An intelligent deception jamming method for intelligent identification of communication modulation patterns

By building and training deep neural networks, signals that can deceive multiple communication modulation patterns are generated, which solves the problem that the existing technology cannot accurately induce specified communication modulation types, realizes intelligent spoof interference with multiple modulation patterns, and improves communication security Sex and reliability.

CN118473589BActive Publication Date: 2024-11-01THE 54TH RESEARCH INSTITUTE OF CHINA ELECTRONICS TECHNOLOGY GROUP CORPORATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410931696.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-01
Estimated Expiration
2044-07-12

AI Technical Summary

Technical Problem

Existing communication modulation style intelligent recognition technology cannot accurately induce non-cooperative artificial intelligence methods to identify signals as specified communication modulation types, and cannot spoof multiple communication modulation styles simultaneously.

Method used

Build a communication modulation style data set, train communication modulation styles to recognize deep neural networks, and construct a spoof interference signal to generate a deep neural network. Through multiple communication signal experiments, a spoof interference is achieved to achieve intelligent spoof interference to multiple modulation styles.

Benefits of technology

It realizes the specified type of spoofing signals of multiple communication modulation types, improves communication security and reliability, and has the "end-to-end" capabilities from perception to generation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118473589B_ABST
    Figure CN118473589B_ABST
Patent Text Reader

Abstract

The present invention discloses an intelligent deception interference method for intelligent identification of communication modulation patterns, which belongs to the intersection of communication confrontation and artificial intelligence. The present invention includes: constructing a communication modulation pattern data set, constructing and training a communication modulation pattern identification deep neural network, constructing and training a communication modulation pattern deception interference signal generation deep neural network, and intelligently generating deception interference signals based on communication signals. Aiming at the objective demand of protecting wireless communication transmission, the present invention trains intelligent deception interference capabilities for the communication modulation pattern intelligent identification model in a targeted manner, and forms an "end-to-end" generation capability from sensing communication signals to generating interference signals. The generated interference signal can simultaneously deceive non-cooperative identification of multiple communication modulation patterns, and trick the identification type into a specified communication modulation pattern type, thereby weakening the intelligence fusion and interference decision-making capabilities, and improving the security and reliability of communications.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention discloses an intelligent deception interference method for intelligent identification of communication modulation patterns, belonging to the intersection field of communication confrontation and artificial intelligence. Background Art

[0002] Communication modulation pattern recognition is an important research topic in the field of communication countermeasures. In particular, intelligent recognition of communication modulation patterns based on deep learning can accurately extract the modulation type of communication signals in complex electromagnetic environments such as low signal-to-noise ratios, thereby guiding communication jammers to use different types of jammers in a targeted manner to improve the bit error rate and thus disrupt communications. Since traditional noise suppression jammers cannot accurately deceive communication modulation pattern recognition to obtain the specified communication type, how to use artificial intelligence methods to deceive reconnaissance equipment to identify the specified communication type has become the current research focus.

[0003] In the article "Research on Adversarial Attack Technology in Electromagnetic Signal Modulation Recognition" published in "Radio Communication Technology", Wang Manxi et al. studied the weak links of modulation recognition processing using multi-layer perceptrons. Based on the concept of adversarial samples, they conducted in-depth research on the impact of adversarial attack technologies such as fast gradient mark attack, basic iterative attack, gradient projection method and momentum iteration method on modulation recognition, which can greatly damage the recognition accuracy under small disturbances. However, this method cannot accurately specify the modulation style.

[0004] In the article "OFDM Signal Generation Based on Generative Adversarial Networks" published by Chen Li et al. in the Journal of University of Electronic Science and Technology of China, a generative adversarial network is used to extract features, learn and reconstruct specific signals to obtain OFDM spoofing signals that can deceive classifiers. However, this method requires generating spoofing signals one by one for different communication modulation styles existing in the communication signals in the environment, and it is impossible to induce the classifier to recognize that the modulation type of the current signal is OFDM for signals of other modulation types.

