A method and system for sound source localization using a difference frequency coherent matched self-product process

Abstract

The application discloses a kind of difference frequency coherent matching self-product processing sound source positioning method and system, the method includes: step S1) Fourier transform is carried out to the sound propagation time domain data received by hydrophone array, the required frequency bandwidth and difference frequency are selected, self-product vector is calculated, and normalization processing is carried out;Step S2) the normalized self-product vector of different difference frequency is spliced to obtain super vector;Step S3) using super vector calculates super cross-spectral density matrix, then carries out normalization processing;Step S4) using sound field model and experimental environment parameter calculates self-product copy vector, and after normalization processing and splicing, obtain super copy vector, then super copy vector is normalized;Step S5) the result of step S4) is matched with the super cross-spectral density matrix of step S3), obtain the ambiguity surface of single frequency point, then through the average of frequency bandwidth, obtain the final ambiguity surface, obtain the position where the peak of ambiguity surface is located, realize sound source positioning.

Description

Technical Field

[0001] This invention belongs to the field of underwater acoustic signal processing technology, and particularly relates to a method and system for sound source localization using difference-frequency coherent matched self-integration processing. Background Technology

[0002] Matched field processing (MRP) is a frequency domain signal processing technique for locating underwater sound sources. It utilizes marine environmental parameters and sound propagation channel characteristics to calculate a predicted copy of the sound field received by an array using an underwater sound field model. This copy is then "matched" with the received data, comparing the similarity between the copied and actual sound fields to achieve passive target localization. However, environmental mismatch (caused by ocean instability and inaccurate environmental parameters), array system mismatch, and environmental noise limit the effectiveness of MRP. Therefore, robust MRP methods have been a key research focus.

[0003] Differential frequency matched field processing (DFF) is an emerging robust matched field processing method. Since the error between the copied sound field and the real sound field increases with distance and signal frequency, DFF utilizes the characteristic that the bandwidth-averaged difference frequency self-integration carries low-frequency sound field information, thus suppressing the effects of time delay mismatch and array tilt mismatch. The disadvantages of DFF are that it sacrifices resolution and has a low main-side lobe ratio in the ambiguity surface. The low resolution is due to the phase sensitivity of the high-frequency sound field, and the spatial resolution of matched field processing decreases with decreasing frequency. The low main-side lobe ratio is caused by the inherent quadratic nonlinearity of the self-integration. Traditional adaptive methods for suppressing sidelobes in matched field processing, such as minimum variance distortionless response and multiple constraint methods, have failed to achieve the same effect when extended to DFF.

[0004] The measurements from different difference-frequency matched-field processing methods are all constructed from the bandwidth-averaged difference-frequency self-integration measured by a hydrophone; the main difference lies in the copy vector. The earliest proposed difference-frequency matched-field processing technique located high-frequency sound sources in shallow waters (a multipath environment dominated by reflection), and its copy vector is the sound pressure field at the difference frequency. However, in deep waters (a multipath environment dominated by refraction), the self-integration does not produce a phase shift at the caustics like the sound field. Therefore, difference-frequency sound source localization matches the measured self-integration with the phase-corrected (phase of the self-integration) copy sound field. In literature studying the self-integration cross term, the weight vector is replaced with the bandwidth-averaged self-integration. Phase-matched self-integration processing uses only the phase information of the bandwidth-averaged self-integration in its copy vector, providing more robust localization results than other difference-frequency passive source localization algorithms. The copy vectors of the above three difference-frequency methods all use the phase of the self-integration and are collectively referred to as matched self-integration processing, applicable to locating sound sources in deep waters. Existing matching self-integration processing methods all employ incoherent approaches, which suffer from common problems in difference-frequency matched field processing, such as low ambiguity surface resolution and low main-side lobe ratio. Summary of the Invention

[0005] To address the issues of low ambiguity, surface resolution, and main-sidelobe ratio in existing incoherent matched self-integration processing for locating broadband sound sources in deep-sea environments, this invention aims to overcome these shortcomings by proposing a difference-frequency coherent matched self-integration processing method for sound source localization.

[0006] To achieve the above objectives, this invention proposes a method for sound source localization using difference-frequency coherent matched self-integration processing, the method comprising:

[0007] Step S1) Perform Fourier transform on the sound propagation time-domain data received by the hydrophone array, select the required frequency bandwidth and difference frequency, calculate the self-integration vector, and perform normalization processing.

[0008] Step S2) Concatenate the normalized self-integral vectors with different difference frequencies to obtain a supervector;

[0009] Step S3) Calculate the super cross-spectral density matrix using hypervectors, and then perform normalization.

