Anti-pulse interference underwater acoustic signal detection method based on intra-cluster abnormal point identification

Abstract

The application relates to a water acoustic signal detection method based on cluster abnormal point identification, and relates to the technical field of water acoustic signal detection.The application is used to solve the problem of low accuracy of existing water acoustic signal detection results.The application comprises the following steps: dividing a water acoustic signal into blocks, constructing a full 1 vector for each block, arranging the blocks to obtain a dictionary matrix P, obtaining the atom most related to the water acoustic signal in the P, and taking the block corresponding to the atom most related to the water acoustic signal as a pulse interference rough positioning area in the water acoustic signal to be detected; performing pulse interference accurate positioning in the pulse interference rough positioning area to obtain a pulse interference cluster abnormal point horizontal coordinate set and a pulse interference cluster abnormal point vertical coordinate set; judging whether the abnormal points in the pulse interference rough positioning area belong to real signal components, and scaling the pulse interference cluster abnormal point vertical coordinates by using the judgment result to suppress the interference pulse; and obtaining a signal detection result by using the water acoustic signal to be detected after the pulse interference is suppressed.The application is used for detecting water acoustic signals.

Description

Technical Field

[0001] This invention relates to the field of underwater acoustic signal detection technology, and in particular to an anti-pulse interference underwater acoustic signal detection method based on intra-cluster anomaly point identification. Background Technology

[0002] Underwater acoustic signal detection systems often operate in environments with pulse interference caused by platform mechanical vibration. This type of pulse interference exhibits heavy-tailed characteristics in probability density similar to common pulse interference (such as shrimp noise), but its time-domain waveform shows a certain time broadening rather than being a transient impulse. Currently, widely used... The stable distribution model (Jiarui Rong, Jingshu Zhang, Huiping Duan. Robust sparse Bayesian learning based on the Bernoulli-Gaussian model of impulsive noise. Digital Signal Processing., 2023;136:1-8.) is based on the instantaneous impulse assumption, which makes it difficult to describe the mechanical vibration pulse interference of a platform with a certain pulse width. As a result, it is difficult to avoid the influence of pulse interference during signal detection. Therefore, how to reduce the impact of pulse interference on signal detection has become a research focus in this field.

[0003] Currently, researchers have proposed various impulse interference suppression techniques. The core idea of ​​existing impulse interference suppression methods is to effectively suppress interference components by identifying and processing abnormal amplitude points in the received signal. Typical operations include clipping and nulling. A widely adopted classic method is to use a median filter to suppress impulse interference (LiuJiaao, Han Xiao, Zhu Guangjun, et al. Impulsive noise suppression for underwater acoustic OFDM communication based on adaptive median filter.Shengxue Xuebao., 2025;50(4):1031-1041.). The median filter uses the median of the signal amplitude in a local window to replace the abnormal points, which significantly improves the robustness of signal detection in the impulse noise environment. However, this method has obvious limitations when the impulse interference and the signal to be detected overlap in the time domain. On the one hand, median filtering will suppress the useful signal energy falling into the window, resulting in distortion of the signal to be detected, and ultimately leading to low accuracy of underwater acoustic signal detection results. On the other hand, when the pulse interference has a certain pulse width, conventional median calculation is difficult to accurately identify the boundary between the signal and the pulse interference. As a result, some pulse interference is not identified as an abnormal signal and is not processed, making it difficult to effectively filter out pulse interference and resulting in low accuracy of underwater acoustic signal detection results. Summary of the Invention

[0004] This invention addresses the problem of low accuracy in existing underwater acoustic signal detection results by proposing an anti-pulse interference underwater acoustic signal detection method based on intra-cluster anomaly point identification.

[0005] A method for detecting underwater acoustic signals against impulse interference based on intra-cluster anomaly identification includes the following steps:

[0006] Step 1: Acquire the underwater acoustic signal to be measured Initialize the iteration label Initialize residuals Initialize the selected atom index set Initialize the selected atom set to an empty set;

[0007] Step 2, for Perform non-overlapping block partitioning, construct an all-one vector for each block, and arrange them according to their positions to obtain a dictionary matrix. Treat each column vector in P as an atom, and then obtain the dictionary matrix. Zhongyu The most relevant atoms will be with The block corresponding to the most relevant atom is used as the coarse localization region for pulse interference in the underwater acoustic signal to be measured, and will be compared with... The most relevant atom labels are added to the selected atom index set. , will with The most relevant atom is added to the selected atom set;

[0008] Step 3: Perform precise pulse interference localization within the coarse localization area to obtain the set of anomalous points in the pulse interference cluster. and the set of ordinates of abnormal points in the pulse interference cluster ;

[0009] Step 4: Utilize the set of anomalous points in the pulse interference cluster by their x-coordinates. Determine whether anomalies in the coarse localization area of ​​pulse interference belong to real signal components, and use the determination results to... The medium amplitude value is scaled to suppress interference pulses, and then step five is executed;

[0010] Step 5: Use the scaled amplitude value from Step 4 to determine whether to terminate pulse interference suppression. If pulse interference suppression is terminated, output the underwater acoustic signal under test after pulse interference suppression. Otherwise, let The underwater acoustic signal under test after pulse interference suppression is used as the residual. And return to step two;

[0011] Step 6: Suppress the underwater acoustic signal to be measured using pulse interference. Obtain the signal detection results.

[0012] Furthermore, the step two... Perform non-overlapping block partitioning, construct an all-one vector for each block, and arrange them according to their positions to obtain a dictionary matrix. Treat each column vector in P as an atom, and then obtain the dictionary matrix. Zhongyu The most relevant atoms will be with The block corresponding to the most relevant atom is used as the coarse localization region for pulse interference in the underwater acoustic signal to be measured, and will be compared with... The most relevant atom labels are added to the selected atom index set. , will with The most relevant atoms are added to the selected atom set, specifically:

[0013] Step 21: Length of transmitted signal As the block length, Divide into multiple non-overlapping blocks, and then divide the contents of a block... The values ​​of each sampling point are replaced with 1, and all other sampling points outside the current block are set to zero, according to each block. The index positions in the dictionary are arranged into a lexicographical matrix. Specifically:

[0014]

[0015] in, It is a dimension of A vector of all 1s Each column in the array is a block, and each block represents an atom.

