A feedback mechanism-based orthogonal matching pursuit sound source identification method

Abstract

The application discloses a kind of orthogonal matching pursuit sound source identification and positioning method based on feedback mechanism, measurement surface and focusing surface are set according to set distance interval, measurement surface and focusing surface are grid division, obtain microphone position and focusing point and Green function transfer matrix;The relationship between microphone measurement sound pressure vector and focusing point sound source intensity is obtained using Green function transfer matrix;The sound source intensity of focusing surface grid point is calculated by orthogonal matching pursuit algorithm based on source intensity priori guidance, and the obtained sound source intensity data is used for accurate identification and positioning of sound source.The application preliminarily selects atom by orthogonal matching pursuit algorithm based on source intensity prior information guidance, then the atom preliminarily selected is fed back to transfer matrix, and the atom closest to measurement sound pressure vector is obtained as the final selected atom, which improves the reconstruction performance of traditional algorithm at low frequency, and improves the sound source identification and positioning ability under the strong correlation between measurement matrix atoms.

Description

Technical Field

[0001] The present invention relates to the recognition and positioning of sound sources, and in particular to an orthogonal matching pursuit sound source recognition method based on a feedback mechanism. Background Art

[0002] In recent years, near-field acoustic holography and beamforming technologies have played an important role in acoustic engineering applications such as noise source identification and location, sound field separation, sound field visualization, and speech enhancement. These two technologies are not affected by the shape or size of the sound source and can easily establish mathematical models and construct transfer matrices. Near-field acoustic holography and beamforming technologies are mostly based on traditional acoustic arrays, such as planar uniform arrays, which collect data based on the Nyquist sampling framework. To ensure that the collected signal does not experience spectral aliasing, the sampling frequency must be greater than twice the highest frequency of the signal when using the above-mentioned regular arrays for sound field data collection. To achieve higher recognition accuracy, the above-mentioned traditional arrays require more microphones, resulting in high spatial complexity and measurement costs. At the same time, the large amount of measurement data increases the difficulty of storage and transmission.

[0003] Compressed sensing technology breaks through the Nyquist sampling theorem. By exploiting the sparsity of signals in the transform domain, it can preserve all the information of the original signal with relatively low sampling frequencies and achieve high-precision reconstruction of the original signal, significantly reducing the number of sampling sensors and lowering system measurement costs. In compressed sensing theory, the orthogonal matching pursuit algorithm, abbreviated as OMP, is the most commonly used reconstruction method for sparse original signals. The OMP algorithm offers advantages such as small imaging sidelobes, minimal artifacts, and fast computational speed. However, the OMP algorithm has shortcomings in practical applications. The OMP algorithm and various other greedy reconstruction algorithms for compressed sensing all expand the atomic support set based on the absolute inner product principle. This criterion measures atomic correlation by measuring the cosine similarity between matrix atoms. As the source frequency decreases, the phase of the Green's function changes more slowly with distance, resulting in a gradual increase in the correlation between atoms and a convergence between two columns of atoms. Furthermore, when the focal plane grid is too densely spaced, the spacing between the focal plane grid points corresponding to two adjacent atoms is very small, which can also lead to high correlation between the two columns of atoms. When the inner product between the atom corresponding to the actual sound source and the residual is smaller than the inner product between the adjacent atom and the residual, the absolute inner product matching criterion cannot expand the correct atom to the support set, inevitably leading to sound source identification failure. Furthermore, for coherent sound sources with low frequencies and close proximity, coherence peaks appear, causing the inner product between the atom corresponding to the coherence peak position in the measurement matrix and the signal to even exceed the inner product between the atom corresponding to the actual sound source and the original signal. This can cause the algorithm to incorrectly select the atom corresponding to the coherence peak position in the first iteration, resulting in sound source identification failure. Summary of the Invention

[0004] In order to avoid the shortcomings of the above-mentioned existing technologies, the present invention provides an orthogonal matching pursuit sound source identification method based on a feedback mechanism, aiming to improve the algorithm reconstruction performance in an environment with strong inter-atomic correlation, and avoid the problem of low sound source identification resolution caused by the increased correlation between adjacent atoms in the transfer matrix due to the encryption of the focal plane grid and the change of the signal phase at low frequency.

