An electromagnetic signal identification method, system, electronic device and medium

Abstract

The present application belongs to the technical field of electromagnetic measurement, and aims to provide an electromagnetic signal identification method, system, electronic device and medium. The present application provides a method for real-time signal sensing in a complex electromagnetic environment, which can be used for identification of unknown electromagnetic signals in non-cooperative communication. Specifically, the present application uses a KNN algorithm to identify the signal label of a to-be-tested sample signal, searches for k nearest neighbor points of the to-be-tested sample signal in the constructed FBST by using the KNN algorithm, and then obtains the signal label of the to-be-tested sample signal according to the signal labels of the k nearest neighbor points. Thus, the system has better robustness. Furthermore, the present application improves the k nearest neighbor algorithm based on the FBST, which can further reduce the calculation amount, improve the identification efficiency of the electromagnetic signal, and solve the problem of low identification efficiency of the signal sensing system in the case of a large sample amount.

Description

Technical Field

[0001] This invention belongs to the field of electromagnetic measurement technology, specifically relating to an electromagnetic signal identification method, system, electronic device, and medium. Background Technology

[0002] In recent years, radar communication technology has developed rapidly, leading to a surge in demand for on-site signal detection. In communication and radar countermeasures, the transmission system is a crucial component. Transmission systems are primarily divided into cooperative and non-cooperative communication. Cooperative communication, due to prior communication between the two parties and the availability of certain prior knowledge, such as signal modulation and demodulation methods, makes signal processing at the receiving end relatively easy. However, in non-cooperative environments, non-cooperative communication is the dominant method, and signal modulation identification is an essential process. In situations where prior knowledge is insufficient or nonexistent, effectively identifying detected enemy signals to determine their modulation type and other information has a decisive impact on subsequent signal processing. Therefore, research on effective signal identification and processing becomes particularly important.

[0003] Electromagnetic signal recognition mainly consists of four steps: first, the received electromagnetic signal is preprocessed; then, feature parameters are extracted from the preprocessed signal; then, a recognition classifier is designed based on the extracted feature parameters; and finally, the recognition classifier is used to make a classification decision to obtain the recognition result.

[0004] In the prior art, Shen Weiguo, Wang Wei, and Xu Xiaofeng of the 36th Research Institute of China Electronics Technology Group Corporation disclosed a signal modulation type identification method and system in Chinese invention patent publication number CN105119862A. The method includes: preprocessing the signal to be identified to extract a predetermined number of feature parameters and using a feature vector composed of the predetermined number of feature parameters to characterize the signal to be identified; using an optimal projection matrix to extract features from the signal to be identified and projecting the signal to be identified onto a low-dimensional feature subspace; wherein the optimal projection matrix is ​​obtained through a local preservation projection algorithm; calculating the Euclidean distance between the signal to be identified in the low-dimensional feature subspace and a training signal with a known signal modulation type, and determining the signal modulation type of the signal to be identified based on a nearest neighbor algorithm using Euclidean distance.

[0005] However, in using the prior art, the inventors discovered at least the following problems:

[0006] The aforementioned prior art uses the nearest neighbor algorithm, which determines the modulation type of the signal under test based on the modulation type of only one nearest neighbor, resulting in poor robustness. Summary of the Invention

[0007] The present invention aims to solve the above-mentioned technical problems to at least a certain extent, and provides an electromagnetic signal identification method, system, electronic device and medium.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] In a first aspect, the present invention provides an electromagnetic signal identification method, comprising:

[0010] Electromagnetic signals were collected from the study area and preprocessed to obtain a dataset with signal labels.

[0011] The specified feature parameters are extracted from the dataset to obtain a feature vector composed of the feature parameters and a feature parameter matrix composed of all feature vectors; wherein the feature vector is used to characterize the corresponding electromagnetic signal and the feature parameter matrix is ​​used to characterize the dataset.

[0012] The dataset is divided into a test sample signal set and a training sample signal set, and the FBST is constructed using the training sample signal set; wherein, the electromagnetic signals in the test sample signal set are test sample signals, and the electromagnetic signals in the training sample signal set are training sample signals.

[0013] The KNN algorithm is used to retrieve the k nearest neighbors of the test sample signal in the constructed FBST;

[0014] The signal labels of the sample signal to be tested are obtained based on the signal labels of the k nearest neighbors.

[0015] This invention provides a method for real-time signal sensing in complex electromagnetic environments, which can be used to identify unknown electromagnetic signals in non-cooperative communication. Specifically, this invention uses the KNN algorithm to identify the signal tags of the sample signal under test. By using the KNN algorithm to retrieve the k nearest neighbors of the sample signal under test in the constructed FBST, and then obtaining the signal tag of the sample signal under test based on the signal tags of the k nearest neighbors, the system can have better robustness. Furthermore, this invention improves the k-nearest neighbor algorithm based on FBST, which can further reduce the amount of computation and improve the recognition efficiency of electromagnetic signals, thus solving the problem of low recognition efficiency of signal sensing systems under large sample sizes.

