A signal sorting method based on membership function and grid density clustering

Abstract

The application belongs to the field of electronic reconnaissance, and is a signal sorting method based on membership function and grid density clustering. Enemy pulse stream data is intercepted and put into a two-dimensional plane, and then normalized and put into an n*n grid plane. A membership function is set to obtain the density weight of each data point in the grid for the four endpoints of the grid, and then the local density value of each grid is calculated by summing the endpoint density weight. The enemy pulse stream data is clustered by setting a boundary factor parameter to form a cluster, and all clusters are obtained by setting different boundary thresholds. Thus, high-density pulse stream is sorted quickly and accurately, and compared with traditional clustering algorithms, the method has higher clustering accuracy, lower computational complexity and does not require prior information such as data set standard classification.

Description

Technical Field

[0001] This application belongs to the field of electronic reconnaissance, and specifically relates to a signal sorting method based on membership function and grid density clustering. Background Technology

[0002] Radar signal sorting technology plays a crucial role in electronic warfare and is an indispensable component of modern electronic reconnaissance systems. By accurately sorting signals intercepted in the complex electromagnetic environment in which the reconnaissance receiver operates, radiation source identification and threat assessment can be achieved, ultimately forming battlefield intelligence.

[0003] With the increasing complexity of the battlefield electromagnetic environment, the growth of pulse currents, and the continuous development of radar technology, signal sorting technology faces numerous challenges. In complex electromagnetic environments, the density of unknown radiation source signals can reach millions per second, and the types are diverse, leading to aliasing in both the time and frequency domains. Furthermore, with the development of counter-reconnaissance technologies such as low intercept (LID) and the limitations of receiver performance, the received signals often contain false pulses and pulse loss. Therefore, traditional sorting methods cannot effectively sort out radiation source signals in complex electromagnetic environments.

[0004] Clustering algorithms are commonly used for sorting radiation source signals. Examples include partition-based K-means, density-based DBSCAN, and OPTICS algorithms. These algorithms utilize parameters such as pulse width, frequency, and angle of arrival of radar signals to cluster received pulse stream data into multiple classes for sorting. However, traditional clustering algorithms have certain drawbacks. For instance, the K-means algorithm requires prior information on the number of standard classifications, and the DBSCAN and OPTICS algorithms involve large computational loads and are not suitable for processing high-density pulse streams in complex electromagnetic environments. Summary of the Invention

[0005] The purpose of this application is to provide a signal sorting method based on membership function and grid density clustering to solve the problem that existing sorting methods cannot effectively sort out radiation source signals in complex electromagnetic environments.

[0006] The technical solution of this application is: a signal sorting method based on membership function and grid density clustering, comprising:

[0007] The system detects enemy pulse flow data in complex electromagnetic environments and saves it to a designated location. It reads enemy pulse flow data from the designated location and then selects two of the parameters, such as pulse width, pulse arrival time, carrier frequency, pulse amplitude, and angle of arrival, as the horizontal and vertical axes to establish a two-dimensional plane. The enemy pulse flow data is then placed into the two-dimensional plane to form a visualized dataset to be clustered.

[0008] Based on the number of enemy pulse stream data, the two-dimensional plane containing the dataset to be clustered is divided into an n*n grid plane. Then, the dataset to be clustered is normalized and multiplied by n and placed into the n*n grid plane.

[0009] By setting membership functions, the relative distance between each data point in the grid and its four endpoints is determined through the membership functions. This results in the density weights assigned to the four endpoints of the grid for each data point. The local density value of each grid is then calculated by summing the endpoint density weights.

[0010] Set boundary factor parameters, then find the local density maximum value in all grids as the density peak value. Determine the boundary threshold of the current cluster by multiplying the boundary factor parameters and the density peak value. Starting from the density peak value grid, merge and cluster the grids that meet the boundary threshold to form a cluster, and mark the grids in the cluster.