[0005] In order to achieve the above-mentioned object of the invention, the present invention adopts the following technical solutions:

[0006] The characteristics of the orthogonal matching pursuit sound source identification method based on the feedback mechanism of the present invention are:

[0007] Setting the measurement plane and the focusing plane at set distance intervals, meshing the measurement plane and the focusing plane, and obtaining the microphone position and the focusing point as well as the Green's function transfer matrix;

[0008] The Green's function transfer matrix is ​​used to obtain the relationship between the sound pressure vector measured by the microphone and the source intensity of the focused sound source;

[0009] The approximate solution of the source strength of the focal point sound source is obtained by calculating the least squares method. The approximate solution of the source strength of the focal point sound source is used as prior information to guide the screening of atoms. The position of the largest element is found as the rough sound source position. Finally, the precise sound source position is obtained by searching near the rough sound source position.

[0010] The orthogonal matching pursuit sound source identification method based on the feedback mechanism of the present invention is characterized by comprising the following steps:

[0011] Step 1: Determine the transfer matrix G between the sound pressure measured by the microphone in the measurement plane and the sound source intensity at the focal point in the focusing plane:

[0012] A measurement surface W is set in the sound field generated by K sound sources, M grid points are evenly distributed in the measurement surface W, and M microphones are arranged on the M grid points in a one-to-one correspondence to form a measurement array;

[0013] Evenly distribute N focusing points on a focusing plane T formed by a sound source calculation plane, and use the N focusing points as potential sound source positions;

[0014] The transfer matrix G between the source intensity of the focused sound source and the measurement surface microphone array is established based on the free-field Green's function:

[0015] Step 2: Use the transfer matrix G to establish the relationship between the source intensity of the focused sound source and the sound pressure measured by the microphone as shown in formula (1):

[0016] p=Gq+ζ (1)

[0017] In formula (1):

[0018] q represents the source intensity vector of the focused sound source; p represents the sound pressure vector measured by the microphone on the measurement surface;

[0019] Let ζ represent the noise vector contained in the sound pressure vector p measured by the microphone;

[0020] Step 3: The orthogonal matching pursuit algorithm based on the feedback mechanism solves the sound source intensity vector and performs sound source localization according to the following process:

[0021] Step 3.1, initialize the residual R0 = p, index set It is an empty set, and the number of iterations k = 1;

[0022] Step 3.2, perform singular value decomposition on the transfer matrix G, calculate the regularization parameter λ, and use the Tikhonov regularization method to solve the approximate solution F of the sound source intensity vector;

[0023] Step 3.3: Use the approximate solution F of the sound source intensity vector as prior information to guide the selection of atoms, and find the location of the largest element according to formula (2):

[0024] i k =arg max x＝1,2,...,N |F(x)| (2)

[0025] In formula (2):

[0026] i k The index of the position of the maximum element at the kth iteration;

[0027] F(x) is the source intensity at the x-th index position of the source intensity vector approximate solution;

[0028] arg max x＝1,2,...,N |F(x)| means finding the value of x that maximizes |F(x)| when x=1,2,...,N;

[0029] Step 3.4: Update the index set to Г k :G k =G k-1 ∪i k

[0030] Among them, Г k-1 is the index set of the optimal atom set that has been screened in the previous iteration;

[0031] Step 3.5: Use index set Г k Take the corresponding atomic composition matrix from the transfer matrix G The sound source intensity q at the corresponding position of the index set is calculated using the least squares method k for: Where, Indicates taking the transfer matrix Gk A new matrix composed of columns denotes the transpose of the matrix ;

[0032] The corresponding source strength vector where: denotes the value of the source strength vector at the position of Г k , and the values at the remaining positions of the source strength vector are all 0;

[0033] Step 3.6, update the residual R according to Equation (3) k as follows:

[0034]

[0035] Step 3.7, when the number of iterations k < K, increase the value of the number of iterations k by 1, return to Step 3.2, and continue the iteration; otherwise, stop the iteration and output the index Г of the maximum element position obtained in each iteration k and the rough solution q of the source strength at the corresponding position k , which is the sound source position and the corresponding source strength obtained by preliminary calculation;

[0036] Step 3.8, search for all combinations δ of the grid point coordinates near the preliminarily selected atoms:

[0037] The coordinates of the Jth sound source position preliminarily selected and the coordinates of all its adjacent grid points are the set O J , J = 1, 2,..., K;

[0038] Take one element from each of the sets O1, O2,..., O k to form a combination H i ;

[0039] δ is all combinations of the grid point coordinates near the preliminarily selected atoms, including all combination cases, that is:8]

[0040] δ = [H1, H2,..., H B

[0041] Denote the ith combination by H i , i = 1, 2,..., B, where B is the total number of all combinations;

