CN113378643A - Signal countermeasure sample detection method based on random transformation and wavelet reconstruction - Google Patents
Signal countermeasure sample detection method based on random transformation and wavelet reconstruction Download PDFInfo
- Publication number
- CN113378643A CN113378643A CN202110525751.5A CN202110525751A CN113378643A CN 113378643 A CN113378643 A CN 113378643A CN 202110525751 A CN202110525751 A CN 202110525751A CN 113378643 A CN113378643 A CN 113378643A
- Authority
- CN
- China
- Prior art keywords
- sample
- signal
- wavelet
- samples
- model
- 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.)
- Granted
Links
- 230000009466 transformation Effects 0.000 title claims abstract description 36
- 238000001514 detection method Methods 0.000 title claims abstract description 14
- 238000000034 method Methods 0.000 claims abstract description 40
- 230000009467 reduction Effects 0.000 claims abstract description 35
- 238000012360 testing method Methods 0.000 claims abstract description 35
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000013145 classification model Methods 0.000 claims abstract description 29
- 230000008569 process Effects 0.000 claims abstract description 8
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 230000006870 function Effects 0.000 claims description 17
- 238000000354 decomposition reaction Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 5
- 239000000126 substance Substances 0.000 claims description 4
- 238000013519 translation Methods 0.000 claims description 4
- 230000001131 transforming effect Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 239000012141 concentrate Substances 0.000 claims description 2
- 230000008602 contraction Effects 0.000 claims description 2
- 238000001914 filtration Methods 0.000 claims description 2
- 230000006872 improvement Effects 0.000 claims description 2
- 238000011176 pooling Methods 0.000 claims description 2
- 230000003042 antagnostic effect Effects 0.000 abstract description 4
- 230000008054 signal transmission Effects 0.000 abstract description 2
- 238000010801 machine learning Methods 0.000 description 4
- 238000013136 deep learning model Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 241000282412 Homo Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000009982 effect on human Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000010363 phase shift Effects 0.000 description 1
- 238000007493 shaping process Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Computer Networks & Wireless Communication (AREA)
- Complex Calculations (AREA)
Abstract
A signal confrontation sample detection method based on random transformation and wavelet reconstruction comprises the following steps: (1) preprocessing the modulation signal data and designing a modulation classification model; (2) designing countermeasure samples according to the signal modulation classifier; (3) designing an enhanced classification model according to a modulation classifier model and a noise reduction method; (4) obtaining a decision threshold according to a normal sample of training set data and a noise reduction sample thereof; (5) and judging the test sample according to the decision threshold value: and judging the sample according to the absolute error value of the prediction results of the test sample and the noise reduction sample of the test sample by the enhanced model, and if the prediction result is higher than the decision threshold value, judging the test sample to be a countermeasure sample, otherwise, judging the test sample to be a normal sample. The method can accurately detect the antagonistic sample in the data, effectively reduce the risk brought by the antagonistic sample in the demodulation process of the signal, and enhance the safety of the signal transmission process.
Description
Technical Field
The invention relates to a signal confrontation sample detection method based on random transformation and wavelet reconstruction, and belongs to the field of information security of machine learning.
Background
With the development of scientific progress and hardware technology, the big data era has come along, and topics such as artificial intelligence, machine learning and the like have become hot spots for the next discussion. The state develops artificial intelligence application pilot work in key industries and fields such as high-end manufacturing, finance, medical treatment, logistics, intelligent home and the like. Especially, deep learning models have achieved great success in the aspect of internal rules and feature extraction of data in recent years. Relevant studies have shown that it is possible to design slightly perturbed, antagonistic samples for known deep learning models, such that the models are identified and classified incorrectly, and although these forged samples have no significant effect on human judgment, they are fatal misleading for deep learning models, often causing deep models to produce human unexpected results. For example, in the field of signal modulation type classification, the classification model judges the label of the BPSK modulation signal as the label of the QPSK modulation type. Recently, a series of antagonistic attacks successfully implemented in the real world have demonstrated that this problem is a safety hazard for all deep learning based systems. Therefore, researches on the sample detection technology attract more and more attention of researchers in the field of machine learning and safety, and particularly, researches and exploration on deep learning have good guiding effects on future application and practice.