[0016] Step 22: Obtain the dictionary matrix Neutral and residual The index corresponding to the most relevant block ,Will The corresponding block serves as the coarse localization area for pulse interference.

[0017] Steps two and three: Set the selected atom indexes together... The corresponding block is added to the selected atom set, and then the dictionary matrix is... middle Set the corresponding block vector to zero.

[0018] Furthermore, the process of obtaining the dictionary matrix in step two-two... Neutral and residual The index corresponding to the most relevant block Specifically:

[0019]

[0020] in, In the dictionary matrix The selected first One atom, This represents the dot product of two vectors. These are atomic labels. This indicates taking the modulus.

[0021] Furthermore, in step three, precise pulse interference localization is performed within the coarse localization region to obtain the set of abscissas of abnormal points in the pulse interference cluster. and the set of ordinates of abnormal points in the pulse interference cluster Specifically:

[0022] Step 3: 1. Initialize the cluster center as the point with the largest amplitude value in the coarse localization region of the pulse interference, and initialize the cluster center iteration label. ;

[0023] Step 3.2: In a two-dimensional space composed of time and amplitude, using a rectangular window of a preset size, search for the set of neighboring points of the cluster center in the coarse localization region of the impulse interference, and obtain the set of time values ​​in the set of neighboring points. and amplitude value set ;

[0024] in, It is the first The set of time values ​​of cluster center neighbors in round iteration. yes The first in One value, It is the first The set of amplitude values ​​of cluster center neighbors in round iteration. yes The first in One value, It is the total number of neighboring points of the cluster center;

[0025] Step 33, Utilize and Obtain the local density distribution within the rectangular window, update the cluster center position using the local density distribution within the rectangular window, and then update the cluster center position. The x-axis includes the set of x-axis coordinates of anomaly points in the pulse interference cluster. The updated density center The ordinate includes the set of ordinates of anomaly points in the pulse interference cluster. ;

[0026] Steps 3 and 4: Determine whether the updated cluster center position is the same as the original cluster center position. If... Then the set of x-coordinates of the abnormal points of the output pulse interference and the set of vertical coordinates of pulse interference anomalies Otherwise, let Then return to step three two.

[0027] Furthermore, the utilization in step three is... and Obtain the local density distribution within the rectangular window, update the cluster center position using the local density distribution within the rectangular window, and then update the cluster center position. The x-axis includes the set of x-axis coordinates of anomaly points in the pulse interference cluster. The updated density center The ordinate includes the set of ordinates of anomaly points in the pulse interference cluster. Specifically:

[0028] First, obtain the data distribution density of the current cluster center, specifically:

[0029]

[0030]

[0031] in, It is an indicator parameter. Pick or , Indicates the time dimension. Indicates the magnitude dimension. Is the cluster center in dimension The value on, It is a data dimension Cluster center data distribution density, It is the total number of points near the cluster center. It is the cluster center neighbor point label. It is a data dimension The first in the set of neighboring points of the upper cluster center A number, when Pick hour, for ,when Pick hour, for , These are sliding window parameters, when Pick hour, For the time-dimension sliding window parameters, when Pick hour, For the sliding window parameters in the amplitude dimension, It is a data dimension; It is a Gaussian kernel function;

[0032] Then, based on the data distribution density, the cluster centers are moved to areas with higher density, thereby updating the cluster center positions and obtaining the updated cluster center positions. ;

[0033] Finally, the updated cluster center The x-axis includes the set of x-axis coordinates of anomaly points in the pulse interference cluster. The updated cluster center The ordinate includes the set of ordinates of anomaly points in the pulse interference cluster. .

[0034] Furthermore, the cluster center position is updated by moving it to a higher density region based on the data distribution density, thus obtaining the updated cluster center position. Specifically:

[0035]

[0036] in, This is the updated cluster center location. Dimension The value on.

[0037] Furthermore, in step four, the set of abscissas of abnormal points in the pulse interference cluster is used. Determine whether anomalies in the coarse localization area of ​​pulse interference belong to real signal components, and use the determination results to... The medium amplitude value is scaled as follows:

[0038]

[0039] in, It is scaled up , It is the average amplitude of all sampling points within the coarse localization area of ​​pulse interference, excluding anomaly points. It means The maximum value in, It means The minimum value in.

[0040] Furthermore, in step five, the amplitude value scaled up in step four is used to determine whether to terminate pulse interference suppression. If pulse interference suppression is terminated, the underwater acoustic signal under test after pulse interference suppression is output. Otherwise, let The underwater acoustic signal under test after pulse interference suppression is used as the residual. Then return to step two, which is as follows:

[0041] First, the amplitude set of sampling points in the coarse localization region of the pulse interference after interference suppression is obtained using the scaled amplitude values ​​from step four. ;

[0042] in, This represents the total number of sampling points in the coarse localization area of ​​pulse interference.

[0043] Then, obtain the amplitude set of sampling points in the coarse localization region of the pulse interference before this round of interference suppression. The amplitude set of sampling points in the coarse localization region of pulse interference after the previous round of interference suppression. , obtain and Local variance , Specifically:

[0044]

[0045]

[0046]

[0047]

[0048] in, , It is the index of sampling points in the coarse localization region of pulse interference. It is the total number of sampling points in the coarse localization area of ​​the pulse interference. It is the first After suppressing wheel interference, the pulse interference coarse localization area is the first The amplitude of each sampling point It is the first Mean amplitude of sampling points in the coarse localization region of pulse interference after wheel interference suppression It is the first Local variance of the coarse localization region for pulse interference after wheel interference suppression. It is the first Wheel interference suppression before pulse interference coarse localization region The amplitude of each sampling point It is the first The mean amplitude of sampling points in the coarse localization region before wheel interference suppression. It is the first Local variance of the coarse localization region before wheel interference suppression and pulse interference;

[0049] Finally, compare and The size, if Then the pulse interference suppression process is terminated, and the underwater acoustic signal under test after pulse interference suppression is output. Otherwise, let The underwater acoustic signal under test after pulse interference suppression is used as the residual. Then return to step two.