[0005] In the patent "A method for generating adversarial sample signal waveforms based on channel pre-compensation" applied by the National University of Defense Technology of the People's Liberation Army of China, with the authorization announcement number CN117478474B, the use of adversarial attack technology to superimpose slight adversarial disturbances in communication signals can effectively weaken the ability of artificial intelligence methods to accurately extract and identify signal modulation while achieving cooperative communication, thereby improving the security and reliability of communication. However, this method is also unable to accurately induce non-cooperative artificial intelligence methods to identify signals as specified communication modulation types.

[0006] In the patent "A Method for Generating Adversarial Sample Signal Waveform Based on Frequency Selection" applied by the National University of Defense Technology of the People's Liberation Army of China, with the authorization announcement number CN117478474B, the adversarial attack technology is used to superimpose slight adversarial disturbances on the communication signal, and the frequency domain constraints are added to limit the generated adversarial disturbances to the frequency band of the signal, thereby enhancing the attack performance of the adversarial sample waveform on the non-cooperative modulation identifier, and in addition, it can reduce the impact on the contract communication. However, this method also cannot accurately induce the non-cooperative artificial intelligence method to identify the signal as the specified communication modulation type. Summary of the invention

[0007] In view of this, the present invention proposes an intelligent deception interference method for intelligent identification of communication modulation patterns. Aiming at the objective demand of protecting wireless communication transmission, the present invention trains the intelligent deception interference capability of the communication modulation pattern intelligent identification model in a targeted manner, and the generated interference signal can deceive the identification of multiple communication modulation patterns at the same time, and trick the identification type into the specified communication modulation pattern type.

[0008] In order to achieve the above object, the technical solution adopted by the present invention is:

[0009] An intelligent deception jamming method for intelligent identification of communication modulation patterns comprises the following steps:

[0010] Step 1: Build a communication modulation style dataset;

[0011] Step 2: Construct a deep neural network for communication modulation pattern recognition;

[0012] Step 3: Train the communication modulation pattern recognition deep neural network;

[0013] Step 4: Construct a communication modulation pattern to deceive the jamming signal to generate a deep neural network;

[0014] Step 5: Train the communication modulation pattern to deceive the jamming signal to generate a deep neural network;

[0015] Step 6: Intelligently generate deceptive jamming signals based on communication signals.

[0016] Furthermore, the specific method of step 1 is:

[0017] Step 1-1: In the test area, communication signal sending equipment, communication reconnaissance equipment and communication anti-reconnaissance equipment are set up. The communication signal sending equipment transmits communication signals of N modulation patterns in total; the communication reconnaissance equipment senses the communication signal and identifies the modulation pattern; the communication anti-reconnaissance equipment generates interference signals based on the communication signal. After the communication signal is superimposed on the interference signal, the communication reconnaissance equipment will generate a specified communication pattern classification, realizing intelligent deception interference of the communication modulation pattern; each modulation pattern of the communication signal is generated M times in total, and the total number of tests is L=M×N;

[0018] After the lth test, the communication signal sample obtained is X l = [ s r , l , s i , l ] 1 × 2 × T , the corresponding communication signal modulation type sample is ,in represents the real data of the communication signal of the first test, It represents the imaginary data of the communication signal of the first test, and the data length is T. Indicates the modulation type corresponding to the communication signal of the first test;

[0019] After L trials, the communication signal sample set is obtained as X = [ X 1 , ⋯ , X l , ⋯ , X L ] L × 1 × 2 × T , the communication signal modulation type set is Y = [ p 1 , ⋯ , p l , ⋯ p L ] L × 1 ;

[0020] Step 1-2: Follow and The communication signal sample set and the communication signal modulation type set are divided into a feature training set , label training set , feature test set and a labeled test set , and their corresponding dimensions are , , and ,in .

