CN112202527B - Intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance - Google Patents
Intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance Download PDFInfo
- Publication number
- CN112202527B CN112202527B CN202011064169.5A CN202011064169A CN112202527B CN 112202527 B CN112202527 B CN 112202527B CN 202011064169 A CN202011064169 A CN 202011064169A CN 112202527 B CN112202527 B CN 112202527B
- Authority
- CN
- China
- Prior art keywords
- matrix
- gradient
- adv
- interference
- signal
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04K—SECRET COMMUNICATION; JAMMING OF COMMUNICATION
- H04K3/00—Jamming of communication; Counter-measures
- H04K3/60—Jamming involving special techniques
- H04K3/65—Jamming involving special techniques using deceptive jamming or spoofing, e.g. transmission of false signals for premature triggering of RCIED, for forced connection or disconnection to/from a network or for generation of dummy target signal
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/0012—Modulated-carrier systems arrangements for identifying the type of modulation
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Noise Elimination (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides an intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance, which determines an optimization direction by solving the gradient of a loss function correctly classified by an identification system relative to a radio frequency signal, and realizes successful interference through multiple iterations, wherein the optimization direction does not depend on the gradient solved in the iteration but is the gradient accumulated in all iterations, so that the optimization is easier to iterate towards a global minimum rather than a local minimum, thereby improving the calculation efficiency and simultaneously improving the interference success rate; the invention greatly reduces the classification accuracy of the identification system under the condition of only slightly disturbing the original signal, thereby achieving the effects of deception and interference.
Description
Technical Field
The invention relates to the field of signal processing, in particular to an intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance. The invention is suitable for interfering an electromagnetic signal modulation recognition system based on a neural network, and under the condition of only slightly disturbing an original signal, the classification accuracy and reliability of the recognition system are greatly reduced, so that the effect of deceptive interference is achieved.
Background
With the deep fusion application of the deep learning technology in the intelligent electromagnetic spectrum sensing, the recognition accuracy of the electromagnetic signal waveform, the modulation parameter and other information is obviously improved, and the method has the characteristics of high classification accuracy, high instantaneity and the like. The intelligent electromagnetic signal identification system poses great threat to the safety and reliability of electromagnetic information transmission. Therefore, from the perspective of a signal transmitting party, how to disturb the learning and reasoning capabilities of the intelligent model, reduce the probability of successful sensing and recognition of the transmitted signal, and ensure the safe transmission of information is a research hotspot and difficult point problem in the field of electromagnetic signal processing.
At present, the existing intelligent system interference research mainly focuses on image recognition, voice signal recognition and the like, but no interference technology aiming at an electromagnetic modulation signal recognition system exists for a while. The basic idea of the interference technology of the intelligent electromagnetic signal identification system is to find potential vulnerability of a neural network optimization learning process, and add small disturbance to an original electromagnetic signal, so that the classification accuracy of the modulation identification system is reduced to the maximum extent under the condition of not influencing the self information expression of the signal. Typical interference techniques include Fast Gradient signal Method (Fast Gradient signal Method), Basic Iterative Method (Basic Iterative Method), and the like. However, the above methods all have the disadvantages that the optimization direction is easy to swing and unstable, so that the generated interference signal data is easy to fall into local minimum, the operation complexity is high, and the success rate is low.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides an intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance, which determines the optimization direction by solving the gradient of a loss function correctly classified by an identification system relative to a radio frequency signal, and realizes successful interference through multiple iterations. The optimization direction of the method does not depend on the gradient solved in the iteration, but the gradient accumulated in all iterations, so that the optimization is easier to iterate towards the global minimum rather than the local minimum, the calculation efficiency is improved, and the interference success rate is also improved; the momentum gradient method optimizes the iteration direction in the iteration process, and greatly improves the calculation efficiency and the interference success rate.
