noise interference.

[0004] In a first aspect, the present invention provides an MVDR beamforming method, which includes obtaining an observed acoustic matrix formed after multiple acoustic sensors receive multiple target signal sources, wherein the multiple acoustic sensors are arranged in an array; drawing a potential function numerical graph based on the observed acoustic matrix to determine the number of target signal sources and the angle of arrival corresponding to each target signal source; calculating a covariance matrix based on the observed acoustic matrix; revising the covariance matrix based on the number of target signal sources to generate an optimized covariance matrix; and obtaining a beamforming result based on the observed acoustic matrix, the optimized covariance matrix and the array response matrix.

[0005] In an optional implementation, the step of drawing a potential function numerical graph based on the observed acoustic matrix specifically includes:

[0006] Perform Fourier transform on the observed acoustic signals collected by any two acoustic sensors to obtain the complex spectrum corresponding to each observed acoustic signal; revise the two complex spectra based on the degree of sign matching between the real part and the imaginary part corresponding to each frequency in the two complex spectra; and draw a potential function numerical graph based on the revised two complex spectra and the potential function, wherein the abscissa of the potential function numerical graph is the angle value and the ordinate of the potential function numerical graph is the potential function value.

[0007] In an optional implementation, the number of target signal sources is determined in the following manner: the number of peaks in the potential function value graph is determined as the number of target signal sources.

[0008] In an optional implementation, the angle value corresponding to each wave crest is determined as the arrival angle; and the arrival angle is subjected to Fourier transformation to generate a response vector to form the array response matrix.

[0009] In an optional embodiment, the covariance matrix R is calculated in the following manner:

[0010] R=X′ T X′ / (N-1);

[0011] X′=[fx 1 ,fx 2 ...,fx n ];

[0012] Among them, fx n is the observed acoustic signal x collected by acoustic sensor n n The transformation result after Fourier transform, N is x n The length value of .

[0013] In an optional embodiment, the optimized covariance matrix is ​​generated in the following manner:

[0014] Based on the covariance matrix, a pseudo-inverse matrix is ​​determined; the pseudo-inverse matrix is ​​decomposed into a first eigenvector matrix and a first eigenvalue matrix; a preset number of eigenvalues ​​are selected from the first eigenvalue matrix to form a second eigenvalue matrix; eigenvectors at corresponding positions are selected from the first eigenvector matrix to form a second eigenvector matrix; based on the second eigenvalue matrix and the second eigenvector matrix, an optimized covariance matrix is ​​reconstructed and generated.

[0015] In an optional embodiment, the two complex spectra are revised in the following manner: a first product between the real part and the imaginary part corresponding to the target frequency in the first complex spectrum is calculated; a second product between the real part and the imaginary part corresponding to the target frequency in the second complex spectrum is calculated; it is determined whether the first product and the second product are both greater than or equal to zero; if not, the real part and the imaginary part corresponding to the target frequency in the first complex spectrum and the second complex spectrum are deleted.

[0016] In a second aspect, the present invention provides an MVDR beamforming device, the device comprising:

[0017] An acquisition module, used to acquire an observation acoustic matrix formed after a plurality of acoustic sensors receive a plurality of target signal sources, wherein the plurality of acoustic sensors are arranged in an array;

[0018] An analysis module is used to draw a potential function numerical graph based on the observed acoustic matrix to determine the number of target signal sources and the arrival angle corresponding to each target signal source;

[0019] A calculation module, used for calculating a covariance matrix based on an observed acoustic matrix;

[0020] A revision module, used for revising the covariance matrix based on the number of target signal sources to generate an optimized covariance matrix;

[0021] The forming module is used to obtain a beamforming result based on an observed acoustic matrix, an optimized covariance matrix and an array response matrix.

[0022] In a third aspect, the present invention provides an electronic device comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus, and the processor executes the machine-readable instructions to perform the steps of any MVDR beamforming method as described in the aforementioned embodiments.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the MVDR beamforming method in any of the aforementioned embodiments are executed.