[0050] Furthermore, in step six, the underwater acoustic signal under test is suppressed using pulse interference. The signal detection results are obtained as follows:

[0051] Step 61: Suppress the underwater acoustic signal to be measured using pulse interference. The channel impulse response is obtained as follows:

[0052] The orthogonal matching pursuit algorithm is used to solve the sparse channel estimation optimization problem and obtain the channel impulse response;

[0053] The sparse channel estimation optimization problem is specifically as follows:

[0054]

[0055]

[0056] in, It is the first The underwater acoustic signal under test after pulse interference suppression at each sampling time. It is the first Signal is transmitted at each sampling time. It is the first Signal is transmitted at each sampling time. The signal is transmitted at the first sampling time. For sparsity parameters, For channel impulse response, It is the first The underwater acoustic signal under test after pulse interference suppression at each sampling time. It is a dictionary matrix;

[0057] Step 62: Obtain the reconstructed signal using the channel impulse response, and then acquire the detection statistics. Detection statistics With preset threshold value In comparison, if This indicates the underwater acoustic signal to be measured. The presence of a transmitted signal indicates that the underwater acoustic signal being measured is present; otherwise, it indicates that the signal is present. There is no transmitted signal.

[0058] Furthermore, the detection statistics Specifically:

[0059]

[0060] in, It's a transpose operation. It is a detection statistic. It is the normalization factor.

[0061] The beneficial effects of this invention are as follows:

[0062] This invention proposes a hierarchical processing scheme for pulse interference signals generated by the mechanical vibration of underwater platforms. First, it quickly locates the underwater acoustic signal segments suspected of interference. Then, it uses a mean-shifting method to accurately identify and eliminate anomalies within the pulse interference cluster. Finally, based on the suppression results of the anomalies within the pulse interference cluster, it completes the final detection of the underwater acoustic signal. This invention suppresses pulse noise while maximizing the preservation of the integrity of the useful signal, avoiding distortion of the underwater acoustic signal to be detected. Furthermore, this invention can accurately identify the boundary between the signal and the pulse interference, improving the accuracy of pulse interference identification, thereby effectively eliminating pulse interference and further improving the accuracy of the underwater acoustic signal detection results. Attached Figure Description

[0063] Figure 1 This is a flowchart of the present invention;

[0064] Figure 2 The probability distribution model values ​​for various types of impulse interference are shown in the diagram.

[0065] Figure 3 Comparison diagram of time-domain impulse interference models;

[0066] Figure 4 This is a diagram showing the coarse localization results of the pulse interference.

[0067] Figure 5 This is a diagram showing the results of identifying abnormal points within a pulse interference cluster.

[0068] Figure 6 This is the signal before pulse interference suppression;

[0069] Figure 7 This is the signal after pulse interference suppression;

[0070] Figure 8 This is a diagram showing the channel estimation results of the present invention;

[0071] Figure 9 The image shows the original signal channel estimation results.

[0072] Figure 10 ROC curves for different methods at different signal-to-noise ratios;

[0073] Figure 11 This is a signal-to-noise ratio versus detection probability curve.

[0074] Figure 12 This is a graph of the measured underwater acoustic signal data before pulse interference suppression;

[0075] Figure 13 This is a graph showing the measured underwater acoustic signal data after pulse interference suppression. Detailed Implementation

[0076] Specific implementation method one: as follows Figure 1 The specific process of the anti-pulse interference underwater acoustic signal detection method based on intra-cluster anomaly identification in the seven embodiments shown is as follows:

[0077] Step 1: Acquire the underwater acoustic signal to be measured Initialize the iteration label Initialize residuals Initialize the selected atom index set Initialize the selected atom set to an empty set and set the sparsity to 1;

[0078] Signal detection is typically modeled as a binary hypothesis model, as follows:

[0079]

[0080] in, , These correspond to scenarios where the transmitted signal is absent and scenarios where it is present, respectively. It is the sampling time. , It is the received signal vector. It is the total length of the received signal. It is the first The received signal at each sampling time, , It is the pulse interference vector. It is the first Pulse interference at each sampling time, , It is a Gaussian noise vector. It is the first Gaussian noise at each sampling time , The signal is obtained after the transmitted signal propagates through the underwater acoustic channel. It is the first The signal is obtained after the transmitted signal at each sampling time propagates through the underwater acoustic channel;

[0081] The pulse interference of this invention refers to pulse interference with a certain pulse width caused by the mechanical vibration of the underwater acoustic platform.

[0082] exist Under the assumption that the received signal can be represented as the superposition of the transmitted signal propagating through the underwater acoustic channel, impulse interference, and Gaussian background noise, where the transmitted signal, after propagation through the underwater acoustic channel, becomes the received signal. ;

[0083]

[0084] in, , It is a dictionary matrix. It is the total length of the transmitted signal. , yes The channel tap coefficients of each atom in the middle;

[0085] In this step, Each column corresponds to one atom; the composition of the acquired underwater acoustic signal is unknown. In subsequent steps, this invention determines whether there is a transmission signal and eliminates pulse interference.