The technical scheme adopted by the invention for solving the technical problem comprises the following specific steps:
the method comprises the following steps: the radio frequency signal to be transmitted is represented as a 2 × N dimensional matrix X (2, N), where N represents time discrete sampling points, two rows of the matrix represent the real part and the imaginary part of the complex signal, respectively, the correct classification category of the matrix X is k, and the modulation identification classifier is represented as f (·), then there are:
f(X)＝k (1)
the gradient accumulation matrix is expressed as a matrix m (2, N) with dimensions of 2 XN, the gradient accumulation matrix is initialized to be a full 0 matrix, and an interference signal matrix X is initialized adv ＝X；
Step two: solving a loss function of a neural network correctly classifying interference signal matrix in the current iteration times, and updating to obtain a gradient accumulation matrix;
step three: obtaining an updated interference signal matrix through the gradient accumulation matrix;
step four: normalizing the updated interference signal matrix obtained in the third step
The method comprises the following steps that Clip (A, max and min) is an upper and lower bound function, elements which are larger than a max value or smaller than a min value in a matrix A are normalized to max or min, and range is a limited disturbance range;
step five: repeating the second step to the fourth step until the following conditions are met:
f(X adv )≠k (5)
i.e. the interference is successful.
In the second step, the loss function is J (X) adv K), then solve the loss function J (X) adv K) with respect to X adv Gradient of (2)
Subjecting the gradient to l 1 Normalizing the norm, and finally adding the norm to the original gradient accumulation matrix m multiplied by the attenuation factor mu to obtain an updated gradient accumulation matrix
Wherein, J (X) adv K) represents the neural network to convert the signal X adv A loss function classified as a kth class,
representational solving with respect to the current interference signal matrix X adv Of the gradient of (c).
In the third step, a gradient accumulation quantity matrix is obtained
Multiplying the sign +1 or-1 by the step factor alpha, and then multiplying the result with X adv Adding to obtain updated interference signal matrix
Wherein sign (·) is a sign function.
The invention has the advantages that by adopting the intelligent electromagnetic signal identification system interference technology based on momentum gradient disturbance, the classification accuracy of the identification system is greatly reduced under the condition of only slightly disturbing the original signal, and the effects of deception and interference are achieved.
Drawings
Fig. 1 is a comparison graph of the time domain of a signal before and after adding interference. Fig. 1(a) is a time domain image of the real part of the complex signal, and fig. 1(b) is a time domain image of the imaginary part of the complex signal.
Fig. 2 is a comparison graph of the frequency domain of a signal before and after adding interference. Fig. 2(a) shows the real part of the complex signal spectrum, and fig. 2(b) shows the imaginary part of the complex signal spectrum.
FIG. 3 is a comparison graph of classification accuracy of an electromagnetic signal identification system before and after a disturbance.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
An 8-layer convolutional neural network is trained as a target modulation identification system, the signal-to-noise ratio of training data is 10dB, the data set comprises 11 common modulation types, namely QPSK, PAM4, AM-DSB, GFSK, QAM64, AM-SSB, 8PSK, QAM16, WBFM, CPFSK and BPSK, the categories are respectively numbered to be 0-10, and the classification accuracy of the system before interference is 75%.
The signal modulation scheme used in this example is QPSK, which corresponds to class 0 in this classification system.
The intelligent electromagnetic signal identification system interference technology based on momentum gradient disturbance comprises the following specific steps:
the method comprises the following steps: the processed rf signal is a 2 × 128 dimensional matrix X (2, N), the number of time discrete sampling points N is 128, two rows of the matrix represent the real part and imaginary part of the complex signal respectively, the correct classification category of the matrix X is k, in this example k is 0, and the identification system is denoted as f (), then there are:
f(X)＝0 (6)
the initial gradient accumulation amount matrix m (2, N) is a full 0 matrix of 2 × 128 dimensions. Initializing an interference signal matrix X adv ＝X。
Step two: the loss function J (X) of the correct classification of the neural network in the current iteration step is obtained adv 0) relative to the signal X adv Is subjected to a gradient of l 1 After norm normalization, the norm is finally added with the gradient accumulation matrix m multiplied by the attenuation factor mu to obtain an updated gradient accumulation matrix
Wherein, J (X) adv 0) represents the neural network to convert the signal X adv A loss function correctly classified into QPSK modulation class,
representational solving with respect to the current interference signal matrix X adv Of the gradient of (c).
Step three: the sign of the gradient accumulation quantity matrix m is obtained, and the sign (+1 or-1) is multiplied by the step factor alpha and then is compared with the signal matrix X adv Adding to obtain updated interference signal matrix X adv ：
X adv ＝X adv +α·sign(m) (8)
Wherein sign (·) is a sign function.
Step four: normalizing the updated interference signal matrix:
[X adv ]＝Clip(X adv ,X+range,X-range) (9)
the Clip (a, max, min) c is an upper and lower bound function, and the elements in the matrix a greater than the max value or less than the min value can be normalized to max or min. range is the limited perturbation range.