[0010] Step S4) Calculate the self-integrated copy vector using the sound field model and experimental environment parameters, and obtain the supercopy vector by normalization and splicing. Then normalize the supercopy vector.

[0011] Step S5) Match the result of step S4) with the super cross-spectral density matrix of step S3) to obtain the ambiguity surface of a single frequency point. Then, by averaging the frequency bandwidth, obtain the final ambiguity surface and obtain the location of the peak of the ambiguity surface to achieve sound source localization.

[0012] As an improvement to the above method, step S1) specifically includes:

[0013] Perform a Fourier transform on the time-domain data of sound propagation received by the hydrophone array, and calculate the difference-frequency self-integration AP(r,Δω,ω) according to the following formula:

[0014] AP(r,Δω,ω)≡P * (r,ω-Δω / 2)P(r,ω+Δω / 2)

[0015] Where ω is the frequency, Δω is the difference frequency, and P(r,ω) is the complex sound pressure level at frequency ω received by the hydrophone at point r, indicated by the superscript (·). * Indicates complex conjugation;

[0016] The source spectrum amplitude is eliminated by normalizing using the 2-norm, and then the source spectrum phase is eliminated by dividing by the self-integration phase of the x-th hydrophone. The self-integration vector corresponds to the element of the l-th hydrophone. for:

[0017]

[0018] Where m, n, and l are the frequency ω, the difference frequency Δω, and the hydrophone number, respectively; 1 ≤ m ≤ M, where M is the total number of frequency points within the frequency bandwidth; 1 ≤ n ≤ N, where N is the total number of difference frequency samples; and 1 ≤ l ≤ L, where L is the total number of hydrophones. Let Green's function be the distance from the actual sound source to the l-th hydrophone. The phase angle is expressed in terms of the Green's function, with the subscript + indicating the high-frequency component in the self-integration, - indicating the complex conjugate of the low-frequency component, and i representing the imaginary unit.

[0019] As an improvement to the above method, the frequency of step S2) is ω. m Hypervector for:

[0020]

[0021] Among them, AP mn This represents the normalized self-integral vector, which is an L×1 dimensional column vector with the superscript (·). T The symbol indicates transpose, and the subscript s indicates the concatenated "hypervector".

[0022] As an improvement to the above method, step S3) specifically includes:

[0023] Through hypervectors The calculated matrix yields the hypercross-spectral density matrix, which is then normalized using the F-norm to obtain the normalized hypercross-spectral density matrix.

[0024]

[0025] Among them, superscript To represent the conjugate transpose, ||·|| F This represents the F-norm.

[0026] As an improvement to the above method, step S4) specifically includes:

[0027] Calculation using sound field model and experimental environment parameters The copied vector yields a frequency ω. m The difference frequency is Δω n The l-th value of the self-integrating copy vector for:

[0028]

[0029] in, To predict the Green's function from the sound source to the l-th hydrophone;

[0030] The supercopy vector is obtained by concatenating the normalized self-integrated copy vectors with different difference frequencies.

[0031]

[0032] According to the following formula Perform 2-norm normalization to obtain the normalized supercopy vector.

[0033]

[0034] Where ||·||2 represents the 2-norm.

[0035] As an improvement to the above method, step S5) specifically includes:

[0036] Calculate the supercopy vector at point r in the predicted sound source grid. hypercopy vector With the hyperspectral density matrix Matching is performed, and then averaging is performed over the entire frequency bandwidth 1≤m≤M to obtain the ambiguity surface B. C,avg (r):

[0037]

[0038] Among them, superscript represents the conjugate transpose, the subscript C indicates coherence, and avg indicates bandwidth average;

[0039] Get B C,avg The location of the peak value of (r) is used to locate the sound source.

[0040] On the other hand, the present invention proposes a difference-frequency coherent matched self-integration processing sound source localization system, the system comprising:

[0041] The self-integration vector calculation module is used to perform Fourier transform on the sound propagation time-domain data received by the hydrophone array, select the required frequency bandwidth and difference frequency, calculate the self-integration vector, and perform normalization processing.

[0042] The hypervector construction module is used to concatenate normalized self-integral vectors with different difference frequencies to obtain a hypervector;

[0043] The hyper-mutual spectral density matrix calculation module is used to calculate the hyper-mutual spectral density matrix using hypervectors and then perform normalization processing.