[0042] Step 3.9, calculate the source strength q' at the coordinate position corresponding to each combination i , and the calculated sound pressure vector p i corresponding to each combination:

[0043] Calculate the source strength q' at the coordinate position corresponding to each combination using the least squares method according to Equation (4) i as follows: ​

[0044]

[0045] in:

[0046] q′ i represents the source intensity at the coordinate position corresponding to the i-th combination;

[0047] Indicates taking the H of the transfer matrix G i The new matrix consists of columns, Representation matrix The transpose of

[0048] The source intensity vector corresponding to each combination is calculated by formula (5):

[0049]

[0050] in:

[0051] Indicates the source intensity vector in H i The value at the position, the rest of the positions of the source intensity vector are 0;

[0052] The calculated sound pressure vector p corresponding to each combination is obtained by formula (6): i :

[0053]

[0054] Step 3.10: Use the maximum likelihood estimation method to calculate and obtain the final solution of the sound source intensity, wherein the final solution of the sound source intensity includes the sound source position coordinate Pos and the sound source intensity corresponding to the sound source position;

[0055] Step 3.11 identifies and locates the sound source according to the sound source position coordinates Pos.

[0056] The orthogonal matching pursuit sound source identification method based on the feedback mechanism of the present invention is also characterized by:

[0057] The final solution of the sound source intensity calculated in step 3.10 is: Using formula (7) and the maximum likelihood estimation method, each calculated sound pressure vector p is calculated. i The index i corresponding to the minimum value of the 2-norm of the difference between the measured sound pressure vector p min :

[0058] i min =arg min i＝1,2,...,B ||p i -p||2 (7)

[0059] in:

[0060] arg min i＝1,2,...,B||p i -p||2 means when i=1,2,...,B, find the value that makes ||p i -p||2The i value when the minimum value is obtained;

[0061] ||p i -p||2 means calculating the sound pressure vector p i The 2-norm of the difference from the measured sound pressure vector p;

[0062] The final calculated sound source position coordinate Pos is represented by formula (8):

[0063]

[0064] The magnitude of the sound source intensity corresponding to the sound source position is

[0065] The orthogonal matching pursuit sound source identification method based on the feedback mechanism of the present invention is also characterized by:

[0066] The transfer matrix G established in step 1 is represented by equation (9):

[0067]

[0068] The elements of the transfer matrix G represented by equation (9) are expressed as g(m,n), then g(m,n) is the transfer function between the m-th microphone and the n-th focus point, m = 1, 2, ..., M, n = 1, 2, ..., N;

[0069] And there are:

[0070]

[0071] Where: j is the imaginary unit, f is the frequency of the sound source;

[0072] d mn is the distance between the mth microphone and the nth focusing point.

[0073] The orthogonal matching pursuit sound source identification method based on the feedback mechanism of the present invention is also characterized by:

[0074] The source intensity vector q of the focused sound source is: q=[SS1, SS2, ..., SS N ], SS a represents the sound source intensity of the ath focus point among N focus points, a=1,2,…,N;

[0075] The sound pressure vector p measured by the microphone on the measuring surface is: p=[SP1, SP2, ..., SP M ], SP brepresents the measured sound pressure of the bth microphone among M microphones, b = 1, 2, …, M.

[0076] Traditional sound source identification algorithms can only assume that the sound source is on the grid points divided by the focal plane. However, in actual sound source identification, the probability of the sound source position being exactly on the grid points is very small, so traditional algorithms can only find the approximate position of the sound source. The method of the present invention effectively solves this problem. It searches for the position of non-grid points near the initial solution through the optimization algorithm to realize the location of the sound source that is separated from the grid point. At the same time, it is necessary to construct a new measurement matrix G to solve the sound source intensity at the corresponding position outside the grid point. Compared with the existing technology, the beneficial effects of the present invention are reflected in:

[0077] 1. The present invention is based on the orthogonal matching pursuit algorithm based on source strength prior, and improves the orthogonal matching pursuit algorithm to achieve accurate sound source localization.