[0011] Find the density peak grid in the unlabeled grid, obtain a new boundary threshold based on the boundary factor parameter and the new density peak grid, merge and cluster the grids that meet the boundary threshold starting from the density peak grid to form a cluster, and label the grids in the cluster; repeat this process until all grids are traversed.

[0012] Preferably, after traversing all grids, a noise threshold is set, the grid where the noise data point is located is determined by the noise threshold, and the noise grid is filtered out.

[0013] Preferably, the specific method for merging and clustering clusters is as follows: starting from the density grid, determine whether the local density values ​​of the surrounding adjacent grids meet the boundary threshold. If so, merge and cluster the grid and mark it; otherwise, do not merge and cluster. Select the marked grids except for the density peak grids, and determine whether the surrounding unmarked grids meet the boundary threshold. If so, merge and cluster the grid and mark it; otherwise, do not merge and cluster. Repeat this process until there are no grids in the surrounding area that meet the boundary threshold.

[0014] Preferably, the membership function is set as follows:

[0015]

[0016] Where, Δd=P d -N d P d N represents the coordinates of a data point in a certain dimension. d Represents the coordinates of the grid endpoints.

[0017] Preferably, the formula for calculating the density weight at a certain endpoint of the grid is as follows:

[0018]

[0019] Where n represents the number of dimensions in the dataset;

[0020] The density weights assigned to any data point P for its four endpoints A, B, C, and D in the grid, obtained through the density weight calculation formula, are as follows:

[0021]

[0022]

[0023]

[0024]

[0025] Where, d P This represents the density of data point P itself, and its value is constant, i.e.

[0026] w A +w B +w C +w D =1.

[0027] Preferably, any local density value of the grid is the sum of the density weights of its four endpoints.

[0028] This application presents a signal sorting method based on membership functions and grid density clustering. The method detects enemy pulse flow data and places it into a two-dimensional plane. After normalization, the data is placed into an n*n grid plane. By setting membership functions, density weights are assigned to the four endpoints of each data point within the grid. The local density value of each grid is calculated by summing these endpoint density weights. Furthermore, boundary factor parameters are set to cluster the enemy pulse flow data into clusters. Different boundary thresholds are used to obtain all the clusters. This method achieves fast and accurate sorting of high-density pulse flows, offering higher clustering accuracy, lower computational complexity, and eliminating the need for prior information such as standard classification of the dataset compared to traditional clustering algorithms. Attached Figure Description

[0029] To more clearly illustrate the technical solutions provided in this application, the accompanying drawings will be briefly described below. Obviously, the drawings described below are merely some embodiments of this application.

[0030] Figure 1 This is a schematic diagram of the overall process of this application;

[0031] Figure 2 This is a schematic diagram of the dataset to be clustered in this application;

[0032] Figure 3 This is a distribution diagram of the dataset after grid partitioning in this application;

[0033] Figure 4 This is a schematic diagram of the peak grid density of this application;

[0034] Figure 5 This is a schematic diagram illustrating the cluster division of this application;

[0035] Figure 6 This is a schematic diagram of the clustering results in this application. Detailed Implementation

[0036] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] A signal sorting method based on membership function and grid density clustering is proposed. It mainly utilizes the pulse width parameter in the time domain, the frequency parameter in the frequency domain, and the angle of arrival parameter in the spatial domain of the radiation source signal to form a classifiable dataset. After reading the dataset, it can perform clustering and sorting of high-density data streams in complex electromagnetic environments through four modules: normalization and grid division, grid density calculation, grid merging and cluster division, and boundary and noise determination. The main advantages of this application are high clustering accuracy, low computational complexity, and no need for prior classification information.

[0038] like Figure 1 As shown, the specific steps of the proposed method are as follows:

[0039] Step S100: Pulse stream data processing.