[0024] The purpose of the embodiment of the present application is to provide an MVDR beamforming method and device, wherein the method includes obtaining an observed acoustic matrix formed after multiple acoustic sensors receive multiple target signal sources, wherein the multiple acoustic sensors are arranged in an array; drawing a potential function numerical diagram based on the observed acoustic matrix to determine the number of target signal sources and the wave arrival angle corresponding to each target signal source; calculating the covariance matrix based on the observed acoustic matrix; revising the covariance matrix based on the number of target signal sources to generate an optimized covariance matrix; obtaining the beamforming result based on the observed acoustic matrix, the optimized covariance matrix and the array response matrix. By introducing the potential function and combining the subspace analysis, the generation of the covariance matrix is ​​optimized, thereby improving the signal extraction quality and accuracy of the MVDR beamforming in a multi-signal source and noise interference environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 A flowchart of an MVDR beamforming method provided in an embodiment of the present application;

[0027] Figure 2A potential function numerical diagram provided in an embodiment of the present application;

[0028] Figure 3 A schematic diagram of an output signal of a target signal source provided in an embodiment of the present application;

[0029] Figure 4 A schematic diagram of an output signal of another target signal source provided in an embodiment of the present application;

[0030] Figure 5 A schematic diagram of the structure of an MVDR beamforming device provided in an embodiment of the present application;

[0031] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0032] In recent years, acoustic beamforming algorithms have attracted attention due to their wide application in many fields, especially in the field of drone detection and prevention. Acoustic beamforming technology adjusts the basic unit parameters of the microphone array so that signals at certain angles obtain constructive interference, while signals at other angles obtain destructive interference, thereby achieving directional reception and enhancement of sound waves in a specific direction. The advantage of this technology is that it can accurately locate and identify the target sound source in a complex acoustic environment, while suppressing interference signals from non-target directions.

[0033] The minimum mean square distortionless response (MVDR) beamforming algorithm was proposed by J. Capon in 1969. The algorithm is based on the minimum mean square error criterion. While constraining the gain in the target direction to remain unchanged, it minimizes the total energy output by the beamformer, that is, it outputs the minimum interference and noise power, suppresses interference and noise signals, and thus recovers the target speech.

[0034] Therefore, a reasonable array response matrix and covariance matrix are very important for the application of the MVDR algorithm waveform forming algorithm.

[0035] Based on this, the present application provides an MVDR beamforming method and device, and proposes to apply the potential function to the MVDR algorithm to solve the deficiency of requiring prior information on the direction of the acoustic signal. The covariance matrix of the signal is analyzed using the subspace to improve the accuracy of the algorithm.

[0036] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0037] Figure 1Flow chart of a MVDR beamforming method provided in an embodiment of the present application. Figure 1 As shown, an MVDR beamforming method provided in an embodiment of the present application includes:

[0038] S1. Obtain an observation acoustic matrix formed after multiple acoustic sensors receive multiple target signal sources, wherein the multiple acoustic sensors are arranged in an array.

[0039] After the array acoustic sensor obtains multiple target signal sources in different directions, the received signals can be expressed in the form of the observed acoustic matrix X as follows:

[0040]

[0041] X=AS

[0042] Among them, A is the mixing matrix, S is the transmission signal matrix corresponding to the target signal source, m is the number of target signal sources, and n is the number of acoustic sensors.

[0043] S2. Draw a potential function numerical graph based on the observed acoustic matrix to determine the number of target signal sources and the arrival angle corresponding to each target signal source.

[0044] In step S2, the step of drawing a potential function numerical graph based on the observed acoustic matrix specifically includes:

[0045] Perform Fourier transform on the observed acoustic signals collected by any two acoustic sensors to obtain the complex spectrum corresponding to each observed acoustic signal.

[0046] Each observed acoustic signal is subjected to a fast Fourier transform (FFT) to obtain the corresponding complex spectrum, which can be expressed as:

[0047]

[0048] Where k is the frequency index, F s is the sampling frequency of the sound sensor, and i is an imaginary value. The main function of FFT is to convert the time domain signal to the frequency domain, so as to analyze the frequency components of the signal.

[0049] The two complex spectra are revised based on the degree of sign matching between the real and imaginary parts corresponding to each frequency in the two complex spectra.

[0050] Through Fourier transform, the original signal is decomposed into a series of sine and cosine wave components of different frequencies, each of which is represented in complex form, containing amplitude and phase information. Specifically, the real and imaginary parts of the complex number correspond to the amplitudes of the cosine and sine waves, respectively, while the phase is given by the inverse tangent of the ratio of the real and imaginary parts. Therefore, if the real and imaginary parts of a frequency point have the same sign, it indicates that the cosine and sine wave components of the frequency point are consistent in phase, that is, they are in phase.

