CN-122001405-A - Synchronous frequency hopping network station sorting method based on multidimensional feature clustering

CN122001405ACN 122001405 ACN122001405 ACN 122001405ACN-122001405-A

Abstract

The invention discloses a synchronous frequency hopping network table sorting method based on multidimensional feature clustering, which comprises the steps of firstly carrying out short-time Fourier transform on received synchronous frequency hopping signals to obtain a time-frequency matrix, calculating to obtain the starting time, the ending time and the center frequency of each frequency hopping signal, then carrying out time interception and filtering treatment on the received signals, extracting multidimensional complementary features aiming at single frequency hopping signals to construct a high-degree feature space, secondly carrying out standardized pretreatment on the features, eliminating dimension differences, then utilizing isolated forests to detect and separate abnormal samples, avoiding interference on clustering results, adopting K-means algorithm to carry out core clustering on normal samples, determining cluster centers, and finally distributing the abnormal samples to clusters closest to the normal samples, thus finishing accurate sorting of the whole samples. The method can effectively inhibit the influence of interference and abnormal samples, improves the precision rate, recall rate and stability of synchronous frequency hopping network station sorting in a complex electromagnetic environment, and meets the index requirements of communication reconnaissance.

Inventors

XIE JIAN
XU JINHUI
WANG LING
GUO ZIXUN
FAN YIFEI
WANG HAITAO
CHEN SHICHAO

Assignees

西北工业大学

Dates

Publication Date: 20260508
Application Date: 20260212

Claims (9)

