Compound jamming signal identification method and system fusing median-of-means skewness estimator and electronic countermeasure equipment

Abstract

The application belongs to the technical field of electronic countermeasures and anti-interference, and particularly relates to a composite jamming signal identification method and system fusing a Median-of-Means skewness estimator and an electronic countermeasure device. The method comprises the following steps: performing feature extraction on a received composite jamming signal, including calculating time domain skewness and frequency domain skewness features of the signal through a Median-of-Means skewness estimator; combining the extracted multiple features into a comprehensive feature vector and inputting the comprehensive feature vector into an IGRF-RFE network for feature screening; and adopting a multilayer perception MLP to classify the screened features and output an interference signal type identification result. The application breaks through the limitation of traditional single-domain analysis, captures the asymmetric features of signals in a multilevel and adaptive manner, and provides a new technical path.

Description

Technical Field

[0001] This invention belongs to the field of electronic countermeasures technology, specifically relating to a method, system, and electronic countermeasures equipment for identifying composite interference signals that integrate a Median-of-Means skewness estimator. Background Technology

[0002] Satellite communication, as a crucial means of wireless communication, has received significant attention due to its advantages such as wide coverage, long-distance transmission, and lack of terrain limitations. However, these advantages also make it more vulnerable to reconnaissance, interception, and jamming. In particular, intelligent jammers with real-time learning capabilities are rapidly developing. To address these challenges, modern anti-jamming satellite communication systems are striving to improve their flexibility and security in complex adversarial environments. Traditional technologies are no longer sufficient to meet these requirements.

[0003] The anti-interference process typically includes four key stages: (1) interference signal identification, (2) anti-interference strategy selection, (3) anti-interference strategy execution, and (4) anti-interference performance evaluation. Figure 1 As shown, interference signal identification is a crucial prerequisite for achieving optimal anti-interference results. Its accuracy directly determines the effectiveness of subsequent anti-interference measures. However, current methods face challenges in distinguishing complex interference types, especially when dealing with multi-type, multi-dimensional, and dynamically changing composite interference. This type of interference is not only highly concealed but also difficult to identify.

[0004] Based on different jamming mechanisms, active jamming techniques can be divided into suppressive jamming and deceptive jamming. Radar jamming has multiple classification dimensions, such as... Figure 2 The diagram summarizes common types of interference in practical engineering applications. Suppressive jamming suppresses target echoes by transmitting noise signals, significantly reducing the signal-to-noise ratio (SNR) and thus disrupting the radar's normal target detection and tracking functions. Deceptive jamming, on the other hand, generates false signals that resemble the characteristics of real targets, inducing the radar system to make incorrect judgments. Deceptive jamming primarily targets radar systems using linear frequency modulation (LFM) signals, while suppressive jamming has a wider range of applications and can effectively interfere with radar systems using various modulation systems, including LFM.

[0005] Notably, in real-world electronic warfare environments, the frequency and threat level of composite suppression jamming are significantly increasing. Existing jamming identification methods often suffer significant performance degradation when facing strongly overlapping suppression jamming and composite suppression jamming due to overlapping signal characteristics and reduced JNR. This limitation severely restricts the effectiveness of subsequent anti-jamming measures, creating a critical technological gap that urgently needs attention.

[0006] In feature extraction, the joint time-frequency analysis method can comprehensively characterize the time and frequency domain characteristics of a signal, providing richer information for interference identification. However, traditional skewness estimation methods based on sample means are extremely sensitive to outliers and are prone to significant deviations in complex electromagnetic environments. Therefore, an innovative multivariate median estimation method is introduced. Through block processing and median calculation, the influence of outliers can be effectively suppressed, improving the robustness of feature estimation. Experiments show that the Median-of-Means method exhibits excellent anti-interference performance in parameter estimation, providing a new technical approach to improving the accuracy of interference identification in complex electromagnetic environments.

[0007] In conclusion, developing a novel interference identification method that incorporates robust statistical features is of significant theoretical and practical value for improving the anti-jamming capabilities of radar systems in complex electromagnetic environments. It is necessary to overcome the bottlenecks in interference identification accuracy and real-time performance of existing methods through innovative technical approaches, providing more effective technical solutions for the field of electronic countermeasures. Summary of the Invention

[0008] The purpose of this invention is to overcome the problem of low accuracy in identifying strongly overlapping composite suppression interference signals under low signal-to-interference ratio conditions, and to provide a composite interference signal identification method that integrates a Median-of-Means skewness estimator, so as to improve the interference perception and countermeasure capabilities of satellite communication systems in complex electromagnetic environments.