[0051] In-phase refers to the synchronization of the oscillations of two signal components in time, that is, they reach their respective maximum and minimum values ​​at the same time. In signal processing, if two signal components (such as cosine waves and sine waves) are in phase at a specific frequency, it means that they are closely related in time and may originate from the same physical process or the same signal source.

[0052] In audio signal processing, two frequency components in the spectrum that are in phase may indicate that they are both produced by the same sound source. The presence of phase provides an important clue for identifying and separating different sound sources.

[0053] Therefore, the two complex spectra can be revised in the following way:

[0054] A first product between the real part and the imaginary part corresponding to the target frequency in the first complex spectrum is calculated.

[0055] A second product between the real part and the imaginary part corresponding to the target frequency in the second complex spectrum is calculated.

[0056] Determine whether the first product and the second product are both greater than or equal to zero.

[0057] If not, the real part and the imaginary part corresponding to the target frequency in the first complex spectrum and the second complex spectrum are deleted.

[0058] That is, traverse k from 1 to For each target frequency in , the signal is clustered through the product relationship between the imaginary part and the real part to identify the components of a single signal source and delete the corresponding complex number from the spectrum.

[0059] Based on the revised two complex spectra and potential function, a potential function numerical graph is plotted, wherein the abscissa of the potential function numerical graph is the angle value, and the ordinate of the potential function numerical graph is the potential function value.

[0060] The potential function here can be defined as:

[0061]

[0062] Δθ (h,j) =θe h -θ j;

[0063]

[0064] h=1,2,…,180;

[0065] j=1,2,...,jmax;

[0066] where jmax is the number of frequency indices in the revised complex spectrum, θe h Take the direction of the traversed clustering plane as arrive interval θ j is the phase angle of the jth time-frequency point, λ is the scale parameter, which is used to adjust the width of the window function, r j is the weight coefficient of the jth time-frequency point, and φ(·) is the window function.

[0067] The potential function method is based on the characteristics of single-source point signal clustering, and analyzes the observed signal in a plane. In this observation plane, the signal shows directional clustering, and the cluster center in each direction corresponds to one of the columns of the mixing matrix, which also means that there is a source signal. The potential function traverses the direction of the clustering plane and outputs the clustering value of each direction. If a certain direction has obvious clustering characteristics, the output of the potential function will reach a local maximum. By analyzing the local maximum of the clustering function, an estimate of the number of sources can be obtained.

[0068] On the other hand, the maximum peak in the potential function image corresponds to the response when the phase of the signal is consistent with that of a certain element in the entire array of microphones. This consistency leads to the best synthesis effect of the signal on the array antenna, thus producing the maximum response value at a specific angle. By finding this maximum peak, the signal's arrival angle can be determined.

[0069] The potential function here is used to evaluate the directionality of the signal. By calculating the value of the potential function at different angles, the independent variable θe can be plotted. h and the strain Φ(θe h ) are the angle and potential function value respectively.

[0070] like Figure 2 As shown, it is a potential function numerical diagram provided by an embodiment of the present application. The horizontal axis of the figure is the angle, ranging from -90 to 90 degrees, indicating the direction or angle range considered in the signal analysis. This is usually related to the direction of the sound source or the direction of signal propagation. At the same time, the vertical axis is the weighted sum, which represents the weighted sum of the signal amplitude at each angle. This value can be interpreted as the "weighted potential value" or "directional strength" of the signal in a specific direction, reflecting the energy distribution of the signal in different directions.

[0071] The angle of arrival can be determined based on the angle corresponding to the wave crest.

[0072] Then, the number of peaks in the potential function numerical graph can be determined as the number of target signal sources, and the angle value corresponding to each peak can be determined as the wave arrival angle to form an array response matrix.

[0073] The formula for calculating the angle between the peak and the trough is:

[0074] C=[θe h ] if Φ(θe h-1 )<Φ(θe h )>Φ(θe h+1 ) h=1,2,...,180

[0075] T=[θe h ] if Φ(θe h-1 )>Φ(θe h )<Φ(θe h+1 ) h=1,2,...,180

[0076] The angle of the peak or trough is determined as follows:

[0077]

[0078] Where tend is the length of the peak set T, cend is the length of the trough set T, the length of the updated peak set T is tend′, and the length of the updated trough set C is cend′.