[0021] Furthermore, in step 2, the input dimension of the communication modulation style recognition deep neural network is consistent with the communication signal sample dimension, and the output dimension is consistent with the communication signal modulation type N; the communication modulation style recognition deep neural network adopts Conv2d convolutional network, ReLU activation function, MaxPool2d maximum pooling layer, ConvTranspose2d transposed convolutional layer and Linear linear transformation layer to construct K feature extraction channels with different receptive fields, and adds the features of different receptive fields and maps them through the Linear linear transformation layer, so that the output is consistent with the communication signal modulation type N.

[0022] Furthermore, the specific method of step 3 is:

[0023] Step 3-1: Set the optimization function and learning rate. The optimization function is Adam optimizer.

[0024] Step 3-2: From the feature training set and labeled training set Randomly extract batch data, where batch is less than or equal to ,composition and , through the communication modulation pattern recognition deep neural network mapping, we get the estimated :

[0025]

[0026] in Deep neural networks for communication modulation pattern recognition, is the network parameter;

[0027] Step 3-3: Set the loss function to the cross entropy loss function to measure the estimated value and labeled training set The difference between them is loss, the error is calculated, and the Adam optimizer is used to optimize the network parameters Optimize

[0028] Step 3-4: From the feature test set and a labeled test set Randomly extract Q data, where Q is less than or equal to ,composition and , through the communication modulation pattern recognition deep neural network mapping, we get the estimated :

[0029]

[0030] statistics and The number of modulation types of the corresponding positions in the same position is divided by Q to get the accuracy , use the accuracy rate to verify the classification performance;

[0031] Step 3-5: Repeat steps 3-2, 3-3, and 3-4, and record the number of repetitions and the maximum accuracy of all previous iterations. , if the accuracy of this iteration is Greater than the maximum accuracy of all previous iterations , then assign the correct rate to , and save the corresponding network parameters ; If the accuracy of this iteration Less than or equal to the maximum value of all correct rates in the previous iteration , then enter the next iteration until When the corresponding number of repetitions step is less than half of the total number of repetitions, the process stops and the network parameters with the best performance are saved.

[0032] Furthermore, in step 4, the input dimension of the deep neural network for generating the communication modulation style deception interference signal is consistent with the communication signal sample dimension, and the output dimension is consistent with the communication signal sample dimension; the deep neural network for generating the communication modulation style deception interference signal adopts Conv2d convolutional network, ReLU activation function, MaxPool2d maximum pooling layer, ConvTranspose2d transposed convolutional layer and Linear linear transformation layer to construct K feature extraction channels with different receptive fields, and adds the features of different receptive fields and maps them through the Linear linear transformation layer, so that the output deception interference signal is consistent with the communication signal sample dimension.

[0033] Furthermore, the specific method of step 5 is:

[0034] Step 5-1: Set the deception type to That is, after the communication signal is superimposed with the interference signal, the goal of obtaining the communication modulation pattern classification result through the communication modulation pattern recognition deep neural network is ;

[0035] Step 5-2: Set the optimization function and learning rate. The optimization function is Adam optimizer.

[0036] Step 5-3: From the feature training set Randomly extract batch data, where batch is less than or equal to ,composition , generate deep neural network mapping by deceiving interference signals through communication modulation patterns, and obtain deceiving interference signals :

[0037]

[0038] in, Generate deep neural networks for communication modulation pattern spoofing jamming signals, is the network parameter;

[0039] By recognizing the deep neural network mapping through the communication modulation pattern, we can estimate :

[0040]

[0041] in, Deep neural networks for communication modulation pattern recognition, is the network parameter;

[0042] Step 5-4: Set the loss function to the cross entropy loss function to measure the estimated value and deception type The difference between them is loss, the error is calculated, and the Adam optimizer is used to optimize the network parameters Optimize

[0043] Step 5-5: From the feature test set Randomly extract Q pieces of data, where Q is less than or equal to ,composition , generate deep neural network mapping by deceiving interference signals through communication modulation patterns, and obtain deceiving interference signals :

[0044]

[0045] By recognizing the deep neural network mapping through the communication modulation pattern, we can estimate :