[0016] In one possible design, the specified feature parameters extracted from the dataset include: the maximum value of the zero-center normalized instantaneous amplitude spectral density, the standard deviation of the absolute value of the zero-center normalized instantaneous amplitude, the standard deviation of the absolute value of the instantaneous frequency of the zero-center normalized non-weak signal segment, and the standard deviation of the instantaneous phase nonlinear component of the zero-center normalized non-weak signal segment.

[0017] The expression for the feature parameter matrix is:

[0018]

[0019] Where Γ (r) Let r represent the r-th feature parameter matrix, where r = 1, 2, ..., R, and R is the specified number of feature parameters; Let represent the r-th feature parameter of the j-th sample of the i-th signal tag, where i = 1, 2, ..., n, n is the number of signal tags, and j = 1, 2, ..., m, m is the number of electromagnetic signal samples of each type of signal tag.

[0020] In one possible design, the training sample signals in the training sample signal set are represented as follows:

[0021]

[0022] w = m × (i - 1) + j;

[0023] Among them, S w This represents the w-th training sample signal, where w is the index of the training sample signal;

[0024] The test sample signals in the test sample signal set are represented as follows:

[0025]

[0026] Among them, T t Let t represent the t-th sample signal to be tested, where t is the index of the sample signal to be tested.

[0027] In one possible design, the method further includes:

[0028] The variance of each characteristic parameter is calculated using the following formula:

[0029]

[0030]

[0031]

[0032] in, Let represent the average value of the j-th column of the r-th eigenvalue matrix, and n represent the number of rows in the r-th eigenvalue matrix. D(z) represents the variance of the j-th column of the r-th eigenparameter matrix. (r) ) represents the variance of the r-th feature parameter matrix, and m represents the number of columns in the r-th feature parameter matrix;

[0033] Then sort the variances from largest to smallest, and modify the training sample signals S according to the order of variances.w With the sample signal T to be tested t The order of the coordinate points is used to obtain the modified training sample signal and the modified test sample signal; whereby the modified training sample signal is represented as:

[0034]

[0035] in, The feature parameter representing the training sample signal represented by the r'th variance in the variance ranking;

[0036] The modified sample signal to be tested is represented as follows:

[0037]

[0038] in, This represents the characteristic parameter of the sample signal to be tested, which is represented by the r'th variance in the variance ranking.

[0039] In one possible design, constructing the FBST using the training sample signal set includes:

[0040] Select all training sample signals S w The first feature parameter is used as the current partition domain of the current partitioning layer of FBST;

[0041] Select the median in the current partition and use the training sample signal it represents as the current node;

[0042] Values ​​less than the median are assigned to the left subtree, and values ​​greater than the median are assigned to the right subtree.

[0043] Select training sample signals S from the left and right subtrees respectively. w The next feature parameter is used as the current partition domain of the current partitioning layer of FBST;

[0044] Reselect the median in the current partition domain until all training sample signals S are obtained. w All are divided to obtain the final FBST.

[0045] In one possible design, the nearest neighbor of the sample signal under test is selected using Euclidean distance, which is calculated using the following formula:

[0046]

[0047] Where, d q Let R be the Euclidean distance between the q-th training sample signal and the test sample signal, and let R be the specified number of feature parameters. Let r be the r-th feature parameter of the sample signal to be tested. Let be the r-th feature parameter of the q-th training sample signal.

[0048] In one possible design, any test sample signal is denoted as the target point; the KNN algorithm is used to retrieve the k nearest neighbors of the test sample signal in the constructed FBST, including:

[0049] Starting from the root node, compare the feature parameters of the target point with the feature parameters of the split point corresponding to the current layer of the FBST. If the feature parameters of the target point are less than the feature parameters of the split point, go to the left subtree; otherwise, go to the right subtree. Then recursively search downwards until a leaf node is reached, which is recorded as the current node.

[0050] Let the minimum distance d be the distance between the target point and the current node. min The distance between the target point and the sibling node of the current node is d. b The distance between the target point and the parent node of the current node is d. f Wherein, the parent node is the parent node of the current node, and the sibling node is the adjacent node that belongs to the same parent node as the current node;

[0051] If d min >d b If d, then update the sibling node to the current node, and simultaneously update the sibling node and parent node of the current node. min <d b Then compare d min With d f ;

[0052] If d min >d f If d, then update the parent node to the current node, and simultaneously update the sibling nodes and parent node of the current node. min <d f If the current node is the nearest neighbor, then the current node is the nearest neighbor.

[0053] Mark the nearest neighbor as D pq And change its distance from the target point to infinity;

[0054] Starting from the root node again and repeating k times, we can find the k nearest neighbors.

[0055] In a second aspect, the present invention provides an electromagnetic signal identification system for implementing the electromagnetic signal identification method as described in any of the preceding claims; the electromagnetic signal identification system includes:

[0056] An electromagnetic signal acquisition module is used to acquire electromagnetic signals from the study area and preprocess the electromagnetic signals to obtain a dataset with signal labels.