1. A synchronous frequency hopping network station sorting method based on multidimensional feature clustering is characterized by comprising the following steps: step 1, synchronizing time-frequency processing of frequency hopping signals and single-hop parameter estimation; step 2, single-hop signal interception and filtering treatment; step 3, extracting multidimensional features; step 4, characteristic standardized pretreatment; Step 5, detecting and separating an isolated forest anomaly sample; Step 6:K-means core clustering and cluster center calculation; And 7, abnormal sample distribution and final sorting.
2. The method for sorting synchronous frequency hopping network stations based on multidimensional feature clustering according to claim 1, wherein the step 1 is specifically as follows: Step 1.1, constructing a time-frequency matrix by short-time Fourier transform; for received synchronous frequency hopping mixed signal Performing short-time Fourier transform, mapping the time domain signal to a time-frequency two-dimensional plane to obtain a time-frequency matrix , wherein, Representing a set of natural numbers, For the number of frequency points, As a point of time in the number of times, Representative of Row of lines The set of natural number matrices of columns, the short-time Fourier transform is defined as: in the formula, As the point of time-sampling, Is the sampling rate; Is a frequency sampling point; Is a window function; is an imaginary unit; is a time domain continuous time variable; Representing a synchronous frequency hopping mixed signal; the elements of the time-frequency matrix are the modular square energy spectrum of the STFT result: in the formula, Representing modulo arithmetic; Step 1.2, binarizing a time-frequency matrix; The method comprises constructing adaptive threshold based on energy distribution of time-frequency matrix to realize signal and noise separation, calculating average energy of time-frequency matrix : Using adaptive thresholds , As a threshold adjustment factor, the binarization rule is: Obtaining a binarized time-frequency matrix ; Step 1.3, calculating a communication area mark and a single-hop parameter; adopts eight-connected region marking algorithm to pair Marking the continuous area with the median value of 1 to obtain a connected area marking matrix Wherein Representing time-frequency points Belonging to the first The number of the communication areas is equal to the number of the communication areas, , Is a background area; for each connected region Three parameters were calculated: (1) Starting time of day The actual time corresponding to the minimum index of the connected area on the time axis is as follows: (2) End time The actual time corresponding to the maximum index of the connected area on the time axis is as follows: (3) Center frequency Calculating the weighted centroid of the connected region on the frequency axis by taking the energy of the time-frequency point as the weight, namely: in the formula, Is the first And (5) collecting time-frequency points of the connected areas.
3. The method for sorting synchronous frequency hopping network stations based on multidimensional feature clustering according to claim 2, wherein the step 2 is specifically as follows: Based on step 1 And For the original received signal , , Time interception is carried out for the total sampling point number to obtain a single-hop signal : In the formula, In order to intercept the start index, For a single-hop signal sample point number, Representing a downward rounding; designing a bandpass filter pair Filtering, the center frequency of the filter is Bandwidth of , To obtain pure single-hop signal after filtering for frequency hopping frequency interval 。
4. The method for sorting synchronous frequency hopping network stations based on multi-dimensional feature clustering according to claim 3, wherein the step 3 is specifically as follows: for each filtered single-hop signal Extracting 6-dimensional multi-domain features and constructing a feature set , Is the first The feature vector of the jump signal has the following calculation formulas: (1) Total energy of signal Reflecting the overall amplitude intensity of the signal, namely: in the formula, Points representing single-hop signals; (2) Real part variance Measuring the discrete degree of a real part of a signal, reflecting stationarity, namely: in the formula, Representing the real part; Mean value representing real part; (3) Spectral skewness Describing the symmetry of the power spectral density distribution, namely: in the formula, For the power spectral density of a single-hop signal, As the mean value of the frequency spectrum, As a standard deviation of the frequency spectrum, Is the number of frequency spectrum points; (4) Spectral entropy Reflecting the dispersion degree of energy in the frequency domain, namely: in the formula, Is a spectral probability distribution; (5) Envelope mean value Reflecting the central trend of the signal envelope, namely: (6) Angle of arrival Estimating by using a MUSIC algorithm, obtaining a signal subspace and a noise subspace through array covariance matrix feature decomposition, and constructing a spectral function by utilizing orthogonality of the signal subspace and the noise subspace, namely: in the formula, As the direction vector of the array, Is the number of the array elements, and is the number of the array elements, The distance between the array elements is the distance between the array elements, As a function of the wavelength of the signal, In the form of a matrix of noise subspaces, Representing conjugate transpose, DOA estimation value is the angle corresponding to the peak value of the spectrum function: 。
5. The method for sorting synchronous frequency hopping network stations based on multidimensional feature clustering according to claim 4, wherein the step 4 is specifically as follows: For feature set Standardized treatment is carried out, the first Dimension features The normalized formula of (2) is: in the formula, Is the first First of jump signal The original characteristic value is maintained and the original characteristic value is obtained, Is the first The mean value of the dimensional features, Is the first The standard deviation of the dimensional characteristics is calculated, Is the normalized characteristic value.
6. The method for sorting synchronous frequency hopping network stations based on multidimensional feature clustering according to claim 5, wherein the step 5 is specifically as follows: Initializing an isolated forest model, setting the number of trees Proportion of abnormal samples Based on standardized feature sets Training a model; Calculating the path length of each sample : In the formula, For the original path length of the sample in a single isolated tree, Represent the first The feature vector of the individual samples is used, For the path length correction value, Representing the total number of samples, Is Euler constant; calculating anomaly score Judging an abnormal sample: in the formula, For the average path length of the sample in all the isolated trees If the sample is abnormal, otherwise, the sample is normal, Separating to obtain normal sample set And an abnormal sample set Recording an abnormal sample index 。
7. The method for sorting synchronous frequency hopping network stations based on multidimensional feature clustering according to claim 6, wherein the step 6 is specifically as follows: obtaining the estimation value of the number of the synchronous frequency hopping network by counting the mode of the number of the connected areas at each moment As the cluster number, pairs Performing K-means clustering; defining a clustering objective function, and minimizing the sum of squares of samples in a cluster: in the formula, Is the first The number of clusters is one, Is the center of the cluster and, Is Euclidean distance; Iteratively updating the cluster center until the convergence condition is met, namely the cluster center variation is smaller than the threshold value Or to a maximum number of iterations The cluster center update formula is: in the formula, Is the first The number of samples of the cluster; Outputting normal sample clustering labels And cluster center matrix 。
8. The method for sorting synchronous frequency hopping network stations based on multidimensional feature clustering according to claim 7, wherein the step 7 is specifically as follows: For each abnormal sample Calculate it and each cluster center Euclidean distance of (c): in the formula, Is the first The first cluster center A dimension characteristic value; Assigning the outlier samples to cluster classes closest to: recombining the final sorting labels, namely filling normal sample labels and abnormal sample labels according to the original sample sequence to obtain And (5) finishing the separation of the synchronous frequency hopping network platform.
9. The method for sorting synchronous frequency hopping network stations based on multi-dimensional feature clustering as set forth in claim 8, wherein said threshold adjustment factor The value is 1.0-1.5.