[0079] Self-set the threshold α for determining the maximum value:

[0080] Cr=[T(ti)]if α<min(Φ(T(ti))-Φ(C(ti)), Φ(T(ti))-Φ(C(ti+1)));

[0081] ti=1,2,...,tend′;

[0082] By calculating the peak and trough of the potential function and comparing whether the difference between the peak and the trough is greater than a threshold, it is determined whether the corresponding peak is the peak of the potential function of the signal source.

[0083] Get the angle θr corresponding to each maximum peak p ∈Cr p=1,2,...,m,m is the number of target signal sources, Cr is the length of the peak set that meets the threshold, and the peak set is used to calculate the MVDR array response matrix W.

[0084] Using the recorded angle of arrival θr p , and create an array response matrix W = [a(θr1 ), a(θr 2 ), ..., a(θr p )], which is obtained by Fourier transform and used in the MVDR beamforming algorithm to enhance the signal in a specific direction. The formula is:

[0085] a(θr p )=exp(-iπ(0,1,…,n-1)·sin(θr p )).

[0086] The array response matrix is ​​a key tool for describing the response characteristics of array antennas to signals. It describes in detail the amplitude and phase response of the antenna array to signals from different directions. This response characteristic is directly affected by the geometric structure of the array and the wavelength of the signal. The array response matrix is ​​used to predict and analyze the reception effect of the array on signals from a specific direction. The size and directivity of the array response matrix directly determine the performance of the array. Through the array response matrix, we can gain an in-depth understanding of how the array performs spatial filtering on signals from different directions, thereby achieving effective extraction and analysis of signals. In beamforming technology, the array response matrix plays a core role. Beamforming technology compensates for the propagation delay of each array element by selecting a suitable weighting vector, so that the array output can be superimposed in phase in a certain desired direction, and a smaller response is generated in other directions. This method can be used to perform beam scanning across the entire space to determine the direction of the signal to be detected. S3. Based on the observed acoustic matrix, the covariance matrix is ​​calculated.

[0087] Here, the covariance matrix R can be calculated as follows:

[0088] R=X′ T X′ / (N-1);

[0089] X′=[fx 1 ,fx 2 …, fx n ];

[0090] Among them, fx n is the observed acoustic signal x collected by acoustic sensor n n The transformation result after Fourier transform, N is x n The length value of .

[0091] S4. Revise the covariance matrix based on the number of target signal sources to generate an optimized covariance matrix.

[0092] In the field of mixed signal processing, the eigenvalue matrix contains the comprehensive statistical information of all components in the signal, where the eigenvalues ​​characterize the significance or energy intensity of each component, and the eigenvectors reveal the intrinsic structure or pattern of these components. The different components of the signal are mutually orthogonal in the feature space, indicating that they are statistically independent. By selecting eigenvectors equal to the number of signals, a signal subspace is constructed that is as close to the true structure of the original signal as possible, while effectively suppressing noise and other interference factors.

[0093] Selecting eigenvalues ​​and their corresponding eigenvectors equal to the number of input signals from the eigenvalue matrix is ​​essentially identifying the dominant components in the signal. These components are usually the parts of the signal that contain key information. By selecting these eigenvalues ​​and eigenvectors and recombining them, the signal is mapped to a new feature space composed of these significant components. This process can maximize the retention of key information in the original signal while minimizing the impact of noise and other non-main components.

[0094] Here the optimized covariance matrix can be generated as follows:

[0095] Based on the covariance matrix, a pseudo-inverse matrix is ​​determined.

[0096] The pseudo-inverse matrix is ​​decomposed into a first eigenvector matrix and a first eigenvalue matrix.

[0097] A preset number of eigenvalues ​​are selected from the first eigenvalue matrix to form a second eigenvalue matrix.

[0098] The eigenvectors at corresponding positions are selected from the first eigenvector matrix to form a second eigenvector matrix.

[0099] The optimized covariance matrix is ​​reconstructed and generated based on the second eigenvalue matrix and the second eigenvector matrix.

[0100] Specifically, the covariance matrix R can be subjected to eigenvalue decomposition to obtain the first eigenvector matrix V and the first eigenvalue matrix D. The mathematical expression of eigenvalue decomposition is:

[0101] VDV T =R.

