A target depth partitioning method based on angle of arrival and target depth mapping

Abstract

The application relates to a target depth division method based on an angle of arrival and target depth mapping, which can convert estimated angle information into target depth information. In a reliable sound path range in a deep sea, there is a vertical linear array, and a sound source is located in a near sea surface area. The actual effective peak value obtained by analyzing the angle of arrival estimation is analyzed, and then the depth information of the sound source target is obtained, that is, the depth of the sound source target is less than a depth judgment critical value or the depth of the sound source target is greater than the depth judgment critical value, so that the target depth division is realized. The application can calculate the depth judgment critical value through simulation, and in addition, the parameter information corresponding to different sound source target positions in a known sea area can be calculated in advance, so that the depth division of the sound source target can be quickly carried out, the calculation efficiency can be improved, and engineering practice is easy.

Description

Technical Field

[0001] This invention falls under the category of underwater acoustic signal processing, specifically marine engineering, underwater acoustic engineering, array signal processing, and sonar technology. It relates to a target depth segmentation method based on an angle-of-arrival and target depth mapping, applicable to the rapid differentiation of target depth regions using a vertical linear array under reliable acoustic paths in the deep sea. Background Technology

[0002] Depth determination of target sound sources is a key and challenging aspect of underwater acoustic signal processing, holding significant value in applications such as target identification, situational awareness, and underwater early warning. Under reliable acoustic path conditions in the deep sea, for receiving arrays near the seabed, acoustic signals radiated from near-surface sound sources primarily propagate to the receiving position via direct waves and reflected waves from the sea surface, forming a multipath sound field with a specific vertical arrival structure. Vertical linear arrays possess spatial sampling capabilities in the vertical direction, thus effectively sensing differences in the acoustic arrival structure of sound sources at different depths. Therefore, by performing a series of processing steps on the target radiation signals received by the vertical linear array, target depth information can be obtained, enabling target depth segmentation.

[0003] To achieve depth segmentation of underwater targets, there are currently two main types of methods. One is direct depth estimation, such as the matched-field processing method proposed by Bucker in 1976 [Homer P. Bucker, Use of calculated soundfields and matched-field detection to locate sound sources in shallow water. The Journal of the Acoustical Society of America, 1976, 59(2): 368-373.]. This method first uses a sound field propagation model to calculate a copy field vector, and then matches the copy field with the measured array data to invert the depth information of the sound source. In 2013, Zurk proposed a modified Fourier transform method [Reid McCargar, Lisa M. Zurk, Depth-based signal separation with vertical line arrays in the deepocean. The Journal of the Acoustical Society of America, 2013, 133(4): 320–325.]. This method is based on the Lloyd mirror theory and performs a modified Fourier transform on the beam output intensity sequence. The peak value of the transform output directly corresponds to the depth of the sound source. Another type is the binary discrimination method, such as the underwater sound source depth discrimination method proposed by Liu Zhitao using warping transform [Liu Zhitao, Guo Lianghao, Yan Chao. Depth discrimination of shallow sea sound sources using warping transform of autocorrelation function. Acta Acustica, 2019, 44(1): 28-38.]. This method uses warping transform to extract the normal mode correlation term in the autocorrelation function and judges the sound source depth by the proportion of the second characteristic frequency of the correlation term.

[0004] In practical applications, if the aperture of a vertical linear array deployed near the seabed is small, conventional beamforming (CBF) methods cannot effectively separate direct waves from sea surface reflected waves for most underwater targets of interest. Therefore, it is difficult to use the vertical arrival structure for depth classification. If a high-resolution method exists that can effectively distinguish between the two, angular domain information can be mapped to the depth domain, thereby achieving classification. Against this backdrop, a target depth classification method based on the mapping of angle of arrival and target depth is proposed, which can achieve efficient and robust underwater sound source depth classification. Summary of the Invention

[0005] Technical problems to be solved To overcome the shortcomings of existing technologies and achieve computationally efficient and robust sound source depth segmentation, this invention proposes a target depth segmentation method based on the mapping between angle of arrival (AHA) and target depth. This method utilizes a vertical linear array deployed in a reliable acoustic path region in the deep sea to receive signals. Then, it cascades two AHA estimation algorithms with different resolution capabilities: the CBF method and the Off-Grid Sparse Bayesian Inference (OGSBI) method. The sound source depth is segmented based on the number of peaks in their output spatial spectra. This invention features fast convergence, strong real-time performance, and high robustness. Furthermore, the critical depth value can be calibrated based on actual marine environmental parameters, demonstrating good environmental adaptability.