Description

Synchronous frequency hopping network station sorting method based on multidimensional feature clustering Technical Field The invention belongs to the technical field of communication, and particularly relates to a synchronous frequency hopping network station sorting method based on multidimensional feature clustering. Background The frequency hopping communication is a core technology of a communication network platform by virtue of the advantages of low interception probability, strong anti-interference performance and the like, the synchronous frequency hopping network platform is widely applied in key tasks due to the characteristics of consistent hopping time and periodical synchronization, and the accurate sorting of the synchronous frequency hopping network platform in communication reconnaissance is a key link for acquiring information, but the prior art has the defects that on one hand, the synchronous frequency hopping network platform is completely consistent in hopping time and period, only the orthogonal distribution of a frequency hopping set is realized, the traditional time feature-dependent sorting method is invalid and is required to be dependent on the multi-dimensional feature sorting network platform, on the other hand, the traditional sorting method is mostly dependent on single or few features (such as signal energy and arrival angle DOA), when the deployment positions of a plurality of network platforms are similar, DOA estimation difference is extremely small, the single features are not effectively distinguished and are easily affected by noise, so that sorting precision is reduced, in addition, a traditional clustering algorithm (such as K-means) is sensitive to an initial clustering center, is easily sunk into a local optimal solution, the multi-dimensional feature space of a synchronous frequency hopping signal is in a spherical distribution, the traditional clustering algorithm is not spherically distributed, meanwhile, the traditional clustering algorithm is not matched with the algorithm, the problem that the prior art is difficult to meet the influence on the sorting error of the complex, and the sorting method is difficult to meet the problem of the overall error-dependent on the high-dimensional noise, and the electromagnetic noise-dependent on the signal, and the noise-dependent on the method is difficult to be easily influenced by the method, and the problem is difficult to be greatly influenced by the noise-based on the method, and has the method. Disclosure of Invention In order to overcome the defects of the prior art, the invention provides a synchronous frequency hopping network table sorting method based on multidimensional feature clustering, which comprises the steps of firstly carrying out short-time Fourier transform (STFT) on received synchronous frequency hopping signals to obtain a time-frequency matrix, carrying out binarization on the time-frequency matrix, then carrying out connected region marking, calculating to obtain the starting time, the ending time and the center frequency of each frequency hopping signal, carrying out time interception and filtering treatment on the received signals according to the starting time, the ending time and the center frequency of each frequency hopping signal, extracting multidimensional complementary features for single frequency hopping signals to construct a high-degree feature space, carrying out standardized pretreatment on the features to eliminate dimension differences, then detecting and separating abnormal samples by utilizing isolated forests to avoid interference on clustering results, carrying out core clustering on normal samples by adopting a K-means algorithm to determine cluster centers, and finally distributing the abnormal samples to cluster types closest to each other to complete accurate sorting of the whole samples. The method can effectively inhibit the influence of interference and abnormal samples, improves the precision rate, recall rate and stability of synchronous frequency hopping network station sorting in a complex electromagnetic environment, and meets the index requirements of communication reconnaissance. The technical scheme adopted for solving the technical problems is as follows: step 1, synchronizing time-frequency processing of frequency hopping signals and single-hop parameter estimation; step 2, single-hop signal interception and filtering treatment; step 3, extracting multidimensional features; step 4, characteristic standardized pretreatment; Step 5, detecting and separating an isolated forest anomaly sample; Step 6:K-means core clustering and cluster center calculation; And 7, abnormal sample distribution and final sorting. Preferably, the step 1 specifically includes: Step 1.1, constructing a time-frequency matrix by short-time Fourier transform; for received synchronous frequency hopping mixed signal Performing short-time Fourier transform, mapping the time domain signal to