[0044] The self-integrated copy vector calculation module is used to calculate the self-integrated copy vector using the sound field model and experimental environment parameters, and then normalize and concatenate it to obtain the supercopy vector, which is then normalized again; and

[0045] The matching and localization module is used to match the result of the self-integrated copy vector calculation module with the super-mutual spectral density matrix of the super-mutual spectral density matrix calculation module to obtain the ambiguity surface of a single frequency point. Then, by averaging the frequency bandwidth, the final ambiguity surface is obtained, and the location of the peak of the ambiguity surface is obtained to realize the sound source localization.

[0046] Compared with the prior art, the advantages of the present invention are:

[0047] The method of the present invention, while maintaining the characteristic of the difference frequency method to carry low-frequency information by utilizing the bandwidth average difference frequency self-integration, suppresses the high side lobes of the ambiguity surface caused by the second nonlinearity of self-integration in the matching self-integration process, improves the resolution, and has stronger positioning robustness. Attached Figure Description

[0048] Figure 1 This is a flowchart of the sound source localization method for difference frequency coherence matching self-integration processing according to the present invention;

[0049] Figure 2 This includes hydrological parameters and sound source-receiver locations from a deep-sea Munk profile environmental simulation.

[0050] Figure 3 shows the ambiguity surface processed by incoherent and coherent matching self-integration, where Figure 3(a) is the incoherent method and Figure 3(b) is the coherent method with normalization using the first hydrophone.

[0051] Figure 4 shows the ambiguity surfaces after coherent and phase-only coherent matching self-integration processing, where Figure 4(a) is the coherent result and Figure 4(b) is the phase-only coherent result.

[0052] Figure 5 It is the change of side lobe value with normalized hydrophone;

[0053] Figure 6 It shows the experimental environment and sound velocity profile in the South China Sea.

[0054] Figure 7 shows the range-depth localization results with an array tilt angle of 0.1°, where Figure 7(a) shows the range result and Figure 7(b) shows the depth result.

[0055] Figure 8 shows the distance-depth positioning results with a water depth mismatch factor of 3%, where Figure 8(a) shows the distance results and Figure 8(b) shows the depth results.

[0056] Figure 9 shows the ambiguity surface of the sound source localization experimental data in the sound shadow area. Figure 9(a) shows the localization result of traditional matching field processing, Figure 9(b) shows the result of incoherent matching self-integration processing, Figure 9(c) shows the result of coherent processing, and Figure 9(d) shows the result of phase-only coherence.

[0057] Figure 10 shows the distance-depth positioning results of the South China Sea experimental data, where Figure 10(a) shows the distance results and Figure 10(b) shows the depth results.

[0058] Figure 11 It refers to the success rate at different acceptable error levels. Detailed Implementation

[0059] Unlike current matched self-integration processing methods, this invention proposes a difference-frequency coherent matched self-integration processing method for broadband signals received by vertical arrays in deep-sea environments. Broadband matched field processing is divided into incoherent and coherent methods. The coherent method utilizes the correlation between different frequencies of the sound field to suppress sidelobes, improve resolution, enhance robustness, and offset some of the negative effects of low signal-to-noise ratio. This invention, based on the original incoherent matched self-integration processing, utilizes the coherence of self-integration in the difference-frequency domain and uses the self-integration phase of a single hydrophone for normalization to eliminate the weighting factor of the source spectrum, thus making it suitable for passive localization of deep-sea sound sources.

[0060] This invention extends the coherent method to matched self-integration processing. The algorithm incorporates the coherence of self-integration in the difference frequency domain while still using incoherent averaging of the ambiguity surface with respect to frequency, thus overcoming the detrimental effects of self-integration cross terms. Simulation results from the deep-sea Munk profile environment and experimental data from the South China Sea demonstrate that coherent matched self-integration processing can suppress sidelobes, improve resolution, and exhibit better robustness.

[0061] Extending coherent matched field processing to incoherent matched self-integration processing to suppress high sidelobes has two problems: 1. Since the difference frequency method involves two frequency sets—signal frequency bandwidth and difference frequency—it is necessary to determine which frequency set of coherent information to use; 2. Coherent algorithms require prior information on the sound source spectrum and are not suitable for passive localization.

[0062] Solutions: 1. Add the use of the coherence of self-integration in the difference frequency domain to the algorithm, while still performing incoherent averaging for the frequency bandwidth; 2. Use the self-integration phase measured by a single hydrophone for normalization during the normalization process, eliminating the phase information of the source spectrum, which is suitable for passive positioning; 3. Based on coherent matching self-integration processing, only use the phase information of self-integration.

[0063] 1. Perform Fourier transform on the sound propagation time-domain data received by the hydrophone array, select the required frequency points or frequency bandwidth and difference frequency, calculate the self-integration, and select a single hydrophone for normalization.