[0006] Technical solution A target depth segmentation method based on the mapping of angle of arrival and target depth is characterized by: deploying a vertical line array on the seabed to receive single-frequency signals emitted by sound sources near the sea surface; the vertical line array consists of... M It consists of array elements, with an element spacing of [missing information]. d Array center depth z r The steps for sound source depth segmentation are as follows: Step 1: Using the CBF method, the angle of arrival (AHA) is estimated by receiving the acoustic signal radiated by the underwater sound source using a vertical linear array, and the CBF normalized AHA estimate spectrum is obtained. P dB_CBF and the estimated angle of arrival of the signal i CBF_esti ; Step 2: Use the estimated angle of arrival obtained in Step 1. i CBF_esti The sound velocity profile of the vertical linear array deployment area, seabed parameters, and depth location information of each array element are input into the Bellhop ray acoustic model to obtain the critical depth corresponding to the CBF method under ideal conditions. D CBF and the critical depth corresponding to the OGSBI method D OGSBI With two critical depths D CBF and D OGSBI As a criterion for determining the depth of the target; Step 3: Use OGSBI to estimate the angle of arrival and obtain the number of valid peaks: Sub-step 1: Based on the estimated angle of arrival from step 1 i CBF_esti The search range for the planned angle is determined, and OGSBI is used for angle of arrival estimation to obtain the OGSBI normalized angle of arrival estimation spectrum. PdB_OGSBI ; Sub-step two: For P dB_OGSBI Perform effective peak statistics to obtain P dB_OGSBI Number of valid peaks in N OGSBI_peaks ; right P dB_OGSBI ( i Peak detection is performed, and the statistical amplitude exceeds a preset threshold. or Number of effective peak values N OGSBI_peaks ; Step 4: Target Depth Division: Based on the number of valid peaks N OGSBI_peaks Perform depth interval partitioning: If N OGSBI_peaks If the value is greater than 1, then the depth of the sound source is determined to satisfy the condition. D CBF > Z s > D OGSBI This indicates that the depth of the sound source is greater than the OGSBI depth threshold. D OGSBI However, it is less than the CBF depth critical value. D CBF ;like N OGSBI_peaks = 1, then the depth of the sound source is determined to satisfy Z s < D OGSBI This indicates that the depth of the sound source is less than the OGSBI depth threshold. D OGSBI .

[0007] The specific steps of step 1 are as follows: The output power of the beamformer in the CBF method is in: Indicates the output power of the beamformer. R y This represents the covariance matrix of the received signal from the array. w ( i ) represents the beam weighting vector, and H represents the transpose sign; The output power is normalized to decibels as follows: ; The angle at which the beam output power reaches its maximum value is the estimated angle of arrival. i CBF_esti : .

[0008] The R y Through formula The calculation yields y( f ) indicates that the frequency of the sound source is f The vertical linear array receives acoustic signals in the frequency domain, where H represents the transpose symbol.

[0009] The w ( i ) through formula The calculation yielded that, c The reference sound speed is defined as the sound speed at the receiving depth of the vertical linear array. M Indicates the number of array elements. d The interval between array elements.

[0010] Two critical depths are obtained in step 2. D CBF and D OGSBI The specific steps are as follows: Within the preset distance-depth two-dimensional search space, traverse the locations of candidate sound sources ( r k , z k Run BELLHOP; for each candidate location, extract the eigenangles of arrival of the direct wave and the first reflected wave. i D and i SR Candidate positions that satisfy the following formula are selected to form a set of angle-matching positions. S :

[0011] In the set S Among them, the distance that appears most frequently is selected as the reference distance. r ref ; Along the reference distance r ref Constructing depth scan sequences Z sim = { z min , z min + z , …, z max},in: z Step size; For each depth z iThe BELLHOP model was used to generate simulated array complex sound pressure data containing direct wave and sea surface reflected wave components, and Gaussian white noise with a preset signal-to-noise ratio was superimposed to form the simulation data. The angle of arrival was estimated using CBF and OGSBI on the simulation data, and the number of effective peaks in the spatial spectrum at each depth was counted. Record N CBF_peaks_ideal ( z i )and N OGSBI_peaks_ideal ( z i As depth changes, N CBF_peaks_ideal The minimum depth required for a stable transition from 1 to 2 is defined as the critical depth. D CBF ,Will N OGSBI_peaks_ideal The minimum depth required for a stable transition from 1 to 2 is defined as the critical depth. D OGSBI ; Effective peak value refers to the amplitude in the spatial spectrum that exceeds a preset threshold. or Local maxima: ; in n = 1, 2, …, N, N This represents the number of angle search grids in the angle of arrival estimation process.