[0046]

[0047] statistics Modulation type and deception type The same number, divided by Q, gives the deception rate , verify the generation performance using the deception rate;

[0048] Step 5-6: Repeat steps 5-3, 5-4, and 5-5, and record the number of repetitions step and the maximum value of all deception rates in the previous iterations. , if the deception rate of this iteration Greater than the maximum value of all deception rates in the previous iteration , then assign the current deception rate to , and save the corresponding network parameters ; If the deception rate of this iteration Less than or equal to the maximum value of all deception rates in the previous iteration , then enter the next iteration until When the corresponding number of repetitions step is less than half of the total number of repetitions, the process stops and the network parameters with the best performance are saved.

[0049] Furthermore, the specific method of step 6 is:

[0050] Step 6-1: Read the network parameters for optimal performance , and loaded into the communication modulation style deception jamming signal generation deep neural network;

[0051] Step 6-2: For the communication signal generated by the communication signal sending device, generate a deep neural network to generate an interference signal by deceiving the interference signal through the communication modulation style, and then superimpose the communication signal on the interference signal and send it to the communication reconnaissance equipment, thereby destroying the communication reconnaissance equipment's ability to identify the communication signal modulation type.

[0052] The beneficial effects of the present invention are:

[0053] 1. The present invention can generate interference signals for communication signals of multiple communication modulation types at the same time, and lure all of them into the classification of the specified communication modulation type.

[0054] 2. The present invention has the "end-to-end" generation capability from sensing communication signals to generating interference signals.

[0055] 3. The present invention provides in detail a specific method for obtaining the optimal decision interference signal generation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is an overall flow chart of an embodiment of the present invention.

[0057] Figure 2 Schematic diagram of the structure of the deep neural network for communication modulation style recognition.

[0058] Figure 3 A diagram showing how the loss on the training set changes with the number of training runs.

[0059] Figure 4 This is a schematic diagram showing how the accuracy on the training set changes with the number of training times.

[0060] Figure 5 A diagram showing how the loss on the test set changes with the number of training cycles.

[0061] Figure 6 This is a schematic diagram showing how the accuracy on the test set changes with the number of training times.

[0062] Figure 7 Schematic diagram of the confusion matrix for communication modulation style identification.

[0063] Figure 8 Schematic diagram of the structure of a deep neural network network for generating jamming signals for communication modulation style spoofing.

[0064] Fig. 9 A diagram showing how the loss on the training set changes with the number of training runs.

[0065] Fig.10 This is a diagram showing how the deception rate on the training set changes with the number of training times.

[0066] Fig.11A diagram showing how the loss on the test set changes with the number of training cycles.

[0067] Fig.12 A diagram showing how the deception rate on the test set changes with the number of training times.

[0068] Fig.13 Schematic diagram of the confusion matrix for communication modulation style spoofing.

[0069] Fig.14 A schematic diagram of a communication signal generated by a communication signal sending device.

[0070] Fig.15 Schematic diagram of intelligently generating deceptive jamming signals for communication anti-reconnaissance equipment. DETAILED DESCRIPTION

[0071] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0072] An intelligent deception jamming method for intelligent identification of communication modulation patterns, such as Figure 1 As shown, the specific steps include:

[0073] Step 1: Build a communication modulation style dataset;

[0074] Step 1-1: Collect communication signals under different modulation types to construct a feature set;

[0075] In the test area, communication signal sending equipment, communication reconnaissance equipment and communication anti-reconnaissance equipment are set up. The communication signal sending equipment transmits a total of N = 11 communication signals with modulation styles, including 8PSK, AM-DSB, AM-SSB, BPSK, CPFSK, GFSK, PAM4, QAM16, QAM64, QPSK and WBFM; the communication reconnaissance equipment senses the communication signal and identifies the modulation style; the communication anti-reconnaissance equipment generates interference signals based on the communication signal. After the communication signal is superimposed on the interference signal, the communication reconnaissance equipment will generate a specified communication style classification, realizing intelligent deception interference of the communication modulation style. Each communication signal with modulation style is generated M = 1000 times, so the total number of experiments is M × N times. Let M × N = L, that is, 11000.