[0064] 2. Concatenate the self-integrated vectors with different difference frequencies to obtain a "hypervector";

[0065] 3. The matrix calculated using the "hypervector" is the "hyper-cross-spectral density matrix", which is normalized using the F-norm;

[0066] 4. The “supercopy vector” is formed by splicing together the self-integrated copy vector calculated using the sound field model and experimental environment parameters, normalizing it using the 2-norm, matching it with the “super cross-spectral density”, and then averaging the obtained ambiguity surface with the frequency bandwidth to obtain the final ambiguity surface.

[0067] 5. The calculation method for phase-coherent matching self-integration processing is the same as that of the coherent method, except that only the phase information of the self-integration is used.

[0068] The following is a detailed description and derivation of the method proposed in this invention, including the bandwidth-averaged difference frequency self-integration carrying low-frequency information and the derivation from incoherent matched self-integration processing to the difference frequency coherent method.

[0069] Bandwidth average difference frequency self-integration carries low-frequency information

[0070] The difference-frequency self-integration (hereinafter referred to as self-integration) AP(r,Δω,ω) refers to the quadratic product of the sound pressure levels at different frequencies measured at the same location in space.

[0071] AP(r,Δω,ω)≡P * (r,ω-Δω / 2)P(r,ω+Δω / 2) (1)

[0072] Where P(r,ω) is the complex sound pressure level at frequency ω received by the hydrophone at point r. The frequency difference Δω used to construct the self-integration is the difference frequency. When the sound field is measured by a signal with sufficient bandwidth, the bandwidth-averaged self-integration is usually used instead of the self-integration itself.

[0073]

[0074] Where S(ω) is the sound source spectrum, and the lower and upper limits of integration are Ω. L +Δω / 2、Ω H -Δω / 2, bandwidth is Ω BW =Ω H -Ω L -Δω.

[0075] The bandwidth-averaged self-product can carry information about the low-frequency sound field, which requires analysis of the self-terms and cross-terms of the self-product, using K... ray The sound field is described analytically by analyzing the incident sound rays.

[0076]

[0077] Among them, A k and τ k Let be the complex amplitude and arrival time of the k-th ray, respectively. The self-integration and its bandwidth average are then...

[0078]

[0079]

[0080] In equations (4) and (5), the first summation term is called the self-term, and the second summation term is called the cross term. It can be seen that the self-term is similar in form to the sound field with the difference frequency. In equation (4), the cross term is related to the frequency ω and is a non-low frequency information term, which is undesirable. However, by comparing equations (4) and (5), the cross term decreases as the average signal bandwidth increases, and bandwidth averaging can suppress the cross term.

[0081] Difference frequency coherent matching self-integration processing

[0082] The difference-frequency coherent matching self-integration processing (hereinafter referred to as coherent matching self-integration processing) is a generalization of its incoherent method. The sound pressure P measured by the first hydrophone... l The self-integral constructed by (ω±Δω / 2) is,

[0083]

[0084] Where m, n, and l are the frequency ω, difference frequency Δω, and hydrophone number, respectively. In the simplified version of equation (6), the subscript + represents the high-frequency component in the self-integration, and - represents the complex conjugate of the low-frequency component. The self-integration vector is normalized using its 2-norm.

[0085]

[0086] Normalization eliminated the amplitude factor of the source spectrum, while the phase factor was retained. The normalized self-integral was matched with the copy vector, and then summed over the hydrophone sequence.

[0087]

[0088] The subscript "IC" indicates the result of the incoherent method. Equation (8) still contains the phase factor of the source spectrum, which can be eliminated by squared modulus.

[0089]

[0090] Equation (9) represents the ambiguity surface under two frequency sets. The incoherent method uses equation (9) to incoherently average the two frequency sets to obtain the final positioning ambiguity surface.

[0091]

[0092] For coherent matched self-integration processing, utilizing the coherence of the self-integration in the difference frequency domain requires converting the incoherent average of the relevant difference frequencies into coherent ones. This changes the process from summing only over the hydrophone number *l* to summing over both the hydrophone number *l* and the difference frequency number *n*. However, using the same normalization method as the incoherent method cannot eliminate the source spectrum phase. To eliminate the source spectrum, the following normalization method can be used: first, normalize the self-integration vector using the 2-norm to eliminate the source spectrum amplitude, then divide by the self-integration phase of a single hydrophone to eliminate the source spectrum phase. Taking the *x*th hydrophone as an example, the *l*th value of the self-integration vector is:

[0093]

[0094] The copy vector is calculated using the formula on the right-hand side of equation (11), and then matched with the normalized self-integral vector.