[0012] The step size z Set according to the required depth resolution. Sub-step one of step 3 obtains the OGSBI normalized angle of arrival estimation spectrum. P dB_OGSBI The specific steps are as follows: (1) The estimated angle of arrival obtained from CBFOGSBI estimation i CBF_esti Plan the search range by angle and construct the array manifold: ① In [ i CBF_esti – i ex , i CBF_esti + i ex Within the range of spacing r generate N grid points { i g1 , i g2 , …, igN}; ② Calculate the guiding vector corresponding to each grid point. a ( i gn ) and its first derivative b ( i gn ) = a `( i gn ), forming a matrix A = [ a ( i g1 ), a ( i g2 ), …, a ( i gN )], B = [ b ( i g1 ), b ( i g2 ), …, b ( i gN )]; (2) SVD dimensionality reduction The array received signal y mentioned above is used as the received data matrix here. Y , ① For the received data matrix Y Perform singular value decomposition: Y=USV H ; ② Then extract before K The right singular value constitutes V 1; ③ Calculate the dimensionality-reduced data matrix: Y SV = YV 1; ④ Data normalization after dimensionality reduction: and the number of quick photos T Replace with K ; in all the following steps y ( t (All refer to) Y The t List, t = 1, …, T ; (3) Hyperparameter initialization ① Set constants: r , c , d Maximum number of iterations iter max Convergence threshold t ; ② Initial value of noise variance:

[0013] ③ Initial values ​​of signal hyperparameters:

[0014] ④ Initial value of deviation from standard: β n , n = 1, …, N ; ⑤ Constructing the matrix: Λ -1 = diag( α 1 -1 , …, α N -1 ); (4) Iterative update Repeat the following sub-steps until the convergence condition is met or the maximum number of iterations is reached; ① Construct the lattice array manifold matrix

[0015] ② Calculate the posterior covariance matrix of the signal.

[0016] ③ Calculate the posterior comean matrix of the signal.

[0017] And make , ,remember U n for U The n ; ④ Update hyperparameters α n and Λ -1

[0018]

[0019] And update Λ -1 = diag(( α 1 new ) -1 , …, ( α N new ) -1 ); ⑤ Update deviation Only for α new The largest K Update the grid points corresponding to each component. For the updated grid points, extract the corresponding columns of A and B. β and Extract the corresponding line. By analyzing the corresponding columns and rows, we obtain the reduced-dimensional matrices, and then calculate:

[0020]

[0021] Then update the off-grid deviation:

[0022] ⑥ Update noise variance

[0023] ⑦ Convergence judgment

[0024] like e < t Stop iteration; otherwise, let α old = α new Return to step ① and continue iterating to update the parameters; (5) Output the estimation results After the iteration is complete, a relative power spectrum can be constructed to find the spectral peaks.

[0025] like K The grid index value corresponding to each peak is g k Then the first k The estimation results of the incident signals are as follows: .

[0026] In practical applications, the output power is often normalized to decibels, i.e. .

[0027] The spacing r The angle search grid spacing can be set according to the required angle resolution accuracy.

[0028] The K Among the right singular values K Pick M 1 or the number of sound sources is known; for the single-target case, take... K = 1.

[0029] A computer program product, characterized in that it includes computer-executable instructions, which, when executed, are used to implement the target depth division method based on the mapping of angle of arrival and target depth as described in claims 1 to 9.

[0030] Beneficial effects This invention proposes a target depth segmentation method based on the mapping between angle of arrival and target depth, which can convert estimated angle information into target depth information. Within the reliable acoustic path range of the deep sea, there is a vertical linear array with a sound source located in the near-surface region. By analyzing the actual effective peak count obtained from the estimated angle of arrival, the depth information of the sound source target is obtained, i.e., whether the sound source target depth is less than or greater than the depth determination threshold, thus achieving target depth segmentation. This invention can calculate the depth determination threshold through simulation. Furthermore, it can pre-calculate the parameter information corresponding to different sound source target locations in known sea areas to facilitate rapid sound source target depth segmentation, improving computational efficiency and facilitating engineering practice. Attached Figure Description

[0031] Figure 1 Munk sound velocity profile used in simulation Figure 2 Scene side view Figure 3 Algorithm Flowchart Figure 4 CBF estimation results when the sound source target is located at depth Zs at a certain moment. Figure 5 Sound source location inversion results based on angle matching criterion Figure 6 Ideally, at a horizontal distance of 7.7 km, the spatial spectrum of estimated CBF angle of arrival for sound source targets at different depths is as follows: Figure 7 Ideally, at a horizontal distance of 7.7 km, the spatial spectrum of OGSBI angle of arrival estimation for sound source targets at different depths. Figure 8 OGSBI estimation results when the sound source target is located at a depth Zs at a certain moment. Detailed Implementation The present invention will now be further described in conjunction with the embodiments and accompanying drawings: A target depth segmentation method based on the mapping of angle of arrival and target depth is characterized by: deploying a vertical linear array on the seabed to receive single-frequency signals emitted by sound sources near the sea surface; the vertical linear array consists of... M It consists of array elements, with an element spacing of [missing information]. d Array center depthz r .