Step five: and repeating the second step, the third step and the fourth step until the interference is successful. In this case:
f(X adv )＝1 (10)
class 1 represents PAM4 modulation in this modulation identification system.
Through the above processes, the system misclassifies the interference signals which should be modulated by QPSK as modulation PAM4, the accuracy rate of classification is reduced from 75% to 28%, and the effect of deceptive interference is achieved.
Fig. 1 and fig. 2 are respectively a time domain and frequency domain signal difference before and after an intelligent electromagnetic signal identification system interference technology based on momentum gradient disturbance is adopted, fig. 1(a) is a time domain image of a real part of a complex signal, fig. 1(b) is a time domain image of an imaginary part of the complex signal, fig. 2(a) is a real part of a complex signal spectrum, and fig. 2(b) is an imaginary part of the complex signal spectrum.
Fig. 3 is a comparison of the classification accuracy of the system before and after the interference, and it can be seen that after the intelligent electromagnetic signal identification system interference technology based on momentum gradient disturbance is adopted, the classification identification accuracy of the system is greatly reduced.
Claims (1)
1. An intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance is characterized by comprising the following steps:
the method comprises the following steps: the radio frequency signal to be transmitted is represented as a 2 × N dimensional matrix X (2, N), where N represents time discrete sampling points, two rows of the matrix represent the real part and the imaginary part of the complex signal, respectively, the correct classification category of the matrix X is k, and the modulation identification classifier is represented as f (·), then there are:
f(X)＝k (1)
the gradient accumulation matrix is expressed as a matrix m (2, N) with dimensions of 2 XN, the gradient accumulation matrix is initialized to be an all-0 matrix, and the interference signal matrix X is initialized adv ＝X；
Step two: solving a loss function of a neural network correctly classifying interference signal matrix in the current iteration times, and updating to obtain a gradient accumulation matrix;
the loss function is J (X) adv K), then solve the loss function J (X) adv K) with respect to X adv Gradient of (2)
Subjecting the gradient to l 1 Normalizing the norm, and finally adding the norm to the original gradient accumulation matrix m multiplied by the attenuation factor mu to obtain an updated gradient accumulation matrix
Wherein, J (X) adv K) represents the neural network to convert the signal X adv A loss function classified as a kth class,
representational solution for a current interference signal matrix X adv A gradient of (a);
step three: obtaining an updated interference signal matrix through the gradient accumulation matrix;
in the third step, a gradient accumulation quantity matrix is obtained
Multiplying the sign +1 or-1 by the step factor alpha, and then multiplying the result with X adv Adding to obtain an updated interference signal matrix
Wherein sign (·) is a sign function;
step four: normalizing the updated interference signal matrix obtained in the third step
The method comprises the following steps that Clip (A, max and min) is an upper and lower bound function, elements which are larger than a max value or smaller than a min value in a matrix A are normalized to max or min, and range is a limited disturbance range;
step five: repeating the second step to the fourth step until the following conditions are met:
f(X adv )≠k (5)
i.e. the interference is successful.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011064169.5A CN112202527B (en) | 2020-10-01 | 2020-10-01 | Intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011064169.5A CN112202527B (en) | 2020-10-01 | 2020-10-01 | Intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112202527A CN112202527A (en) | 2021-01-08 |
| CN112202527B true CN112202527B (en) | 2022-09-13 |
Family
ID=74013944
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011064169.5A Active CN112202527B (en) | 2020-10-01 | 2020-10-01 | Intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112202527B (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106301419A (en) * | 2015-06-04 | 2017-01-04 | 华为技术有限公司 | Radio frequency interference removing method and device |
| CN106603135A (en) * | 2016-12-29 | 2017-04-26 | 重庆邮电大学 | Iterative interference alignment method based on double-layer precoding in multi-user MIMO interference system |
| WO2017195654A1 (en) * | 2016-05-12 | 2017-11-16 | シャープ株式会社 | Transmission device, reception device, and communication method |
| CN109302262A (en) * | 2018-09-27 | 2019-02-01 | 电子科技大学 | A kind of communication anti-interference method determining Gradient Reinforcement Learning based on depth |