[0095]

[0096] The subscript "C" indicates coherence, and N and L are the total number of difference frequency samples and the total number of hydrophones, respectively, serving as normalization factors to ensure |b C (r,ω m The value is between 0 and 1. Equation (12) is then incoherently averaged over the frequency bandwidth to obtain the ambiguity surface obtained by coherent matching self-integration.

[0097]

[0098] Phase-only coherent matched self-integration processing is an algorithm based on coherent matched self-integration processing. It only uses the phase information of the self-integration in the coherent method, and the l-th value of its self-integration vector is...

[0099]

[0100] The calculation steps are the same as those for the coherent method.

[0101] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0102] Example 1

[0103] Embodiment 1 of the present invention proposes a sound source localization method using difference frequency coherence matching self-integration processing.

[0104] The description of this invention has detailed the algorithm. However, in practice, the summation form is often less concise and easier to implement than the matrix form. Therefore, the implementation method of this invention is described in matrix form.

[0105] 1. Perform a Fourier transform on the time-domain data of sound propagation received by the hydrophone array, and select the required frequency point or bandwidth Ω. H -Ω L ω m=(Ω) L +Δω n / 2)+[(m-1) / (M-1)](Ω H -Ω L -Δω n ) is to subtract Δω in the frequency band n The sampling frequencies are uniformly distributed within the range, such that for any m (1≤m≤M), the sound pressure P for each frequency pair is... l (ω m ±Δω n / 2) All are within the frequency band. Δω n =ΔΩ L +[(n-1) / (N-1)](ΔΩ H -ΔΩ L ) is the sampling difference frequency selected by the user, and 1≤n≤N. The self-integration is calculated using equation (1), and the self-integration vector is normalized using equation (11).

[0106] 2. Concatenate the self-integrated vectors with different difference frequencies to obtain a "hypervector," using the abbreviation rules described in the description.

[0107]

[0108] AP mn This represents the normalized self-integral vector, which is an L×1 dimensional column vector with the superscript (·). T This indicates transpose.

[0109] 3. The matrix calculated using the "hypervector" is the "hyper-cross-spectral density matrix," which is normalized using the F-norm. The superscript... This indicates the conjugate transpose.

[0110]

[0111] 4. The "hypercopy vector" is formed by concatenating the self-integrated copy vectors calculated using the sound field model and experimental environment parameters, using the same calculation method as in step 2. The "hypercopy vector" is normalized using the 2-norm, specifically:

[0112] Calculation using sound field model and experimental environment parameters The copied vector yields a frequency ω. m The difference frequency is Δω n The l-th value of the self-integrating copy vector for:

[0113]

[0114] in, To predict the Green's function from the sound source to the l-th hydrophone;

[0115] The supercopy vector is obtained by concatenating the normalized self-integrated copy vectors with different difference frequencies.

[0116]

[0117] According to the following formula Perform 2-norm normalization to obtain the normalized supercopy vector.

[0118]

[0119] Where ||·||2 represents the 2-norm.

[0120] Then, it is matched with the "super cross-spectral density matrix", and the resulting ambiguity surface is averaged over the frequency bandwidth to obtain the final ambiguity surface. The location of the peak of the ambiguity surface is obtained to realize the sound source localization.

[0121]

[0122] 5. The calculation method for phase coherent matching self-integration is the same as that for coherent method, except that only the phase information of self-integration is used, and Equation (14) is used for calculation.

[0123] Best mode for achieving the present invention / Technical effects of this application

[0124] This invention addresses the problem of locating broadband signal sources using a vertical array in deep-sea environments. Simulations compare the ambiguity surface differences between this invention and existing algorithms in locating sources near the sound channel axis in deep-sea Munk profile environments. For South China Sea experimental data: simulations were performed to address array tilt mismatch and water depth mismatch, comparing the robustness differences between this invention and existing algorithms; and, using only hydrological data near the array, a copy field calculated through a sound field model was matched with measured values, and statistical parameters were used to describe the location distance-depth results.

[0125] Example 1: Deep-sea Munk Profile Environment

[0126] The deep-sea Munk profile environment is a typical deep-sea environment. Figure 2 The simulation environment is shown, with a flat, decompression-releasing surface and a half-space at the bottom, where the P-wave velocity is 1600 m / s. The sound source is near the deep-sea acoustic axis (depth 1000 m), propagating a broadband signal of 201-300 Hz in 1 Hz increments. A typical Munk profile is shown on the right. The selected difference frequencies are 1-5 Hz in 1 Hz increments. No noise was added to the measurements, and the predicted sound source distance is 2-200 km, with a depth of 40-2000 m. Both the measurement and copy vectors are constructed using Green's functions calculated from the KRAKEN model.