[0032] 1. Deep-sea environment, sound source, and vertical linear array configuration of the receiver in this embodiment: To verify the effectiveness of the method of this invention, a computer simulation experiment was conducted. This embodiment considers a typical deep-sea environment at a depth of 5000m, and the Munk sound velocity profile used is as follows. Figure 1 As shown. The sound speed at the sea surface is 1548 m / s, the acoustic axial depth is 1300 m, the acoustic axial velocity is 1500 m / s, the critical depth is 4800 m, and the sound speed on the seabed is 1552 m / s. The model above the sea surface is a vacuum environment, and the seabed is an acoustic half-space. Specific seabed sediment parameters are: seabed sound speed is 1600 m / s, and seabed density is 1.8 g / cm³. 3 The substrate attenuation coefficient is 0.8 dB / λ.

[0033] sound source and M = The side view of the relative positions of the 56-yuan vertical line array is as follows Figure 2 As shown, the sound source remains stationary and continuously radiates frequencies into the vertical linear array. f = A 300Hz single-frequency acoustic signal. Horizontal distance between the sound source and the array. r s = 7.7km, sound source depth z s = 200m, array center depth z r = 4837m, array element spacing d = 2.5m.

[0034] 2. Vertical linear array receiving signals This embodiment assumes the sound source signal is a single-frequency signal. The simulation method for the array receiving signal is as follows: First, for the sound source position at a certain moment, the frequency domain sound field generated by the sound source on the receiving array (all sensors) is calculated using the BELLHOP sound field model. Then, the frequency domain sound field is used to construct a steering vector to generate a frequency domain snapshot signal. Finally, independent Gaussian white noise between array elements is added according to a given signal-to-noise ratio to form the array receiving signal. y Assume that the received signal-to-noise ratio (SNR) on each array element is SNR = 10dB.

[0035] The steps of the target depth segmentation method based on the mapping between angle of arrival and target depth are as follows: like Figure 3 As shown, the specific implementation process of the target depth division method based on the mapping of angle of arrival and target depth proposed in this invention is as follows.

[0036] Step 1: Use the CBF method to estimate the angle of arrival. The acoustic signal radiated by the underwater sound source was received using a vertical linear array, and the angle of arrival (Angle of Arrival) was estimated using the CBF method to obtain the CBF-normalized Angle of Arrival (Angle of Arrival) estimation spectrum. P dB_CBF and the estimated angle of arrival of the signal i CBF_esti .

[0037] Perform conventional beamforming on the array-received signal and calculate the target angle of arrival estimate. i CBF_esti The specific process is as follows: The CBF method is calculated using the following formula: (1) in P ( i The ) represents the output power of the beamformer. R y This represents the covariance matrix of a 3-array signal. w ( i ) represents the beam weighting vector, and H represents the transpose symbol.

[0038] In practical applications, the output power of the beamformer is often normalized to decibels, i.e. (2) The angle at which the beam output power reaches its maximum value is the estimated angle of arrival. i CBF_esti ,Right now (3) The R y It is calculated using the following formula. (4) in, y ( f ) indicates that the frequency of the sound source is f The vertical linear array receives acoustic signals in the frequency domain. The w ( i ) is calculated using the following formula, (5) in, c The reference sound speed is defined as the sound speed at the receiving depth of the vertical linear array.

[0039] In a specific embodiment: (1) Receive the array signal y Substituting into equation (4), the covariance matrix of the array received signal is calculated. R y ; (2) The number of array elements M = 56. Array element spacing d = 2.5m, signal frequency f = 300Hz and reference speed of sound c Substituting 1548m / s into equation (5) yields the beam weighting vector. w ( i ); (3) The covariance matrix R y With beam weighting vector w ( i Substituting into equation (1), we obtain the beamformer output power. P CBF ( i ); (4) The output power of the beamformer P ( i Substituting into equation (2), the normalized output power of the beamformer is calculated in decibels. P dB_CBF ( i ); (5) According to formula (3), the angle at which the beam output power reaches its maximum value is the estimated angle of arrival. i CBF_esti .

