A communication anti-interference method and device based on elbow method and DBSCAN

Abstract

This invention provides a communication anti-interference method and apparatus based on the elbow method and DBSCAN. The elbow method is used to determine the number of clusters, which is then used as the selection criterion for clustering algorithm parameters. The DBSCAN, DPC, and K-means clustering algorithms are combined to improve the clustering algorithm. Optimal DBSCAN clustering algorithm parameters are selected using known signal information to find the optimal noise point filtering result, improving the accuracy of the mixing matrix estimation and thus correcting the target signal separation result. Even with inappropriate signal preprocessing thresholds, this invention can still achieve accurate separation of the target signal and improve accuracy, demonstrating practical significance and promising application prospects.

Description

Technical Field

[0001] This invention belongs to the field of communication anti-interference technology, and specifically relates to a communication anti-interference method and device based on the elbow method and DBSCAN. Background Technology

[0002] As the spectrum usage continues to expand, receivers are susceptible to various intentional or unintentional interferences in practical applications. The presence of interference signals can cause erroneous transmission of useful signals, severely impacting communication performance. To effectively separate interference from the signal and accurately estimate parameters and extract information from the target signal, interference suppression and separation are essential.

[0003] The interference suppression method combining traditional sparse component analysis with compressed sensing model can achieve separation of interference signals and target signals under underdetermined conditions. This process requires clustering to obtain the mixing matrix, and the required clustering algorithms include K-means clustering, DPC (density peak) clustering, and DBSCAN (density-based spatial clustering with noise).

[0004] K-means clustering requires manually setting the number of source signals, and when the signal preprocessing threshold is inappropriate, it is difficult to accurately determine the number of clusters from the graph. Random selection of initial cluster centers makes the clustering results prone to getting trapped in local optima, and the presence of individual discrete stray points affects the accuracy of the mixture matrix estimation. While DPC (Density Peak) clustering can provide relatively accurate cluster centers based on density peaks, its selection of points with high local density and relatively large relative distances on the decision graph is too subjective, making it difficult to accurately determine the criteria for high local density and relatively large relative distances. Furthermore, the selection of thresholds in the signal preprocessing part also affects the decision graph results. Although DBSCAN clustering can classify data by using density to determine connectivity, it can determine the number of data categories and the category of each data point, and filter out noise points. However, the classification results depend on parameter selection. Inappropriate parameter selection can lead to errors in the number of categories, thus affecting the results. Even when the number of categories is correct, different parameter selections can lead to different noise points being filtered out, thus affecting the clustering results. Combining the DBSCAN clustering algorithm with the particle swarm optimization algorithm can solve the problem that clustering results are prone to getting trapped in local optima, but there is also the problem that inappropriate parameter selection can affect the results.

[0005] In summary, various problems in clustering the mixture matrix lead to a decrease in the accuracy of target signal separation, which is detrimental to anti-interference in communication transmission. Summary of the Invention

[0006] To address the aforementioned problems in the prior art, this invention provides a communication anti-interference method and apparatus based on the elbow method and DBSCAN. The technical problem to be solved by this invention is achieved through the following technical solution:

[0007] In a first aspect, the present invention provides a communication anti-interference method based on the elbow method and DBSCAN, comprising:

[0008] S100: Acquire the observation signal and preprocess the observation signal to obtain preprocessed data;

[0009] S200, determine the number of categories using the elbow method, and determine the initial cluster centers based on the number of categories;

[0010] S300, the preprocessed data is clustered once using the DBSCAN algorithm, and the clustering parameters of the DBSCAN algorithm are changed to execute the next loop, or the initial clustering centers of the K-means clustering algorithm are replaced with the initial clustering centers, and the final clustering centers are determined using the K-means clustering algorithm, by comparing whether the number of clusters obtained by clustering is the same as the number of categories.

[0011] S400, the source signal is reconstructed using the final cluster center and compressed sensing reconstruction algorithm to obtain the separation result, and the error in the number of zero crossings of the target signal is determined based on the separation result;

[0012] S500, repeat S300 to S400 until the maximum number of cycles is reached, and determine the separation result corresponding to the minimum error as the optimal separation result of the observed signal.