[0076] After the lth test, the communication signal sample obtained is X l = [ s r , l , s i , l ] 1 × 2 × T , the corresponding communication signal modulation type sample is ,in represents the real data of the communication signal of the first test, It represents the imaginary data of the communication signal in the first test, and the data length is T=128. Indicates the modulation type corresponding to the communication signal of the lth test.

[0077] After L trials, the communication signal sample set can be obtained as X = [ X 1 , ⋯ , X l , ⋯ , X L ] 11000 × 1 × 2 × 128 , the communication signal modulation type set is Y = [ p 1 , ⋯ , p l , ⋯ p L ] 11000 × 1 .

[0078] Step 1-2: Split the feature set to construct the training set and validation set;

[0079] according to and The communication signal sample set and the communication signal modulation type set are divided into a feature training set , label training set , feature test set and a labeled test set , and their corresponding dimensions are , , and ,in , =0.8, =0.2.

[0080] Step 2: Construct a deep neural network for communication modulation pattern recognition

[0081] Construct a deep neural network for communication modulation pattern recognition. The specific structure is as follows: Figure 2 The parameters of each layer are shown in Table 1. The input dimension of the deep neural network for communication modulation pattern recognition is consistent with the dimension of the communication signal sample, and the output dimension is consistent with the communication signal modulation type N. Conv2d convolutional network, ReLU activation function, MaxPool2d maximum pooling layer, ConvTranspose2d transposed convolution layer and Linear linear transformation layer are used to construct K feature extraction channels with different receptive fields. The features of different receptive fields are added and mapped through the Linear linear transformation layer so that the output is consistent with the number of communication signal modulation types.

[0082] Table 1 Parameters of each layer of the deep neural network for communication modulation pattern recognition

[0083]

[0084] Step 3: Train a deep neural network for communication modulation pattern recognition

[0085] Step 3-1: Set the optimization function to Adam optimizer and the learning rate to 0.0003;

[0086] Step 3-2: Using the communication modulation pattern recognition deep neural network to obtain the communication modulation pattern classification result;

[0087] From the feature training set and labeled training set Randomly extract batch = 64 data, where batch is less than or equal to ,composition and . By recognizing the deep neural network mapping through the communication modulation pattern, we can estimate ,Right now

[0088]

[0089] in Deep neural networks for communication modulation pattern recognition, is the network parameter;

[0090] Step 3-3: Set the loss function, calculate the error and optimize the network parameters; set the loss function to the cross entropy loss function to measure the estimated value and labeled training set The gap between them is loss, and the Adam optimizer is used to optimize the network parameters Make optimizations.

[0091] Step 3-4: Use the accuracy rate to verify the classification performance;

[0092] From the feature test set and a labeled test set Randomly extract Q pieces of data, where Q is less than or equal to ,composition and , through the communication modulation pattern recognition deep neural network mapping, we get the estimated

[0093]

[0094] statistics and The number of modulation types of the corresponding positions in the same position is divided by Q to get the accuracy .

[0095] Step 3-5: Save the network parameters with the best performance;

[0096] Repeat 3-2, 3-3 and 3-4, record the number of repetitions step and update the maximum accuracy. Greater than the maximum accuracy of all previous iterations , then assign the correct rate to , and save the corresponding network parameters ; If the accuracy of this iteration Less than or equal to the maximum value of all correct rates in the previous iteration , then enter the next iteration until the maximum accuracy Stop when the corresponding number of repetitions step is less than half of the total number of repetitions. Figure 3 To show how the loss on the training set changes with the number of training times, Figure 4 To show that the accuracy on the training set varies with the number of training times, Figure 5 To show how the loss on the test set changes with the number of training times, Figure 6 To show how the accuracy on the test set changes with the number of training times, Figure 7 The confusion matrix for communication modulation pattern recognition. After training, the accuracy of modulation type recognition for communication signals reached 80.71%;