Although Convolutional Neural Networks (CNN), Deep Neural Networks (DNN) perform well on some complex problems such as speech recognition, signal modulation type classification, etc., they are susceptible to well-designed perturbations; typically, these perturbations are imperceptible to humans, but they can make the model misjudge with a higher confidence. In practical application, a signal transmitting base station transmits radio signals to a target base station, the signals have great application value, if the signals are intercepted maliciously in the middle and are judged by a machine learning means to be transmitted again after being adjusted elaborately (the signals are called as countermeasure signals), great potential threats are brought to a receiver of the signals, and therefore, it is important to detect which signals are the countermeasure signals.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a method for detecting a signal countercheck sample based on random transformation and wavelet reconstruction, which can accurately detect a countercheck sample in data, effectively reduce the risk brought by the countercheck sample in the signal demodulation process and strengthen the safety of the signal transmission process.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a signal confrontation sample detection method based on random transformation and wavelet reconstruction comprises the following steps:
(1) preprocessing the modulation signal data and designing a modulation classification model
And carrying out normalization processing on the existing data, and simultaneously dividing a training set and a test set. Constructing a signal modulation classifier according to the type and the characteristics of signal data;
(2) designing countermeasure samples from signal modulation classifiers
Selecting a signal countermeasure sample generation method based on optimization attack, and adjusting the gradient direction of input data according to model parameters of a signal modulation classifier, so that the signal modulation classifier generates wrong class marks for generated countermeasure samples under the condition that the input samples are slightly changed;
(3) designing an enhanced classification model from a modulation classifier model and a noise reduction method
Firstly, randomly transforming training data, primarily destroying structural disturbance, then performing wavelet decomposition on a sample after random transformation, then performing wavelet reconstruction, further destroying structural disturbance to achieve the purpose of noise reduction, and performing retraining on an original classification model by using the sample after twice transformation to obtain an enhanced classification model;
(4) obtaining decision threshold value according to normal sample of training set data and noise reduction sample thereof
Firstly, predicting an original sample of training set data and a noise reduction sample corresponding to the original sample by a model to respectively obtain prediction probability values of the model, and counting an average absolute error value of predicted values before and after sample transformation to be used as a decision threshold;
(5) determining test samples based on decision thresholds
And judging the sample according to the absolute error value of the prediction results of the test sample and the noise reduction sample of the test sample by the enhanced model, and if the prediction result is higher than the decision threshold value, judging the test sample to be a countermeasure sample, otherwise, judging the test sample to be a normal sample.
In the invention, signal modulation type data is divided into a training set and a test set, and a modulation classification model is designed according to the data information of the training set; designing a countermeasure sample generator with malicious information samples by utilizing a countermeasure sample generation method based on optimization attack (CW) and combining a modulation classification model; carrying out random transformation and wavelet reconstruction on the training set data to obtain a noise reduction sample, and training the modulation classification model again based on the noise reduction sample to obtain an enhanced modulation classification model; and predicting the training set data and the noise reduction samples according to the enhanced classification model, counting the average absolute error of the predicted value of the training set data and the noise reduction samples as a decision threshold of the detection samples, predicting the test samples (including the test set and the countermeasure samples thereof) and the noise reduction samples thereof by using the enhanced classification model to obtain the absolute error between the test samples and the noise reduction samples, and comparing the absolute error with the decision threshold, wherein the samples are countermeasure samples when the absolute error is larger than the decision threshold, and the samples are normal samples when the absolute error is larger than the decision threshold.
The invention has the following beneficial effects: obtaining a denoised sample after normal signal denoising by using a random transformation and wavelet reconstruction method; secondly, training the model again by using the noise reduction sample to obtain an enhanced model; taking the average absolute error value of the training sample and the predicted value of the noise reduction sample thereof as a decision threshold value through the enhanced classification model; and finally, detecting the test sample and the transformed noise reduction sample through a decision threshold, wherein the samples higher than the decision threshold are confrontation samples, and the samples are normal samples otherwise. The invention can effectively detect the correctness of the modulation type of the signal while classifying the modulation type of the signal data, and enhances the safety performance and the secret performance in the aspect of signal demodulation.