[0013] Secondly, the present invention provides a communication anti-interference device based on the elbow method and DBSCAN, comprising:

[0014] The acquisition module is configured to acquire the observation signal and preprocess the observation signal to obtain preprocessed data;

[0015] The determination module is configured to determine the number of classes using the elbow method and to determine the initial cluster centers based on the number of classes;

[0016] The clustering module is configured to perform a clustering operation on the preprocessed data using the DBSCAN algorithm, and determine whether to change the clustering parameters of the DBSCAN algorithm to execute the next loop or replace the initial clustering centers of the K-means clustering algorithm with the initial clustering centers by comparing whether the number of clusters obtained by clustering is the same as the number of categories, and then use the K-means clustering algorithm to determine the final clustering centers.

[0017] The reconstruction module is configured to reconstruct the source signal using the final cluster centers and the compressed sensing reconstruction algorithm to obtain the separation result, and to determine the error in the number of zero-crossing points of the target signal based on the separation result;

[0018] The loop module is configured to repeat the implementation process of the clustering module and the reconstruction module until the maximum number of loops is reached, and the separation result corresponding to the minimum error is determined as the optimal separation result of the observed signal.

[0019] Beneficial effects:

[0020] This invention provides a communication anti-interference method and device based on the elbow method and DBSCAN. The elbow method is used to determine the number of clusters, which is then used as the selection criterion for clustering algorithm parameters. The DBSCAN clustering algorithm, DPC (Density Peak Count) clustering algorithm, and K-means clustering algorithm are combined to improve the clustering algorithm. Optimal DBSCAN clustering algorithm parameters are selected using known signal information to find the optimal noise point filtering result, improving the accuracy of the mixing matrix estimation and thus correcting the target signal separation result. Even with inappropriate signal preprocessing thresholds, this invention can still achieve accurate separation of the target signal and improve accuracy, demonstrating practical significance and promising application prospects.

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating a communication anti-interference method based on the elbow method and DBSCAN provided by the present invention.

[0023] Figure 2 This is a flowchart illustrating the implementation of the communication anti-interference method based on the elbow method and DBSCAN of the present invention.

[0024] Figure 3 Simulation diagram comparing the normalized mean square error of the mixture matrix estimates and the true mixture matrix of the present invention with those of other methods under different single-source point screening thresholds;

[0025] Figure 4 Simulation diagram comparing the correlation coefficients between the target signal and the source signal separated by this invention and other methods under different single-source point screening thresholds; Detailed Implementation

[0026] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0027] First, the technical solution of the present invention will be introduced in general.

[0028] This invention uses the elbow method to determine the number of clusters, ensuring accurate determination even when the signal preprocessing threshold is improperly set. It combines DBSCAN (Density-Based Spatial Clustering with Noise), DPC (Density Peak Clustering), and K-means clustering algorithms for clustering. The number of clusters k determined by the elbow method is used as the criterion for selecting points with high local density and relatively large relative distances in the decision graph constructed by the DPC (Density Peak Clustering) algorithm; that is, the first k points with relatively high local density sums are selected as the initial cluster centers. The algorithm sets a maximum number of iterations, selects a small maximum radius and minimum point as initial values ​​for the DBSCAN (Noisy Density-Based Spatial) clustering algorithm, and sets a range for the maximum radius. If the number of clusters identified by the DBSCAN algorithm with the current maximum radius and minimum point is the same as the number identified by the elbow method, then further K-means clustering and compressed sensing reconstruction are performed to separate the signal. The error or similarity between the target signal separation result and known signal information is calculated. If the number of clusters identified by the DBSCAN algorithm with the current maximum radius and minimum point is different from the number identified by the elbow method, then the error or similarity is set to a default value, the current iteration ends, and the next iteration begins. The maximum radius and minimum point parameters are changed after each iteration until the maximum number of iterations is reached. This effectively avoids errors in estimating the number of clusters caused by improper selection of DBSCAN algorithm parameters and correctly removes noise points, ensuring the accuracy of target signal separation.

[0029] By utilizing a portion of known signal information, the optimal DBSCAN (density-based spatial clustering with noise) clustering algorithm parameters are selected. This allows for adjustments to the noise point screening results, effectively improving the accuracy of the mixing matrix estimation and thus correcting the target signal separation results, thereby enhancing the accuracy of target signal separation.