[0102] The inverse matrix R of the covariance matrix -1 Calculate this and adjust it by adding a small identity matrix I, the formula is R -1 = I / R. However, since the covariance matrix R may be close to singular (i.e., not rank-full or the determinant is close to zero), direct inversion may be numerically unstable.

[0103] Select the largest m eigenvalues ​​from the first eigenvalue matrix D, which is the number of known signal sources calculated by the potential function, and use them as the second eigenvalue matrix D signal At the same time, the corresponding m eigenvectors are selected from the first eigenvector matrix V to form the second eigenvector matrix V signal .in,

[0104] D signal =[λ 1 ,λ 2 , ..., λ m ];

[0105] V signal =[v 1 , v 2 , .., v m ];

[0106] Among them, λ i is the i-th largest eigenvalue in D, v i is the corresponding eigenvector.

[0107] Using the signal subspace V signal and D signal To estimate the optimal covariance matrix R est :

[0108] R est =V signal D signal V signal T ;

[0109] Where D signal Scaling is done as the middle diagonal matrix.

[0110] S5. Obtain beamforming results based on the observed acoustic matrix, optimized covariance matrix and array response matrix.

[0111] The MVDR algorithm is an adaptive filtering technology widely used in the field of signal processing. Its core goal is to minimize the impact of noise or interference while keeping the target signal undistorted. This algorithm has important applications in many fields such as array signal processing, speech enhancement, noise suppression, and directional microphones. The core idea of ​​the MVDR algorithm is to adjust the weight of the receiving array so that the signal in the desired direction passes without distortion, while suppressing noise or interference in other directions.

[0112] Specifically, the MVDR algorithm constructs a covariance matrix that contains the characteristics of the signal and noise. The algorithm then calculates a weight vector that minimizes the variance of the array output while ensuring that the signal response in the target direction remains unchanged. In practical applications, the MVDR algorithm first needs to estimate the covariance matrix of the signal, which is usually achieved by collecting signal samples for a certain period of time. The algorithm then calculates the corresponding weight vector based on the incident angle of the target signal. This weight vector is used to process the received signal to enhance the target signal and suppress interference.

[0113] A key feature of the MVDR algorithm is that it can effectively reduce the noise level while maintaining the target signal without distortion. This is because the algorithm takes into account the directionality of the target signal when it is designed, so that it can effectively suppress noise in other directions without distorting the target signal. In addition, the MVDR algorithm also has a high resolution and can distinguish close signal sources. This is because the MVDR algorithm uses a smaller beam width, which improves the resolution.

[0114] The array response matrix W and the covariance matrix R after subspace analysis est Substituting into the MVDR weight vector formula, we can get the modified formula:

[0115] The array response matrix a(θ 0 ) and the covariance matrix R after subspace analysis est Substituting into the MVDR weight vector formula, we can get the modified formula:

[0116]

[0117] The MVDR beamforming weight vector wp is multiplied by the observed acoustic matrix X to obtain the beamforming output signal xout. p :

[0118] xoutt p =w p X.

[0119] like Figure 3 and Figure 4 As shown in FIG. 1 , there are output signals of two target signal sources formed by the aforementioned method.

[0120] The present application provides an MVDR beamforming method, which reduces the MVDR algorithm's reliance on prior information of the sound source direction by introducing a potential function. This adaptive adjustment method allows the algorithm to work under unknown or changing sound source directions, thereby broadening the scope of application of the algorithm and making it efficient even without accurate sound source localization information. The potential function allows the algorithm to dynamically adjust according to the characteristics of the current signal, improving the adaptability of the algorithm. This flexibility enables the algorithm to better respond to environmental changes and fluctuations in signal characteristics. The application of the potential function enhances the robustness of the algorithm, especially in the face of noise and interference. Even when the signal source direction information is inaccurate or partially missing, the algorithm can maintain stable performance and ensure reliability in complex environments. Through the optimization of the potential function, the MVDR algorithm can more accurately identify the target signal and suppress background noise and other interference, thereby improving the signal-to-noise ratio. This is crucial to improving the clarity and accuracy of signal extraction, especially in complex environments with high noise levels. In environments with multiple signal sources and noise interference, the application of the potential function helps to enhance the algorithm's anti-interference ability. Through optimization, the algorithm can more effectively suppress interference and improve the extraction quality of the target signal.