[0040] The CBF output power normalized to decibels and the estimated angle of arrival spectrum are as follows: Figure 4 As shown. The estimated pitch angle at this point is 59.9°.

[0041] Step 2: Depth Range Threshold Setting Simulations were performed using the Bellhop ray acoustic model to obtain the critical depth corresponding to the CBF method under ideal conditions. D CBF and the critical depth corresponding to the OGSBI method D OGSBI .

[0042] Using the angle of arrival estimate obtained in step 1 i CBF_esti The system combines marine environmental information such as the sound velocity profile of the array deployment area, seabed parameters, and the depth location information of each array element with the input of the BELLHOP ray acoustic model. Within a preset distance-depth two-dimensional search space, it traverses the locations of candidate sound sources (…). r k , z k Run BELLHOP. For each candidate location, extract the intrinsic angles of arrival of the direct wave and the first reflected wave. i D and i SR Candidate positions that satisfy the following formula are selected to form a set of angle-matching positions. S (6) In the set S Among them, the distance that appears most frequently is selected as the reference distance. r ref A depth scan sequence is constructed along this distance. Z sim = { z min , z min + z , …, z max}, where step size z This can be set according to the required depth resolution. For each depth... z i The BELLHOP model was used to generate simulated array complex sound pressure data containing direct wave and sea surface reflected wave components, and Gaussian white noise with a preset signal-to-noise ratio was superimposed.

[0043] Angle of arrival estimation was performed on the simulation data using CBF and OGSBI, and the number of effective peaks in the spatial spectrum at each depth was counted. Recording... N CBF_peaks ( z i )and N OGSBI_peaks ( z i (This varies with depth.) N CBF_peak The minimum depth required for a stable transition from 1 to 2 is defined as the critical depth. D CBF ,Will N OGSBI_peak The minimum depth required for a stable transition from 1 to 2 is defined as the critical depth. D OGSBI The effective peak value refers to the amplitude in the spatial spectrum that exceeds a preset threshold. or The local maximum point.

[0044] (7) In a specific embodiment: (1) Settings i = 2°, minimum depth Z min = 5m, maximum depth Z max = 500m, depth variation step size Z = 5m, effective peak threshold or = -3dB.

[0045] (2) Combining the estimated elevation angle, sound velocity profile data, substrate parameter data, and array position data obtained in step 1, the possible location set of the sound source is calculated using the BELLHOP model. S The obtained position inversion results are as follows Figure 5 As shown. By Figure 5 It can be seen that the distance with the highest frequency of occurrence is approximately 7.7km, therefore, the distance selected is... r ref = 7.7km was used as the reference distance. The BELLHOP sound field model was used to generate simulated array complex sound pressure data containing direct wave and sea surface reflected wave components at different depths, and Gaussian white noise with a preset signal-to-noise ratio was added to form the array received signal.

[0046] (3) The simulation data in (2) were processed using the CBF and OGSBI angle of arrival estimation methods respectively, and the spatial spectrum of the estimated angle of arrival of sound source targets at different depths was obtained when using CBF or OGSBI. The results are as follows: Figure 6 and Figure 7 As shown. Analysis Figure 6 and Figure 7 It can be seen that when using the CBF method, the critical depth is... D CBF = 230m; This indicates that the critical depth is [value missing] when using the OGSBI method. D OGSBI = 60m.

[0047] Step 3: Use OGSBI to estimate the angle of arrival and obtain the number of valid peaks. Sub-step 1: Use OGSBI to estimate the angle of arrival. Angle of arrival estimated based on CBF estimation i CBF_esti The angle search range was redefined, and OGSBI was used to estimate the angle of arrival within this range to obtain the OGSBI normalized angle of arrival estimation spectrum. P dB_OGSBI .

[0048] (1) The estimated angle of arrival obtained from OGSBI estimation i CBF_esti Re-plan the angle search range and construct the array manifold.

[0049] ① In [ i CBF_esti – i ex , i CBF_esti + iex Within the range of spacing r generate N grid points { i g1 , i g2 , …, i gN}

[0050] ② Calculate the guiding vector corresponding to each grid point based on the array geometry. a ( i gn ) and its first derivative b ( i gn )= a `( i gn ).

[0051] ③ Obtain the matrix A = [ a ( i g1 ), a ( i g2 ), …, a ( i gN )], B = [ b ( i g1 ), b ( i g2 ), …, b ( i gN )).

[0052] (2) SVD dimensionality reduction The array received signal y mentioned above is used as the received data matrix here. Y ; ① For the received data matrix Y Perform singular value decomposition: Y=USV H .