[0097] Step 4: Construct a communication modulation pattern to deceive the jamming signal to generate a deep neural network;

[0098] Construct communication modulation style to deceive interference signal generation deep neural network network specific structure is as follows Figure 8 The parameters of each layer are shown in Table 2. The input dimension of the deep neural network for generating the communication modulation style deception interference signal is consistent with the communication signal sample dimension, and the output dimension is consistent with the communication signal sample dimension. Conv2d convolutional network, ReLU activation function, MaxPool2d maximum pooling layer, ConvTranspose2d transposed convolutional layer and Linear linear transformation layer are used to construct K=3 feature extraction channels with different receptive fields. The features of different receptive fields are added and mapped through the Linear linear transformation layer, so that the output deception interference signal is consistent with the communication signal sample dimension.

[0099] Table 2 Parameters of each layer of the deep neural network for communication modulation pattern deception jamming signal generation

[0100]

[0101] Step 5: Train the communication modulation pattern to deceive the jamming signal to generate a deep neural network;

[0102] Step 5-1: Set the deception type;

[0103] The modulation type recognition result obtained after setting the classification is =5, corresponding to the GFSK communication signal modulation type, then after the communication signal is superimposed with the interference signal, the communication modulation pattern classification result is obtained through the communication modulation pattern recognition deep neural network. .

[0104] Step 5-2: Set the optimization function and learning rate;

[0105] Set the optimization function to Adam optimizer and learning rate

[0106] Step 5-3: Generate a deep neural network using the communication modulation pattern to deceive the interference signal and obtain the communication modulation pattern to deceive the interference signal;

[0107] From the feature training set Randomly extract batch data, where batch is less than or equal to ,composition . Generate a deep neural network mapping by deceiving the jamming signal through the communication modulation style to obtain the deceiving jamming signal ,Right now

[0108]

[0109] in Generate deep neural networks for communication modulation pattern spoofing jamming signals, is the network parameter;

[0110] By recognizing the deep neural network mapping through the communication modulation pattern, we can estimate ,Right now

[0111]

[0112] in Deep neural networks for communication modulation pattern recognition, is the network parameter;

[0113] Step 5-4: Based on the set loss function, calculate the error and optimize the parameters of the deep neural network for generating the communication modulation pattern deception interference signal;

[0114] Set the loss function, calculate the error and optimize the network parameters; set the loss function to the cross entropy loss function to measure the estimated value and deception type The gap between them is loss, and the Adam optimizer is used to optimize the network parameters Make optimizations.

[0115] Step 5-5: Verify the generation performance using the deception rate;

[0116] From the feature test set Randomly extract Q pieces of data, where Q is less than or equal to ,composition , generate deep neural network mapping by deceiving interference signals through communication modulation patterns, and obtain deceiving interference signals ,Right now

[0117]

[0118] By recognizing the deep neural network mapping through the communication modulation pattern, we can estimate ,Right now

[0119]

[0120] statistics Modulation type and deception type The same number, divided by Q, gives the deception rate .

[0121] Step 5-6: Save the generated deep neural network parameters with the best performance;

[0122] Repeat 5-3, 5-4 and 5-5, record the number of repetitions step and update the maximum deception rate. Greater than the maximum value of all deception rates in the previous iteration , then assign the current deception rate to , and save the corresponding network parameters ; If the deception rate of this iteration Less than or equal to the maximum value of all deception rates in the previous iteration , then enter the next iteration until the maximum deception rate Stop when the corresponding number of repetitions step is less than half of the total number of repetitions. Fig. 9 To show how the loss on the training set changes with the number of training times, Fig.10 To change the deception rate on the training set with the number of training times, Fig.11 To show how the loss on the test set changes with the number of training times, Fig.12 To change the deception rate on the test set with the number of training times, Fig.13 This is the communication modulation style deception confusion matrix. After training, the deception rate of the modulation type recognition of the communication signal reaches 99.99%;

[0123] Step 6: Intelligently generate deception jamming signals based on communication signals;

[0124] Step 6-1: Read the network parameters corresponding to the optimal performance;

[0125] Read the optimal network parameters , and loaded into the communication modulation style deception jamming signal generation deep neural network;

[0126] Step 6-2: Sense the communication signal and generate a deceptive jamming signal.