[0086] Step 2: Use the BOMP algorithm to... Perform non-overlapping block partitioning, construct an all-one vector for each block, and arrange them according to their positions to obtain a dictionary matrix. Treat each column vector in P as an atom, and then obtain the dictionary matrix. Zhongyu The most relevant atoms will be with The block corresponding to the most relevant atom is used as the coarse localization region for pulse interference in the underwater acoustic signal to be measured, and will be compared with... The most relevant atom labels are added to the selected atom index set. , will with The most relevant atoms are added to the selected atom set, specifically:

[0087] Step 21, As the block length, Divide into multiple non-overlapping blocks, and then divide the contents of a block... The values ​​of each sampling point are replaced with 1, and all other sampling points outside the current block are set to zero. The blocks are then arranged into a dictionary matrix according to their index positions in the original signal. In the corresponding column, the specific form is:

[0088]

[0089] in, It is a dimension of A vector of all 1s Each column in the array represents one atom. It is the floor function;

[0090] Step 22: Using the dictionary matrix obtained in Step 21 Construct a sparse reconstruction problem, solve it, and obtain the dictionary matrix. The middle and current residuals The index of the most relevant atom ,Will The corresponding atoms are used as the coarse localization region for pulse interference, specifically:

[0091] First, we construct the sparse reconstruction problem, specifically:

[0092]

[0093] in, , It is a sparse coefficient vector;

[0094] Then, the sparse reconstruction problem is solved to obtain the dictionary matrix. The middle and current residuals The index of the most relevant atom Specifically:

[0095]

[0096] in, In the dictionary matrix The selected first One atom, This represents the dot product of two vectors. These are atomic labels. Indicates modulo;

[0097] Steps two and three: Add to the selected atom index set, Add the corresponding atom to the selected atom set, and then process the dictionary matrix. middle Set the corresponding atom to zero.

[0098] In this step, the index of the strongest remaining impulse interference in the signal is determined in the dictionary, and then the atom with the corresponding index is set to zero in the dictionary matrix to avoid the problem of repeated selection in subsequent iterations. Figure 4 The results of coarse localization of impulse interference are shown, and the proposed method can accurately identify the impulse interference components with larger amplitudes.

[0099] For clustered pulse interference, conventional median filtering methods require a trade-off between pulse interference suppression and waveform detail preservation. Furthermore, these methods cannot identify the fine structure of the pulse interference, making it difficult to effectively distinguish between pulse interference and useful signal sampling points. This invention designs a two-level structured anomaly point identification method within pulse interference clusters. Through a progressively refined detection strategy, it achieves accurate identification of interference signals, ultimately enabling efficient and reliable interference suppression. The block orthogonal matching pursuit algorithm, as a type of compressed sensing algorithm, can identify signal components that are clustered and have certain structural priors, while also possessing the advantage of low computational complexity.

[0100] Step 3: Use the mean-shift algorithm to perform precise pulse interference localization in the coarse localization area of ​​pulse interference, and obtain the set of abscissas of anomaly points in the pulse interference cluster. and the set of ordinates of anomaly points in the pulse interference cluster Specifically:

[0101] Step 3: 1. Initialize the cluster center as the point with the largest amplitude value in the coarse localization region of the pulse interference, and initialize the cluster center iteration label. ;

[0102] Step 3.2: In a two-dimensional space composed of time and amplitude, using a rectangular window of a preset size, search for the set of neighboring points of the cluster center in the coarse localization region of the impulse interference, and obtain the set of time values ​​in the set of neighboring points. and amplitude value set ;

[0103] in, It is the first The set of time values ​​of cluster center neighbors in round iteration. yes The first in One value, It is the first The set of amplitude values ​​of cluster center neighbors in round iteration. yes The first in One value, It is the total number of neighboring points of the cluster center;

[0104] Step 33, Utilize and Obtain the local density distribution within the rectangular window, update the cluster center position using the local density distribution within the rectangular window, and then update the cluster center position. The x-axis includes the set of x-axis coordinates of anomaly points in the pulse interference cluster. The updated density center The ordinate includes the set of ordinates of abnormal points in the pulse interference cluster. Specifically:

[0105] First, obtain the data distribution density at the current cluster center, specifically:

[0106]

[0107]

[0108] in, It is an indicator parameter. Pick or , Indicates the time dimension. Indicates the magnitude dimension. Is the cluster center in dimension The value, It is a data dimension Cluster center data distribution density, It is the total number of points near the cluster center. It is the cluster center neighbor point label. It is a data dimension The first in the set of neighboring points of the upper cluster center A number, when Pick hour, for ,when Pick hour, for , These are sliding window parameters, when Pick hour, For the time-dimension sliding window parameters, when Pick hour, For the sliding window parameters in the amplitude dimension, It is a data dimension; It is a Gaussian kernel function;

[0109] This step, The kernel function determines the search range of the mean shift algorithm. This defines the degree of influence of each data point within the window on the current center point, and this degree decreases as the distance from the center point increases. Currently, the center of the data cluster in this round... This step calculates the drift vector for each data point, which indicates the rising direction of the probability density gradient within the local window.

[0110] Then, based on the data distribution density, the cluster centers are moved to areas with higher density, and the cluster center positions are updated using the mean shift vector. The updated cluster center positions are... It can be obtained through the following formula:

[0111]

[0112] in, This is the updated cluster center location. Dimension The value on;

[0113] Finally, the updated cluster center The x-axis includes the set of x-axis coordinates of anomaly points in the pulse interference cluster. The updated cluster center The ordinate includes the set of ordinates of abnormal points in the pulse interference cluster. ;

[0114] Steps three and four: Determine whether the updated cluster center position is the same as the original cluster center position. If... Then the set of x-coordinates of the abnormal points of the output pulse interference and the set of vertical coordinates of pulse interference anomalies Otherwise, let Then return to step three two.

[0115] In this step, given the inconsistency in the dimensions of time and amplitude coordinates in the time domain signal, this paper adopts a rectangular search window instead of the traditional circular search area, thereby enabling more independent adjustment and optimization of the two dimensions.