[0127] Figure 3 shows a comparison of the ambiguity surfaces simulated by the incoherent and coherent methods. Figure 3(a) shows the incoherent method, and Figure 3(b) shows the coherent method with normalization using the first hydrophone. The dynamic range of the ambiguity surfaces is set to 12 dB for both methods. Real sound sources are marked with "○", and predicted sound sources are marked with "×". The coherent method shows results in suppressing sidelobes and improving resolution. However, the sidelobe height of the coherent method varies with the normalized hydrophone. When normalization is performed using a hydrophone close to the sound source depth, the ambiguity surface images of the coherent and incoherent methods are basically the same.

[0128] Since coherent methods require normalization using a single hydrophone, the self-integration phase accuracy measured by a single hydrophone has a significant impact on the localization performance of coherent methods. For sound sources near the acoustic channel axis, their energy primarily propagates within the deep-sea acoustic channel, resulting in a high signal-to-noise ratio for data measured by hydrophones at the same depth, but leading to high sidelobes on the ambiguity surface. In this case, phase-coherent matched self-integration processing alone still provides results with a high main-to-sidelobe ratio.

[0129] Figure 4(a) shows the coherent results, and Figure 4(b) shows the phase-only coherent results. Both were normalized using hydrophone number 11 (depth 1050m). Figure 4(a) shows that when the depth of a single hydrophone is close to the depth of the sound source, the blurred surface of the coherent method is almost identical to that of the incoherent method in Figure 3(a), and there is no effect on suppressing sidelobes. However, phase-only coherence achieves the effect of suppressing sidelobes and improving localization resolution by using only phase information.

[0130] Figure 5 The variation of sidelobe values ​​with a normalized hydrophone (peak values ​​are all 1) is shown. Traditional matched-field processing and incoherent matched-self-integration processing lack a single-hydrophone normalization step, resulting in constant sidelobe values. Incoherent matched-self-integration processing, due to its quadratic nonlinearity, produces significantly higher sidelobes than traditional matched-field processing. When the difference between the depth of the single hydrophone and the sound source depth is large, coherent methods can suppress sidelobes. Phase coherence alone provides the best sidelobe suppression.

[0131] Example 2: Experimental Environment in the South China Sea

[0132] Figure 6The South China Sea experiment shown expands the application scenario of the difference frequency method, with the sound source not near the acoustic channel axis. The surface is a flat pressure relief surface, and the bottom consists of a sediment layer and a half-space. The sound source is dragged from the first convergence zone towards the array, propagating a bandwidth signal of 251-350Hz in increments of 1Hz. Traditional matched field processing is used as a basis, and compared with incoherent, coherent, and phase-only coherent matched self-integration processing to simulate two common mismatch types. All difference frequency methods use difference frequencies of 1-5Hz with 1Hz increments, and the coherent methods are normalized using hydrophone number 14. The predicted sound source grid has a distance of 0.2-115km and a depth of 5-500m. Because the ambiguity surface peaks and sidelobes calculated by matched self-integration processing exhibit a periodic structure similar to the sound field, even if the sound source is located within the first convergence zone, the predicted distance range needs to reach near the second convergence zone.

[0133] Array tilt mismatch best demonstrates the robustness enhancement of the difference frequency method by utilizing low-frequency information. The localization results with a tilt angle of 0.1° are shown in Figure 7, where Figure 7(a) shows the distance result and Figure 7(b) shows the depth result. When using a traditional matched field, array tilt has a significant impact on localization, while the localization results of the three difference frequency methods are generally accurate, proving that the coherent method and the phase-only coherent method still maintain the robustness of the difference frequency method. Array tilt leads to the difference between the actual and model propagation paths, resulting in phase difference. Δs represents the acoustic path difference. If this distance is on the same order of magnitude as the effective wavelength, the localization method based on the matching field will fail. However, since the difference frequency method extracts low-frequency information, the effective wavelength λ... e =2πc / Δω is relatively long, so the phase difference becomes Therefore, it is very robust to array tilt mismatch.