[0078] 2. The present invention searches for grid points near the sound source position obtained by the orthogonal matching pursuit algorithm based on the source strength prior to form a set of approximate solutions. The source strength at the corresponding position is obtained by multiplying the pseudo-inverse of the matrix composed of the columns of the transfer matrix corresponding to the position with the sound pressure vector measured by each microphone, that is, the sound source strength of each approximate solution is obtained. The transfer matrix is ​​multiplied by the sound source strength of each approximate solution to obtain the sound pressure vector at the measurement point position of the feedback calculation. The second norm of the difference between each feedback sound pressure vector and the sound pressure vector measured by the actual microphone is calculated. This value reflects the degree of proximity between the feedback sound pressure vector and the actual sound pressure vector. Finally, the combination with the smallest second norm of the difference is selected as the optimal solution, and its sound source strength is the final solution. When the correlation between the atoms of the measurement matrix is ​​strong, the sound source position reconstructed by the algorithm guided by the source strength prior will have a certain deviation from the actual sound source position, and the calculated sound source position will be near the actual sound source position. The present invention further improves the reconstruction accuracy of the algorithm when the atomic correlation is strong by searching and optimizing at nearby positions, thereby widening the applicable frequency range of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a simplified flow chart of the method of the present invention;

[0080] Figure 2 Schematic diagram of a sound source identification model composed of a microphone array and a focusing plane in the present invention;

[0081] Figure 3 It is the measurement array position map;

[0082] Figure 4a This is the sound source localization effect diagram of the standard OMP algorithm when the sound source frequency is 300Hz;

[0083] Figure 4b This is a sound source localization effect diagram of the present invention when the sound source frequency is 300 Hz;

[0084] Figure 5a The sound source localization effect diagram of the standard OMP algorithm when the sound source frequency is 1700Hz;

[0085] Figure 5b This is a sound source localization effect diagram of the present invention when the sound source frequency is 1700 Hz;

[0086] Figure 6a This is the sound source localization effect diagram when the standard OMP algorithm is used for positioning with a grid spacing of 0.01m on the focal plane;

[0087] Figure 6b This is a sound source localization effect diagram with a grid spacing of 0.01m on the focusing surface after adopting the present invention;

[0088] Figure 7a This is the sound source localization effect diagram when the standard OMP algorithm is used;

[0089] Figure 7b This is a diagram showing the sound source localization effect after adopting the present invention. DETAILED DESCRIPTION

[0090] See also Figure 1 and Figure 2 In this embodiment, the orthogonal matching pursuit sound source identification method based on the feedback mechanism is performed in the following steps:

[0091] Step 1: Determine the transfer matrix G between the sound pressure measured by the microphone in the measurement plane and the sound source intensity at the focal point in the focusing plane:

[0092] A measurement surface W is set in the sound field generated by K sound sources. A rectangular area with a length of Lx and a width of Ly in the measurement surface W is divided into M grid points with U rows and V columns evenly distributed. M microphones are arranged on the M grid points in a one-to-one correspondence to form a measurement array.

[0093] On the focal plane T formed by the sound source calculation plane, a rectangular area with a length of Nx and a width of Ny in the focal plane T is divided into N focal points evenly distributed in X1 rows and Y1 columns, and the N focal points are used as potential sound source positions;

[0094] According to the free-field Green's function, the transfer matrix G between the source intensity of the focused sound source and the measurement surface microphone array is established as:

[0095]

[0096] Denote the elements of the transfer matrix G as g(m,n), where g(m,n) is the transfer function between the mth microphone and the nth focal point, with m = 1, 2, ..., M and n = 1, 2, ..., N. Furthermore, Where j is the imaginary unit, f is the frequency of the sound source, and d mn is the distance between the mth microphone and the nth focusing point, and c is the speed of sound. Since the number of grid points M on the measurement surface is much smaller than the number of grid points N on the focusing surface, the transfer matrix G is a matrix with much fewer rows than columns. Each column of the transfer matrix G is called an atom.

[0097] Step 2: Use the transfer matrix G to establish the relationship between the source intensity of the focused sound source and the sound pressure measured by the microphone as shown in formula (1):

[0098] p=Gq+ζ (1)

[0099] In formula (1):

[0100] The source intensity vector of the focused sound source is represented by q, q=[SS1,SS2,...,SS N ], SS a represents the sound source intensity of the ath focus point among N focus points, a=1,2,…,N;

[0101] Let p represent the sound pressure vector measured by the microphone on the measurement surface, p=[SP1,SP2,...,SP M ], SP b represents the measured sound pressure of the bth microphone among M microphones, b = 1, 2, ..., M;

[0102] Let ζ represent the noise vector contained in the sound pressure vector p measured by the microphone, and ζ has the same dimension as p.