[0030] When the DBSCAN (Density-Based Spatial Clustering with Noise) algorithm's parameters are appropriately selected to correctly estimate the number of clustered signals, different parameter selections will result in different noise points being selected, leading to different clustering results and thus varying target signal separation effects. This invention considers the impact of noise point selection on the clustering results. By comparing the magnitude of the error or similarity between the reconstructed target signal and a known subset of signal information under different DBSCAN parameter combinations, and identifying the DBSCAN parameter combination with the minimum error or maximum similarity, the noise point selection results are adjusted to improve the accuracy of the mixing matrix estimation, thereby correcting the target signal separation results. This improves the accuracy of target signal separation.

[0031] The implementation details of the present invention are described below.

[0032] Example 1

[0033] Combination Figure 1 and Figure 2 This invention provides a communication anti-interference method based on the elbow method and DBSCAN, comprising:

[0034] S100: Acquire the observation signal and preprocess the observation signal to obtain preprocessed data;

[0035] S200, determine the number of categories using the elbow method, and determine the initial cluster centers based on the number of categories;

[0036] S300, the preprocessed data is clustered once using the DBSCAN algorithm, and the clustering parameters of the DBSCAN algorithm are changed to execute the next loop, or the initial clustering centers of the K-means clustering algorithm are replaced with the initial clustering centers, and the final clustering centers are determined using the K-means clustering algorithm, by comparing whether the number of clusters obtained by clustering is the same as the number of categories.

[0037] S400, the source signal is reconstructed using the final cluster center and compressed sensing reconstruction algorithm to obtain the separation result, and the error in the number of zero crossings of the target signal is determined based on the separation result;

[0038] S500, repeat S300 to S400 until the maximum number of cycles is reached, and determine the separation result corresponding to the minimum error as the optimal separation result of the observed signal.

[0039] Example 2

[0040] As an optional embodiment of the present invention, S100 includes:

[0041] S110, acquire the observed signal x(t);

[0042] S120, The observed signal is transformed into a time-frequency domain signal using the short-time Fourier transform method;

[0043] This invention employs the Short-Time Fourier Transform (STFT) method to perform time-frequency transformation on the signal. This method involves dividing the given signal into frames and applying windowing, performing time-frequency transformation on each frame, and finally transforming the time-domain signal to the time-frequency domain. The Short-Time Fourier Transform formula is as follows:

[0044]

[0045] Where w(t) is the window function, and a Hamming window is chosen as the window function, X(t,f)=[X1(t,f),…X m (t,f)] T The observed signal is x(t) = [x1(t),…x m (t)] T The short-time Fourier transform.

[0046] S130, remove non-single-source points from the time-frequency domain signal to obtain the filtered signal;

[0047] After performing STFT (Short Time Fourier Transform) on the observed signal to transform it into the time-frequency domain, although the sparsity of the signal is enhanced, there are still many spurious points that do not meet the sparsity requirement. Therefore, single-source point screening is performed after the STFT to further improve the sparsity of the signal.

[0048] The formula for determining whether a point is a single-source point is as follows:

[0049]

[0050] Where ε is the threshold for single-source filtering, which is a constant and ε∈(0,1).

[0051] S140, low-energy points are filtered out from the screening signal to obtain the filtered signal;

[0052] The formula for determining whether a point is a low-energy point that needs to be filtered out is as follows:

[0053]

[0054] Where β is the threshold for filtering out low-energy points, which is a constant and usually takes a value between 0.01 and 0.3.

[0055] S150, the filtered signal is normalized to obtain preprocessed data.

[0056] The data normalization formula of this invention is as follows:

[0057]

[0058] Example 3

[0059] As an optional embodiment of the present invention, S200 includes:

[0060] S210, Use the elbow method to determine the number of categories k;

[0061] The elbow method is set to classify classes from 1 to 8. The distortion degree of each class is equal to the sum of the squares of the distances between the centroid of that class and the positions of its internal members. The formula for calculating the total distortion degree (aggregation coefficient) is as follows:

[0062]

[0063] Where J is the aggregation coefficient, K is the number of categories, and C k For the k-th class (k = 1, 2, ..., K), u k This refers to the center of gravity position of this type.