Drawings
FIG. 1 is a diagram of a modulation signal classification neural network;
FIG. 2 is a schematic flow chart of a quadratic training model;
FIG. 3 is a schematic diagram of a wavelet transform denoising process;
fig. 4 is a schematic diagram of a decision threshold detection flow.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
Referring to fig. 1 to 4, a method for detecting a signal countercheck sample based on random transformation and wavelet reconstruction includes the following steps:
(1) preprocessing the modulation signal data and designing a modulation classification model
And carrying out normalization processing on the existing data, and simultaneously dividing a training set and a test set. Constructing a signal modulation classifier according to the type and the characteristics of signal data;
the process of the step (1) is as follows:
1.1: normalizing the modulation signal data set and dividing the modulation signal data set D into training sets DtrainAnd test set DtestWherein the data set D { (X)1,Y1),(X2,Y2),…,(Xn+m,Yn+m)}，Dtrain＝{(X1,Y1),(X2,Y2),…,(Xn,Yn)}，Dtest＝{(Xn+1,Yn+1),(Xn+2,Yn+2),…,(Xn+m,Yn+m)}，Xi＝(xi1,xi2,…,xid) D represents XiThe length of the data of (a) is,
c represents the modulation type number, and the normalization formula is:
wherein the content of the first and second substances,
denotes normalized normal samples, XiDenotes normal samples, min (X)i) Denotes the minimum value of the normal sample, max (X)i) Represents the maximum value of normal samples;
1.2: using training set DtrainTraining the constructed classification model by data:
modulation signal classification model:
wherein F (-) is a modulation classification model, theta is a model parameter, and YiIs the output vector value of the model.
(2) Designing countermeasure samples from signal modulation classifiers
Selecting a signal countermeasure sample generation method based on optimization attack, and adjusting the gradient direction of input data according to model parameters of a signal modulation classifier, so that the signal modulation classifier generates wrong class marks for generated countermeasure samples under the condition that the input samples are slightly changed;
in the step (2), the confrontation sample
Is defined as:
wherein, deltaiIs a perturbation added to the original sample;
the optimization function is defined as:
where dis (-) represents the distance between the original sample and the challenge sample, phi (-) is the objective function, Z (-) represents the output vector of the model SoftMax layer, c is the selection constant for the binary search, and k is the confidence constant for controlling the attack.
(3) Designing an enhanced classification model from a modulation classifier model and a noise reduction method
Firstly, randomly transforming training data, primarily destroying structural disturbance, then performing wavelet decomposition on a sample after random transformation, then performing wavelet reconstruction, further destroying structural disturbance to achieve the purpose of noise reduction, and performing retraining on an original classification model by using the sample after twice transformation to obtain an enhanced classification model;
the process of the step (3) is as follows:
3.1: and (3) carrying out random transformation on the normal samples:
samples of the modulated signal are
Traversing each value in the sample in turn, and using a random number r e (0,1) and a threshold t e (0,1) as the scaling condition, where the t value is 0.15, as follows:
wherein the content of the first and second substances,
the value of the sample after scaling is represented, min _ t and max _ t represent scaling coefficients, the value of min _ t is 0.85, and the value of max _ t is 1.15;
3.2: performing wavelet reconstruction on the samples after random transformation:
the wavelet transform is an improvement on the fourier transform in nature, the basis function used in the fourier transform is fixed, and has a great disadvantage on the extraction of non-stationary aperiodic signal frequency information, and the wavelet transform replaces an infinite-length trigonometric function basis used in the fourier transform with a finite-length attenuation wavelet basis, so that frequency information can be obtained and time information can be located.