[0053] ② Then extract before K The right singular value constitutes V 1, of which K Pick M 1 or the number of known sound sources (for the single-target case, take...) K = 1).

[0054] ③ Calculate the dimensionality-reduced data matrix: Y SV= YV 1.

[0055] ④ Data normalization after dimensionality reduction: and the number of quick photos T Replace with K In all the following steps y ( t (All refer to) Y The t List, t = 1, …, T .

[0056] (3) Hyperparameter initialization ① Set constants: r , c , d Maximum number of iterations iter max Convergence threshold t .

[0057] ② Initial value of noise variance:

[0058] ③ Initial values ​​of signal hyperparameters:

[0059] ④ Initial value of deviation from standard: β n , n = 1, …, N .

[0060] ⑤ Constructing the matrix: Λ -1 = diag( α 1 -1 , …, α N -1 ).

[0061] (4) Iterative update Repeat the following sub-steps until the convergence condition is met or the maximum number of iterations is reached.

[0062] ① Construct the lattice array manifold matrix (8) ② Calculate the posterior covariance matrix of the signal. (9) ③ Calculate the posterior comean matrix of the signal. (10) And make , ,remember U n for U The n OK.

[0063] ④ Update hyperparameters α n and Λ -1 (11) (12) And update Λ -1 = diag(( α 1 new ) -1 , …, ( α N new ) -1 ).

[0064] ⑤ Update deviation Only for α new The largest K Update the grid points corresponding to each component. For the grid points that need to be updated, extract the corresponding columns of A and B, and then extract... β and Extract the corresponding line. By analyzing the corresponding columns and rows, we obtain the reduced-dimensional matrices, and then calculate: (13) (14) Then update the off-grid deviation: (15) ⑥ Update noise variance (16) ⑦ Convergence judgment (17) like e < t Stop iteration; otherwise, let α old = α new Return to step ① and continue iterating to update the parameters.

[0065] (5) Output the estimation results After the iteration is complete, a relative power spectrum can be constructed to find the spectral peaks. (18) like K The grid index value corresponding to each peak isg k Then the first k The estimation results of the incident signals are as follows: .

[0066] In practical applications, the output power is often normalized to decibels, i.e. (19) Sub-step 2: Obtain the number of valid peak values N OGSBI_peaks Regarding the aforementioned P dB_OGSBI Perform effective peak statistics to obtain P dB_OGSBI Number of valid peaks in N OGSBI_peaks .

[0067] right P dB_OGSBI ( i Peak detection is performed, and the statistical amplitude exceeds a preset threshold. or Number of effective peak values N OGSBI_peaks .

[0068] In a specific embodiment: (1) Take i ex = 10°, grid spacing r = 0.1°, generate 200 grid points within the range [50°, 70°], and calculate the guiding vector corresponding to each grid point. a ( i gn ) and its first derivative b ( i gn ), further obtaining the matrix A and B .

[0069] (2) For the received data matrix Y Perform singular value decomposition and take K = 2, construct V 1. Subsequently, the dimensionality-reduced data is normalized.

[0070] (3) Set constants r = 0.01, c = d = 0.0001, maximum number of iterations iter max = 500, convergence threshold t=0.001. Then, initialize according to the initialization formulas for noise variance, signal hyperparameters, and deviation in step 4.

[0071] (4) Iterate and update according to equations (8)-(17) until the convergence factor is reached. e < t Or the number of iterations reaches the maximum number of iterations. iter max Stop iteration.

[0072] (5) Based on the final iteration results, calculate the relative power spectrum according to equation (18). P OGSBI The azimuth estimation spectrum after normalization to decibels is calculated according to equation (19) as follows: Figure 8 As shown. At this time, the estimated angle of arrival of the direct wave is 61°, and the estimated angle of arrival of the sea surface reflected wave is 58.6°.

[0073] (6) Analysis Figure 8 The number of effective peaks is calculated according to formula (7), which gives the number of effective peaks in the OGSBI azimuth estimation spectrum at this time. N OGSBI_peaks = 2.

[0074] Step 4: Target Depth Segmentation According to the above P dB_OGSBI Number of valid peaks N OGSBI_peaks Perform depth interval partitioning: if N OGSBI_peaks If the value is greater than 1, then the depth of the sound source is determined to satisfy the condition. D CBF > Z s > D OGSBI This indicates that the depth of the sound source is greater than the OGSBI depth threshold. D OGSBI However, it is less than the CBF depth critical value. D CBF ;like N OGSBI_peaks = 1, then the depth of the sound source is determined to satisfy Z s < D OGSBI This indicates that the depth of the sound source is less than the OGSBI depth threshold. D OGSBI .