[0064] Plot a line graph with the number of clusters K as the x-axis and the aggregation coefficient J as the y-axis. The number of clusters corresponding to the flattening of the downward trend of the line is determined as the number of categories k determined by the elbow method.

[0065] S220, the number of categories k is used as a constraint of the DPC clustering algorithm, and the initial cluster centers are determined in the preprocessed data using the constrained DPC clustering algorithm.

[0066] Example 4

[0067] As an optional embodiment of the present invention, S220 includes:

[0068] S221, Generate a distance matrix from all preprocessed data;

[0069] S222, using the distance matrix and the preset cutoff distance, calculate the local density of the preprocessed data and the relative distance of the data;

[0070] S223, Construct a decision map using the local density and the relative distance;

[0071] S224: Select the k points with the highest local density and the highest relative distance as the initial cluster centers.

[0072] This invention generates a distance matrix based on preprocessed data and calculates the local density and relative distance of all data to construct a decision graph. The number of clusters determined by the elbow method is used as the criterion for selecting points with high local density and relatively large relative distances in the decision graph constructed by the DPC (Density Peak) clustering algorithm. The first k points with relatively high local density and relative distances are selected as the initial cluster centers.

[0073] Example 5

[0074] As an optional embodiment of the present invention, S300 includes:

[0075] S310, The preprocessed data is clustered once using the DBSCAN algorithm to obtain a clustered dataset with multiple categories;

[0076] This step sets the maximum radius, minimum point, and maximum radius search range for the DBSCAN algorithm; the DBSCAN algorithm is then used to perform a clustering operation on the preprocessed data to identify noise points, resulting in a clustered dataset with multiple categories.

[0077] The algorithm selects the smaller maximum radius and minimum point as the initial values ​​for the DBSCAN (Noisy Density-Based Spatial Clustering) algorithm, and sets the range of the maximum radius value. It then enters a loop to start the DBSCAN algorithm and perform a judgment. The specific process is as follows:

[0078] (1) Calculate the maximum radius and minimum point based on the current loop count, select any point, and determine whether the point is the core point;

[0079] (2) If the point is not the core point, determine the next point;

[0080] (3) If the point is a core point, and the point does not belong to any existing category, a new category is created and the neighborhood is determined according to the maximum radius. All points that are directly accessible from the density of the point are assigned to the new category. If the point already belongs to a certain category, the neighborhood is determined according to the maximum radius. All points that are directly accessible from the density of the point are assigned to that category.

[0081] (4) Repeat (1)-(3) until all points have been filtered and points that do not belong to any class are identified as noise points;

[0082] (5) Output the clustered dataset after classification.

[0083] S320, if the number of categories in the clustering dataset is not the same as the number of categories, then label the current clustering iteration with an error, change the clustering parameters, and return to S310; wherein, the clustering parameters include the maximum radius and the minimum point;

[0084] S330, if the number of categories in the clustered dataset is the same as the number of categories, then the initial cluster centers of the K-means clustering algorithm are replaced with the initial cluster centers, and the K-means clustering algorithm is used to re-cluster the clustered dataset with noise points removed to determine the final cluster centers.

[0085] This invention compares the number of clusters in the dataset classified by the DBSCAN (density-based spatial clustering with noise) algorithm with the number of clusters determined by the elbow method. If they are the same, the K-means clustering algorithm is used to cluster again to determine the final cluster centers.

[0086] Example 6

[0087] As an optional embodiment of the present invention, S330 includes:

[0088] S331, If ​​the number of categories in the clustered dataset is the same as the number of categories, then the noise points are removed from the clustered dataset to obtain the denoised dataset;

[0089] S332, the initial cluster centers are used as the initial cluster centers of the K-means clustering algorithm, and the number of categories is determined as the number of clusters in the K-means clustering algorithm;

[0090] S333, The K-means clustering algorithm is used to re-cluster the denoised dataset to obtain the final cluster centers.