| CN110048978A (en) * | 2019-04-09 | 2019-07-23 | 西安电子科技大学 | A kind of signal modulate method |
| CN111638488A (en) * | 2020-04-10 | 2020-09-08 | 西安电子科技大学 | Radar interference signal identification method based on LSTM network |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10271233B2 (en) * | 2013-03-15 | 2019-04-23 | DGS Global Systems, Inc. | Systems, methods, and devices for automatic signal detection with temporal feature extraction within a spectrum |
| CN110086737B (en) * | 2019-03-13 | 2021-07-02 | 西安电子科技大学 | Communication signal modulation mode identification method based on graph neural network |
| CN111092836B (en) * | 2019-12-13 | 2022-05-17 | 中国人民解放军空军工程大学 | Signal modulation mode identification method and device |
-
2020
- 2020-10-01 CN CN202011064169.5A patent/CN112202527B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106301419A (en) * | 2015-06-04 | 2017-01-04 | 华为技术有限公司 | Radio frequency interference removing method and device |
| WO2017195654A1 (en) * | 2016-05-12 | 2017-11-16 | シャープ株式会社 | Transmission device, reception device, and communication method |
| CN106603135A (en) * | 2016-12-29 | 2017-04-26 | 重庆邮电大学 | Iterative interference alignment method based on double-layer precoding in multi-user MIMO interference system |
| CN109302262A (en) * | 2018-09-27 | 2019-02-01 | 电子科技大学 | A kind of communication anti-interference method determining Gradient Reinforcement Learning based on depth |
| CN110048978A (en) * | 2019-04-09 | 2019-07-23 | 西安电子科技大学 | A kind of signal modulate method |
| CN111638488A (en) * | 2020-04-10 | 2020-09-08 | 西安电子科技大学 | Radar interference signal identification method based on LSTM network |
Non-Patent Citations (2)
| Title |
|---|
| SAR Interference Suppression Based on Signal Synthesis from Joint Time-Frequency Distribution;Jia Su等;《IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium》;20191114;全文 * |
| 基于深度学习的数字调制信号识别算法研究;王咏实;《中国优秀硕士学位论文全文数据库》;20200115;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112202527A (en) | 2021-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Güner et al. | Automatic digital modulation classification using extreme learning machine with local binary pattern histogram features | |
| CN112418014A (en) | Modulation signal identification method based on wavelet transformation and convolution long-short term memory neural network | |
| Ma et al. | Automatic modulation classification under non-Gaussian noise: A deep residual learning approach | |
| CN112039820B (en) | Communication signal modulation and identification method for quantum image group mechanism evolution BP neural network | |
| CN108549832B (en) | Low-interception radar signal classification method based on full-connection neural network | |
| CN113094993B (en) | Modulation signal denoising method based on self-coding neural network | |
| CN114595732B (en) | Radar radiation source sorting method based on depth clustering | |
| KR102073935B1 (en) | Modulation recognition for radil signal | |
| CN113055107B (en) | Interference strategy generation method for radio station with unknown communication mode | |
| CN114531274B (en) | Intelligent countermeasure method, system, medium and equipment for communication signal modulation recognition | |
| CN114881092A (en) | Signal modulation identification method based on feature fusion | |
| CN112305506B (en) | CNN-based LPI radar signal detection method | |
| CN110751049B (en) | Defense method facing signal sampling gradient attack | |
| CN116866129A (en) | Wireless communication signal detection method | |
| CN114726692B (en) | SERESESESENet-LSTM-based radiation source modulation mode identification method | |
| CN113378644A (en) | Signal modulation type recognition attack defense method based on generative countermeasure network | |
| CN111159891B (en) | Single-sample polarization filter coefficient prediction method based on DNN neural network | |
| CN114595729A (en) | Communication signal modulation identification method based on residual error neural network and meta-learning fusion | |
| CN112202527B (en) | Intelligent electromagnetic signal identification system interference method based on momentum gradient disturbance | |
| Abdelmutalab et al. | Automatic modulation classification using polynomial classifiers | |
| Cheng et al. | An Optimized Neural Network Classifier for Automatic Modulation Recognition | |
| CN115834310A (en) | Communication signal modulation identification method based on LGTransformer | |
| CN113343924B (en) | Modulation signal identification method based on cyclic spectrum characteristics and generation countermeasure network | |
| Ghauri et al. | -PAM signals classification using modified Gabor filter network | |
| CN113537025A (en) | Electromagnetic modulation signal deep denoising method and system based on self-supervision learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |