A multi-station time difference signal sorting method based on adaptive stream clustering

Abstract

The application discloses a multi-station time difference signal sorting method based on adaptive flow clustering and relates to a multi-station time difference signal sorting method. The application is used to solve the problem of insufficient real-time performance of the current multi-station time difference sorting. The steps of the application comprise the following steps: step 1, extracting multi-station time difference data; step 2, calculating the error of time difference estimation; step 3, static clustering and setting an adaptive bias parameter; step 4, detecting and merging radar clusters; step 5, adaptively setting a flow clustering threshold; and step 6, flow clustering and real-time updating of a clustering result. The application belongs to the technical field of electronic reconnaissance.

Description

Technical Field

[0001] This invention relates to a multi-station time difference signal sorting method, belonging to the field of electronic reconnaissance technology. Background Technology

[0002] Modern radar signals are highly dense and interwoven, and radar signal modulation methods are becoming increasingly complex and varied, posing a significant challenge to the sorting of complex radar source signals. Furthermore, the rapidly changing nature of the electronic battlefield places ever-higher demands on the rapid and accurate sorting of radar source signals. To address the ineffectiveness of single-station sorting methods in handling pulse-agile signals, multi-station time-difference cooperative signal sorting has been proposed in recent years. This method utilizes the time difference information of the arrival of pulses transmitted from the same source at different stations for sorting. It depends only on the locations of the source and receiving stations, and is independent of any pulse modulation pattern of the source, effectively handling pulse-agile radar signals.

[0003] However, existing clustering-based multi-station time difference (TDOA) signal sorting methods require batch processing of a sufficient number of accumulated TDOA samples over a period of time. This presents challenges in meeting the high real-time requirements and providing insufficient usable data samples in the highly competitive electromagnetic spectrum battlefield. Compared to traditional batch processing clustering algorithms, flow clustering algorithms perform real-time online clustering of radar radiation source signals pulse by pulse, updating the clusters in real-time with data input. This algorithm can uncover the dynamic characteristics of pulse information, such as the evolution of radar radiation sources' appearance, disappearance, and reappearance. Therefore, applying flow clustering algorithms to TDOA signal sorting can solve the problems of insufficient real-time performance and insufficient usable sample data in batch clustering TDOA signal sorting. Based on these requirements, a novel TDOA signal sorting method based on flow clustering needs to be developed.

[0004] Flow clustering sorting algorithms typically suffer from several issues. For example, they are prone to batch-addition problems when faced with data imbalance caused by the coexistence of high and low repetition rate signals. Furthermore, the accuracy of flow clustering sorting algorithms is easily affected by the threshold setting. To address these problems, time difference measurement error information from radar radiation sources can be incorporated to adaptively detect and merge the flow clusters, effectively suppressing radar batch-addition phenomena. Simultaneously, adaptive setting of the sorting threshold improves the algorithm's adaptability to the signal environment and its speed in detecting and responding to unknown radar clusters. Summary of the Invention

[0005] To address the current problem of insufficient real-time performance in multi-station time difference sorting, this invention proposes a multi-station time difference signal sorting method based on adaptive flow clustering.

[0006] The technical solution adopted by the present invention to solve the above problems is as follows: The steps of the present invention include:

[0007] Step 1: Extract time difference data from multiple stations;

[0008] Step 2: Calculate the error in time difference estimation;

[0009] Step 3: Static clustering and setting adaptive bias parameters;

[0010] Step 4: Detection and merging of previously approved radar clusters;

[0011] Step 5: Adaptive setting of flow clustering threshold;

[0012] Step 6: Perform streaming clustering and update the clustering results in real time.

[0013] Furthermore, step 2, which involves calculating the time difference estimation error, includes:

[0014] Step 201: Calculate the time synchronization error, which is the error caused by time synchronization.

[0015] Step 202: Calculate the quantization error:

[0016] The accuracy period of the receiver clock counting pulse is limited, which leads to quantization error in TOA parameter measurement. The formula for calculating the quantization error is:

[0017]

[0018] In formula (1), T0 represents the period of the counting pulse;

[0019] Step 203: Calculate the error in the noise bandwidth.

[0020] When measuring the pulse TOA, the noise bandwidth error is affected by the steepness of the pulse envelope leading edge and the signal-to-noise ratio. The formula is as follows:

[0021]

[0022] In formula (2), k represents the correction coefficient, which is usually 0.35, B represents the minimum bandwidth of the receiver system, S / N represents the signal-to-noise ratio of the signal, and k / B is the rise time of the leading edge of the pulse envelope, which is usually the rise time of the pulse amplitude from 10% to 90%.

[0023] Step 204: In the actual reception process, the positions of each receiving station relative to the radiation source are different, resulting in different amplitudes of the same pulse signal received by different radiation sources. The formula is:

[0024]

[0025] In formula (3), A1 and A2 represent the amplitudes of the same pulse signal received by the two receivers, At Indicates a fixed detection threshold;

[0026] Step 205: Calculate the TDOA time difference error:

[0027]

[0028] In formula (4), σ1 represents timing error, σ2 represents quantization error, σ3 represents noise bandwidth error, and σ4 represents amplitude error.

[0029] Furthermore, step 3, which involves static clustering and setting adaptive bias parameters, is as follows:

[0030] Select the first T0 sample points to perform static clustering; let S(x) i ,x k ) is the data point x i Using data point x k To measure the similarity of cluster centers, negative Euclidean distance is typically used, and the similarity is calculated using the formula:

[0031]

[0032] In formula (5), the data point x i The x and y coordinates are respectively x i and y i Data point x k The x and y coordinates are respectively x k and y k p i For x i The default bias parameter for a point, p i The larger the value, the more clustering results are produced.

[0033] Similarity S(x) of the same point i ,x i Also known as the bias parameter P, where P(P={p i},p i =S(x i x i The clustering algorithm iterates as follows:

[0034]

[0035] In formula (6), x i Let x represent the i-th data point. k Let x represent the k-th data point, i′ represent any point other than the i-th data point, k′ represent any point other than the k-th data point, and r(i,k) represent any data point x. k As data point x i The attractiveness of cluster centers, where a(i,k) represents the data point x.i Select data point x k The suitability of cluster centers.

[0036] Furthermore, the steps for batch radar cluster detection and merging in step 4 are as follows:

[0037] Assume there are multiple radar clusters at time t, where the absolute similarity between radar cluster 1 and radar cluster 2 is less than 3σ. TDOA If the two clusters are considered to be additional clusters from the same radar radiation source, they are merged. The formula for merging the additional radar clusters is:

[0038]

[0039] In formula (7), This represents the representative point of the i-th class at time t. This indicates that at time t, excluding the representative point of class i, N represents the j-th data point at time t. i N represents the number of points in the i-th class. i′ This represents the number of points excluding the i-th type. This represents the timestamp of the last class assigned to the i-th class up to time t. This represents the timestamp of the last class assigned to a class other than the i-th class up to time t.

[0040] Furthermore, the steps for setting the clustering threshold in step 5 are as follows:

[0041] The flow clustering threshold parameter θ varies with the TDOA estimation error σ TDOA Adaptation, in which The threshold parameter is defined as an estimate of θ, as follows:

[0042]

[0043] Furthermore, in step 6, the flow clustering and real-time updating of the clustering results are as follows:

[0044] Step 601: Determine whether the cluster is in a cluster or an outlier; if it is in a cluster, determine its clustering result library and update the corresponding cluster summary; if it is an outlier, update the outlier library.

[0045] Step 602: Detect the presence of unknown clusters in the outlier database; if the number of outlier samples exceeds the threshold or meets other triggering conditions, perform static clustering and input the clustering structure into the clustering result database for summary update.

[0046] The beneficial effects of this invention are: Attached Figure Description

[0047] Figure 1 This is a flowchart of the present invention;

[0048] Figure 2 This is a schematic diagram of the adaptive bias parameter operation;

[0049] Figure 3 This is a schematic diagram of adaptive merging and batching radar clusters;

[0050] Figure 4 This is a schematic diagram of the flow clustering process. Detailed Implementation

[0051] Specific implementation method one: Combining Figure 1 This embodiment describes a multi-station time difference signal sorting method based on adaptive flow clustering, comprising the following steps:

[0052] Step 1: Extract time difference data from multiple stations;

[0053] Step 2: Calculate the error in time difference estimation;

[0054] Step 3: Static clustering and setting adaptive bias parameters;

[0055] Step 4: Detection and merging of previously approved radar clusters;

[0056] Step 5: Adaptive setting of flow clustering threshold;

[0057] Step 6: Perform streaming clustering and update the clustering results in real time.

[0058] Specific Implementation Method Two: Combining Figure 1 This embodiment describes a multi-station time difference signal sorting method based on adaptive flow clustering, where step 2 involves calculating the time difference estimation error. The steps include:

[0059] Step 201: Calculate the time synchronization error, which is the error caused by time synchronization.

[0060] Step 202: Calculate the quantization error:

[0061] The accuracy period of the receiver clock counting pulse is limited, which leads to quantization error in TOA parameter measurement. The formula for calculating the quantization error is:

[0062]

[0063] In formula (1), T0 represents the period of the counting pulse;

[0064] Step 203: Calculate the error in the noise bandwidth.

[0065] When measuring the pulse TOA, the noise bandwidth error is affected by the steepness of the pulse envelope leading edge and the signal-to-noise ratio. The formula is as follows:

[0066]

[0067] In formula (2), k represents the correction coefficient, which is usually 0.35, B represents the minimum bandwidth of the receiver system, S / N represents the signal-to-noise ratio of the signal, and k / B is the rise time of the leading edge of the pulse envelope, which is usually the rise time of the pulse amplitude from 10% to 90%.

[0068] Step 204: In the actual reception process, the positions of each receiving station relative to the radiation source are different, resulting in different amplitudes of the same pulse signal received by different radiation sources. The formula is:

[0069]

[0070] In formula (3), A1 and A2 represent the amplitudes of the same pulse signal received by the two receivers, A t Indicates a fixed detection threshold;

[0071] Step 205: Calculate the TDOA time difference error:

[0072]

[0073] In formula (4), σ1 represents timing error, σ2 represents quantization error, σ3 represents noise bandwidth error, and σ4 represents amplitude error.

[0074] Specific implementation method three: Combining Figure 2 This embodiment describes step 3 of a multi-station time difference signal sorting method based on adaptive flow clustering, specifically the step of static clustering and setting adaptive bias parameters:

[0075] Select the first T0 sample points to perform static clustering; let S(x) i ,x k ) is the data point x i Using data point x k To measure the similarity of cluster centers, negative Euclidean distance is typically used, and the similarity is calculated using the formula:

[0076]

[0077] In formula (5), the data point x i The x and y coordinates are respectively x i and y i Data point x k The x and y coordinates are respectively x k and y k p i For x i The default bias parameter for a point, p i The larger the value, the more clustering results are produced.

[0078] Similarity S(x) of the same point i ,x i Also known as the bias parameter P, where P(P={p i},p i =S(x i x i The clustering algorithm iterates as follows:

[0079]

[0080] In formula (6), the step of batch radar cluster detection and merging in step 4 of the multi-station time difference signal sorting method of flow clustering is as follows:

[0081] Assume there are multiple radar clusters at time t, where the absolute similarity between radar cluster 1 and radar cluster 2 is less than 3σ. TDOA If the two clusters are considered to be additional clusters from the same radar radiation source, they are merged. The formula for merging the additional radar clusters is:

[0082]

[0083] In formula (7), This represents the representative point of the i-th class at time t. This indicates that at time t, excluding the representative point of class i, N represents the j-th data point at time t. i N represents the number of points in the i-th class. i′ This represents the number of points excluding the i-th type. This represents the timestamp of the last class assigned to the i-th class up to time t. This represents the timestamp of the last class assigned to a class other than the i-th class up to time t.

[0084] Specific Implementation Method Five: Combining Figure 1 In this embodiment, step 5 of the multi-station time difference signal sorting method based on adaptive flow clustering, the step of setting the clustering threshold, is as follows:

[0085] The flow clustering threshold parameter θ varies with the TDOA estimation error σ TDOA Adaptation, in which The threshold parameter is defined as an estimate of θ, as follows:

[0086]

[0087] Specific Implementation Method Six: Combination Figure 1 In this embodiment, step 6 of the multi-station time difference signal sorting method based on adaptive flow clustering, involving flow clustering and real-time updating of the clustering results, is as follows:

[0088] Step 601: Determine whether the cluster is in a cluster or an outlier; if it is in a cluster, determine its clustering result library and update the corresponding cluster summary; if it is an outlier, update the outlier library.

[0089] Step 602: Detect the presence of unknown clusters in the outlier database; if the number of outlier samples exceeds the threshold or meets other triggering conditions, perform static clustering and input the clustering structure into the clustering result database for summary update.

[0090] Example

[0091] Step 1: Extraction of multi-station time difference data; Multiple receiving stations in the electromagnetic environment generate multiple sets of arrival times; the time difference is the difference in arrival time of the same radiation source signal to different receiving stations; the extraction of multi-station time difference data provides input samples for the multi-station time difference signal sorting method;

[0092] Let the duration of a single reconnaissance operation by a multi-station reconnaissance system be T, and the distance from the radiation source to the main station be d. OA The distance to the secondary station is d. OI I = 1, 2, L; the radar pulse propagates to the main and auxiliary stations at the speed of light c, t A , t I These represent the time required for the same pulse signal to reach the main and secondary stations, respectively.

[0093]

[0094] The locations of the main and secondary stations and the target radiation source form a triangle. From the relationship between the three sides of a triangle, we can deduce:

[0095]

[0096] Where, Δt AI Represents the time required for an electromagnetic wave to propagate from the secondary station I to the primary station A; the nth pulse in the pulse stream of the signal received by the primary station A. A The arrival time of each pulse is denoted as The pulse stream of the signal received by secondary station I may contain... The arrival time range of the pulse signals received by master station A from the same radiation source is:

[0097]

[0098] ∑(n A The determined time range is that of the main station A. The time difference window corresponds to the pulse received at each time; the time difference window is used to constrain the paired pulses between the master and slave stations, so that the master station A can achieve the same pulse at each time. Coarse matching between the received single pulse and the pulse stream received by the secondary station I;

[0099] Step 2: Calculate the time difference estimation error; the time difference TDOA is obtained by directly subtracting the arrival time (ToA) measurements from each station. Its measurement error mainly includes the following components:

[0100] Time-related errors

[0101] When a multi-station reconnaissance system starts operating, the clocks of multiple receivers need to be synchronized before they are deployed to different locations; therefore, the system error caused by clock synchronization is called timing error.

[0102] Quantization error

[0103] The accuracy of the receiver clock counting pulse is limited, which leads to quantization error in the TOA parameter measurement; where the period of the counting pulse is T0, the quantization error formula is:

[0104]

[0105] Noise bandwidth error

[0106] When measuring the pulse TOA, the noise bandwidth error is affected by factors such as the steepness of the pulse envelope leading edge and the signal-to-noise ratio; therefore, the noise bandwidth error formula is:

[0107]

[0108] Where k is the correction factor, usually 0.35, B is the minimum bandwidth of the receiver system, S / N is the signal-to-noise ratio of the signal, and k / B is the rise time of the leading edge of the pulse envelope, usually the rise time of the pulse amplitude from 10% to 90%.

[0109] Amplitude error

[0110] In actual reception, the different positions of each receiving station relative to the radiation source result in different amplitudes of the same pulse signal received by different radiation sources. For fixed threshold detection, the variation in amplitude leads to different pulse rise times, thus generating amplitude errors, as shown in the formula:

[0111]

[0112] A1 and A2 represent the amplitudes of the same pulse signal received by the two receivers. t To fix the detection threshold, the measurement error of TDOA is calculated as follows:

[0113]

[0114] Where σ1 is the timing error, σ2 is the quantization error, σ3 is the noise bandwidth error and σ4 is the amplitude error; in the actual multi-station time difference sorting process, it is necessary to estimate parameters such as pulse signal-to-noise ratio and signal bandwidth, and substitute them into the above formula to approximately calculate the standard deviation of the multi-station time difference estimation error;

[0115] Step 3: Static clustering. Select the first T0 sample points to arrive for static clustering; the clustering results serve as initialization information for the effective library summary. The static clustering process of stream clustering uses the nearest neighbor propagation clustering algorithm. The core idea of ​​this algorithm is to pass messages between data points, including attraction information r(i,k) and membership information a(i,k), ultimately selecting representative elements as cluster centers to complete the clustering. The calculation of attraction information is based on similarity calculation, let S(x... i x k ) is the data point x i Using data point x k To measure the similarity between cluster centers, negative Euclidean distance is typically used; the similarity is defined as follows:

[0116]

[0117] Similarity S(x) of the same point i x i Also known as the bias parameter P, where P(P={p i},p i =S(x i x i The clustering algorithm iterates as follows:

[0118]

[0119] Adaptive bias parameter setting: In static clustering algorithms, the bias parameter is set as a fixed constant, assuming that each point has the same probability of becoming a cluster center. However, when sorting multi-station time difference data samples, since the probability of each point becoming a cluster center is not the same, setting the same bias parameter will ignore the data distribution information, thus reducing clustering performance and leading to excessive unnecessary computation. This paper sets an adaptive bias parameter based on the "3σ" principle of time difference estimation error distribution.

[0120] like Figure 2 As shown, at data point x i Centered on 3σ TDOA There are other data points x within the radius. i ', then x i For a group of points, its bias parameter is the similarity S(x). i ,x i The average value of '); the adaptive bias parameter is defined as:

[0121]

[0122] In the above formula, μ is the unit step function, and the similarity S(x) i ,x i' The calculation uses negative Euclidean distance; when the data point x i and x i' Euclidean distance greater than 3σ TDOA Time step function ε(S(x) i ,x i' )+3σ TDOA The result is 0 when the data point x is zero. i and x i' Euclidean distance less than or equal to 3σ TDOA Time step function ε(S(x) i ,x i' )+3σ TDOA Therefore, at data point x, we get 1. i The radius of the center is 3σ TDOA Range, if in multiple x i' Point, P i For the corresponding S(x) i ,x i' The average of ); otherwise, it means x i The radius of the center point is 3σ TDOA There are no other points in the range, x i Point P is an outlier in an independent cluster. i It is 0.

[0123] Step 4: Detect and merge additional radar clusters; such as... Figure 3 As shown, assume there are multiple radar clusters at time t, where the absolute similarity between radar cluster 1 and radar cluster 2 is less than 3σ. TDOA When the two clusters are considered to be additional clusters from the same radar radiation source, they are merged; the MAX function selects the maximum value within the parentheses; the formula after merging the additional radar clusters is:

[0124]

[0125] Step 5: Setting the flow clustering threshold; When determining whether a data point is a cluster point, x needs to be calculated. t and cluster centers within the cluster summary similarity Determine if the similarity is within the threshold parameter θ; if the parameter θ is set too small, the data point x t Data points are easily misassigned to the outlier database; the parameter θ is set too high, resulting in data points x... t It is easy to be misassigned to the clustering result library; the time difference characteristics in the streaming clustering process also conform to the "3σ" principle of Gaussian distribution, where the similarity between data points and the center of clusters is greater than 3σ.TDOA When a data point is considered irrelevant to a cluster, the time-varying estimation error σ of the flow clustering threshold parameter is considered to be... TDOA Adaptation; then the flow clustering threshold parameter θ varies with the TDOA estimation error σ TDOA Adaptation, in which The threshold parameter is defined as an estimate of θ, as follows:

[0126]

[0127] Step 6: Perform streaming clustering and update the clustering results in real time;

[0128] like Figure 4 As shown, after the flow clustering begins, each incoming pulse multi-station time difference feature sample is sorted and its evolution is detected pulse by pulse. It is determined whether the sample is a cluster point or a discrete point. If it is a cluster point, it is determined to belong to the clustering result library, and the corresponding cluster summary is updated. If it is an outlier, the outlier library is updated. The outlier library is checked for the presence of unknown clusters. If the number of outlier samples exceeds the threshold or meets other triggering conditions, static clustering is performed, and the clustering results are input into the clustering result library for summary update.

[0129] This embodiment performs simulation verification of the above-mentioned multi-station time difference signal sorting method based on adaptive flow clustering, according to the situation of three observation stations;

[0130] The simulation conditions are as follows:

[0131] For the sake of simplicity, the following assumptions are made about this algorithm model:

[0132] 1. All engineering errors are added to the time difference error, and all observation stations have the same accuracy;

[0133] 2. Assume the target is stationary or moving at an extremely low speed;

[0134] 3. The time difference error follows a Gaussian distribution with a mean of zero, and the time difference errors of each observation station are independent of each other;

[0135] 4. To simplify the model, all observation stations and targets are located in the XOY plane.

[0136] The parameter settings are as follows:

[0137] The location information of the main station is [0,0,0] km; the location information of secondary station 1 is [10,0,5] km; and the location information of secondary station 2 is [0,15,-10] km. The time difference window can be calculated using the location information of the main and secondary stations. The time difference window is the range of time difference between the arrival of the same pulse at the main and secondary stations. Specifically, the time difference window between the main station and secondary station 1 is [-37.27,37.27] μs, and the time difference window between the main station and secondary station 2 is [-60.09,60.09] μs. The coordinates of the six targets are [50,50,30] km, [60,80,60] km, [70,50,70] km, [130,150,80] km, [90,100,120] km, and [60,120,165] km.

[0138] Comparative simulations were performed using the radar data in Table 1. E1 and E2 are data-sparse radars with imbalances of 1.65 and 1, respectively. E3-E6 are data-dense radars, used to verify the sorting algorithm's performance on complex waveform agile radars.

[0139] Table 1:

[0140]

[0141] 1. Using 100 measurement data as one simulation data, Table 2 shows the success rate of sorting six radar radiation sources when the time difference error is 20ns, 40ns and 60ns respectively.

[0142] Table 2:

[0143]

[0144]

[0145] 2. Using 100 measurement data as one simulation data, Table 3 shows the sorting success rate of the six radar radiation sources when the data loss rate is 5%, 10%, and 20%, respectively.

[0146] Table 3:

[0147]

[0148] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention, and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A multi-station time difference signal sorting method based on adaptive flow clustering, characterized in that: The steps of the multi-station time difference signal sorting method based on adaptive flow clustering include: Step 1: Extract time difference data from multiple stations; Step 2: Calculate the error in time difference estimation; Step 3: Static clustering and setting adaptive bias parameters; specific steps include: Select the first to reach Static clustering is performed on _ sample points; let _ ... Data points With data points To measure the similarity of cluster centers, negative Euclidean distance is typically used, and the similarity is calculated using the formula: （5）， In formula (5), data points The x and y coordinates are respectively and data points The x and y coordinates are respectively and , for The default bias parameter of the point. The larger the value, the more clustering results are produced. Similarity of the same point Also known as bias parameter ,in The clustering algorithm iterates as follows: （6）， In formula (6), Indicates the first Data points, Indicates the first Data points, Indicates except the first Points other than those points Indicates except the first Points other than those points Representing data points As data points The attractiveness of cluster centers Representing data points Select data points The suitability of a cluster center; Step 4: Detection and merging of previously approved radar clusters; specific steps include: exist There are multiple radar clusters in total, among which the absolute similarity between radar cluster 1 and radar cluster 2 is less than 1. If the two clusters are considered to be additional clusters from the same radar radiation source, they are merged. The formula for merging the additional radar clusters is: （7）， In formula (7), express Time of the first The representative point of the class, express Except for the first moment The representative point of the class, express Time of the first Data points, Indicates the first The number of class points, Indicates except the first The number of class points, Indicates as of The last point was awarded to the [number]th [unit]. timestamp of class, Indicates as of The last one to be assigned except for the first The timestamp of the class; Step 5: Adaptive setting of flow clustering threshold; Step 6: Perform streaming clustering and update the clustering results in real time.

2. The multi-station time difference signal sorting method based on adaptive flow clustering according to claim 1, characterized in that: Step 2, which involves calculating the time difference estimation error, includes the following steps: Step 201: Calculate the time synchronization error, which is the error caused by time synchronization. Step 202: Calculate the quantization error: The accuracy period of the receiver clock counting pulse is limited, which leads to quantization error in TOA parameter measurement. The formula for calculating the quantization error is: （1）， In formula (1), Indicates the period of the counting pulse; Step 203: Calculate the error in the noise bandwidth. When measuring the pulse TOA, the noise bandwidth error is affected by the steepness of the pulse envelope leading edge and the signal-to-noise ratio. The formula is as follows: （2）， In formula (2), This represents the correction factor, typically 0.

35. This indicates the minimum bandwidth of the receiver system. Indicates the signal-to-noise ratio (SNR) of the signal. This is the rise time of the leading edge of the pulse envelope, typically the rise time of 10% to 90% of the pulse amplitude. Step 204: In the actual reception process, the positions of each receiving station relative to the radiation source are different, resulting in different amplitudes of the same pulse signal received by different radiation sources. The formula is: （3）， In formula (3), and This indicates the amplitude of the same pulse signal received by two receivers. Indicates a fixed detection threshold; Step 205: Calculate the TDOA time difference error: （4）， In formula (4), Indicates timing error, Indicates quantization error, Indicates noise bandwidth error. This indicates the amplitude error.

3. The multi-station time difference signal sorting method based on adaptive flow clustering according to claim 1, characterized in that: The steps for setting the clustering threshold in step 5 are as follows: Flow clustering threshold parameter With TDOA estimation error Adaptation, in which for The estimated value, the threshold parameter is defined as follows: （8）。 4. The multi-station time difference signal sorting method based on adaptive flow clustering according to claim 1, characterized in that: In step 6, the flow clustering and real-time updating of the clustering results are as follows: Step 601: Determine whether the point is in a cluster or an outlier; If it is a cluster point, determine its belonging to the cluster result library and update the corresponding cluster summary; if it is an outlier point, update the outlier library. Step 602: Detect the presence of unknown clusters in the outlier database; If the number of outlier samples exceeds the threshold or other triggering conditions are met, static clustering is performed, and the clustering structure is input into the clustering result library for summary updating.

Images