[0040] Enemy pulse flow data in complex electromagnetic environments is detected and saved to a designated location. Enemy pulse flow data is then read from this location. Two parameters—pulse width, pulse arrival time, carrier frequency, pulse amplitude, and angle of arrival—are selected as the horizontal and vertical axes to create a two-dimensional plane. The enemy pulse flow data is then placed onto this two-dimensional plane to form a visualized dataset for clustering. Figure 2 .

[0041] That is, all enemy pulse flow data are placed in a unified coordinate system to prepare for finding further features of the data.

[0042] Step S200: Normalization and mesh generation.

[0043] Based on the quantity of enemy pulse stream data, the two-dimensional plane containing the dataset to be clustered is divided into an n*n grid plane. Then, the normalized dataset to be clustered, multiplied by n, is placed into the n*n grid plane, as shown below. Figure 3 This forms an 11x11 grid.

[0044] By normalizing and meshing the enemy flow data, the calculation of subsequent mesh density can be simplified and standardized.

[0045] Step S300: Calculate the mesh density.

[0046] By setting membership functions, the relative distance between each data point in a grid and its four endpoints is determined. This results in the density weights assigned to the four endpoints of the grid for each data point. The local density value of each grid is then calculated by summing the endpoint density weights.

[0047] Preferably, the membership function is set as follows:

[0048]

[0049] Where, Δd=P d -N d P d N represents the coordinates of a data point in a certain dimension. d Represents the coordinates of the grid endpoints.

[0050] The sum of the density weights assigned to each data point at the four endpoints of the grid is the same. The closer a data point is to an endpoint, the greater the density weight it is assigned; the farther away it is, the smaller the density weight it is assigned.

[0051] The formula for calculating the density weight at a certain endpoint of the grid is as follows:

[0052]

[0053] Where n represents the number of dimensions in the dataset;

[0054] For example, if the coordinates of a data point P are (5.2, 5.3), then the density weights assigned to the four endpoints A(5,5), B(5,6), C(6,5), and D(6,6) of the grid containing P are calculated based on the membership function using the following formula:

[0055] w A =(1-5.2-5) 2 )×(1-5.3-5 2 ) = 0.8736

[0056] w B =(1-5.2-5) 2 )×(5.3-6-1) 2=0.0864

[0057] w C =(1-5.3-5) 2 )×(5.2-6-1) 2 =0.0364

[0058] w D =(5.3-6-1) 2 ×(5.2-6-1) 2 =0.0036

[0059] Therefore, using the density weight calculation formula, the density weight assigned to any data point P for its four endpoints A, B, C, and D of the grid is:

[0060]

[0061]

[0062]

[0063]

[0064] Where, d P This represents the density of data point P itself, generally d P Taking 1, it can be seen from the above formula that the sum of the weighted density values ​​contributed by a data point to each endpoint of its grid is 1, that is...

[0065] w A +w B +w C +w D =1

[0066] Therefore, the grid density calculation process is simplified, requiring only the sum of the total weight values ​​of the four endpoints of each grid.

[0067] Furthermore, the calculations above show that for a given endpoint, the larger its density weight, the more neighboring data points it has; and the smaller its density weight, the fewer neighboring data points it has.

[0068] For a radar radiation source signal, the pulse stream data emitted by it will have the following characteristics: the closer to the center of the radiation source, the more pulse stream data there will be; the farther away from the center of the radiation source, the fewer pulse stream data there will be. This characteristic is the same as that of density weight. Therefore, by obtaining the density weight of a certain endpoint, it is possible to accurately determine whether it is the center of the radiation source.

[0069] Step S400: Merge meshes and class clusters.

[0070] The main idea of ​​this step is to start by detecting the surrounding grids one by one from the grid with a large local density value until the cluster boundary is reached.

[0071] Set the boundary factor parameters, and then find the local density maximum value in all grids as the density peak, such as... Figure 4 The local density value is the sum of the density weights at any endpoint. The boundary threshold of the current cluster is determined by multiplying the boundary factor parameter with the density peak value. Starting from the density peak grid, the grids that meet the boundary threshold are merged and clustered to form a cluster, and the grids in the cluster are marked.