[0009] In view of this, the present invention proposes a method for identifying composite interference signals by incorporating a Median-of-Means skewness estimator, comprising: Feature extraction is performed on the received composite interference signal, including calculating the time-domain and frequency-domain skewness features of the signal using a Median-of-Means skewness estimator; The extracted features are combined into a comprehensive feature vector, which is then input into the IGFRF-RFE network for feature filtering. Multilayer perceptron (MLP) is used to classify the filtered features and output the interference signal type identification results.

[0010] Preferably, the extracted features further include: The frequency domain carrier factor coefficient C, and the maximum power spectral density of the normalized central instantaneous amplitude. The ratio of the maximum amplitude to the second highest amplitude in the spectrum The width exceeding the normalized instantaneous frequency distribution average value Time domain parameters Frequency domain parameters and average spectral flatness coefficient .

[0011] Preferably, the calculation process using the Median-of-Means skewness estimator includes: The signal samples are divided into multiple blocks; Calculate the mean, second central moment, and third central moment for each block; Take the median of the second central moments of each block. and the median of the third central moments ; The skewness estimate for each block is calculated using the following formula. :

[0012] in, Indicates the first One block; The median of all block skewness estimates is taken as the final skewness estimate.

[0013] Preferably, the Median-of-Means skewness estimator is performed in both the time and frequency domains to obtain the time-domain skewness features. Frequency domain skewness characteristics .

[0014] Preferably, the IGRRF-RFE network includes: The Information Gain (IG) calculation module is used to calculate the information gain of each feature with respect to the classification label. The Random Forest Importance (RF) module is used to calculate the importance score of each feature in the random forest; and The Recursive Feature Elimination (RFE) module is used to iteratively eliminate features of low importance to obtain feature selection results.

[0015] Preferably, the multilayer perceptron (MLP) includes an input layer, at least one hidden layer, and an output layer, wherein the hidden layer uses the ReLU activation function and the output layer uses the Softmax activation function.

[0016] Preferably, the method is applicable to the identification of composite interference signals including at least one of the following suppressed interferences: single-tone interference signal STJ, narrowband interference signal NBJ, comb spectrum interference signal CSJ, and sweep frequency interference signal SJ. On the other hand, the present invention provides a composite interference signal identification system that integrates a Median-of-Means skewness estimator, comprising: The feature extraction module is used to extract features from the received composite interference signal, including calculating the time-domain skewness and frequency-domain skewness features of the signal using a Median-of-Means skewness estimator. The feature selection module combines multiple extracted features into a comprehensive feature vector, which is then input into the IGRRF-RFE network for feature selection; and The type recognition module is used to classify the filtered features using a multilayer perceptron (MLP) and output the interference signal type recognition result.

[0017] Thirdly, the present invention provides an electronic countermeasures device, including the aforementioned composite interference signal identification system.

[0018] Compared with the prior art, the advantages of the present invention are: 1. The skewness estimator proposed in this invention significantly improves the robustness of skewness estimation by effectively reducing the influence of outliers. Traditional skewness features are easily affected by outliers or noise, leading to distorted estimation results. In contrast, the proposed estimator can maintain accuracy in assessing signal distribution asymmetry even in the presence of outliers, thereby improving the system's ability to identify interference signals.

[0019] 2. Robust Local Feature Capture Based on Block Mechanism. When faced with multiple superimposed interference signals, traditional methods often become dominated by strong interference signals, neglecting the key features of weak interference signals. The method of this invention can effectively capture robust features of signals within local regions. Even if the features of the interference signals are aliased, the statistical properties of local sub-sample blocks can still reveal the asymmetry of the overall signal distribution, thus helping to identify subtle features that are usually masked by strong interference signals.

[0020] 3. Adaptability to different JNR conditions. When the JNR is below 0dB, traditional skewness features are easily distorted by extreme values, leading to feature failure. The method proposed in this invention can extract weak interference signal skewness features in extreme noise environments ranging from -20dB to 0dB. Through the adjustment of hidden layer weights in a multilayer perceptron (MLP), the discriminative power of weak signals is effectively amplified, enabling the differentiation of different interference signals even in complex noise environments. Attached Figure Description