[0075] In a specific embodiment: at this time, N OGSBI_peaks >1, then we can know that at this time D CBF >Z s > D OGSBI That is, the depth of the sound source target. Z s <Critical Depth D CBF And the depth of the sound source target Z s Critical Depth D OGSBI .

[0076] This invention is applicable to areas where the target and receiving array are distributed in a reliable acoustic path region in the deep sea.

[0077] This invention is applicable not only to OGSBI-based angle of arrival estimation methods, but also to other high-resolution angle of arrival estimation methods.

[0078] This invention is applicable to vertical line arrays.

Claims

1. A target depth segmentation method based on the mapping between angle of arrival and target depth, characterized in that: A vertical line array is deployed on the seabed to receive single-frequency signals emitted by sound sources near the sea surface; the vertical line array consists of... M It consists of array elements, with an element spacing of [missing information]. d Array center depth z r The steps for sound source depth segmentation are as follows: Step 1: Using the CBF method, the angle of arrival (AHA) is estimated by receiving the acoustic signal radiated by the underwater sound source using a vertical linear array, and the CBF normalized AHA estimate spectrum is obtained. P dB_CBF and the estimated angle of arrival of the signal θ CBF_esti ; Step 2: Use the estimated angle of arrival obtained in Step 1. θ CBF_esti The sound velocity profile of the vertical linear array deployment area, seabed parameters, and depth location information of each array element are input into the Bellhop ray acoustic model to obtain the critical depth corresponding to the CBF method under ideal conditions. D CBF and the critical depth corresponding to the OGSBI method D OGSBI With two critical depths D CBF and D OGSBI As a criterion for determining the depth of the target; Step 3: Use OGSBI to estimate the angle of arrival and obtain the number of valid peaks: Sub-step 1: Based on the estimated angle of arrival from step 1 θ CBF_esti The search range for the planned angle is determined, and OGSBI is used for angle of arrival estimation to obtain the OGSBI normalized angle of arrival estimation spectrum. P dB_OGSBI ; Sub-step two: For P dB_OGSBI Perform effective peak statistics to obtain P dB_OGSBI Number of valid peaks in N OGSBI_peaks ; right P dB_OGSBI ( θ Peak detection is performed, and the statistical amplitude exceeds a preset threshold. η Number of effective peak values N OGSBI_peaks ; Step 4: Target Depth Division: Based on the number of valid peaks N OGSBI_peaks Perform depth interval partitioning: If N OGSBI_peaks If the value is greater than 1, then the depth of the sound source is determined to satisfy the condition. D CBF > Z s > D OGSBI This indicates that the depth of the sound source is greater than the OGSBI depth threshold. D OGSBI However, it is less than the CBF depth critical value. D CBF ;like N OGSBI_peaks = 1, then the depth of the sound source is determined to satisfy Z s < D OGSBI This indicates that the depth of the sound source is less than the OGSBI depth threshold. D OGSBI .

2. The target depth segmentation method based on the mapping between angle of arrival and target depth according to claim 1, characterized in that: The specific steps of step 1 are as follows: The output power of the beamformer in the CBF method is in: Indicates the output power of the beamformer. R y This represents the covariance matrix of the received signal from the array. w ( θ ) represents the beam weighting vector, and H represents the transpose sign; The output power is normalized to decibels as follows: ; The angle at which the beam output power reaches its maximum value is the estimated angle of arrival. θ CBF_esti : 。 3. The target depth segmentation method based on the mapping between angle of arrival and target depth according to claim 2, characterized in that: The R y Through formula The calculation yields y( f ) indicates that the frequency of the sound source is f The vertical linear array receives acoustic signals in the frequency domain, where H represents the transpose symbol.

4. The target depth segmentation method based on the mapping between angle of arrival and target depth according to claim 2, characterized in that: The w ( θ ) through formula The calculation yielded that, c The reference sound speed is defined as the sound speed at the receiving depth of the vertical linear array. M Indicates the number of array elements. d The interval between array elements.