[0116] Step 4: Utilize the set of abscissas of anomaly points in the pulse interference cluster Determine whether anomalies in the coarse localization area of ​​pulse interference belong to real signal components, and use the determination results to... The medium amplitude value is scaled to suppress interference pulses, and then step five is executed, specifically:

[0117]

[0118] in, It is scaled up , It is the average amplitude of all sampling points within the coarse localization area of ​​pulse interference, excluding anomaly points. It means The maximum value in, It means The minimum value in;

[0119] In this step, given that the pulse interference signal may overlap with the actual received signal in the time domain, to effectively distinguish between the two and avoid unnecessary reduction of the actual signal energy during suppression, this step utilizes the characteristic that the pulse width of the pulse interference is much smaller than that of the actual signal, and uses the time span of the extracted anomaly point set distributed on the horizontal axis for discrimination. If this time span is greater than a preset threshold... If the outlier in the cluster is determined to be a real signal component, no further processing is required. Similarly, to avoid excessive reduction of the real signal energy, this step does not use the conventional method of directly setting the outlier to zero. Instead, it calculates the average amplitude of the remaining sampling points in the block (excluding the outlier) as a reference, and uses the ratio of the largest outlier to this average as a uniform scaling factor to proportionally scale all outliers in the cluster. Figure 5 This demonstrates the effectiveness of anomaly identification within a cluster of impulse interference. In the simulated scenario, the impulse interference overlaps with the useful signal. Figure 5 As can be seen, even if the amplitude of the pulse interference is comparable to that of the useful signal, the present invention can still accurately distinguish between the two, identifying only the interference sampling points as abnormal, without misjudging the sampling points of the useful signal.

[0120] Step 5: Use the scaled amplitude value from Step 4 to determine whether to terminate pulse interference suppression. If pulse interference suppression is terminated, output the underwater acoustic signal under test after pulse interference suppression. Otherwise, let The underwater acoustic signal under test after pulse interference suppression is used as the residual. Then return to step two, which is as follows:

[0121] First, the amplitude set of sampling points in the coarse localization region of the pulse interference after interference suppression is obtained using the scaled amplitude values ​​from step four. ;

[0122] in, This represents the total number of sampling points in the coarse localization region of the pulse interference, and the set of amplitudes of the sampling points in the coarse localization region of the pulse interference before interference suppression. The set of amplitude values ​​of sampling points in the coarse localization region of pulse interference after interference suppression is as follows: The set of amplitude values ​​of the sampling points after impulse interference suppression includes the amplitude values ​​scaled in step four and the amplitude values ​​that do not require scaling.

[0123] Then, obtain the amplitude set of sampling points in the coarse localization region of the pulse interference before this round of interference suppression. The amplitude set of sampling points in the coarse localization region of pulse interference after the previous round of interference suppression. , obtain and Local variance , Specifically:

[0124]

[0125]

[0126]

[0127]

[0128] in, , It is the index of sampling points in the coarse localization region of pulse interference. It is the total number of sampling points in the coarse localization area of ​​the pulse interference. It is the first After suppressing wheel interference, the pulse interference coarse localization area is the first The amplitude of each sampling point It is the first Mean amplitude of sampling points in the coarse localization region of pulse interference after wheel interference suppression It is the first Local variance of the coarse localization region for pulse interference after wheel interference suppression. It is the first Wheel interference suppression before pulse interference coarse localization region The amplitude of each sampling point It is the first The mean amplitude of sampling points in the coarse localization region before wheel interference suppression. It is the first Local variance of the coarse localization region before wheel interference suppression and pulse interference;

[0129] Finally, compare and The size, if Then the pulse interference suppression process is terminated, and the underwater acoustic signal under test after pulse interference suppression is output. Otherwise, let The underwater acoustic signal under test after pulse interference suppression is used as the residual. And return to step two;

[0130] The underwater acoustic signal under test after pulse interference suppression includes the block after pulse interference suppression and the block without other processing.

[0131] In this step, given the randomness of the occurrence of pulse interference signals and the unknown number of pulse interference signals in the signal to be processed, this invention introduces local variance as the criterion for terminating the iteration, ensuring that the algorithm can automatically terminate after all pulse interference has been accurately searched. If the local variance of the current block before pulse interference suppression is lower than the local variance after pulse interference suppression in the previous iteration, it indicates that there is no significant pulse interference in the block, and therefore no further processing is required. Figure 6 , Figure 7 The overall received signal characteristics before and after impulse interference suppression are shown separately. A comparison reveals that... Figure 6 Most of the larger amplitude pulse interference in the middle Figure 7 The interference has been removed, leaving only a small amount of interference components with an amplitude close to that of the useful signal. This result fully verifies the pulse interference suppression effect of the present invention.

[0132] Step 6: Suppress the underwater acoustic signal to be measured using pulse interference. The signal detection results are obtained as follows:

[0133] Step 61: Suppress the underwater acoustic signal to be measured using pulse interference. The channel impulse response is obtained as follows:

[0134] First, the underwater acoustic signal under test after pulse interference suppression. Modeling, specifically:

[0135]

[0136] in, It is the first The underwater acoustic signal under test after pulse interference suppression at each sampling time. It is the first The signal is obtained after the transmitted signal at each sampling time propagates through the underwater acoustic channel. It is the first Gaussian noise at each sampling time;

[0137] Then, a sparse channel estimation optimization problem is constructed, and the orthogonal matching pursuit algorithm is used to solve the sparse channel estimation optimization problem to obtain the channel impulse response, specifically:

[0138]

[0139] in, For sparsity parameters, For channel impulse response, When the objective function is minimized Values, It is the L2 norm. yes The number of non-zero elements in the array;

[0140] In this step, the orthogonal matching pursuit algorithm first iteratively... The most relevant atoms to the residual signal are selected, and these atoms are used to gradually approximate the original signal to estimate the channel impulse response. Based on this, the estimated channel impulse response is convolved with the original clean signal to reconstruct the received signal under multipath conditions. To address the issue of unknown sparsity in channel estimation, this paper uses the BIC criterion as the termination criterion for the iterative process. This method seeks a balance between model fit and complexity, achieving adaptive estimation of sparsity. The orthogonal matching pursuit algorithm has good channel estimation capabilities under Gaussian noise, but its performance deteriorates significantly in non-Gaussian impulse interference environments. After employing the impulse interference suppression method and inverse preprocessing proposed in this paper, impulse interference in the received signal is effectively suppressed. Applying the orthogonal matching pursuit algorithm then yields significantly better channel estimation results than directly processing the original signal with impulse interference. Figure 8 , Figure 9 The channel estimation performance under conditions without impulse interference suppressor is demonstrated.