Let fi(t) represents a timing signal
Representing the added disturbance deltai，gi(t) denotes challenge samples
Then there are:
gi(t)＝fi(t)+ei(t),i＝1,2,…,n+m (7)
to fi(t) performing wavelet transform, wherein the formula can be expressed as:
wherein, a represents a transformation scale for controlling the expansion and contraction of the wavelet function; tau represents translation amount and is used for controlling the translation of the wavelet function; ψ (-) is a wavelet function.
Hypothesis perturbation ei(t) is a mean of zero and a variance of σ2White noise of (2). Let WTe(a, τ) is ei(t) is the wavelet transform of:
the energy formula can be inferred from the assumptions:
the main purpose of noise reduction of the signal is to suppress the disturbing part e as much as possiblei(t) obtaining a true signal fi(t) of (d). After the signal is subjected to wavelet transform, the signal f can be removed to the maximum extentiThe correlation of (t) concentrates most of the energy on a few wavelet coefficients of relatively large magnitude. And disturbance eiAnd (t) after wavelet transformation, the wavelet transformation is distributed on all time axes at all scales, and the amplitude is not very large. The aim of noise reduction can be achieved by a threshold filtering method by utilizing different characteristics of noise and signals on each scale of wavelet transformation.
The steps of the threshold noise reduction method are as follows:
3.2.1, for sample signal fi(t) performing wavelet transform. I.e. selecting an orthogonal wavelet and the number N of decomposition layers, for signal fi(t) performing N-layer wavelet decomposition.
3.2.2, for sample signal fiAnd (t) carrying out nonlinear threshold processing on the wavelet transform coefficients. And processing the high-frequency coefficient of each layer from the first layer to the N layers through a threshold function, and not processing the low-frequency coefficient of each layer. The threshold function is formulated as follows:
where w is the wavelet coefficient and λ is the selected threshold.
The threshold value is selected by using a common minimum maximum variance threshold value, and the formula is as follows:
μ＝middle(W1,k)/0.6745,0≤k≤2a-1-1 (13)
wherein N is the number of wavelet coefficients on the corresponding scale; w1,kRepresents wavelet coefficients of scale 1; mu is the standard deviation of the noise signal, namely the median value is obtained after the absolute value of the first-level wavelet coefficient decomposed by the signal is obtained.
3.3: and reconstructing the processed wavelet coefficient. Low frequency of Nth layer according to wavelet decompositionCarrying out signal reconstruction on the coefficients and the processed high-frequency coefficients of the first layer to the N layers so as to obtain a noise-reduced sample signal fi wt(t), then:
carrying out random transformation and wavelet reconstruction on the denoised sample
Inputting the data into an original model F (-) and performing optimization training of parameters such as convolution, pooling and the like again to obtain an enhanced model Fwt(-) making the model more robust to denoised samples. The formula is as follows:
wherein, Fwt(. to) is an enhanced modulation classification model,. theta'. is an enhanced model parameter,. YiIs the output vector value of the model.
(4) Obtaining decision threshold value according to normal sample of training set data and noise reduction sample thereof
Firstly, predicting an original sample of training set data and a noise reduction sample corresponding to the original sample by a model to respectively obtain prediction probability values of the model, and counting an average absolute error value of predicted values before and after sample transformation to be used as a decision threshold;
enhanced model FwtNormal samples of training set
Predicted probability value pi，Fwt(. to noise-reduced samples
Predicted probability value pi wt. The formula is as follows:
wherein Z (-) represents the output vector of the model SoftMax layer.
Using p of all samplesiAnd pi wtThe average absolute error value between the two is used as a decision threshold, and those greater than the decision threshold T are confrontation samples, otherwise, they are normal samples. The decision threshold calculation formula is as follows:
(5) determining test samples based on decision thresholds
And judging the sample according to the absolute error value of the prediction results of the test sample and the noise reduction sample of the test sample by the enhanced model, and if the prediction result is higher than the decision threshold value, judging the test sample to be a countermeasure sample, otherwise, judging the test sample to be a normal sample.
Confrontation sample
Obtaining a noise reduction sample after random transformation and wavelet reconstruction are carried out in the step (3)
Then using the enhanced model fwt(. to predict the contrast sample and its noise reduced sample to obtain predicted value p'iAnd
then, an absolute error value D is calculatedi，DiIs expressed as follows:
if D isiIf the decision threshold value T is larger than the threshold value T, the samples are countersamples, otherwise, the samples are normal samples. Is represented as follows:
wherein 0 represents a challenge sample; 1 represents a normal sample;
and counting the detection result of the method on the confrontation sample according to the prediction result.
Example (c): data in actual experiments
1) Selecting experimental data
The data set of the experimental signal is data.mat, and the specific conditions comprise 12 small categories of phase shift keying modulation, frequency shift keying modulation, quadrature amplitude modulation and pulse amplitude modulation: BPSK, QPSK, 8PSK, OQPSK, 2FSK, 4FSK, 8FSK, 16QAM, 32QAM, 64QAM, 4PAM and 8 PAM. The original data is randomly generated to ensure equal probability of transmitting bits. The pulse shaping filter adopts a raised cosine filter and a roll coefficient, and a random value is extracted within the range of [0.2 and 0.7 ]. The phase deviation is randomly selected within the range of [ -pi, pi ], and the normalized carrier frequency offset is randomly selected within the range of [ -0.1,0.1 ]. The signal-to-noise ratio for each modulation class is evenly distributed from-20 dB to 30 dB. Each data sample is an IQ signal, comprising 64 symbols, with 8 sample points per symbol, and thus 512 sample points per sample. The training set and the test set are 312,000 and 156,000 in size, respectively, and the amount of samples of each type of modulation signal is the same.
2) Results of the experiment
And obtaining a confrontation sample based on an optimized attack method, and obtaining a final detection result of the confrontation sample by using the method. And finally, the effect of the experiment is judged by using the ACC index. As can be seen from table 1, the method for detecting the countersample of the signal based on the random transform and the wavelet reconstruction can effectively detect the countersample.
| Model | Attack | Acc |
| Alexnet | CW | 89.98％ |
TABLE 1
The proposed detection method (RWT) is shown in table 2 in comparison to bayesian uncertainty detection method (BUE), nuclear density detection method (KDE), Local Intrinsic Dimension (LID) based methods.
| Method | RWT | LID | KDE | BUE |
| ACC | 89.98％ | 87.20％ | 51.31％ | 44.08％ |
Table 2.
The embodiments described in this specification are merely illustrative of implementations of the inventive concepts, which are intended for purposes of illustration only. The scope of the present invention should not be construed as being limited to the particular forms set forth in the examples, but rather as being defined by the claims and the equivalents thereof which can occur to those skilled in the art upon consideration of the present inventive concept.
Claims (6)
1. A method for detecting a countersample of a signal based on random transformation and wavelet reconstruction, the method comprising the steps of:
(1) preprocessing the modulation signal data and designing a modulation classification model
Performing normalization processing on the existing data, simultaneously dividing a training set and a test set, and constructing a signal modulation classifier according to the type and the characteristics of signal data;
(2) designing countermeasure samples from signal modulation classifiers
Selecting a signal countermeasure sample generation method based on optimization attack, and adjusting the gradient direction of input data according to model parameters of a signal modulation classifier, so that the signal modulation classifier generates wrong class marks for generated countermeasure samples under the condition that the input samples are slightly changed;
(3) designing an enhanced classification model from a modulation classifier model and a noise reduction method
Firstly, randomly transforming training data, primarily destroying structural disturbance, then performing wavelet decomposition on a sample after random transformation, then performing wavelet reconstruction, further destroying structural disturbance to achieve the purpose of noise reduction, and performing retraining on an original classification model by using the sample after twice transformation to obtain an enhanced classification model;
(4) obtaining decision threshold value according to normal sample of training set data and noise reduction sample thereof
Firstly, predicting an original sample of training set data and a noise reduction sample corresponding to the original sample by a model to respectively obtain prediction probability values of the model, and counting an average absolute error value of predicted values before and after sample transformation to be used as a decision threshold;
(5) determining test samples based on decision thresholds
And judging the sample according to the absolute error value of the prediction results of the test sample and the noise reduction sample of the test sample by the enhanced model, and if the prediction result is higher than the decision threshold value, judging the test sample to be a countermeasure sample, otherwise, judging the test sample to be a normal sample.
2. The method for detecting the countermeasures to the signals based on the random transformation and the wavelet reconstruction as claimed in claim 1, wherein the process of the step (1) is as follows:
1.1: normalizing the modulation signal data set and dividing the modulation signal data set D into training sets DtrainAnd test set DtestWherein the data set D { (X)1,Y1),(X2,Y2),…,(Xn+m,Yn+m)}，Dtrain＝{(X1,Y1),(X2,Y2),…,(Xn,Yn)}，Dtest＝{(Xn+1,Yn+1),(Xn+2,Yn+2),…,(Xn+m,Yn+m)}，Xi＝(xi1,xi2,…,xid) D represents XiThe length of the data of (a) is,
c represents the modulation type number, and the normalization formula is:
wherein the content of the first and second substances,
denotes normalized normal samples, XiDenotes normal samples, min (X)i) Denotes the minimum value of the normal sample, max (X)i) Represents the maximum value of normal samples;
1.2: using training set DtrainScore of data pair constructionTraining the class model:
modulation signal classification model:
wherein F (-) is a modulation classification model, theta is a model parameter, and YiIs the output vector value of the model.
3. The method for detecting the countermeasures to the signals based on the stochastic transform and the wavelet reconstruction as claimed in claim 1 or 2, wherein in the step (2), the countermeasures are taken
Is defined as:
wherein, deltaiIs a perturbation added to the original sample;
the optimization function is defined as:
where dis (-) denotes the distance between the original sample and the challenge sample,
for the objective function, Z (-) represents the output vector of the model SoftMax layer, c selects a constant for the dichotomy search, and k is the confidence constant for controlling the attack.
4. A method for detecting signal countersamples based on random transformation and wavelet reconstruction as claimed in claim 1 or 2, wherein the procedure of said step (3) is:
3.1: and (3) carrying out random transformation on the normal samples:
samples of the modulated signal are
Each value in the sample is traversed in turn and scaled using a random number r e (0,1) and a threshold t e (0,1) as conditions, as follows:
wherein the content of the first and second substances,
representing the value of the sample after scaling, and min _ t and max _ t represent scaling coefficients;
3.2: performing wavelet reconstruction on the samples after random transformation:
the wavelet transform is an improvement on the Fourier transform in nature, the basis function used in the Fourier transform is fixed, and the method has great disadvantages for extracting non-stationary aperiodic signal frequency information; the wavelet transformation is to replace an infinite-length trigonometric function base used in Fourier transformation with a finite-length attenuation wavelet base, so that frequency information can be obtained and time information can be positioned;
let fi(t) represents a timing signal
ei(t) represents the added disturbance δi，gi(t) denotes challenge samples
Then there are:
gi(t)＝fi(t)+ei(t),i＝1,2,…,n+m (7)
to fi(t) performing wavelet transform, wherein the formula is as follows:
wherein, a represents a transformation scale for controlling the expansion and contraction of the wavelet function; tau represents translation amount and is used for controlling the translation of the wavelet function; ψ (-) is a wavelet function;
hypothesis perturbation ei(t) is a mean of zero and a variance of σ2White noise of (1), let WTe(a, τ) is ei(t) is the wavelet transform of:
the energy formula can be inferred from the assumptions:
the main purpose of noise reduction of the signal is to suppress the disturbing part e as much as possiblei(t) obtaining a true signal fi(t) after wavelet transform, the signal f can be removed to the maximum extenti(t) correlation, which concentrates most of the energy on a few wavelet coefficients of relatively large amplitude, and perturbation ei(t) after wavelet transform, the noise is distributed on all time axes under each scale, the amplitude is not very large, and the purpose of noise reduction can be achieved by a threshold filtering method by utilizing different characteristics of noise and signals expressed on each scale of wavelet transform;
the steps of the threshold noise reduction method are as follows:
3.2.1, for sample signal fi(t) performing a wavelet transform, i.e. selecting an orthogonal wavelet and the number of decomposition levels N, on the signal fi(t) performing N-layer wavelet decomposition;