[0103] Step 3: The orthogonal matching pursuit algorithm based on the feedback mechanism solves the sound source intensity vector and performs sound source localization according to the following process:

[0104] Step 3.1, initialize the residual R0 = p, index set It is an empty set, and the number of iterations k = 1;

[0105] Step 3.2, perform singular value decomposition on the transfer matrix G, calculate the regularization parameter λ, and use the Tikhonov regularization method to solve the approximate solution F of the sound source intensity vector;

[0106] Step 3.3: Use the approximate solution F of the sound source intensity vector as prior information to guide the selection of atoms and find the location of the largest element according to formula (2):

[0107] i k =arg max x＝1,2,...,N |F(x)| (2)

[0108] In formula (2):

[0109] i kThe index of the position where the maximum element is located at the k-th iteration;

[0110] F(x) is the source strength at the x-th index position of the approximate solution of the source strength vector;

[0111] |F(x)| represents taking the absolute value of F(x);

[0112] arg max x＝1,2,...,N |F(x)| means that when x = 1, 2,..., N, find the x value that makes |F(x)| reach the maximum;

[0113] Step 3.4: Add the newly selected atom to the index set and update the index set to Г k : Г k = Г k-1 ∪i k

[0114] where Г k-1 is the index set of the optimal atom set that has been screened in the previous iteration;

[0115] Step 3.5: Take out the corresponding atoms from the transfer matrix G through the index set Г k to form a matrix Calculate the sound source strength q at the corresponding positions of the index set by using the least squares method k as: In the formula represents taking the Г k columns of the transfer matrix G to form a new matrix, represents the matrix transpose;

[0116] corresponding source strength vector where: represents the value of the source strength vector at the Г k position, and the values at the other positions of the source strength vector are all 0;

[0117] Step 3.6: Update the residual R according to Equation (3) k as:

[0118]

[0119] Step 3.7: When the iteration number k < K, increase the value of the iteration number k by 1, return to Step 3.2, and continue the iteration; otherwise, stop the iteration and output the index Г of the maximum element position obtained in each iteration k and the rough solution q of the source strength at the corresponding positions k , which are the initially calculated sound source position and the corresponding source strength;

[0120] Step 3.8: Search all combinations δ of the grid point coordinates near the initially selected atoms:

[0121] The coordinates of the J-th sound source position and all its adjacent grid points are initially selected as the set O J , J=1,2,...,K;

[0122] From O1, O2, ..., O K Take one element from each set to form a combination H i , H i Contains K different elements, each element is the index of a grid point;

[0123] δ is all combinations of the coordinates of the grid points near the atoms that are initially selected, including all combinations, namely:

[0124] δ=[H1,H2,...,H B ]

[0125] H i represents the i-th combination, i = 1, 2, ..., B, B is the total number of all combinations, equal to the product of the number of elements in each set;

[0126] Step 3.9: Calculate the source intensity q at the corresponding coordinate position of each combination i ′, and the calculated sound pressure vector p corresponding to each combination i :

[0127] According to formula (4), the source intensity q′ at the coordinate position corresponding to each combination is calculated using the least squares method. i :

[0128]

[0129] in:

[0130] q′ i represents the source intensity at the coordinate position corresponding to the i-th combination;

[0131] Indicates taking the H of the transfer matrix G i The new matrix consists of columns, Representation matrix The transpose of

[0132] Express request The inverse matrix of

[0133] The source intensity vector corresponding to each combination is calculated by formula (5):

[0134]

[0135] in:

[0136] Indicates the source intensity vector in H i The value at the position, the rest of the positions of the source intensity vector are 0;

[0137] The calculated sound pressure vector p corresponding to each combination is obtained by formula (6): i :

[0138]

[0139] Step 3.10: Use the maximum likelihood estimation method to calculate the final solution of the sound source strength:

[0140] The calculated sound pressure of each combination of grid points near the rough solution has been obtained. The actual sound source position is approximated by calculating the calculated sound pressure closest to the measured sound pressure, and the degree of proximity of the two vectors is measured using the maximum likelihood estimation method. Using formula (7) and the maximum likelihood estimation method, each calculated sound pressure vector p is obtained. i The index i corresponding to the minimum value of the 2-norm of the difference between the measured sound pressure vector p min :

[0141] i min =arg min i＝1,2,...,B ||p i -p||2 (7)

[0142] in:

[0143] arg min i＝1,2,...,B ||p i -p||2 means when i=1,2,...,B, find the value that makes ||p i -p||2The i value when the minimum value is obtained;

[0144] ||p i -p||2 means calculating the sound pressure vector p i The 2-norm of the difference from the measured sound pressure vector p;

[0145] The final calculated sound source position coordinate Pos is represented by formula (8):

[0146]

[0147] Pos contains K different elements, each element is the index of a grid point. The sound source strength corresponding to the sound source position is There are also K elements, each of which corresponds one-to-one to an element in Pos, indicating the magnitude of the sound source at the position of the element in Pos;

[0148] 3.11 The sound source is identified and located using the final calculated sound source position coordinates Pos. The final determined sound source position is the grid point corresponding to the element in Pos.