[0127] The communication signal generated by the communication signal sending device is as follows: Fig.14 As shown, the communication modulation type is 8PSK, and the communication anti-reconnaissance equipment intelligently generates deception interference signals such as Fig.15As shown, it is sent to the communication reconnaissance equipment, and the modulation type obtained by intelligent modulation pattern recognition is GFSK, thereby achieving the purpose of destroying the communication reconnaissance equipment's ability to recognize the communication signal modulation type.

[0128] In summary, the present invention is aimed at the objective demand of protecting wireless communication transmission, and specifically trains intelligent deception and interference capabilities for the intelligent recognition model of communication modulation patterns, forming an "end-to-end" generation capability from sensing communication signals to generating interference signals. It can realize intelligent deception and interference for intelligent recognition of communication modulation patterns. The generated interference signal can deceive non-cooperative recognition of multiple communication modulation patterns at the same time, and trick the recognition type into a specified communication modulation pattern type, thereby weakening the intelligence fusion and interference decision-making capabilities, and improving the security and reliability of communications.

Claims

1. An intelligent deception jamming method for intelligent identification of communication modulation patterns, characterized in that: The following steps are involved: Step 1: Construct a communication modulation pattern dataset; the specific method is: Step 1-1: In the test area, communication signal sending equipment, communication reconnaissance equipment and communication anti-reconnaissance equipment are set up. The communication signal sending equipment transmits communication signals of N modulation patterns in total; the communication reconnaissance equipment senses the communication signal and identifies the modulation pattern; the communication anti-reconnaissance equipment generates interference signals based on the communication signal. After the communication signal is superimposed on the interference signal, the communication reconnaissance equipment will generate a specified communication pattern classification, realizing intelligent deception interference of the communication modulation pattern; each modulation pattern of the communication signal is generated M times in total, and the total number of tests is L=M×N; After the lth test, the communication signal sample obtained is , the corresponding communication signal modulation type sample is ,in represents the real data of the communication signal of the first test, It represents the imaginary data of the communication signal of the first test, and the data length is T. Indicates the modulation type corresponding to the communication signal of the first test; After L trials, the communication signal sample set is obtained as , the communication signal modulation type set is ; Step 1-2: Follow and The communication signal sample set and the communication signal modulation type set are divided into a feature training set , label training set , feature test set and a labeled test set , and their corresponding dimensions are , , and ,in ; Step 2: Construct a deep neural network for communication modulation pattern recognition; the input dimension of the deep neural network for communication modulation pattern recognition is consistent with the communication signal sample dimension, and the output dimension is consistent with the communication signal modulation type N; the deep neural network for communication modulation pattern recognition uses Conv2d convolutional network, ReLU activation function, MaxPool2d maximum pooling layer, ConvTranspose2d transposed convolutional layer and Linear linear transformation layer to construct K feature extraction channels with different receptive fields, and adds the features of different receptive fields and maps them through the Linear linear transformation layer, so that the output is consistent with the communication signal modulation type N; Step 3: Train the communication modulation pattern recognition deep neural network; the specific method is: Step 3-1: Set the optimization function and learning rate. The optimization function is Adam optimizer. Step 3-2: From the feature training set and labeled training set Randomly extract batch data, where batch is less than or equal to ,composition and , through the communication modulation pattern recognition deep neural network mapping, we get the estimated : in Deep neural networks for communication modulation pattern recognition, is the network parameter; Step 3-3: Set the loss function to the cross entropy loss function to measure the estimated value and labeled training set The difference between them is loss, the error is calculated, and the Adam optimizer is used to optimize the network parameters Optimize Step 3-4: From the feature test set and a labeled test set Randomly extract Q data, where Q is less than or equal to ,composition and , through the communication modulation pattern recognition deep