[0057] The feature extraction module is communicatively connected to the electromagnetic signal acquisition module and is used to extract specified feature parameters from the dataset, and obtain a feature vector composed of the feature parameters, and a feature parameter matrix composed of all feature vectors; wherein, the feature vector is used to characterize the corresponding electromagnetic signal, and the feature parameter matrix is ​​used to characterize the dataset;

[0058] The FBST construction module, which is communicatively connected to the feature extraction module, is used to divide the dataset into a test sample signal set and a training sample signal set, and to construct the FBST using the training sample signal set; wherein, the electromagnetic signals in the test sample signal set are test sample signals, and the electromagnetic signals in the training sample signal set are training sample signals.

[0059] The feature matching module is communicatively connected to the FBST construction module and is used to retrieve the k nearest neighbors of the test sample signal in the constructed FBST using the KNN algorithm.

[0060] The signal label determination module is communicatively connected to the feature matching module and is used to obtain the signal label of the sample signal to be tested based on the signal labels of the k nearest neighbors.

[0061] Thirdly, the present invention provides an electronic device, comprising:

[0062] Memory, used to store computer program instructions; and,

[0063] A processor for executing the computer program instructions to perform the operation of the electromagnetic signal identification method as described in any of the preceding claims.

[0064] Fourthly, the present invention provides a computer-readable storage medium for storing computer-readable computer program instructions configured to perform operations of the electromagnetic signal identification method as described in any of the preceding claims when executed. Attached Figure Description

[0065] Figure 1 This is a flowchart of the electromagnetic signal identification method in this invention;

[0066] Figure 2 This is a flowchart of the procedure for retrieving FBST in this invention;

[0067] Figure 3 This is a baseband signal diagram of the 2ASK (Amplitude Shift Keying, ASK) modulated signal used in this invention;

[0068] Figure 4 This is a diagram of the 2ASK modulation signal used in this invention;

[0069] Figure 5 This is a baseband signal diagram of the 2PSK (Phase Shift Keying, PSK) modulated signal used in this invention;

[0070] Figure 6 This is a diagram of the 2PSK modulation signal used in this invention;

[0071] Figure 7 This is a baseband signal diagram of the 2FSK (Frequency Shift Keying, FSK) modulated signal used in this invention;

[0072] Figure 8 This is a diagram of the 2FSK modulation signal used in this invention;

[0073] Figure 9 This is a baseband signal diagram of the 4ASK modulation signal used in this invention;

[0074] Figure 10 This is a diagram of the 4ASK modulation signal used in this invention;

[0075] Figure 11 This is a diagram of the AM (Amplitude Modulation, AM) radar signal used in this invention;

[0076] Figure 12 This is a diagram of the LFM (Linear Frequency Modulation) radar signal used in this invention;

[0077] Figure 13 This is a diagram of the CW (Continuous Waves, CW) radar signal used in this invention;

[0078] Figure 14 This is a comparison chart of the detection results of nine test sample signals detected according to the method disclosed in this invention and the actual modulation type of the test sample signals;

[0079] Figure 15 This is a comparison chart of the detection results of 99 test sample signals detected according to the method disclosed in this invention and the actual modulation type of the test sample signals;

[0080] Figure 16 This is a block diagram of the electromagnetic signal recognition system in this invention. Detailed Implementation

[0081] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0082] Example 1:

[0083] This embodiment discloses an electromagnetic signal identification method, which can be executed by, but is not limited to, a computer device or virtual machine with certain computing resources, such as a personal computer, smartphone, personal digital assistant or wearable device, or by a virtual machine.

[0084] like Figure 1 As shown, an electromagnetic signal identification method may include, but is not limited to, the following steps:

[0085] S1. Electromagnetic signal acquisition: Electromagnetic signals are acquired from the study area and preprocessed to obtain a dataset with signal labels.

[0086] In this embodiment, electromagnetic signals are acquired through on-site acquisition. Specifically, when acquiring electromagnetic signals from the study area, a signal propagation model is first constructed based on the electromagnetic propagation environment of the study area. Then, electromagnetic signals are radiated within the study area and received from a receiver located within the study area. The electromagnetic signals are then preprocessed to obtain a dataset with signal tags. Furthermore, the methods for receiving electromagnetic signals by the receiver include, but are not limited to, far-field measurement, near-field measurement, and compacted-field measurement. Near-field measurement methods include, but are not limited to, planar near-field measurement, cylindrical near-field measurement, and spherical near-field measurement. As an example, in this embodiment, a total of 162 sets of electromagnetic signals are acquired from the study area. After preprocessing, these 162 sets of electromagnetic signals are labeled and used to construct a dataset.

[0087] In this embodiment, the signal tag includes, but is not limited to, signal modulation type and / or individual type of signal radiation source.