[0091] This embodiment uses the dataset of noisy points filtered out by the DBSCAN (Density-Based Space with Noise) clustering algorithm as the denoised dataset to reduce the impact of noise points on the estimation of the mixture matrix. The number of clusters k determined by the elbow method is set as the number of clusters in the K-means clustering algorithm, improving upon the shortcomings of the traditional K-means clustering algorithm, which requires manual setting of the number of clusters and is difficult to determine when the signal preprocessing threshold is inappropriate. The cluster centers selected in the DPC (Density Peak) clustering algorithm are used as the initial cluster centers in the K-means clustering algorithm, replacing the randomly selected initial cluster centers in the traditional K-means clustering algorithm. K-means clustering is then performed to obtain the final cluster centers and clustering results.

[0092] Example 7

[0093] As an optional embodiment of the present invention, S400 includes:

[0094] S410, the final cluster centers are deformed, and the deformation result is used as the observation matrix of the compressed sensing reconstruction algorithm;

[0095] S420, the source signal is reconstructed using the observation matrix and the compressed sensing reconstruction algorithm to obtain the separation result, which includes the separated target signal and interference signal;

[0096] This step uses compressed sensing reconstruction algorithms to reconstruct and recover the source signal, such as the OMP algorithm (Orthogonal Matching Pursuit algorithm), to separate the target signal from the interference signal. The reconstructed target signal is...

[0097] S430, calculate the difference between the first number of zero crossings and the second number of zero crossings, and determine the difference as the error in the number of zero crossings of the target signal;

[0098] Wherein, the first number of zero-crossing points is the number of zero-crossing points of the target signal in the separation result, and the second number of zero-crossing points is the known number of zero-crossing points of the target signal;

[0099] The target signal obtained by this step through reconstruction is: The difference between the number of zero-crossing points in the target signal and the known number of zero-crossing points in the target signal is calculated, and the difference data for this loop is recorded. Of course, the information type of the known target signal can be changed. If the calculation is not of error but of similarity, then a smaller similarity is used by default.

[0100] S440, record the error in the number of zero-crossing points of the target signal.

[0101] Example 8

[0102] As an optional embodiment of the present invention, S500 includes:

[0103] S510, repeat S300 to S400 until the maximum number of loops is reached;

[0104] S520, determine the number of iterations when the error in determining the number of zero-crossing points of the target signal is minimized, and determine the parameters corresponding to the maximum radius and minimum point of this iteration as the optimal parameters of the DBSCAN clustering algorithm;

[0105] This invention determines whether the number of iterations is less than or equal to the maximum number of iterations. If it is true, it returns to S300; if it is not true, it finds the number of iterations corresponding to the minimum error in the record. The maximum radius and minimum point parameters corresponding to this number of iterations are the selected optimal DBSCAN (density-based spatial clustering with noise) clustering algorithm parameters.

[0106] S530, the final cluster center calculated using the optimal parameters is determined as the optimal cluster center;

[0107] S540, the separation result reconstructed through the optimal cluster center is determined as the optimal separation result;

[0108] S550, the optimal separation result is determined as the optimal separation result of the observed signal.

[0109] This invention calculates the optimal DBSCAN (density-based spatial clustering with noise) clustering algorithm parameters from S300 to S400, identifies the optimal noise point selection results, the corresponding optimal mixture matrix estimates (optimal cluster centers), and the optimal target signal separation results. As the final result of target signal separation, the target signal separation result is corrected.

[0110] This invention provides a communication anti-interference device based on the elbow method and DBSCAN, comprising:

[0111] The acquisition module is configured to acquire the observation signal and preprocess the observation signal to obtain preprocessed data;

[0112] The determination module is configured to determine the number of classes using the elbow method and to determine the initial cluster centers based on the number of classes;

[0113] The clustering module is configured to perform a clustering operation on the preprocessed data using the DBSCAN algorithm, and determine whether to change the clustering parameters of the DBSCAN algorithm to execute the next loop or replace the initial clustering centers of the K-means clustering algorithm with the initial clustering centers by comparing whether the number of clusters obtained by clustering is the same as the number of categories, and then use the K-means clustering algorithm to determine the final clustering centers.

[0114] The reconstruction module is configured to reconstruct the source signal using the final cluster centers and the compressed sensing reconstruction algorithm to obtain the separation result, and to determine the error in the number of zero-crossing points of the target signal based on the separation result;

[0115] The loop module is configured to repeat the implementation process of the clustering module and the reconstruction module until the maximum number of loops is reached, and the separation result corresponding to the minimum error is determined as the optimal separation result of the observed signal.