3.2.2, for sample signal fi(t) performing nonlinear thresholding on the wavelet transform coefficients, performing thresholding on each high-frequency coefficient of the first layer to the N layers, and not performing thresholding on the low-frequency coefficient of each layer, wherein the formula of the thresholding function is as follows:
wherein w is a wavelet coefficient and λ is a selected threshold;
the threshold value is selected by using a common minimum maximum variance threshold value, and the formula is as follows:
μ＝middle(W1,k)/0.6745,0≤k≤2a-1-1 (13) wherein N is the number of wavelet coefficients on the corresponding scale; w1,kRepresents wavelet coefficients of scale 1; mu is the standard deviation of the noise signal, namely, the median value is obtained after the absolute value of the first-level wavelet coefficient decomposed by the signal is obtained;
3.3: reconstructing the processed wavelet coefficient, and performing signal reconstruction according to the low-frequency coefficient of the Nth layer of wavelet decomposition and the high-frequency coefficient from the processed first layer to the N layer, thereby obtaining a denoised sample signal fi wt(t), then:
carrying out random transformation and wavelet reconstruction on the denoised sample
Inputting the data into an original model F (-) and performing convolution and optimization training of pooling parameters again to obtain an enhanced model Fwt(-) to make the model more robust to denoised samplesRobust, the formula is:
wherein, Fwt(. to) is an enhanced modulation classification model,. theta'. is an enhanced model parameter,. YiIs the output vector value of the model.
5. The method for detecting the countermeasures to the signals based on the stochastic transform and the wavelet reconstruction as claimed in claim 1 or 2, wherein in the step (4), the model F is enhancedwtNormal samples of training set
Predicted probability value pi，Fwt(. to noise-reduced samples
Predicted probability value pi wtThe formula is as follows:
wherein Z (-) represents the output vector of the model SoftMax layer;
using p of all samplesiAnd pi wtThe average absolute error value between the two is taken as a decision threshold, the samples which are greater than the decision threshold T are confrontation samples, otherwise, the samples are normal samples, and the decision threshold calculation formula is as follows:
6. the method for detecting the countermeasures to the signals based on the stochastic transform and the wavelet reconstruction as claimed in claim 1 or 2, wherein in the step (5), the countermeasures are taken
Obtaining a noise reduction sample after random transformation and wavelet reconstruction are carried out in the step (3)
Then using the enhanced model fwt(. to predict the contrast sample and its noise reduced sample to obtain predicted value p'iAnd
then, an absolute error value D is calculatedi，DiIs expressed as follows:
Di＝|p′i-pi'wt|,i＝n+1,n+2，...,n+m (19)
if D isiIf the decision threshold value T is larger than the threshold value T, the samples are countersamples, otherwise, the samples are normal samples, and the expression is as follows:
wherein 0 represents a challenge sample; 1 represents a normal sample;
and counting the detection result of the method on the confrontation sample according to the prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110525751.5A CN113378643B (en) | 2021-05-14 | 2021-05-14 | Signal countermeasure sample detection method based on random transformation and wavelet reconstruction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110525751.5A CN113378643B (en) | 2021-05-14 | 2021-05-14 | Signal countermeasure sample detection method based on random transformation and wavelet reconstruction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113378643A true CN113378643A (en) | 2021-09-10 |
| CN113378643B CN113378643B (en) | 2024-03-22 |
Family
ID=77571080
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110525751.5A Active CN113378643B (en) | 2021-05-14 | 2021-05-14 | Signal countermeasure sample detection method based on random transformation and wavelet reconstruction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113378643B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113723358A (en) * | 2021-09-15 | 2021-11-30 | 中国电子科技集团公司第三十六研究所 | Countermeasure signal detection method and device based on generation of countermeasure network and electronic equipment |