[0088] S2. Feature Extraction: Extract specified feature parameters from the dataset and obtain feature vectors composed of the feature parameters, and feature parameter matrix composed of all feature vectors; wherein the feature vectors are used to characterize the corresponding electromagnetic signals, and the feature parameter matrix is ​​used to characterize the dataset.

[0089] In step S2, the specified feature parameters extracted from the dataset include: the maximum value of the zero-center normalized instantaneous amplitude spectral density γ. max Zero-center normalized instantaneous amplitude absolute value standard deviation σ aa The standard deviation σ of the absolute value of the instantaneous frequency of the zero-center normalized non-weak signal segment af Among them, the non-weak signal is the signal greater than the normalized instantaneous amplitude mean, and the standard deviation σ of the instantaneous phase nonlinear component of the zero-center normalized non-weak signal segment. dp ;

[0090] Specifically, in this embodiment, the maximum instantaneous amplitude spectral density γ centered at zero is... max for:

[0091]

[0092] in, For Fourier operators;

[0093] Zero-center normalized instantaneous amplitude absolute value standard deviation σ aa for:

[0094]

[0095] The standard deviation σ of the absolute value of the instantaneous frequency of a non-weak signal segment normalized to zero center af for:

[0096]

[0097] Zero-center normalized standard deviation of instantaneous phase nonlinear component σ of weak signal segment dp for:

[0098]

[0099] In the above formula, N s Let m be the number of signal sampling points, u be the index of the sampling point in the dataset, a(u) be the instantaneous amplitude of the signal, and m be the number of sampling points. a Let a be the instantaneous average amplitude of the signal. n (u) represents the normalized instantaneous amplitude of the signal, a cn (u) represents the zero-center normalized instantaneous amplitude of the signal; Let c be the normalized instantaneous mean amplitude of the signal, and c be the mean amplitude across all N signals. s The number of non-weak signal value sampling points in the sampling points, where f(u) is the instantaneous frequency of the sampled signal. R is the instantaneous frequency of the normalized sampled signal. s The signal symbol rate; f is the instantaneous phase of the signal. s f c These are the signal frequency and the carrier frequency, respectively. The nonlinear phase component of the signal is as follows:

[0100]

[0101]

[0102] a cn (u)=a n (u)-1

[0103]

[0104]

[0105] The expression for the feature parameter matrix is:

[0106]

[0107] Where Γ (r) Let R represent the r-th feature parameter matrix, where r = 1, 2, ..., R, and R is the specified number of feature parameters. As an example, this embodiment takes four feature parameters, i.e., R = 4. Let represent the r-th feature parameter of the j-th sample of the i-th signal tag, where i = 1, 2, ..., n, n is the number of signal tags, and j = 1, 2, ..., m, m is the number of electromagnetic signal samples of each type of signal tag.

[0108] In step S2, after extracting the feature parameters, the feature parameters are used to construct a feature vector, so that the training sample signals in the training sample signal set are represented as follows:

[0109]

[0110] w = m × (i - 1) + j;

[0111] Among them, S w Let w represent the w-th training sample signal, where w is the sequence number of the training sample signal; as an example, in this embodiment, the value range of w is (1, 63).

[0112] As an example, in this embodiment, the number of specified feature parameters R = 4, the number of signal tags n = 7, and the number of electromagnetic signal samples for each signal tag m = 9, then the training sample signals are represented as follows:

[0113]

[0114] w = 9 × (i - 1) + j.

[0115] The test sample signals in the test sample signal set are represented as follows:

[0116]

[0117] Among them, T t Let t represent the t-th sample signal to be tested, where t is the index of the sample signal to be tested; as an example, in this embodiment, the value range of t is (1, 99).

[0118] As an example, in this embodiment, the signal of the sample to be tested is represented as:

[0119]

[0120] Furthermore, in step S2 of this embodiment, the method further includes:

[0121] The variance of each characteristic parameter is calculated using the following formula:

[0122]

[0123]

[0124]

[0125] in, Let represent the average value of the j-th column of the r-th eigenvalue matrix, and n represent the number of rows in the r-th eigenvalue matrix. D(z) represents the variance of the j-th column of the r-th eigenparameter matrix. (r) ) represents the variance of the r-th feature parameter matrix, and m represents the number of columns in the r-th feature parameter matrix;

[0126] As an example, in this embodiment, the variance of each characteristic parameter is calculated using the following formula:

[0127]

[0128]

[0129]

[0130] Then sort the variances from largest to smallest, and modify the training sample signals S according to the order of variances. w With the sample signal T to be tested t The order of the coordinate points is used to obtain the modified training sample signal and the modified test sample signal, so as to determine the feature parameters of FBST layer by layer from top to bottom according to the order of variance from largest to smallest; wherein, the modified training sample signal is expressed as:

[0131]

[0132] in, The feature parameter representing the training sample signal represented by the r'th variance in the variance ranking;

[0133] The modified sample signal to be tested is represented as follows:

[0134]

[0135] in, This represents the characteristic parameter of the sample signal to be tested, which is represented by the r'th variance in the variance ranking.