[0141] Step 62: Obtain the reconstructed signal using the channel impulse response, and then acquire the detection statistics. Detection statistics With preset threshold value In comparison, if This indicates the underwater acoustic signal to be measured. The presence of a transmitted signal indicates that the underwater acoustic signal being measured is present; otherwise, it indicates that the signal is present. There is no transmitted signal.

[0142] The channel impulse response detection statistics are obtained through the following method:

[0143]

[0144] in, It's a transpose operation. It is a detection statistic. It is a normalization factor;

[0145] Among them, the normalization factor The received signal can be adjusted to a stable amplitude range, allowing for subsequent signal processing on a relatively uniform scale. If the assumption holds true, a false alarm occurs if the detection statistic exceeds a preset threshold, and the false alarm probability is: .exist When the assumption holds true, if the detection statistic exceeds a certain threshold, it is considered a correct detection, and the detection probability is... .

[0146] This step can fully tap into and utilize the multipath energy of the signal to effectively accumulate gain, thereby achieving a significant improvement in detection performance compared to traditional methods.

[0147] Example: To verify the beneficial effects of the present invention, the following experiments were conducted in this example:

[0148] Model the mechanical vibration pulse interference of underwater platforms and simulate and generate the pulse interference;

[0149] Underwater pulse interference signals, such as shrimp noise signals, typically exhibit a "heavy-tailed" probability distribution, hence they are often used... A stable distribution describes it. However, The process of generating noise interference in a stable distribution is considered an independent and identically distributed process, which can only generate isolated, instantaneous spikes. The pulse interference caused by the mechanical vibration of the underwater platform, which this invention focuses on, oscillates continuously over a period of time, making... Stable distributions are insufficient for characterizing the time-domain waveform structure of impulse interference signals. This invention proposes a composite probability density function model, which models the occurrence time and amplitude of impulse interference signals as Bernoulli-Gaussian distributions to describe their randomness, suddenness, and impulsiveness, while modeling the duration of the impulse interference as a Rayleigh distribution to reflect its temporal continuity.

[0150] First of all, let This indicates the probability of pulse interference occurring, and correspondingly, This represents the probability of no pulse interference. Based on this, if a pulse does occur, its signal amplitude follows a mean. Zero, variance is The Gaussian distribution.

[0151]

[0152]

[0153] in, It is a random amplitude. It is the amplitude of the pulse interference. It is an indicator function;

[0154] Then, the Rayleigh distribution is a strictly non-negative distribution, with values ​​always being positive. This property is very suitable for describing the duration of impulse interference signals. Let... Let the scale parameter of the Rayleigh distribution be denoted by the probability distribution function:

[0155]

[0156] in, It is a random pulse width. It is the pulse interference width;

[0157] The probability of a pulse occurring within the pulse duration. The value is always 1, which means that the pulse interference during this period still follows a Gaussian distribution, but its intensity is much greater than the background noise.

[0158] The pulse interference generated using the above model not only retains the "heavy-tailed" distribution characteristics of underwater pulse interference, but also achieves a much more accurate representation of the waveform structure in the time domain than traditional models. To verify the accuracy of the proposed model, this embodiment compares and evaluates the model from both probability density and time-domain waveform perspectives. First... Figure 2 The probability density function curves of the measured data and the simulated signals generated according to different distributions were plotted. Figure 2 The probability density of the Gaussian distribution decreases sharply in the large-amplitude region, while the proposed model and... The stable distribution models all closely match the distribution of real underwater pulse interference data, and both exhibit obvious "heavy-tailed" statistical characteristics. Next, we compare the time-domain waveform characteristics of the measured data with those of signals generated by different distributions. Figure 3 (a) shows the pulse interference caused by the measured mechanical vibration of an underwater platform. This type of interference exhibits typical cluster characteristics, that is, in the time domain, it manifests as a transient pulse train with violent oscillations over a short period of time, and the signal amplitude is significantly higher than the ambient background noise. Figure 3 As shown in (b) above, based on Stable distributed pulse interference exhibits a series of isolated impulse sampling points in the time domain. This is because each sampling point is statistically independent during simulation, resulting in a lack of continuity in its signal structure, which significantly differs from the clustered characteristics of the pulse interference signal studied in this invention. In contrast, the pulse interference signal generated using the model proposed in this invention exhibits a more consistent waveform in the time domain (e.g., ...). Figure 3 The cluster structure and energy accumulation phenomenon of this type of pulse interference are reproduced in (c) shown in the figure.

[0159] Next, a simulation analysis was performed on the pulse interference suppression method of the present invention, with a sampling rate of 60 kHz, a transmitted signal frequency band of 5 kHz to 10 kHz, and a transmitted signal pulse width of 0.01 s. The probability of pulse interference occurrence was set... The scale parameter of the Rayleigh distribution is ;