[0121] In addition, by combining the subspace analysis method, the algorithm can more accurately identify and extract the signal source. The subspace method helps the algorithm distinguish different signal sources by decomposing the covariance matrix of the signal, thereby improving the extraction accuracy. This method can improve the accuracy of signal processing in a multi-signal source environment.

[0122] The introduction of potential functions and the combination of subspace analysis enable the MVDR algorithm to adapt to a variety of signal environments, improving the versatility of the algorithm. This improvement in versatility allows the algorithm to be applied to a wider range of fields and scenarios, improving the practicality and flexibility of the algorithm. Through potential functions and subspace analysis, the decision-making process of the algorithm becomes clearer and the interpretability of the algorithm is enhanced. This not only helps debug and optimize the algorithm, but also makes the behavior of the algorithm more transparent and easier to understand and analyze.

[0123] like Figure 5 As shown, based on the same inventive concept, an MVDR beamforming device is also provided in an embodiment of the present application, and the device 50 includes:

[0124] An acquisition module 510 is used to acquire an observation acoustic matrix formed after a plurality of acoustic sensors receive a plurality of target signal sources, wherein the plurality of acoustic sensors are arranged in an array;

[0125] An analysis module 520 is used to draw a potential function numerical graph based on the observed acoustic matrix to determine the number of target signal sources and the angle of arrival corresponding to each target signal source;

[0126] A calculation module 530, for calculating a covariance matrix based on the observed acoustic matrix;

[0127] A revision module 540, configured to revise the covariance matrix based on the number of target signal sources to generate an optimized covariance matrix;

[0128] The forming module 550 is used to obtain a beamforming result based on the observed acoustic matrix, the optimized covariance matrix and the array response matrix.

[0129] In a preferred embodiment, the analysis module 520 is specifically used to perform Fourier transform on the observed acoustic signals collected by any two acoustic sensors to obtain the complex spectrum corresponding to each observed acoustic signal; based on the degree of sign matching between the real part and the imaginary part corresponding to each frequency in the two complex spectra, the two complex spectra are revised; based on the revised two complex spectra and the potential function, a potential function numerical graph is drawn, wherein the horizontal axis of the potential function numerical graph is the angle value, and the vertical axis of the potential function numerical graph is the potential function value.

[0130] In a preferred embodiment, the analysis module 520 determines the number of target signal sources in the following manner: determining the number of peaks in the potential function value graph as the number of target signal sources.

[0131] In a preferred embodiment, the analysis module 520 is further used to determine the angle value corresponding to each wave crest as the wave arrival angle; perform Fourier transform on the wave arrival angle to generate a response vector to form the array response matrix.

[0132] In a preferred embodiment, the calculation module 530 calculates the covariance matrix R in the following manner:

[0133] R=X′ T X′ / (N-1);

[0134] X′=[fx 1 ,fx 2 …, fx n ];

[0135] Among them, fx n is the observed acoustic signal x collected by acoustic sensor n n The transformation result after Fourier transform, N is x n The length value of .

[0136] In a preferred embodiment, the revision module 540 generates the optimized covariance matrix in the following manner:

[0137] Based on the covariance matrix, a pseudo-inverse matrix is ​​determined; the pseudo-inverse matrix is ​​decomposed into a first eigenvector matrix and a first eigenvalue matrix; a preset number of eigenvalues ​​are selected from the first eigenvalue matrix to form a second eigenvalue matrix; eigenvectors at corresponding positions are selected from the first eigenvector matrix to form a second eigenvector matrix; based on the second eigenvalue matrix and the second eigenvector matrix, an optimized covariance matrix is ​​reconstructed and generated.

[0138] In a preferred embodiment, the analysis module 520 revises the two complex spectra in the following manner: calculating a first product between the real part and the imaginary part corresponding to the target frequency in the first complex spectrum; calculating a second product between the real part and the imaginary part corresponding to the target frequency in the second complex spectrum; determining whether the first product and the second product are both greater than or equal to zero; if not, deleting the real part and the imaginary part corresponding to the target frequency in the first complex spectrum and the second complex spectrum.

[0139] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 6 As shown in , the electronic device 600 includes a processor 610 , a memory 620 and a bus 630 .

[0140] The memory 620 stores machine-readable instructions executable by the processor 610. When the electronic device 600 is running, the processor 610 communicates with the memory 620 through a bus 630. When the machine-readable instructions are executed by the processor 610, steps of an MVDR beamforming method in the above-mentioned method embodiment can be executed. The specific implementation method can be found in the method embodiment, which will not be repeated here.

[0141] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of an MVDR beamforming method in the above method embodiment can be executed. The specific implementation method can be found in the method embodiment and will not be repeated here.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0143] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0144] In addition, the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] Furthermore, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0146] It should be noted that if the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM) random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0147] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0148] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An MVDR beamforming method, characterized in that: The method comprises: Acquiring an observation acoustic matrix formed after a plurality of acoustic sensors receive a plurality of target signal sources, wherein the plurality of acoustic sensors are arranged in an array; Drawing a potential function numerical graph based on the observed acoustic matrix to determine the number of target signal sources and the angle of arrival corresponding to each target signal source; Based on the observed acoustic matrix, calculating a covariance matrix; Revising the covariance matrix based on the number of target signal sources to generate an optimized covariance matrix; A beamforming result is obtained based on the observed acoustic matrix, the optimized covariance matrix and the array response matrix.

2. The method according to claim 1, characterized in that: The step of drawing a potential function numerical graph based on the observed acoustic matrix specifically includes: Performing Fourier transform on the observed acoustic signals collected by any two acoustic sensors to obtain the complex spectrum corresponding to each observed acoustic signal; revising the two complex spectra based on the degree of sign matching between the real part and the imaginary part corresponding to each frequency in the two complex spectra; Based on the revised two complex frequency spectra and potential function, a potential function numerical graph is plotted, wherein the abscissa of the potential function numerical graph is the angle value, and the ordinate of the potential function numerical graph is the potential function value.

3. The method according to claim 1, characterized in that The number of target signal sources is determined in the following manner: the number of peaks in the potential function value graph is determined as the number of target signal sources.

4. The method according to claim 3, characterized in that Determine the angle value corresponding to each wave crest as the wave arrival angle; The arrival angle is Fourier transformed to generate a response vector to form the array response matrix.

5. The method according to claim 1, characterized in that The covariance matrix R is calculated as follows: R=X′ T X′ / (N-1); X′=[fx1,fx2...,fx n ]; Among them, fx n is the observed acoustic signal x collected by acoustic sensor n n The transformation result after Fourier transform, N is x n The length value of .

6. The method according to claim 1, characterized in that The optimized covariance matrix is ​​generated by: Based on the covariance matrix, determining a pseudo-inverse matrix; Decomposing the pseudo inverse matrix into a first eigenvector matrix and a first eigenvalue matrix; Selecting a preset number of eigenvalues ​​from the first eigenvalue matrix to form a second eigenvalue matrix; Selecting eigenvectors at corresponding positions from the first eigenvector matrix to form a second eigenvector matrix; The optimized covariance matrix is ​​reconstructed and generated based on the second eigenvalue matrix and the second eigenvector matrix.

7. The method according to claim 2, characterized in that The two complex spectra are revised in the following way: Calculating a first product between the real part and the imaginary part corresponding to the target frequency in the first complex spectrum; calculating a second product between the real part and the imaginary part corresponding to the target frequency in the second complex spectrum; determining whether the first product and the second product are both greater than or equal to zero; If not, the real part and the imaginary part corresponding to the target frequency in the first complex spectrum and the second complex spectrum are deleted.

8. An MVDR beamforming device, characterized in that: The device comprises: An acquisition module, used to acquire an observation acoustic matrix formed after a plurality of acoustic sensors receive a plurality of target signal sources, wherein the plurality of acoustic sensors are arranged in an array; An analysis module is used to draw a potential function numerical graph based on the observed acoustic matrix to determine the number of target signal sources and the angle of arrival corresponding to each target signal source; A calculation module, used for calculating a covariance matrix based on the observed acoustic matrix; A revision module, used for revising the covariance matrix based on the number of target signal sources to generate an optimized covariance matrix; A forming module is used to obtain a beamforming result based on the observed acoustic matrix, the optimized covariance matrix and the array response matrix.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the MVDR beamforming method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the MVDR beamforming method according to any one of claims 1 to 7 are executed.

Images