[0116] The effects of the present invention are further illustrated below using simulation diagrams:

[0117] 1. Simulation conditions

[0118] The parameters used in this simulation are shown in the table below:

[0119]

[0120] 2. Simulation Content

[0121] To verify the effectiveness and necessity of this invention, the simulation includes the following: The DBSCAN (Noisy Density-Based Space) clustering algorithm is randomly selected with maximum radius and minimum point values ​​under different single-source point detection thresholds. The simulation also examines the number of clusters obtained when the parameters are inappropriately selected. This demonstrates the necessity of using the elbow method to determine the number of clusters, serving as the standard for selecting DBSCAN parameters to ensure accurate target signal separation. Furthermore, the simulation compares the optimal DBSCAN parameters obtained by this invention under different single-source point thresholds with the performance of three other methods, demonstrating that the algorithm effectively improves the accuracy of the mixture matrix estimation and the accuracy of target signal separation.

[0122] simulation:

[0123] (1) Under different single-source point screening thresholds, the traditional DBSCAN (density-based spatial clustering with noise) clustering algorithm randomly selects the maximum radius and minimum point value. The following are the results of the number of classifications when the parameters are not selected appropriately:

[0124] Single-source filtering threshold 9e-1 9e-1 9e-1 5e-1 5e-1 5e-1 Maximum radius 0.01 0.02 0.01 0.01 0.015 0.02 minimum point 10 20 130 10 10 200 DBSCAN category count 13 6 3 9 7 3 Single-source filtering threshold 1e-1 1e-1 5e-2 5e-2 1e-2 1e-2 Maximum radius 0.01 0.02 0.01 0.03 0.01 0.025 minimum point 10 80 5 60 5 20 DBSCAN category count 5 3 5 3 5 3

[0125] Simulation results show that inappropriate selection of the two parameters, maximum radius and minimum point, in the DBSCAN (Noisy Density-Based Spatial) clustering algorithm can lead to errors in estimating the number of clustered signals and in identifying noise points. This demonstrates the necessity of using the elbow method to determine the number of clusters in this invention and using it as the parameter selection criterion for the DBSCAN (Noisy Density-Based Spatial) clustering algorithm to ensure the accuracy of target signal separation.

[0126] (2) The performance of the method of the present invention compared with three other methods under different single-source point thresholds. The four methods are as follows:

[0127] (1) Clustering results are obtained using only the DPC (Density Peak) clustering algorithm.

[0128] (2) After obtaining the initial cluster centers using the DPC (Density Peak) clustering algorithm, the K-means (K-means) clustering algorithm is used for clustering.

[0129] (3) Clustering is performed by combining DBSCAN (density-based spatial clustering with noise) clustering algorithm (parameters are randomly selected when the number of categories is correct), DPC (density peak) clustering algorithm, and K-means (K-means) clustering algorithm.

[0130] (4) The algorithm of the present invention.

[0131] like Figure 3 and 4The following table shows the parameter combinations for the DPC_DBSCAN_K-means method in the dataset, where the DBSCAN (density-based spatial clustering with noise) clustering algorithm achieves the correct number of classifications under different single-source point screening thresholds:

[0132] Single-source detection threshold 9e-1 5e-1 1e-1 5e-2 1e-2 Randomly select the maximum radius 0.01 0.015 0.015 0.02 0.025 Randomly select the minimum point 100 100 50 40 10

[0133] The simulation results of the optimal DBSCAN (density-based spatial clustering with noise) clustering algorithm parameters obtained by this invention are shown in the following table:

[0134] Single-source detection threshold 9e-1 5e-1 1e-1 5e-2 1e-2 Maximum radius 0.025 0.025 0.014 0.029 0.027 minimum point 255 212 14 45 9

[0135] Simulation results comparing the normalized mean square error of the mixture matrix estimated by the method of this invention with that of three other methods under different single-source screening thresholds are shown in the appendix. Figure 3 The simulation results of the correlation coefficient between the separated target signal and the real target signal are shown in the appendix. Figure 4 From the appendix Figure 3 and attached Figure 4 The following conclusions can be drawn from this:

[0136] Under different single-source point screening thresholds, if the optimal DBSCAN (density-based spatial noisy) clustering algorithm parameters are not selected to adjust the noise point screening results, and instead the parameters when the DBSCAN (density-based spatial noisy) clustering algorithm correctly classifies the number of points are randomly selected as the final parameters for screening noise points (e.g.) Figure 3 and Figure 4 The DPC_DBSCAN_K-means method in this paper results in a larger normalized mean square error between the estimated and true mixture matrices compared to the simple DPC (Density Peak) clustering algorithm or the DPC_K-means (K-means) clustering algorithm. Furthermore, the correlation coefficient between the separated target signal and the true target signal is smaller compared to the DPC (Density Peak) clustering algorithm or the DPC_K-means (K-means) clustering algorithm. This indicates that the traditional DBSCAN (Noisy Density-Based Spatial) clustering algorithm's random selection of parameters to remove noise points may lead to worse target signal separation. This demonstrates the necessity of this invention to select the optimal DBSCAN (Noisy Density-Based Spatial) clustering algorithm parameters using a known subset of signal information, thereby adjusting the noise point selection results and correcting the target signal separation results.

[0137] Under different single-source point screening thresholds, the algorithm of this invention yields smaller normalized mean square errors for both the estimated and true mixture matrices compared to the other three algorithms, and the separated target signals have larger correlation coefficients with the true target signals. This demonstrates that the present invention can effectively improve the accuracy of the mixture matrix estimation, thereby correcting the target signal separation results and enhancing the accuracy of target signal separation.

[0138] In summary, this invention can effectively guarantee and improve the accuracy of target signal separation. It has practical significance and promising application prospects.

[0139] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0140] Although this application has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings, the disclosure, and the appended claims in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality.

[0141] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A communication anti-interference method based on the elbow method and DBSCAN, characterized in that, include: S100: Acquire the observation signal and preprocess the observation signal to obtain preprocessed data; S200, determine the number of categories using the elbow method, and determine the initial cluster centers based on the number of categories; S300, the preprocessed data is clustered once using the DBSCAN algorithm, and the clustering parameters of the DBSCAN algorithm are changed to execute the next loop, or the initial clustering centers of the K-means clustering algorithm are replaced with the initial clustering centers, and the final clustering centers are determined using the K-means clustering algorithm, by comparing whether the number of clusters obtained by clustering is the same as the number of categories. S400, the source signal is reconstructed using the final cluster center and compressed sensing reconstruction algorithm to obtain the separation result, and the error in the number of zero crossings of the target signal is determined based on the separation result; S500, repeat S300 to S400 until the maximum number of cycles is reached, and determine the separation result corresponding to the minimum error as the optimal separation result of the observed signal; The S300 includes: S310, The preprocessed data is clustered once using the DBSCAN algorithm to obtain a clustered dataset with multiple categories; S320, if the number of categories in the clustering dataset is not the same as the number of categories, then label the current clustering iteration with an error, change the clustering parameters, and return to S310; wherein, the clustering parameters include the maximum radius and the minimum point; S330, if the number of categories in the clustered dataset is the same as the number of categories, then the initial cluster centers of the K-means clustering algorithm are replaced with the initial cluster centers, and the K-means clustering algorithm is used to re-cluster the clustered dataset to determine the final cluster centers.

2. The communication anti-interference method based on the elbow method and DBSCAN according to claim 1, characterized in that, S100 includes: S110, acquire the observation signal; S120, The observed signal is transformed into a time-frequency domain signal using the short-time Fourier transform method; S130, remove non-single-source points from the time-frequency domain signal to obtain the filtered signal; S140, low-energy points are filtered out from the screening signal to obtain the filtered signal; S150, the filtered signal is normalized to obtain preprocessed data.

3. The communication anti-interference method based on the elbow method and DBSCAN according to claim 1, characterized in that, S200 includes: S210, Determining the number of categories using the elbow method. ; S220, the number of categories As a constraint on the DPC clustering algorithm, the initial cluster centers are determined in the preprocessed data using the constrained DPC clustering algorithm.