[0136] As an example, in this embodiment, the modified training sample signal and the modified test sample signal are represented as follows:

[0137]

[0138]

[0139] S3. FBST (Full Binary Search Tree) Construction: The dataset is divided into a test sample signal set and a training sample signal set, and an FBST is constructed using the training sample signal set; wherein, the electromagnetic signals in the test sample signal set are the test sample signals, and the electromagnetic signals in the training sample signal set are the training sample signals; as an example, in this embodiment, the dataset is divided into a test sample signal dataset including 99 sets of electromagnetic signals, and a training sample signal set including 63 sets of electromagnetic signals. The number of layers K of the constructed FBST is 6, the number of signals of each modulation type in the training sample signal set is 9, and the number of modulation types in the training sample signal set is 7.

[0140] In this embodiment, the number of signal labels in the training sample signal set, the number of electromagnetic signals corresponding to the signal labels, and the number of FBST layers satisfy the following conditions:

[0141] n×m=2 K -1;

[0142] In the formula, n is the number of signal tags of electromagnetic signals in the training sample signal set, m is the number of electromagnetic signals corresponding to each signal tag in the training sample signal set, K is the number of layers of FBST, and K is a positive integer.

[0143] In step S3, the FBST is constructed using the training sample signal set, including:

[0144] S301. Select all training sample signals S w The first feature parameter is used as the current partition domain of the current partitioning layer of FBST;

[0145] S302. Select the median in the current partition and use the training sample signal it represents as the current node;

[0146] S303. Values ​​less than the median are assigned to the left subtree, and values ​​greater than the median are assigned to the right subtree;

[0147] S304. Select training sample signals S from the left and right subtrees respectively. w The next feature parameter is used as the current partition domain of the current partitioning layer of FBST;

[0148] S305. Reselect the median in the current partition, i.e., continuously repeat steps S302-S304 until all training sample signals S w All are divided to obtain the final FBST.

[0149] S4.FBST feature matching: such as Figure 2 As shown, the KNN (K-Nearest Neighbor) algorithm is used to retrieve the k nearest neighbors of the test sample signal in the constructed FBST, that is, the k training sample signals that are closest to the test sample signal. The nearest neighbors of the test sample signal are the nodes that are closest to the test sample signal.

[0150] In this embodiment, the nearest neighbor point of the sample signal to be tested is selected using Euclidean distance. The formula for calculating Euclidean distance is as follows:

[0151]

[0152] Where, d q Let R be the Euclidean distance between the q-th training sample signal and the test sample signal, and let R be the specified number of feature parameters. Let r be the r-th feature parameter of the sample signal to be tested. Let be the r-th feature parameter of the q-th training sample signal.

[0153] As an example, in this embodiment, the formula for calculating the Euclidean distance is as follows:

[0154]

[0155] Specifically, in step S4, if any test sample signal is designated as the target point, the KNN algorithm is used to retrieve the k nearest neighbor points of the test sample signal in the constructed FBST, including:

[0156] S401. Starting from the root node, compare the feature parameters of the target point with the feature parameters of the split point corresponding to the current layer of the FBST. If the feature parameters of the target point are less than the feature parameters of the split point, go to the left subtree; otherwise, go to the right subtree. Then recursively search downwards until a leaf node is reached. This leaf node is recorded as the current node.

[0157] S402. Let the minimum distance d be the distance between the target point and the current node. min The distance between the target point and the sibling node of the current node is d. b The distance between the target point and the parent node of the current node is d. f Wherein, the parent node is the parent node of the current node, and the sibling node is the adjacent node that belongs to the same parent node as the current node;

[0158] S403. If d min >d b If d min <d b Then compare d min With d f ;

[0159] S404. If d min >d f If d min <d f If the current node is the nearest neighbor, then the current node is the nearest neighbor.

[0160] S405. Mark the nearest neighbor as D. pq And change its distance from the target point to infinity;

[0161] S406. Start from the root node again and repeat k times, that is, repeat steps S401 to S405 k times to retrieve the k nearest neighbor points, that is, the k training sample signals that are closest to the test sample signal.

[0162] S5. Determine the signal label of the sample signal to be tested: Based on the signal labels of the k nearest neighbors, obtain the signal label of the sample signal to be tested. The signal label of the sample signal to be tested is the signal label that has the largest proportion among the k nearest neighbors.

[0163] Specifically, in this embodiment, a voting method is used to determine the signal label of the sample to be tested. Specifically, the signal label of the sample signal to be tested is determined based on the labels of the k nearest neighbors, including:

[0164] The k nearest neighbors in step S4 are voted on to obtain the signal label of the sample signal to be tested.

[0165] It should be noted that the voting principle is as follows: among the k nearest neighbors, for each training sample signal with a certain signal label, the vote count for that signal label is incremented by one, and the signal label with the highest number of votes is finally output as the signal label of the sample signal to be tested.