[0072] The specific design is as follows: For the first cluster, firstly... Figure 4 A peak search is performed to find the grid with the highest density value. Then, grids adjacent to this peak grid (within 1 unit) are marked into the same cluster. A boundary threshold is used to determine if the density of adjacent grids is less than the threshold. If it is, the adjacent grid is considered a cluster boundary; otherwise, the search continues with surrounding grids. A cluster boundary is defined as a grid whose density is less than the product of the current boundary factor parameter and the density peak value. Starting with the merged grid, its adjacent grids are merged into the current cluster until the current cluster boundary is reached. Classified grids are marked to avoid duplicate divisions. New density peak grids are then searched, and the cluster number is incremented by 1. This process of finding grids of the same type is repeated until all grids are marked.

[0073] Step S500: Mark all clusters.

[0074] Find the density peak grid in the unlabeled grid, obtain a new boundary threshold based on the boundary factor parameter and the new density peak grid, merge and cluster the grids that meet the boundary threshold starting from the density peak grid to form a cluster, and label the grids in the cluster; repeat this process until all grids are traversed.

[0075] Through the above search, most of the pulse stream data is divided into clusters, each cluster representing a radar radiation source, while the remaining pulse stream data is generally noise.

[0076] At the same time, noise reduction design is also required.

[0077] For datasets without noise, individual data points will be classified into the cluster of their nearest labeled grid. For noisy datasets, unclassified sparse data points may belong to cluster boundaries or be noisy data. Further judgment is needed; by setting boundary thresholds, cluster edges can be easily detected during grid merging and cluster partitioning.

[0078] For noisy meshes, whose density values ​​are very small, even much smaller than the product of the boundary factor and the current peak mesh density, this type of mesh can be easily filtered out by setting a noise threshold.

[0079] Therefore, the strategy for terminating peak search is as follows: after traversing all grids, a noise threshold is set, the grid where the noise data point is located is determined by the noise threshold, and the noise grid is filtered out.

[0080] The strategy for terminating peak search is to check if the density of the current peak grid is greater than the noise threshold. If the density of the current peak grid is less than the noise threshold, it is marked and filtered out in step S500. Figure 6 , where ① and ② are the clusters formed by clustering, and ③ is noise.

[0081] This step allows us to detect noise points with densities much lower than the boundary conditions. Whether it's white noise or non-white noise, data points contained in the grid with lower density values ​​will be classified as noise. This completes the division of all clusters and the removal of noise.

[0082] This application detects enemy pulse flow data and places it into a two-dimensional plane. After normalization, the data is placed into an n*n grid plane. By setting membership functions, density weights are assigned to the four endpoints of each data point within the grid. The local density value of each grid is then calculated by summing the endpoint density weights. Furthermore, the enemy pulse flow data is clustered by setting boundary factor parameters to form clusters. Different boundary thresholds are used to obtain all clusters. This achieves fast and accurate sorting of high-density pulse flows, offering the following advantages over traditional clustering algorithms: no prior information such as standard classification numbers is required; noise in the pulse flow can be filtered out by setting density thresholds; computational complexity is reduced to O(n) by eliminating the need to calculate Euclidean distance between data points; and density values ​​are calculated based on fuzzy approximate soft decision membership functions, rather than by calculating the number of data points in the grid, thus improving clustering accuracy.

[0083] The table below compares the running times of the DBSCAN algorithm, the K-means algorithm, and the algorithm of this invention under different numbers of pulses, demonstrating the advantage of this invention in terms of running time.

[0084] Table 1 Running time of different algorithms under different pulse flow densities.

[0085]

[0086] The table below compares the running time of the DBSCAN algorithm, the K-means algorithm, and the algorithm of this invention under different types of radiation sources, demonstrating the advantage of this invention in clustering accuracy.