4. The communication anti-interference method based on the elbow method and DBSCAN according to claim 3, characterized in that, S220 includes: S221, Generate a distance matrix from all preprocessed data; S222, using the distance matrix and the preset cutoff distance, calculate the local density of the preprocessed data and the relative distance of the data; S223, Construct a decision map using the local density and the relative distance; S224, Select the sorted order first. Points with local density and relative distance are used as initial cluster centers.

5. The communication anti-interference method based on the elbow method and DBSCAN according to claim 1, characterized in that, S310 includes: S311, set the maximum radius, minimum point, and maximum radius search range of the DBSCAN algorithm; S312, The DBSCAN algorithm is used to perform a clustering operation on the preprocessed data to identify noise points, resulting in a clustered dataset with multiple categories.

6. The communication anti-interference method based on the elbow method and DBSCAN according to claim 1, characterized in that, The S330 includes: S331, If ​​the number of categories in the clustered dataset is the same as the number of categories, then the noise points are removed from the clustered dataset to obtain the denoised dataset; S332, the initial cluster centers are used as the initial cluster centers of the K-means clustering algorithm, and the number of categories is determined as the number of clusters in the K-means clustering algorithm; S333, The K-means clustering algorithm is used to re-cluster the denoised dataset to obtain the final cluster centers.

7. The communication anti-interference method based on the elbow method and DBSCAN according to claim 1, characterized in that, The S400 includes: S410, the final cluster centers are deformed, and the deformation result is used as the observation matrix of the compressed sensing reconstruction algorithm; S420, the source signal is reconstructed using the observation matrix and the compressed sensing reconstruction algorithm to obtain the separation result, which includes the separated target signal and interference signal; S430, calculate the difference between the first number of zero crossings and the second number of zero crossings, and determine the difference as the error in the number of zero crossings of the target signal; Wherein, the first number of zero-crossing points is the number of zero-crossing points of the target signal in the separation result, and the second number of zero-crossing points is the known number of zero-crossing points of the target signal; S440, record the error in the number of zero-crossing points of the target signal.

8. The communication anti-interference method based on the elbow method and DBSCAN according to claim 1, characterized in that, The S500 includes: S510, repeat S300 to S400 until the maximum number of loops is reached; S520, determine the number of iterations when the error in determining the number of zero-crossing points of the target signal is minimized, and determine the parameters corresponding to the maximum radius and minimum point of this iteration as the optimal parameters of the DBSCAN clustering algorithm; S530, the final cluster center calculated using the optimal parameters is determined as the optimal cluster center; S540, the separation result reconstructed through the optimal cluster center is determined as the optimal separation result; S550, the optimal separation result is determined as the optimal separation result of the observed signal.

9. A communication anti-interference device based on the elbow method and DBSCAN, characterized in that, include: The acquisition module is configured to acquire the observation signal and preprocess the observation signal to obtain preprocessed data; The determination module is configured to determine the number of classes using the elbow method and to determine the initial cluster centers based on the number of classes; The clustering module is configured to perform a clustering operation on the preprocessed data using the DBSCAN algorithm, and determine whether to change the clustering parameters of the DBSCAN algorithm to execute the next loop or replace the initial clustering centers of the K-means clustering algorithm with the initial clustering centers by comparing whether the number of clusters obtained by clustering is the same as the number of categories, and then use the K-means clustering algorithm to determine the final clustering centers. The reconstruction module is configured to reconstruct the source signal using the final cluster centers and the compressed sensing reconstruction algorithm to obtain the separation result, and to determine the error in the number of zero-crossing points of the target signal based on the separation result; The loop module is configured to repeat the implementation process of the clustering module and the reconstruction module until the maximum number of loops is reached, and the separation result corresponding to the minimum error is determined as the optimal separation result of the observed signal; The clustering module is configured as follows: S310, The preprocessed data is clustered once using the DBSCAN algorithm to obtain a clustered dataset with multiple categories; S320, if the number of categories in the clustering dataset is not the same as the number of categories, then label the current clustering iteration with an error, change the clustering parameters, and return to S310; wherein, the clustering parameters include the maximum radius and the minimum point; S330, if the number of categories in the clustered dataset is the same as the number of categories, then the initial cluster centers of the K-means clustering algorithm are replaced with the initial cluster centers, and the K-means clustering algorithm is used to re-cluster the clustered dataset to determine the final cluster centers.

Images