5. The target depth segmentation method based on the mapping between angle of arrival and target depth according to claim 1, characterized in that: Two critical depths are obtained in step 2. D CBF and D OGSBI The specific steps are as follows: Within the preset distance-depth two-dimensional search space, traverse the locations of candidate sound sources ( r k , z k Run BELLHOP; for each candidate location, extract the eigenangles of arrival of the direct wave and the first reflected wave. θ D and θ SR ; Select candidate positions that satisfy the following formula to form a set of angle-matching positions. S : In the set S Among them, the distance that appears most frequently is selected as the reference distance. r ref ; Along the reference distance r ref Constructing depth scan sequences Z sim = { z min , z min + z , …, z max },in: z Step size; For each depth z i The BELLHOP model was used to generate simulated array complex sound pressure data containing direct wave and sea surface reflected wave components, and Gaussian white noise with a preset signal-to-noise ratio was superimposed to form the simulation data. The angle of arrival was estimated using CBF and OGSBI on the simulation data, and the number of effective peaks in the spatial spectrum at each depth was counted. Record N CBF_peaks_ideal ( z i )and N OGSBI_peaks_ideal ( z i As depth changes, N CBF_peaks_ideal The minimum depth required for a stable transition from 1 to 2 is defined as the critical depth. D CBF ,Will N OGSBI_peaks_ideal The minimum depth required for a stable transition from 1 to 2 is defined as the critical depth. D OGSBI ; Effective peak value refers to the amplitude in the spatial spectrum that exceeds a preset threshold. η Local maxima: ； in n = 1, 2, …, N, N This represents the number of angle search grids in the angle of arrival estimation process.

6. The target depth segmentation method based on the mapping between angle of arrival and target depth according to claim 5, characterized in that: The step size z Set according to the required depth resolution.

7. The target depth segmentation method based on the mapping between angle of arrival and target depth according to claim 1, characterized in that: Sub-step one of step 3 obtains the OGSBI normalized angle of arrival estimation spectrum. P dB_OGSBI The specific steps are as follows: (1) The estimated angle of arrival obtained from OGSBI estimation θ CBF_esti Plan the search range by angle and construct the array manifold: ① In [ θ CBF_esti – θ ex , θ CBF_esti + θ ex Within the range of spacing r generate N grid points { θ g1 , θ g2 , …, θ gN }; ② Calculate the guiding vector corresponding to each grid point. a ( θ gn ) and its first derivative b ( θ gn ) = a `( θ gn ), forming a matrix A =[ a ( θ g1 ), a ( θ g2 ), …, a ( θ gN )], B = [ b ( θ g1 ), b ( θ g2 ), …, b ( θ gN )]; (2) SVD dimensionality reduction The array received signal y mentioned above is used as the received data matrix here. Y ; ① For the received data matrix Y Perform singular value decomposition: Y = USV H ; ② Then extract before K The right singular value constitutes V 1; ③ Calculate the dimensionality-reduced data matrix: Y SV = YV 1; ④ Data normalization after dimensionality reduction: and the number of quick photos T Replace with K ; in all the following steps y ( t (All refer to) Y The t List, t = 1, …, T ; (3) Hyperparameter initialization ① Set constants: ρ , c , d Maximum number of iterations iter max Convergence threshold τ ; ② Initial value of noise variance: ③ Initial values ​​of signal hyperparameters: ④ Initial value of deviation from standard: β n , n = 1, …, N ; ⑤ Constructing the matrix: Λ -1 = diag( α 1 -1 , …, α N -1 ); (4) Iterative update Repeat the following sub-steps until the convergence condition is met or the maximum number of iterations is reached; ① Construct the lattice array manifold matrix ② Calculate the posterior covariance matrix of the signal. ③ Calculate the posterior comean matrix of the signal. And make , ,remember U n for U The n ; ④ Update hyperparameters α n and Λ -1 And update Λ -1 = diag(( α 1 new ) -1 , …, ( α N new ) -1 ); ⑤ Update deviation Only for α new The largest K Update the grid points corresponding to each component. For the updated grid points, extract the corresponding columns of A and B. β and Extract the corresponding line. By analyzing the corresponding columns and rows, we obtain the reduced-dimensional matrices, and then calculate: Then update the off-grid deviation: ⑥ Update noise variance ⑦ Convergence judgment like ε < τ Stop iteration; otherwise, let α old = α new Return to step ① and continue iterating to update the parameters; (5) Output the estimation results After the iteration is complete, a relative power spectrum can be constructed to find the spectral peaks. like K The grid index value corresponding to each peak is g k Then the first k The estimation results of the incident signals are as follows: ; In practical applications, the output power is often normalized to decibels, i.e. 。 8. The target depth segmentation method based on the mapping between angle of arrival and target depth according to claim 7, characterized in that: The spacing r The angle search grid spacing can be set according to the required angle resolution accuracy.

9. The target depth segmentation method based on the mapping between angle of arrival and target depth according to claim 7, characterized in that: The K Among the right singular values K Pick M 1 or the number of sound sources is known; for the single-target case, take... K = 1.

10. A computer program product, characterized in that... It includes computer-executable instructions, which, when executed, are used to implement the target depth division method based on the angle of arrival and target depth mapping as described in claims 1 to 9.

Images