| CN113723358B (en) * | 2021-09-15 | 2024-07-09 | 中国电子科技集团公司第三十六研究所 | Method and device for detecting countermeasure signal based on generated countermeasure network and electronic equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101420758A (en) * | 2008-11-26 | 2009-04-29 | 北京科技大学 | Method for resisting simulated main customer attack in cognitive radio |
| US20200220751A1 (en) * | 2019-01-09 | 2020-07-09 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Method for bandpass sampling by position modulated wavelets |
| CN112347844A (en) * | 2020-09-21 | 2021-02-09 | 浙江工业大学 | Signal countermeasure sample detector design method based on LID |
-
2021
- 2021-05-14 CN CN202110525751.5A patent/CN113378643B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101420758A (en) * | 2008-11-26 | 2009-04-29 | 北京科技大学 | Method for resisting simulated main customer attack in cognitive radio |
| US20200220751A1 (en) * | 2019-01-09 | 2020-07-09 | Commissariat A L'energie Atomique Et Aux Energies Alternatives | Method for bandpass sampling by position modulated wavelets |
| CN112347844A (en) * | 2020-09-21 | 2021-02-09 | 浙江工业大学 | Signal countermeasure sample detector design method based on LID |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113723358A (en) * | 2021-09-15 | 2021-11-30 | 中国电子科技集团公司第三十六研究所 | Countermeasure signal detection method and device based on generation of countermeasure network and electronic equipment |
| CN113723358B (en) * | 2021-09-15 | 2024-07-09 | 中国电子科技集团公司第三十六研究所 | Method and device for detecting countermeasure signal based on generated countermeasure network and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113378643B (en) | 2024-03-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108234370B (en) | Communication signal modulation mode identification method based on convolutional neural network | |
| CN111582320B (en) | Dynamic individual identification method based on semi-supervised learning | |
| US11700156B1 (en) | Intelligent data and knowledge-driven method for modulation recognition | |
| Hong-hai et al. | Radar emitter multi-label recognition based on residual network | |
| CN112087774A (en) | Communication radiation source individual identification method based on residual error neural network | |
| CN110276256A (en) | Based on the low signal-to-noise ratio Modulation Recognition of Communication Signal method and device for adjusting ginseng accidental resonance | |
| CN111985349B (en) | Classification recognition method and system for radar received signal types | |
| CN111680737A (en) | Radar radiation source individual identification method under differential signal-to-noise ratio condition | |
| CN112347844B (en) | LID-based signal countermeasure sample detector design method | |
| CN116612308A (en) | Abnormal data detection method, device, equipment and storage medium | |
| Ghasemzadeh et al. | GGCNN: An efficiency-maximizing gated graph convolutional neural network architecture for automatic modulation identification | |
| Ali et al. | Modulation format identification using supervised learning and high-dimensional features | |
| CN113076925B (en) | M-QAM signal modulation mode identification method based on CNN and ELM | |
| Essai et al. | Automatic Modulation Classification: Convolutional Deep Learning Neural Networks Approaches | |
| CN113378643B (en) | Signal countermeasure sample detection method based on random transformation and wavelet reconstruction | |
| CN111950635A (en) | Robust feature learning method based on hierarchical feature alignment | |
| Zhang et al. | Deep learning‐based digital signal modulation identification under different multipath channels | |
| CN104408715A (en) | SAR (Synthetic Aperture Radar) image analysis method based on self-adaptive fuzzy C mean-value clustering fuzzification | |
| CN111404852B (en) | Modulation mode identification method based on amplitude and spectral amplitude characteristics | |
| Guo-Xi | Research on a Convolutional Neural Network Method for Modulation Waveform Classification. | |
| CN103049887B (en) | Curvelet transformation domain image denoising method based on fuzzy support vector machine | |
| Gao et al. | Evolutionary Neural Network based on Quantum Elephant Herding Algorithm for Modulation Recognition in Impulse Noise | |
| CN113723358B (en) | Method and device for detecting countermeasure signal based on generated countermeasure network and electronic equipment | |
| Zhang et al. | Modulation recognition of 5G signals based on AlexNet convolutional neural network | |
| Zhang et al. | GAN-SNR-shrinkage-based network for modulation recognition with small training sample size |
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 |