[0166] This embodiment provides a method for real-time signal sensing in complex electromagnetic environments, which can be used to identify unknown electromagnetic signals in non-cooperative communication. Specifically, this embodiment uses the KNN algorithm to identify the signal label of the sample signal to be tested. By using the KNN algorithm to retrieve the k nearest neighbors of the sample signal to be tested in the constructed FBST, the signal label of the sample signal to be tested is obtained according to the signal labels of the k nearest neighbors. This makes the system more robust. Furthermore, this embodiment improves the k-nearest neighbor algorithm based on FBST, which can further reduce the amount of computation and improve the recognition efficiency of electromagnetic signals, thus solving the problem of low recognition efficiency of signal sensing systems under large sample sizes.

[0167] It should be noted that this embodiment uses the KNN algorithm to retrieve the k nearest neighbors of the test sample signal in the constructed FBST. The KNN algorithm means that each test sample signal can be represented by k training sample signals that are closest to it. The core idea of ​​the KNN algorithm is to consider the signal label with the largest weight among the k training sample signals that are closest to the test sample signal as the signal label of the test sample signal, which has better robustness. Since the KNN algorithm has the disadvantage of large computation and long processing time when there are many training samples, this embodiment further adopts the FBST implementation method to solve this problem, thereby effectively reducing the computation and shortening the running time.

[0168] The effects of this invention can be further illustrated by the following simulation results:

[0169] 1. Experimental conditions and contents:

[0170] Experimental conditions:

[0171] Hardware: Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz, NVIDIA GeForce GTX 1650.

[0172] Software: Windows 10, Matlab R2020a.

[0173] Experiment content:

[0174] (1) First, electromagnetic signals of seven modulation types, including 2ASK, 2PSK, 4ASK, 2FSK, AM, LFM and CW, are radiated and collected in the study area. Nine samples are collected for each modulation type. The above 63 sets of signals are labeled to form a training sample signal set. Then, 99 sets of signals to be tested are randomly collected and labeled to form a test sample signal set.

[0175] (2) Following step S2 of this embodiment, calculate the maximum instantaneous amplitude spectral density γ of the zero-center normalized instantaneous amplitude spectral density for both the training sample signal set and the test sample signal set. max Zero-center normalized instantaneous amplitude absolute value standard deviation σ aa The standard deviation σ of the absolute value of the instantaneous frequency of the zero-center normalized non-weak signal segment af The standard deviation σ of the instantaneous phase nonlinear component of the zero-center normalized non-weak signal segment dp There are four feature parameters in total. Then, a feature parameter matrix is ​​generated, and this matrix is ​​converted into an FBST structure according to step S3 of this embodiment. Subsequently, the FBST feature matching algorithm in step S4 of this embodiment sequentially retrieves the k nearest neighbor nodes of the test sample signal in the FBST. Finally, the modulation type of the test signal is determined by a voting method.

[0176] 2. Analysis of experimental results:

[0177] Reference Figure 3 and Figure 4 The diagram shows the information of five symbols of the 2ASK baseband signal and modulation signal collected in the experiment. The horizontal axis represents the time range [0, 0.05] in microseconds (μs), and the vertical axis represents the amplitude in volts (V).

[0178] Reference Figure 5 and Figure 6 The figure shows the information of two symbols of the 2PSK baseband signal and modulation signal collected in the experiment. The horizontal axis represents the time range [0, 0.02] in microseconds (μs), and the vertical axis represents the amplitude in volts (V).

[0179] Reference Figure 7 and Figure 8 The diagram shows the information of five symbols of the 2FSK baseband signal and modulation signal collected in the experiment. The horizontal axis represents the time range [0, 0.05] in microseconds (μs), and the vertical axis represents the amplitude in volts (V).

[0180] Reference Figure 9 and Figure 10The diagram shows the information of 10 symbols of the 4ASK baseband signal and modulation signal collected in the experiment. The horizontal axis represents the time range [0, 0.1] in microseconds (μs), and the vertical axis represents the amplitude in volts (V).

[0181] Reference Figure 11 , represents the AM radar signal and its envelope obtained in the experiment. The horizontal axis represents time, ranging from [0,8], in microseconds (μs), and the vertical axis represents amplitude, in volts (V).

[0182] Reference Figure 12 , represents the LFM radar signal collected in the experiment. The horizontal axis represents time, ranging from [0, 2.5], with the unit being microseconds (μs). The vertical axis represents amplitude, with the unit being volts (V).

[0183] Reference Figure 13 , represents the CW radar signal and its envelope obtained in the experiment. The horizontal axis represents time, ranging from [0,8], in nanoseconds (ns), and the vertical axis represents amplitude, in volts (V).