[0021] Figure 1 This is a flowchart of the anti-interference process; Figure 2 This is a classification diagram of common interference signals in engineering applications; Figure 3 This is a model structure diagram of the composite interference signal identification method that integrates Median-of-Means skewness estimator according to the present invention; Figure 4 This is a flowchart of the Median-of-Means skewness estimator algorithm of the present invention; Figure 5 This is a flowchart of the IRGRF-RFE identification network of the present invention; Figure 6 This is a three-dimensional time-frequency characteristic diagram of the LFM signal of the present invention when it is subjected to single-tone interference; Figure 7 This is a three-dimensional time-frequency characteristic diagram of the LFM signal of the present invention when it is subjected to narrowband interference; Figure 8 This is a three-dimensional time-frequency characteristic diagram of the LFM signal of the present invention when it is subjected to comb spectrum interference; Figure 9 This is a three-dimensional time-frequency characteristic diagram of the LFM signal of the present invention when subjected to frequency sweep interference; Figure 10 This is a trend chart showing the recognition accuracy of the six composite interference signals of this invention under different JNR conditions. Detailed Implementation

[0022] This invention discloses a composite interference signal identification method integrating a Median-of-Means skewness estimator. The network comprises a skewness feature estimator and an IGFRF-RFE network. The estimator employs multivariate mean-median estimation, reducing the impact of outliers on sample skewness features through a more robust approach. The IGFRF-RFE network is a hybrid feature selection method based on MLP.

[0023] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0024] Example 1 like Figure 3 As shown, Embodiment 1 of the present invention proposes a method for identifying composite interference signals by integrating a Median-of-Means skewness estimator. The method includes: inputting the received composite interference signal into a pre-constructed and trained interference identification network, extracting nine features including skewness features, using an IGFRF-RFE network to identify the type of interference signal, obtaining the identification result, and realizing the identification of composite interference signals.

[0025] Nine features, including skewness features, were extracted. Specifically, the temporal skewness of the signal was calculated using a Median-of-Means skewness estimator. and frequency domain skewness Features; extract the frequency domain carrier factor coefficient C of the signal, and the maximum power spectral density of the normalized central instantaneous amplitude. The ratio of the maximum amplitude to the second highest amplitude in the signal spectrum The width exceeding the normalized instantaneous frequency distribution average value Time-domain parameters of the signal Frequency domain parameters of the signal Average spectral flatness coefficient The above features are combined into a comprehensive feature vector, which is then input into the IGRRF-RFE network for identification, and the identification result of the interference signal is output.

[0026] The IGRF-RFE network performs feature filtering by calculating the information gain (IG) of each feature to the identification label and its importance in the random forest (RF). Then, it uses ensemble feature selection and recursive feature elimination (RFE) to remove unimportant features, and inputs them into a three-layer MLP recognition network to obtain the recognition result.

[0027] The method includes the following steps: 1. Skewness feature extraction Based on the sub-Gaussian mean estimation framework proposed by Lugosi & Mendelson, this algorithm utilizes the Median-of-Means method to calculate signal skewness in both the time and frequency domains. The estimator is based on the novel concept of multivariate median. Under the condition of second-order moments, it achieves sub-Gaussian performance. The algorithm includes signal sample partitioning, intra-block statistic calculation, robust estimation, and skewness estimation output. The detailed algorithm flowchart is shown below. Figure 4 As shown.

[0028] Given a sample of N independent and identically distributed random vectors Divide the samples into k A block, denoted as ,in k These are parameters of the estimator, and their values ​​depend on the desired confidence level. The size of each block is:

[0029] For each block Calculate the block average:

[0030] Without loss of generality, assume that N can be k Divisible, therefore , .

[0031] For each ,make:

[0032] For all :

[0033] Define the mean estimator as:

[0034] in:

[0035] choose Make all middle Reaching the minimum value. Let... From point These points constitute the condition that, for most blocks, .

[0036] For each block This method requires calculation: Calculate the second central moment:

[0037] Calculate the third-order central moments:

[0038] The median of the block moments is obtained as follows:

[0039]

[0040] in, This represents the median.

[0041] The skewness estimator is given by the following formula:

[0042] The median of the block skewness estimators was used to calculate the final estimate:

[0043] The Median-of-Means skewness estimator performs the above calculations in both the time and frequency domains to obtain the time-domain skewness. and frequency domain skewness The estimated value.

[0044] Among them, calculation It requires a continuous space search. The time complexity is exponential, and it cannot be computed in polynomial time. However... This is a non-convex combinatorial problem, which is transformed into a solvable semidefinite programming problem.

[0045] For any direction, at least half of the blocks have the average value. satisfy:

[0046] in, express 3D space A unit vector ( ), This represents the Euclidean inner product.

[0047] Define binary variables , indicating a Is it beyond the distance? r Therefore, this problem is equivalent to an optimization problem:

[0048] The above system of equations is transformed into a semidefinite programming problem.

[0049]

[0050] in , , . yes A real matrix of dimension 1.

[0051] The constraints are:

[0052]

[0053]

[0054]

[0055] in , Indicates the first The arithmetic mean of all samples in the group. yes matrix The OK, Representation matrix The traces.