neural network mapping, we get the estimated : statistics and The number of modulation types of the corresponding positions in the same position is divided by Q to get the accuracy , use the accuracy rate to verify the classification performance; Step 3-5: Repeat steps 3-2, 3-3, and 3-4, and record the number of repetitions and the maximum accuracy of all previous iterations. , if the accuracy of this iteration is Greater than the maximum accuracy of all previous iterations , then assign the correct rate to , and save the corresponding network parameters ; If the accuracy of this iteration Less than or equal to the maximum value of all correct rates in the previous iteration , then enter the next iteration until When the corresponding number of repetitions step is less than half of the total number of repetitions, stop and save the network parameters with the best performance; Step 4: Construct a deep neural network for generating a communication modulation style deception interference signal; wherein the input dimension of the deep neural network for generating a communication modulation style deception interference signal is consistent with the communication signal sample dimension, and the output dimension is consistent with the communication signal sample dimension; the deep neural network for generating a communication modulation style deception interference signal adopts a Conv2d convolutional network, a ReLU activation function, a MaxPool2d maximum pooling layer, a ConvTranspose2d transposed convolutional layer, and a Linear linear transformation layer to construct K feature extraction channels with different receptive fields, and adds the features of different receptive fields and maps them through a Linear linear transformation layer, so that the output deception interference signal is consistent with the communication signal sample dimension; Step 5: Train the communication modulation pattern to deceive the jamming signal to generate a deep neural network; the specific method is: Step 5-1: Set the deception type to That is, after the communication signal is superimposed with the interference signal, the goal of obtaining the communication modulation pattern classification result through the communication modulation pattern recognition deep neural network is ; Step 5-2: Set the optimization function and learning rate. The optimization function is Adam optimizer. Step 5-3: From the feature training set Randomly extract batch data, where batch is less than or equal to ,composition , generate deep neural network mapping by deceiving interference signals through communication modulation patterns, and obtain deceiving interference signals : in, Generate deep neural networks for communication modulation style spoofing jamming signals, is the network parameter; By recognizing the deep neural network mapping through the communication modulation pattern, we can estimate : in, Deep neural networks for communication modulation pattern recognition, is the network parameter; Step 5-4: Set the loss function to the cross entropy loss function to measure the estimated value and deception type The difference between them is loss, the error is calculated, and the Adam optimizer is used to optimize the network parameters Optimize Step 5-5: From the feature test set Randomly extract Q data, where Q is less than or equal to ,composition , generate deep neural network mapping by deceiving interference signals through communication modulation patterns, and obtain deceiving interference signals : By recognizing the deep neural network mapping through the communication modulation pattern, we can estimate : statistics Modulation type and deception type The same number, divided by Q, gives the deception rate , verify the generation performance using the deception rate; Step 5-6: Repeat steps 5-3, 5-4, and 5-5, and record the number of repetitions step and the maximum value of all deception rates in the previous iterations. , if the deception rate of this iteration Greater than the maximum value of all deception rates in the previous iteration , then assign the current deception rate to , and save the corresponding network parameters ; If the deception rate of this iteration Less than or equal to the maximum value of all deception rates in the previous iteration , then enter the next iteration until When the corresponding number of repetitions step is less than half of the total number of repetitions, stop and save the network parameters with the best performance; Step 6: Intelligently generate deception jamming signals based on communication signals; the specific method is as follows: Step 6-1: Read the network parameters for optimal performance , and loaded into the communication modulation style deception jamming signal generation deep neural network; Step 6-2: For the communication signal generated by the communication signal sending device, generate a deep neural network to generate an interference signal by deceiving the interference signal through the communication modulation style, and then superimpose the communication signal on the interference signal and send it to the communication reconnaissance equipment, thereby destroying the communication reconnaissance equipment's ability to identify the communication signal modulation type.