[0087] Table 2. Clustering accuracy of different algorithms for different types of radiation sources.

[0088]

[0089]

[0090] Experiments using different datasets show that the proposed clustering method based on membership function and grid density outperforms traditional K-means, DBSCAN, and Leader methods in terms of clustering accuracy and sorting time for pulse flows of different densities, and is applicable to datasets of arbitrary shapes.

[0091] Finally, the following points should be noted: First, in the description of this application, it should be noted that, unless otherwise specified and limited, the terms "installation", "connection", and "linkage" should be interpreted broadly, and can be mechanical or electrical connections, or internal connections between two components, or direct connections. "Up", "down", "left", "right", etc. are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may change.

[0092] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0093] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A signal sorting method based on membership function and grid density clustering, characterized in that, include: The system detects enemy pulse flow data in complex electromagnetic environments and saves it to a designated location. It reads enemy pulse flow data from the designated location and then selects two of the parameters, such as pulse width, pulse arrival time, carrier frequency, pulse amplitude, and angle of arrival, as the horizontal and vertical axes to establish a two-dimensional plane. The enemy pulse flow data is then placed into the two-dimensional plane to form a visualized dataset to be clustered. Based on the number of enemy pulse stream data, the two-dimensional plane containing the dataset to be clustered is divided into an n*n grid plane. Then, the dataset to be clustered is normalized and multiplied by n and placed into the n*n grid plane. By setting membership functions, the relative distance between each data point in the grid and its four endpoints is determined through the membership functions. This results in the density weights assigned to the four endpoints of the grid for each data point. The local density value of each grid is then calculated by summing the endpoint density weights. Set boundary factor parameters, then find the local density maximum value in all grids as the density peak value. Determine the boundary threshold of the current cluster by multiplying the boundary factor parameters and the density peak value. Starting from the density peak value grid, merge and cluster the grids that meet the boundary threshold to form a cluster, and mark the grids in the cluster. Find density peak grids in unlabeled grids, obtain new boundary thresholds based on boundary factor parameters and new density peak grids, merge and cluster grids that meet the boundary thresholds starting from density peak grids to form a cluster, and label the grids in the cluster. Repeat this process until all grids have been traversed.

2. The signal sorting method based on membership function and grid density clustering as described in claim 1, characterized in that: After traversing all grids, a noise threshold is set. The grid where the noisy data points are located is determined by the noise threshold, and the noisy grids are filtered out.

3. The signal sorting method based on membership function and grid density clustering as described in claim 1, characterized in that: The specific method for merging and clustering clusters is as follows: Starting from the density grid, determine whether the local density values ​​of the surrounding adjacent grids meet the boundary threshold. If so, merge and cluster the grid and mark it; otherwise, do not merge and cluster. Select the marked grids except for the density peak grids, and starting from the grid, determine whether the surrounding unmarked grids meet the boundary threshold. If so, merge and cluster the grid and mark it; otherwise, do not merge and cluster. Repeat this process until there are no grids in the surrounding area that meet the boundary threshold.

4. The signal sorting method based on membership function and grid density clustering as described in claim 1, characterized in that: The membership function is set as follows: ； in, , This represents the coordinates of a data point in a certain dimension. Represents the coordinates of the grid endpoints.

5. The signal sorting method based on membership function and grid density clustering as described in claim 4, characterized in that: The formula for calculating the density weight at a certain endpoint of the grid is as follows: ； in This indicates the number of dimensions in the dataset; The density weight calculation formula is used to obtain any data point. The density weights assigned to the four endpoints A, B, C, and D of the grid are: ； ； ； ； in, Representing data points Its own density is a constant, that is... 。 6. The signal sorting method based on membership function and grid density clustering as described in claim 1, characterized in that: The local density value of any mesh is the sum of the density weights of its four endpoints.

Images