[0160] This embodiment compares the performance of a combination of median and matched filters with the impulse interference suppression algorithm of this invention. The comparison algorithm first uses a median filter to suppress impulse interference in the received signal, and then completes the signal detection process using a matched filter. This embodiment uses detection probability as the performance evaluation index in the simulation. In the simulation experiment, the signal-to-interference ratio (SNR) is fixed at -30dB, and the changes in detection probability under different SNR conditions are analyzed. First, the ROC curves are calculated for SNRs of -20dB, -18dB, and -16dB, respectively, and then the detection performance of each algorithm is analyzed. 300 Monte Carlo experiments are performed under each SNR condition. Figure 10 The performance comparison results show that the detection performance of both algorithms improves with increasing signal-to-noise ratio (SNR). Under the same SNR conditions, the ROC curve of this invention is consistently superior to the comparison algorithm. In the application scenario targeted by this invention, the signal detection performance mainly depends on the effectiveness of impulse interference suppression. When facing clustered impulse interference, traditional sliding window median filtering methods for individual sampling points can only replace the current point based on local information within the window, essentially processing point by point, thus lacking the ability to distinguish between real signals and impulse interference anomalies. This invention, however, can identify all abnormal sampling points in the entire interference cluster at once, thereby effectively distinguishing real signals from interference using the signal structure in the time domain, achieving accurate identification and targeted suppression of anomalies while preserving the real signal. To quantitatively evaluate the detection performance of each algorithm, this embodiment uses a constant false alarm rate (CFAR) detector. With the false alarm probability fixed at 0.01, the detection probabilities corresponding to different SNRs are calculated, and 300 Monte Carlo simulations are performed for each SNR condition. Figure 11 As shown, the algorithm with interference suppression processing exhibits an increasing detection probability as the signal-to-noise ratio (SNR) increases, while the present invention demonstrates significant performance advantages across the entire SNR range. Especially under a low SNR condition of -16dB, the detection probability of the present invention reaches 97.3%, fully demonstrating its excellent performance in non-Gaussian environments.

[0161] Next, the performance of the present invention will be verified using the measured data from the Songhua Lake test in November 2024. Figure 12 The waveform of the received signal after bandpass filtering is shown, and Figure 13 This demonstrates the effect of using the pulse interference suppression processing of the present invention. From Figure 12-13 The results show that the present invention effectively suppresses outliers in pulse interference signals while perfectly preserving useful signal components. This result verifies that the present invention achieves a good balance between pulse interference suppression and waveform detail preservation.

[0162] The experiment in this embodiment verifies that the pulse interference model established in this embodiment closely approximates the characteristics of clustered pulse signals caused by real underwater mechanical vibration. The pulse interference suppression method of this invention is effective in handling such interference. This advantage is directly reflected in the performance comparison of the subsequent detection stage. That is, under the constant false alarm detection condition with a false alarm probability set to 0.01, this invention achieves a 4dB performance improvement in detection performance compared to the traditional algorithm.

Claims

1. A method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification, characterized in that... The specific process of the method is as follows: Step 1: Acquire the underwater acoustic signal to be measured Initialize the iteration label Initialize residuals Initialize the selected atom index set Initialize the selected atom set to an empty set; Step 2, for Perform non-overlapping block partitioning, construct an all-one vector for each block, and arrange them according to their positions to obtain a dictionary matrix. ,Will Each column vector is treated as an atom, and then the dictionary matrix is ​​obtained. Zhongyu The most relevant atoms will be with The block corresponding to the most relevant atom is used as the coarse localization region for pulse interference in the underwater acoustic signal to be measured, and will be compared with... The most relevant atom labels are added to the selected atom index set. , will with The most relevant atom is added to the selected atom set; Step 3: Perform precise pulse interference localization within the coarse localization area to obtain the set of anomalous points in the pulse interference cluster. and the set of ordinates of abnormal points in the pulse interference cluster ; Step 4: Utilize the set of anomalous points in the pulse interference cluster by their x-coordinates. Determine whether anomalies in the coarse localization area of ​​pulse interference belong to real signal components, and use the determination results to... The medium amplitude value is scaled to suppress interference pulses, and then step five is executed; Step 5: Use the scaled amplitude value from Step 4 to determine whether to terminate pulse interference suppression. If pulse interference suppression is terminated, output the underwater acoustic signal under test after pulse interference suppression. Otherwise, let The underwater acoustic signal under test after pulse interference suppression is used as the residual. And return to step two; Step 6: Suppress the underwater acoustic signal to be measured using pulse interference. Obtain the signal detection results.

2. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 1, characterized in that: The second step is to Perform non-overlapping block partitioning, construct an all-one vector for each block, and arrange them according to their positions to obtain a dictionary matrix. Treat each column vector in P as an atom, and then obtain the dictionary matrix. Zhongyu The most relevant atoms will be with The block corresponding to the most relevant atom is used as the coarse localization region for pulse interference in the underwater acoustic signal to be measured, and will be compared with... The most relevant atom labels are added to the selected atom index set. , will with The most relevant atoms are added to the selected atom set, specifically: Step 21: Length of transmitted signal As the block length, Divide into multiple non-overlapping blocks, and then divide the contents of a block... The values ​​of each sampling point are replaced with 1, and all other sampling points outside the current block are set to zero, according to each block. The index positions in the dictionary are arranged into a lexicographical matrix. Specifically: in, It is a dimension of A vector of all 1s Each column in the array is a block, and each block represents an atom. Step 22: Obtain the dictionary matrix Neutral and residual The index corresponding to the most relevant block ,Will The corresponding block serves as the coarse localization area for pulse interference. Steps two and three: Set the selected atom indexes together... The corresponding block is added to the selected atom set, and then the dictionary matrix is... middle Set the corresponding block vector to zero.

3. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 2, characterized in that: The step 22 involves obtaining the dictionary matrix. Neutral and residual The index corresponding to the most relevant block Specifically: in, In the dictionary matrix The selected first One atom, This represents the dot product of two vectors. These are atomic labels. This indicates taking the modulus.

4. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 3, characterized in that: In step three, precise pulse interference localization is performed within the coarse localization region to obtain the set of abscissas of abnormal points in the pulse interference cluster. and the set of ordinates of abnormal points in the pulse interference cluster Specifically: Step 3:

1. Initialize the cluster center as the point with the largest amplitude value in the coarse localization region of the pulse interference, and initialize the cluster center iteration label. ; Step 3.2: In a two-dimensional space composed of time and amplitude, using a rectangular window of a preset size, search for the set of neighboring points of the cluster center in the coarse localization region of the impulse interference, and obtain the set of time values ​​in the set of neighboring points. and amplitude value set ; in, It is the first The set of time values ​​of cluster center neighbors in round iteration. yes The first in One value, It is the first The set of amplitude values ​​of cluster center neighbors in round iteration. yes The first in One value, It is the total number of neighboring points of the cluster center; Step 33, Utilize and Obtain the local density distribution within the rectangular window, update the cluster center position using the local density distribution within the rectangular window, and then update the cluster center position. The x-axis includes the set of x-axis coordinates of anomaly points in the pulse interference cluster. The updated density center The ordinate includes the set of ordinates of anomaly points in the pulse interference cluster. ; Steps 3 and 4: Determine whether the updated cluster center position is the same as the original cluster center position. If... Then the set of x-coordinates of the abnormal points of the output pulse interference and the set of vertical coordinates of pulse interference anomalies Otherwise, let Then return to step three two.

5. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 4, characterized in that: The utilization in step three-three and Obtain the local density distribution within the rectangular window, update the cluster center position using the local density distribution within the rectangular window, and then update the cluster center position. The x-axis includes the set of x-axis coordinates of anomaly points in the pulse interference cluster. The updated density center The ordinate includes the set of ordinates of anomaly points in the pulse interference cluster. Specifically: First, obtain the data distribution density of the current cluster center, specifically: in, It is an indicator parameter. Pick or , Indicates the time dimension. Indicates the magnitude dimension. Is the cluster center in dimension The value on, It is a data dimension Cluster center data distribution density, It is the total number of points near the cluster center. It is the cluster center neighbor point label. It is a data dimension The first in the set of neighboring points of the upper cluster center A number, when Pick hour, for ,when Pick hour, for , These are sliding window parameters, when Pick hour, For the time-dimension sliding window parameters, when Pick hour, For the sliding window parameters in the amplitude dimension, It is a data dimension; It is a Gaussian kernel function; Then, based on the data distribution density, the cluster centers are moved to areas with higher density, thereby updating the cluster center positions and obtaining the updated cluster center positions. ; Finally, the updated cluster center The x-axis includes the set of x-axis coordinates of anomaly points in the pulse interference cluster. The updated cluster center The ordinate includes the set of ordinates of anomaly points in the pulse interference cluster. .

6. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 5, characterized in that: The cluster center position is updated by moving it to a higher density region based on the data distribution density, thus obtaining the updated cluster center position. Specifically: in, This is the updated cluster center location. Dimension The value on.

7. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 6, characterized in that: Step four involves using the set of anomalous points from pulse interference clusters. Determine whether anomalies in the coarse localization area of ​​pulse interference belong to real signal components, and use the determination results to... The medium amplitude value is scaled as follows: in, It is scaled up , It is the average amplitude of all sampling points within the coarse localization area of ​​pulse interference, excluding anomaly points. It means The maximum value in, It means The minimum value in.

8. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 7, characterized in that: In step five, the amplitude value scaled up in step four is used to determine whether to terminate pulse interference suppression. If pulse interference suppression is terminated, the underwater acoustic signal under test after pulse interference suppression is output. Otherwise, let The underwater acoustic signal under test after pulse interference suppression is used as the residual. Then return to step two, which is as follows: First, the amplitude set of sampling points in the coarse localization region of the pulse interference after interference suppression is obtained using the scaled amplitude values ​​from step four. ; in, This represents the total number of sampling points in the coarse localization area of ​​pulse interference. Then, obtain the amplitude set of sampling points in the coarse localization region of the pulse interference before this round of interference suppression. The amplitude set of sampling points in the coarse localization region of pulse interference after the previous round of interference suppression. , obtain and Local variance , Specifically: in, , It is the index of sampling points in the coarse localization region of pulse interference. It is the total number of sampling points in the coarse localization area of ​​the pulse interference. It is the first After suppressing wheel interference, the pulse interference coarse localization area is the first The amplitude of each sampling point It is the first Mean amplitude of sampling points in the coarse localization region of pulse interference after wheel interference suppression It is the first Local variance of the coarse localization region for pulse interference after wheel interference suppression. It is the first Wheel interference suppression before pulse interference coarse localization region The amplitude of each sampling point It is the first The mean amplitude of sampling points in the coarse localization region before wheel interference suppression. It is the first Local variance of the coarse localization region before wheel interference suppression and pulse interference; Finally, compare and The size, if Then the pulse interference suppression process is terminated, and the underwater acoustic signal under test after pulse interference suppression is output. Otherwise, let The underwater acoustic signal under test after pulse interference suppression is used as the residual. Then return to step two.

9. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 8, characterized in that: The step six involves suppressing the underwater acoustic signal under test using pulse interference. The signal detection results are obtained as follows: Step 61: Suppress the underwater acoustic signal to be measured using pulse interference. The channel impulse response is obtained as follows: The orthogonal matching pursuit algorithm is used to solve the sparse channel estimation optimization problem and obtain the channel impulse response; The sparse channel estimation optimization problem is specifically as follows: in, It is the first The underwater acoustic signal under test after pulse interference suppression at each sampling time. It is the first Signal is transmitted at each sampling time. It is the first Signal is transmitted at each sampling time. The signal is transmitted at the first sampling time. For sparsity parameters, For channel impulse response, It is the first The underwater acoustic signal under test after pulse interference suppression at each sampling time. It is a dictionary matrix; Step 62: Obtain the reconstructed signal using the channel impulse response, and then acquire the detection statistics. Detection statistics With preset threshold value In comparison, if This indicates the underwater acoustic signal to be measured. The presence of a transmitted signal indicates that the underwater acoustic signal being measured is present; otherwise, it indicates that the signal is present. There is no transmitted signal.

10. The method for detecting underwater acoustic signals against pulse interference based on intra-cluster anomaly identification according to claim 9, characterized in that: The detection statistics Specifically: in, It's a transpose operation. It is a detection statistic. It is the normalization factor.

Images