[0149] Simulation Example 1: Verify that compared with the standard OMP algorithm positioning method, the present invention has better reconstruction performance for low-frequency sparse sound source signals.

[0150] Simulation process: Assume that two monopole point sound sources of equal intensity are located on the focusing plane T. The size of the focusing plane T is 1m×1m, and a total of 21×21 grid points are evenly divided into 441 focusing positions. The coordinate positions of the monopole sound sources in space are (-0.15, 0, 0.2)m and (0.15, 0, 0.2)m. The measurement array is located on the grid points of the measurement plane W. The measurement plane is divided into 11×11 grid points, and the 49 measurement array elements used in the measurement are distributed on the grid points, as shown in the figure. Figure 3 As shown. The distance between the measurement plane W and the focusing plane T is 0.2m. To simulate the noise conditions in the actual measurement process, 25dB Gaussian white noise is added to the acoustic signal measured by the array. In order to quantitatively measure the reconstruction effect of the standard OMP algorithm and the improved OMP algorithm, the error definition method is given and represented by the following formula:

[0151]

[0152] Where: err is the source strength reconstruction error, q bul is the source intensity reconstruction vector, q sou is the standard source intensity position vector, and ||·||2 is the vector 2-norm.

[0153] Figure 4a This is the sound source localization effect diagram of the standard OMP algorithm when the sound source frequency is 300Hz. Figure 4b The figure below is a sound source localization effect diagram of the present invention when the sound source frequency is 300 Hz. In the image of the sound source recognition result, the "*" indicates the actual location of the sound source point.

[0154] When the sound source frequency is 300 Hz, the atoms corresponding to the sound source location have strong coherence with their neighboring atoms. The traditional OMP algorithm, influenced by the coherence peak, selects the wrong atom, resulting in a sound source identification error of 1.556 and failure. However, the improved OMP algorithm of the present invention accurately locates the sound source with an error of 0.034. This demonstrates that the method of the present invention outperforms the standard OMP algorithm in sound source identification at low and medium frequencies.

[0155] In order to compare the sound source identification performance of the standard OMP algorithm and the improved OMP algorithm at high frequencies, the sound source frequency was changed to 1700 Hz and the spatial positions of the two monopole sound sources were changed to (-0.1, 0, 0.2) m and (0.1, 0, 0.2) m, while keeping other parameters unchanged. Figure 5a The sound source localization effect diagram of the standard OMP algorithm, with a sound source identification error of 0.083; Figure 5b This is a sound source localization effect diagram of the present invention when the sound source frequency is 1700 Hz, and the sound source recognition error is 0.010; it can be seen that the sound source recognition effect of the present invention at high frequencies is slightly better than that of the standard OMP algorithm.

[0156] Simulation Example 2: Verify that the sound source localization resolution of the method of the present invention is higher than that of the standard OMP algorithm.

[0157] Simulation process: The denser the focus plane is divided, the higher the required sound source identification resolution. To compare the sound source localization resolution of the proposed method and the standard OMP algorithm, the focus plane was divided into a uniform 101*101 grid with a spacing of 0.01m between adjacent grids. The sound source frequency was set to 2000Hz, and the coordinate positions of the two monopole sound sources were set to (-0.05, 0, 0.2)m and (0.05, 0, 0.2)m, respectively. The remaining simulation conditions were the same as in Simulation 1. Figure 6a This is the sound source localization effect diagram when using the standard OMP algorithm. Figure 6b The following figure shows the sound source localization effect after using the present invention. When the spacing between adjacent grids on the focal plane is 0.01m, the reduced spacing between focal points leads to increased coherence between atoms in the corresponding transfer matrix. The standard OMP algorithm selects the atom at the coherence peak and generates a false sound source, with a sound source identification error of 1.342. However, the improved OMP algorithm using the present invention can still accurately locate the sound source, with a sound source identification error of 0.024. This shows that the present method significantly outperforms the standard OMP algorithm in sound source localization resolution.

[0158] Simulation Example 3: Verifying that the present invention can better identify sound source location information than traditional sound source localization methods in an environment with four adjacent sound sources. Simulation process: The four monopole sound sources are located at the spatial coordinates of (-0.1, 0, 0.2) m, (0.1, 0, 0.2) m, (0, 0.1, 0.2) m, and (0, -0.1, 0.2) m. The sound source frequency is 1500 Hz, and all other parameters are the same as in Simulation 1. Figure 7a This is the sound source localization effect diagram when using the standard OMP algorithm. Figure 7bThis is a diagram showing the effect of sound source localization after adopting the present invention. It can be clearly seen from the figure that due to the close distance between the four sound sources and the strong coherence between the atoms, the traditional OMP algorithm is difficult to accurately distinguish the positions of each sound source in such a high coherence environment. Traditional algorithms often have positioning errors when processing multiple sound sources that are close to each other, and the positioning accuracy is significantly reduced. However, after adopting the improved method proposed in this article, the position of each sound source can still be accurately identified even under conditions of strong coherence. This shows that the method of the present invention has better robustness and anti-interference ability, and can effectively overcome the limitations of traditional algorithms when faced with complex sound source configurations. Therefore, the method of this article not only surpasses the traditional sound source localization algorithm in accuracy, but also provides stronger adaptability to application scenarios, especially when processing multiple sound sources that are close to each other and have large signal interference, it can significantly improve the reliability and accuracy of sound source localization.

Claims

1. A method for identifying sound sources by orthogonal matching pursuit based on a feedback mechanism, characterized by: The measurement surface and the focusing surface are set at a set distance interval, and the measurement surface and the focusing surface are grid-divided to obtain the microphone position and the focusing point as well as the Green's function transfer matrix; the relationship between the microphone measurement sound pressure vector and the source intensity of the focusing point sound source is obtained by using the Green's function transfer matrix; the approximate solution of the source intensity of the focusing point sound source is obtained by the least squares method, and the approximate solution of the source intensity of the focal point sound source is used as a priori information to guide the screening of atoms, and the position of the largest element is found as the rough sound source position, and finally the precise sound source position is obtained by searching near the rough sound source position; the sound source obtained by the orthogonal matching pursuit algorithm based on the source intensity prior is obtained. A search is performed on the grid points near the position to form a set of approximate solutions. The source strength at the corresponding position is obtained by multiplying the pseudo-inverse of the matrix composed of the columns of the transfer matrix corresponding to the position with the sound pressure vector measured by each microphone, and the sound source strength of each approximate solution is obtained; the transfer matrix is ​​multiplied by the sound source strength of each approximate solution to obtain the sound pressure vector at the measurement point position of the feedback calculation, and the dinorm of the difference between each feedback sound pressure vector and the sound pressure vector measured by the actual microphone is calculated. This value reflects the degree of proximity between the feedback sound pressure vector and the actual sound pressure vector. Finally, the combination with the smallest dinorm of the difference is selected as the optimal solution, and its sound source strength is the final solution.