[0056] The optimal solution to this SDP can be obtained. The estimated value.

[0057] 2. Other Feature Extraction Carrier factor coefficient C The ratio between the maximum and second-highest values ​​of frequency domain data is expressed by the following formula:

[0058] in and Y is the index of the maximum and second-largest values ​​of the signal spectrum in the frequency domain. Y is the dataset after the Fast Fourier Transform operation.

[0059] It is the maximum power spectral density of the normalized center instantaneous amplitude, calculated by the following formula:

[0060] in The normalized central instantaneous amplitude value, . The instantaneous amplitude of the signal. N This represents the number of signal points.

[0061] It is the ratio of the maximum amplitude to the second highest amplitude in the signal spectrum.

[0062] It is the width that exceeds the average value of the normalized instantaneous frequency distribution.

[0063] This is a time-domain parameter, representing the degree to which the signal's spectral amplitude fluctuates over time. The calculation formula is:

[0064] in, and These represent the mean and variance of the spectral amplitudes for multiple short-time frames, respectively.

[0065] This is a frequency domain parameter, representing the degree of fluctuation in the signal's spectral amplitude at different frequencies. The calculation formula is:

[0066] in, and These are the mean and variance of the spectrum of that frame on the frequency axis, respectively.

[0067] This is the average spectral flatness coefficient, representing the standard deviation of the difference between the original power spectrum and the windowed smoothed signal power spectrum. It indicates the degree of variation in signal amplitude in the power spectral domain, and its calculation formula is:

[0068] in, The difference between the original power spectrum and the power spectrum of the windowed and smoothed signal is denoted as N, where N is the total number of frequency points. express The average value.

[0069] 3. IRGRF-RFE Identification Network The IGRF-RFE recognition network is a hybrid feature selection method that includes information gain, random forest feature importance, recursive feature elimination (RFE), and an MLP classifier. A flowchart is shown below. Figure 5 As shown.

[0070] The network includes an Information Gain (IG) module for calculating the information gain of each feature to the classification label; a Random Forest Importance (RF) module for calculating the importance score of each feature in the random forest; and an Integrated Recursive Feature Elimination (RFE) module for iteratively eliminating features with low importance to obtain the feature selection results.

[0071] The feature selection results are input into the MLP classifier. The specific flowchart is as follows: Figure 5 As shown.

[0072] MLP is a fully connected feedforward neural network architecture with a directed acyclic graph structure, where information propagates unidirectionally from the input layer through the hidden layers to the output layer. It uses ReLU as the activation function for the hidden layers and Softmax as the activation function for the final output layer.

[0073] Before inputting the MLP recognition network, calculate the Indicative Significance (IG) of each feature to the classification label and its importance in the Random Field (RF). Perform feature filtering. Then use the Regression-Evaluation (RFE) method to remove unimportant features.

[0074] IG is a univariate filter feature selection method based on information entropy. Each feature... First, based on its IG relative to category tags Y Conduct an assessment:

[0075] in It is Shannon entropy. Y Represents a category vector. This represents the eigenvector.

[0076] Random forest is a machine learning method based on multiple decision trees. Features The importance score is defined as the average decrease in node impurity:

[0077] in T This represents a tree in a random forest. This represents a single decision tree in a forest. Representation of features In the tree t The amount of impurity reduction contributed by the middle.

[0078] Example 2 Embodiment 2 of the present invention provides a composite interference signal identification system that integrates a Median-of-Means skewness estimator, implemented based on the method of Embodiment 1, including: The feature extraction module is used to extract features from the received composite interference signal, including calculating the time-domain skewness and frequency-domain skewness features of the signal using a Median-of-Means skewness estimator. The feature selection module is used to combine multiple extracted features into a comprehensive feature vector, which is then input into the IGFRF-RFE network for feature selection. The type recognition module is used to classify the filtered features using a multilayer perceptron (MLP) and output the interference signal type recognition result.

[0079] Example 3 Embodiment 3 of the present invention provides an electronic countermeasures device, characterized in that it includes a composite interference signal identification system as described in Embodiment 2.

[0080] Simulation example: The results of this invention can be verified using the following simulation data: A linear frequency modulation (LFM) signal with a bandwidth of 10MHz, a pulse width of 20μs, and a sampling frequency of 20MHz is set. The types of interference signals to be suppressed include single-tone interference (STJ), narrowband interference (NBJ), comb-spectrum interference (CSJ), and swept-frequency interference (SJ). The three-dimensional time-frequency characteristics of the LFM signal under the interference of the suppression signals are shown in the figures below. Figure 6 , Figure 7 , Figure 8 and Figure 9 As shown in Table 1, the specific parameter settings for suppressing interference signals are as follows: Table 1 Parameter settings for suppressing interference signals

[0081] Six strongly overlapping composite interference signals were generated by targeting combinations of STJ, NBJ, CSJ, and SJ suppression interference signals. The channel noise was additive white Gaussian noise. The JNR value ranged from -20 dB to 20 dB.