[0134] Water depth mismatch is another common type of mismatch. Mismatch is simulated by increasing the percentage of the actual water depth. Figure 8 shows the distance-depth localization results with a mismatch factor of 3%, where Figure 8(a) shows the distance result and Figure 8(b) shows the depth result. Overall, the difference frequency method is more robust than the traditional matched field processing, but its superiority is not as significant as that of array tilt mismatch. Regarding distance, the traditional matched field processing exhibits a situation where the entire localization is misaligned into the second sound shadow region in the shadow region. Incoherent and coherent matched self-integration processing suppress this phenomenon, while phase-coherent localization alone has a higher success rate. Regarding depth, both the traditional matched field processing and the incoherent matched self-integration processing exhibit a localization bias towards the surface in the first sound shadow region. The coherent method has a certain suppressive effect on this phenomenon, while the phase-coherent method alone is generally accurate in localization except for the second convergence region.

[0135] The effectiveness of the invention and its superiority over traditional methods were verified using experimental data from the South China Sea in 2016. In the experiment, the source level was 193dB, the sampling rate was 16000Hz, and a hyperbolic frequency modulated signal was transmitted. The experimental vessel towed the sound source towards the receiving array. Each signal lasted 20 seconds, with a 20-second interval between signals. Four signals were grouped together, with a 50-second interval between each group. The ambiguity surface for sound source localization in the sound shadow zone is shown in Figure 9, where the actual sound source location is marked "○" and the predicted sound source location is marked "×".

[0136] Figure 9(a) shows the localization result of traditional matched field processing, Figure 9(b) shows the result of incoherent matched self-integration processing, Figure 9(c) shows the result of coherent processing, and Figure 9(d) shows the result of phase-only coherence processing. Compared with traditional matched field processing, the incoherent algorithm has extremely high sidelobes, and both methods localize the sound source at the sidelobes. The coherent method has a certain effect in suppressing the sidelobes at the location errors, but it is not enough to achieve successful localization. Although the ambiguity surface of the phase-only coherence method has a lower ambiguity peak value and dynamic range, it successfully localizes the sound source due to its superior sidelobe suppression effect. This may be due to the presence of mismatch and the fact that using only phase information is more sensitive to noise, thus increasing the background level of the ambiguity surface.

[0137] Figure 10 shows the sound source localization results of four algorithms within 58 km of the array in the experiment, where Figure 10(a) shows the distance results and Figure 10(b) shows the depth results. All three difference-frequency algorithms showed stronger localization performance than the traditional matched-field algorithm, with better distance localization results than depth localization. Only the phase-coherent matched self-integration method achieved almost complete localization success in terms of distance. The localization performance of the coherent and incoherent methods was similar. The depth localization performance of all four algorithms was not very accurate, with localization failures mainly occurring in the sound shadow region.

[0138] Success rate, mean absolute percentage error (MAPE), mean absolute error (MAE), and root mean square error (RMSE) were selected as metrics for evaluating positioning results. MAPE was used only for distance error assessment. RMSE, MAE, and success rate were used to evaluate distance and depth errors, respectively. Statistical results are shown in Table 1, where the subscripts r and d represent the parameters calculated for distance and depth, respectively. Successful positioning was categorized into five levels: distance, defined as relative errors of 10%, 12.5%, 15%, 17.5%, and 20%; and depth, defined as absolute errors of 60m, 75m, 90m, 105m, and 120m. Figure 11 It displays different levels of success rate.

[0139] Table 1. Statistical Results of South China Sea Experimental Data Location

[0140]

[0141] All five statistical parameters reflect the positioning error trend of "phase coherence only < coherence < incoherence < traditional matched field", which, together with the positioning success rate trend of "phase coherence only > coherence > incoherence > traditional matched field", demonstrates the superiority of the coherent method over the incoherent method and the best positioning effect of phase coherence only.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for source localization using difference-frequency coherent matched self-integration processing, the method comprising: Step S1) Perform Fourier transform on the sound propagation time domain data received by the hydrophone array, select the required frequency bandwidth and difference frequency, calculate the self-integration vector, and use the coherence of the self-integration in the difference frequency domain to normalize the hydrophone self-integration phase to eliminate the weighting factor of the source spectrum phase. Step S2) Concatenate the normalized self-integral vectors with different difference frequencies to obtain a supervector; Step S3) Calculate the super cross-spectral density matrix using hypervectors, and then perform normalization. Step S4) Calculate the self-integrated copy vector using the sound field model and experimental environment parameters, and obtain the supercopy vector by normalization and splicing. Then normalize the supercopy vector. Step S5) Match the result of step S4) with the super cross-spectral density matrix of step S3) to obtain the ambiguity surface of a single frequency point. Then, by averaging the frequency bandwidth, obtain the final ambiguity surface and obtain the location of the peak of the ambiguity surface to achieve sound source localization.