[0184] 3. Summary of Experimental Results:

[0185] Figure 14 This is a comparison chart of the results obtained from judging 99 test sample signals and the actual modulation type of the test sample signals. Figure 14 The horizontal axis represents the sequence number of the sample signal under test, ranging from [1, 99], and the vertical axis represents the modulation type of the sample signal under test, namely 2ASK, 2PSK, 4ASK, 2FSK, AM, LFM and CW. Figure 14 The black dots in the image represent the actual modulation type of the sample signal under test. Figure 14 The modulation type of the test sample signal determined by the FBST algorithm in this embodiment is indicated by a black box. (Refer to...) Figure 14 As can be seen, this embodiment uses the FBST-optimized KNN algorithm to determine the signal modulation type, achieving an accuracy of 86.87%. Furthermore, the single-sample detection time is achieved through the system's internal clock reading algorithm, referencing... Figure 15 It can be seen that in the 10 random single sample detections, the response time is in the millisecond range, and the average single sample detection response time is 2.37974ms.

[0186] The FBST-optimized KNN algorithm in this embodiment has a high success rate in judging the sample signal under test, and has good real-time performance. It can accurately and efficiently judge the modulation type of the signal in complex electromagnetic environment.

[0187] Example 2:

[0188] This embodiment discloses an electromagnetic signal identification system for implementing the electromagnetic signal identification method in Embodiment 1; such as Figure 16 As shown, the electromagnetic signal identification system includes:

[0189] An electromagnetic signal acquisition module is used to acquire electromagnetic signals from the study area and preprocess the electromagnetic signals to obtain a dataset with signal labels.

[0190] The feature extraction module is communicatively connected to the electromagnetic signal acquisition module and is used to extract specified feature parameters from the dataset, and obtain a feature vector composed of the feature parameters, and a feature parameter matrix composed of all feature vectors; wherein, the feature vector is used to characterize the corresponding electromagnetic signal, and the feature parameter matrix is ​​used to characterize the dataset;

[0191] The FBST construction module, which is communicatively connected to the feature extraction module, is used to divide the dataset into a test sample signal set and a training sample signal set, and to construct the FBST using the training sample signal set; wherein, the electromagnetic signals in the test sample signal set are test sample signals, and the electromagnetic signals in the training sample signal set are training sample signals.

[0192] The feature matching module is communicatively connected to the FBST construction module and is used to retrieve the k nearest neighbors of the test sample signal in the constructed FBST using the KNN algorithm.

[0193] The signal label determination module is communicatively connected to the feature matching module and is used to obtain the signal label of the sample signal to be tested based on the signal labels of the k nearest neighbors.

[0194] Example 3:

[0195] Based on Embodiment 1 or 2, this embodiment discloses an electronic device, which may be a smartphone, tablet computer, laptop computer, or desktop computer, etc. The electronic device may be referred to as a terminal, portable terminal, desktop terminal, etc., and includes:

[0196] Memory, used to store computer program instructions; and,

[0197] A processor is used to execute the computer program instructions to perform the operation of the electromagnetic signal identification method as described in any of Embodiment 1.

[0198] Example 4:

[0199] Based on any one of the embodiments 1 to 3, this embodiment discloses a computer-readable storage medium for storing computer-readable computer program instructions, which are configured to perform the operation of the electromagnetic signal identification method as described in Embodiment 1 when executed.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An electromagnetic signal identification method, characterized in that: include: Electromagnetic signals were collected from the study area and preprocessed to obtain a dataset with signal labels. The specified feature parameters are extracted from the dataset to obtain a feature vector composed of the feature parameters and a feature parameter matrix composed of all feature vectors; wherein the feature vector is used to characterize the corresponding electromagnetic signal and the feature parameter matrix is ​​used to characterize the dataset. The dataset is divided into a test sample signal set and a training sample signal set, and the FBST is constructed using the training sample signal set; wherein, the electromagnetic signals in the test sample signal set are test sample signals, and the electromagnetic signals in the training sample signal set are training sample signals. The KNN algorithm is used to retrieve the k nearest neighbors of the test sample signal in the constructed FBST; The signal labels of the sample signal to be tested are obtained based on the signal labels of the k nearest neighbors. Constructing an FBST using the training sample signal set includes: Select all training sample signals The first feature parameter is used as the current partition domain of the current partitioning layer of FBST; Select the median in the current partition and use the training sample signal it represents as the current node; Values ​​less than the median are assigned to the left subtree, and values ​​greater than the median are assigned to the right subtree. Select training sample signals from the left and right subtrees respectively. The next feature parameter is used as the current partition domain of the current partitioning layer of FBST; Reselect the median in the current partition region until all training sample signals are received. All are divided to obtain the final FBST; Let any test sample signal be denoted as the target point; use the KNN algorithm to retrieve the k nearest neighbor points of the test sample signal in the constructed FBST, including: Starting from the root node, compare the feature parameters of the target point with the feature parameters of the split point corresponding to the current layer of the FBST. If the feature parameters of the target point are less than the feature parameters of the split point, go to the left subtree; otherwise, go to the right subtree. Then recursively search downwards until a leaf node is reached, which is recorded as the current node. The minimum distance is defined as the distance between the target point and the current node. d min The distance between the target point and the sibling node of the current node is d b The distance between the target point and the parent node of the current node is d f Wherein, the parent node is the parent node of the current node, and the sibling node is the adjacent node that belongs to the same parent node as the current node; like d min > d b If so, then update the sibling node to the current node, and simultaneously update the sibling node and parent node of the current node. d min < d b Then compare d min and d f ; like d min > d f If so, then update the parent node to the current node, and simultaneously update the sibling nodes and parent node of the current node. d min < d f If the current node is the nearest neighbor, then the current node is the nearest neighbor. Mark the nearest neighbor as D pq And change its distance from the target point to infinity; Starting from the root node again and repeating k times, we can find the k nearest neighbors.