2. The orthogonal matching pursuit sound source identification method based on feedback mechanism according to claim 1 is characterized in that The steps include: Step 1: Determine the transfer matrix G between the sound pressure measured by the microphone in the measurement plane and the sound source intensity at the focal point in the focusing plane: A measurement surface W is set in the sound field generated by K sound sources, M grid points are evenly distributed in the measurement surface W, and M microphones are arranged on the M grid points in a one-to-one correspondence to form a measurement array; Evenly distribute N focusing points on a focusing plane T formed by a sound source calculation plane, and use the N focusing points as potential sound source positions; The transfer matrix G between the source intensity of the focused sound source and the measurement surface microphone array is established based on the free-field Green's function: Step 2: Use the transfer matrix G to establish the relationship between the source intensity of the focused sound source and the sound pressure measured by the microphone as shown in formula (1): p=Gq+ζ (1) In formula (1): q represents the source intensity vector of the focused sound source; p represents the sound pressure vector measured by the microphone on the measurement surface; Let ζ represent the noise vector contained in the sound pressure vector p measured by the microphone; Step 3: The orthogonal matching pursuit algorithm based on the feedback mechanism solves the sound source intensity vector and performs sound source localization according to the following process: Step 3.1, initialize the residual R0 = p, index set It is an empty set, and the number of iterations k = 1; Step 3.2, perform singular value decomposition on the transfer matrix G, calculate the regularization parameter λ, and use the Tikhonov regularization method to solve the approximate solution F of the sound source intensity vector; Step 3.3: Use the approximate solution F of the sound source intensity vector as prior information to guide the selection of atoms, and find the location of the largest element according to formula (2): i k =arg max x＝1,2,...,N |F(x)| (2) In formula (2): i k The index of the position of the maximum element at the kth iteration; F(x) is the source intensity at the x-th index position of the source intensity vector approximate solution; arg max x＝1,2,...,N |F(x)| means finding the value of x that maximizes |F(x)| when x=1,2,...,N; Step 3.4: Update the index set to Г k :G k =G k-1 ∪i k Among them, Г k-1 is the index set of the optimal atom set that has been screened in the previous iteration; Step 3.5: Use index set Г k Take the corresponding atomic composition matrix from the transfer matrix G The sound source intensity q at the corresponding position of the index set is calculated using the least squares method k for: Where, Indicates taking the transfer matrix G k The new matrix consists of columns, Representation matrix The transpose of The corresponding source intensity vector in: Indicates the source intensity vector in Г k The value at the position, the rest of the positions of the source intensity vector are 0; Step 3.6: Update the residual R according to formula (3) k for: Step 3.7: When the iteration number k < K, increase the value of the iteration number k by 1, return to Step 3.2, and continue the iteration; otherwise, stop the iteration and output the index Г of the maximum element position obtained in each iteration k and the rough solution q of the source strength at the corresponding position k , which are the sound source position and the corresponding source strength obtained by preliminary calculation; Step 3.8: Search for all combinations of grid point coordinates near the initially selected atoms δ: The coordinates of the J-th sound source position and all its adjacent grid points are initially selected as the set O J , J=1,2,...,K; From O1, O2, ..., O K Take one element from each set to form a combination H i ; δ is all combinations of the coordinates of the grid points near the atoms that are initially selected, including all combinations, namely: δ=[H1,H2,...,H B ] H i represents the i-th combination, i = 1, 2, ..., B, B is the total number of all combinations; Step 3.9: Calculate the source intensity q′ at the corresponding coordinate position of each combination i , and the calculated sound pressure vector p corresponding to each combination i : According to formula (4), the source intensity q′ at the coordinate position corresponding to each combination is calculated using the least squares method. i : in: q i ′ represents the source intensity at the coordinate position corresponding to the i-th combination; Indicates taking the H of the transfer matrix G i The new matrix consists of columns, Representation matrix The transpose of The source intensity vector corresponding to each combination is calculated by formula (5): in: Indicates the source intensity vector in H i The value at the position, the rest of the positions of the source intensity vector are 0; The calculated sound pressure vector p corresponding to each combination is obtained by formula (6): i : Step 3.10: Use the maximum likelihood estimation method to calculate and obtain the final solution of the sound source intensity, wherein the final solution of the sound source intensity includes the sound source position coordinate Pos and the sound source intensity corresponding to the sound source position; Step 3.11 identifies and locates the sound source according to the sound source position coordinates Pos.

3. The orthogonal matching pursuit sound source identification method based on feedback mechanism according to claim 2, characterized in that: The final solution of the sound source intensity calculated in step 3.10 is: Using formula (7) and the maximum likelihood estimation method, each calculated sound pressure vector p is calculated. i The index i corresponding to the minimum value of the 2-norm of the difference between the measured sound pressure vector p min : in min =my anger i＝1,2,...,B ||p i -p||2 (7) in: arg min i＝1,2,...,B ||p i -p||2 means when i=1,2,...,B, find the value that makes ||p i -p||2The i value when the minimum value is obtained; ||p i -p||2 means calculating the sound pressure vector p i The 2-norm of the difference from the measured sound pressure vector p; The final calculated sound source position coordinate Pos is represented by formula (8): Pos=H imin (8) The magnitude of the sound source intensity corresponding to the sound source position is 4. The orthogonal matching pursuit sound source identification method based on feedback mechanism according to claim 2, characterized in that: The transfer matrix G established in step 1 is represented by equation (9): The elements of the transfer matrix G represented by equation (9) are expressed as g(m,n), then g(m,n) is the transfer function between the m-th microphone and the n-th focus point, m = 1, 2, ..., M, n = 1, 2, ..., N; And there are: Where: j is the imaginary unit, f is the frequency of the sound source; d mn is the distance between the mth microphone and the nth focusing point.

5. The orthogonal matching pursuit sound source identification method based on feedback mechanism according to claim 2, characterized in that: The source intensity vector q of the focused sound source is: q=[SS1, SS2, ..., SS N ], SS a represents the sound source intensity of the ath focus point among N focus points, a=1,2,…,N; The sound pressure vector p measured by the microphone on the measuring surface is: p=[SP1, SP2, ..., SP M ], SP b represents the measured sound pressure of the bth microphone among M microphones, b = 1, 2, …, M.

Images