2. The sound source localization method using difference-frequency coherent matched self-integration processing according to claim 1, characterized in that, Step S1) specifically includes: Perform a Fourier transform on the time-domain data of sound propagation received by the hydrophone array, and calculate the difference-frequency self-integration AP(r,Δω,ω) according to the following formula: AP(r,Δω,ω)≡P * (r,ω-Δω / 2)P(r,ω+Δω / 2) Where ω is the frequency, Δω is the difference frequency, and P(r,ω) is the complex sound pressure level at frequency ω received by the hydrophone at point r, indicated by the superscript (·). * Indicates complex conjugation; The source spectrum amplitude is eliminated by normalizing using the 2-norm, and then the source spectrum phase is eliminated by dividing by the self-integration phase of the x-th hydrophone. The self-integration vector corresponds to the element of the l-th hydrophone. for: Where m, n, and l are the frequency ω, the difference frequency Δω, and the hydrophone number, respectively; 1 ≤ m ≤ M, where M is the total number of frequency points within the frequency bandwidth; 1 ≤ n ≤ N, where N is the total number of difference frequency samples; and 1 ≤ l ≤ L, where L is the total number of hydrophones. Let Green's function be the distance from the actual sound source to the l-th hydrophone. The phase angle is expressed in terms of the Green's function, with the subscript + indicating the high-frequency component in the self-integration, - indicating the complex conjugate of the low-frequency component, and i representing the imaginary unit.

3. The sound source localization method using difference-frequency coherent matched self-integration processing according to claim 2, characterized in that, The frequency of step S2) is ω m Hypervector for: Among them, AP mn This represents the normalized self-integral vector, which is an L×1 dimensional column vector with the superscript (·). T This indicates transpose, and the subscript s indicates the concatenated "hypervector".

4. The sound source localization method using difference-frequency coherent matched self-integration processing according to claim 3, characterized in that, Step S3) specifically includes: Through hypervectors The calculated matrix yields the hypercross-spectral density matrix, which is then normalized using the F-norm to obtain the normalized hypercross-spectral density matrix. Among them, superscript To represent the conjugate transpose, ||·|| F This represents the F-norm.

5. The sound source localization method using difference-frequency coherence matched self-integration processing according to claim 3, characterized in that, Step S4) specifically includes: Calculation using sound field model and experimental environment parameters The copied vector yields a frequency ω. m The difference frequency is Δω n The l-th value of the self-integrating copy vector for: in, To predict the Green's function from the sound source to the l-th hydrophone; The supercopy vector is obtained by concatenating the normalized self-integrated copy vectors with different difference frequencies. According to the following formula Perform 2-norm normalization to obtain the normalized supercopy vector. Where ||·||2 represents the 2-norm.

6. The sound source localization method using difference-frequency coherence matched self-integration processing according to claim 5, characterized in that, Step S5) specifically includes: Calculate the supercopy vector at point r in the predicted sound source grid. hypercopy vector With the hyperspectral density matrix Matching is performed, and then averaging is performed over the entire frequency bandwidth 1≤m≤M to obtain the ambiguity surface B. C,avg (r): Among them, superscript represents the conjugate transpose, the subscript C indicates coherence, and avg indicates bandwidth average; Get B C,avg The location of the peak value of (r) is used to locate the sound source.

7. A difference-frequency coherent matched self-integration processing sound source localization system, characterized in that, The system includes: The self-integration vector calculation module is used to perform Fourier transform on the acoustic propagation time-domain data received by the hydrophone array, select the required frequency bandwidth and difference frequency, calculate the self-integration vector, and use the coherence of the self-integration in the difference frequency domain to normalize the hydrophone self-integration phase and eliminate the weighting factor of the source spectrum phase. The hypervector construction module is used to concatenate normalized self-integral vectors with different difference frequencies to obtain a hypervector; The hyper-mutual spectral density matrix calculation module is used to calculate the hyper-mutual spectral density matrix using hypervectors and then perform normalization processing. The self-integrated copy vector calculation module is used to calculate the self-integrated copy vector using the sound field model and experimental environment parameters, and then normalize and concatenate it to obtain the supercopy vector, which is then normalized again; and The matching and localization module is used to match the result of the self-integrated copy vector calculation module with the super-mutual spectral density matrix of the super-mutual spectral density matrix calculation module to obtain the ambiguity surface of a single frequency point. Then, by averaging the frequency bandwidth, the final ambiguity surface is obtained, and the location of the peak of the ambiguity surface is obtained to realize the sound source localization.

Images