2. The electromagnetic signal identification method according to claim 1, characterized in that: The specified feature parameters extracted from the dataset include: the maximum value of the zero-center normalized instantaneous amplitude spectral density, the standard deviation of the absolute value of the zero-center normalized instantaneous amplitude, the standard deviation of the absolute value of the instantaneous frequency of the zero-center normalized non-weak signal segment, and the standard deviation of the instantaneous phase nonlinear component of the zero-center normalized non-weak signal segment. The expression for the feature parameter matrix is: ； in Indicates the first r A matrix of characteristic parameters, r =1, 2, ..., R , R The number of specified feature parameters; Indicates the first i The first type of signal tag j The first sample r One feature parameter, i =1, 2, ..., n , n The number of signal tags. j =1, 2, ..., m , m This represents the number of electromagnetic signal samples for each type of signal tag.

3. The electromagnetic signal identification method according to claim 2, characterized in that: The training sample signals in the training sample signal set are represented as follows: ； ； in, Indicates the first w One training sample signal, w The sequence number of the training sample signal; The test sample signals in the test sample signal set are represented as follows: ； in, Indicates the first t One sample signal to be tested, t The sequence number of the sample signal to be tested.

4. The electromagnetic signal identification method according to claim 3, characterized in that: The method further includes: The variance of each characteristic parameter is calculated using the following formula: ； ； ； in, Indicates the first r The first eigenparameter matrix of the feature parameter matrix j The average of the column, n Indicates the first r The number of rows in the feature parameter matrix. Indicates the first r The first eigenparameter matrix of the feature parameter matrix j The variance of the column, Indicates the first r The variance of the characteristic parameter matrix, m represents the variance of the characteristic parameter matrix. r The number of columns in the feature parameter matrix; Then sort the variances from largest to smallest, and modify the training sample signals according to the order of variance. With the signal of the sample to be tested The order of the coordinate points is used to obtain the modified training sample signal and the modified test sample signal; whereby the modified training sample signal is represented as: ； in, Indicating the order of variance size r’ The characteristic parameters of the training sample signal represented by each variance; The modified sample signal to be tested is represented as follows: ； in, Indicating the order of variance size r’ The variance represents the characteristic parameters of the sample signal to be tested.

5. The electromagnetic signal identification method according to claim 1, characterized in that: The nearest neighbor of the sample signal under test is selected using Euclidean distance. The formula for calculating Euclidean distance is as follows: ； in, d q For the first q The Euclidean distance between a training sample signal and a test sample signal R The specified number of feature parameters, The first of the sample signals to be tested r One feature parameter, For the first q The training sample signal of the th training sample signal r Each feature parameter.

6. An electromagnetic signal identification system, characterized in that: For implementing the electromagnetic signal identification method as described in any one of claims 1 to 5; the electromagnetic signal identification system comprises: An electromagnetic signal acquisition module is used to acquire electromagnetic signals from the study area and preprocess the electromagnetic signals to obtain a dataset with signal labels. The feature extraction module is communicatively connected to the electromagnetic signal acquisition module and is used to extract specified feature parameters from the dataset, and obtain a feature vector composed of the feature parameters, and a feature parameter matrix composed of all feature vectors; wherein, the feature vector is used to characterize the corresponding electromagnetic signal, and the feature parameter matrix is ​​used to characterize the dataset; The FBST construction module, which is communicatively connected to the feature extraction module, is used to divide the dataset into a test sample signal set and a training sample signal set, and to construct the FBST using the training sample signal set; wherein, the electromagnetic signals in the test sample signal set are test sample signals, and the electromagnetic signals in the training sample signal set are training sample signals. The feature matching module is communicatively connected to the FBST construction module and is used to retrieve the k nearest neighbors of the test sample signal in the constructed FBST using the KNN algorithm. The signal label determination module is communicatively connected to the feature matching module and is used to obtain the signal label of the sample signal to be tested based on the signal labels of the k nearest neighbors.

7. An electronic device, characterized in that: include: Memory is used to store computer program instructions; as well as, A processor is configured to execute the computer program instructions to perform the operation of the electromagnetic signal identification method as described in any one of claims 1 to 5.

8. A computer-readable storage medium for storing computer-readable computer program instructions, characterized in that: The computer program instructions are configured to perform the operation of the electromagnetic signal identification method as described in any one of claims 1 to 5 when the program is executed.

Images