[0082] Accuracy is used to quantitatively evaluate the algorithm's performance in recognizing complex interference signals. Accuracy refers to the ratio of correctly identified samples to the total number of samples.

[0083]

[0084] Wherein, TP represents the number of positive samples correctly predicted as positive, TN represents the number of negative samples correctly predicted as negative, FP represents the number of negative samples incorrectly predicted as positive, and FN represents the number of positive samples incorrectly predicted as negative.

[0085] like Figure 10As shown in the figure, this figure illustrates the recognition accuracy trends of six composite interference signals under different JNR conditions. It can be seen that with the increase of JNR, the recognition accuracy of all composite interference signals increases, but the performance of different combinations varies significantly. The composite interference signal recognition method incorporating the Median-of-Means skewness estimator exhibits stable performance under both noise and interference conditions across a wide dynamic range.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art should understand that modifications or equivalent substitutions to the technical solutions of the present invention do not depart from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying composite interference signals by integrating a Median-of-Means skewness estimator, comprising: Feature extraction is performed on the received composite interference signal, including calculating the time-domain and frequency-domain skewness features of the signal using a Median-of-Means skewness estimator; The extracted features are combined into a comprehensive feature vector, which is then input into the IGFRF-RFE network for feature filtering. Multilayer perceptron (MLP) is used to classify the filtered features and output the interference signal type identification results.

2. The method for identifying composite interference signals by fusing a Median-of-Means skewness estimator according to claim 1, characterized in that, The extracted features also include: The frequency domain carrier factor coefficient C, and the maximum power spectral density of the normalized central instantaneous amplitude. The ratio of the maximum amplitude to the second highest amplitude in the spectrum The width exceeding the normalized instantaneous frequency distribution average value Time domain parameters Frequency domain parameters and average spectral flatness coefficient .

3. The method for identifying composite interference signals by fusing a Median-of-Means skewness estimator according to claim 1, characterized in that, The calculation process using the Median-of-Means skewness estimator includes: The signal samples are divided into multiple blocks; Calculate the mean, second central moment, and third central moment for each block; Take the median of the second central moments of each block. and the median of the third central moments ; The skewness estimate for each block is calculated using the following formula. : ； in, Indicates the first One block; The median of all block skewness estimates is taken as the final skewness estimate.

4. The method for identifying composite interference signals by fusing a Median-of-Means skewness estimator according to claim 3, characterized in that, The Median-of-Means skewness estimator is performed in both the time and frequency domains to obtain the time-domain skewness features. Frequency domain skewness characteristics .

5. The method for identifying composite interference signals by fusing a Median-of-Means skewness estimator according to claim 1, characterized in that, The IRGRF-RFE network includes: The Information Gain (IG) calculation module is used to calculate the information gain of each feature with respect to the classification label. The Random Forest Importance (RF) module is used to calculate the importance score of each feature in the random forest; and The Recursive Feature Elimination (RFE) module is used to iteratively eliminate features of low importance to obtain feature selection results.

6. The method for identifying composite interference signals by fusing a Median-of-Means skewness estimator according to claim 1, characterized in that, The multilayer perceptron (MLP) includes an input layer, at least one hidden layer, and an output layer, wherein the hidden layer uses the ReLU activation function and the output layer uses the Softmax activation function.

7. The method for identifying composite interference signals by fusing a Median-of-Means skewness estimator according to claim 1, characterized in that, The method is applicable to the identification of composite interference signals including at least one of the following suppression interferences: single-tone interference signal STJ, narrowband interference signal NBJ, comb spectrum interference signal CSJ, and sweep frequency interference signal SJ.

8. A composite interference signal identification system integrating a Median-of-Means skewness estimator, characterized in that, include: The feature extraction module is used to extract features from the received composite interference signal, including calculating the time-domain skewness and frequency-domain skewness features of the signal using a Median-of-Means skewness estimator. The feature selection module combines multiple extracted features into a comprehensive feature vector, which is then input into the IGRRF-RFE network for feature selection; and The type recognition module is used to classify the filtered features using a multilayer perceptron (MLP) and output the interference signal type recognition result.

9. An electronic countermeasures device, characterized in that, Including the composite interference signal identification system as described in claim 8.

Images