CN115270866A - Short-wave communication behavior identification method based on self-correlation high-order spectrum features - Google Patents
Short-wave communication behavior identification method based on self-correlation high-order spectrum features Download PDFInfo
- Publication number
- CN115270866A CN115270866A CN202210867401.1A CN202210867401A CN115270866A CN 115270866 A CN115270866 A CN 115270866A CN 202210867401 A CN202210867401 A CN 202210867401A CN 115270866 A CN115270866 A CN 115270866A
- Authority
- CN
- China
- Prior art keywords
- autocorrelation
- order
- communication behavior
- matrix
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000004891 communication Methods 0.000 title claims abstract description 35
- 238000001228 spectrum Methods 0.000 title claims abstract description 30
- 239000011159 matrix material Substances 0.000 claims abstract description 31
- 230000004913 activation Effects 0.000 claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 23
- 238000005311 autocorrelation function Methods 0.000 claims abstract description 15
- 230000009466 transformation Effects 0.000 claims abstract description 11
- 238000005070 sampling Methods 0.000 claims abstract description 8
- 238000010606 normalization Methods 0.000 claims abstract description 7
- 230000006399 behavior Effects 0.000 claims description 33
- 230000006870 function Effects 0.000 claims description 31
- 238000002474 experimental method Methods 0.000 claims description 14
- 238000012360 testing method Methods 0.000 claims description 10
- 238000013527 convolutional neural network Methods 0.000 claims description 8
- 238000004519 manufacturing process Methods 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000008569 process Effects 0.000 claims description 6
- 239000000654 additive Substances 0.000 claims description 3
- 230000000996 additive effect Effects 0.000 claims description 3
- 238000004088 simulation Methods 0.000 claims description 3
- 230000003595 spectral effect Effects 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 4
- 238000003062 neural network model Methods 0.000 abstract description 3
- 230000005855 radiation Effects 0.000 description 6
- 230000000052 comparative effect Effects 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000001149 cognitive effect Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000003542 behavioural effect Effects 0.000 description 1
- 230000019771 cognition Effects 0.000 description 1
- 238000005314 correlation function Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000000717 retained effect 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
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Algebra (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a short wave communication behavior identification method based on self-correlation high-order spectrum characteristics, which comprises the following steps: solving an autocorrelation function of the time domain signal to obtain an autocorrelation sequence; performing bispectrum transformation on the autocorrelation sequence, and calculating third-order cumulant; sampling the obtained third-order cumulant matrix, converting the three-order cumulant matrix into a three-channel format similar to a picture, and constructing a data set; designing two paths of input convolution neural network models, and further improving the models by increasing the depth of the models, introducing a Batch Normalization layer and improving an activation function to obtain a final network model; and inputting the data set into a final network model to complete the tasks of training and classification. The method can avoid a series of operations such as demodulation, de-interleaving, decoding and the like on the signals, can directly judge the communication behavior by identifying the physical layer signals, and has good identification effect under the environment with low signal-to-noise ratio.
Description
Technical Field
The invention belongs to the field of intelligent communication countermeasure, and particularly relates to a short-wave communication behavior identification method based on autocorrelation high-order spectrum characteristics.
Background
With the development and development of cognitive electronic warfare, the application of "cognition" in the field of electronic warfare is increasingly required. The communication countermeasure and the radar countermeasure develop towards intellectualization. The communication radiation source such as a short-wave radio station, an ultrashort-wave radio station, a satellite link and a tactical data link plays a crucial role in an information battlefield, and the communication radiation source has an operator, so that the communication radiation source bears a loaded behavior intention and reflects behavior information such as actions, situations and states of non-cooperative parties to a certain extent. By analyzing the behavior level of the intercepted communication radiation source signal, the behavior information of the communication radiation source operator can be reversely mined, and the communication countermeasure method has the advantages of communication countermeasure.
Behavioral level cognitive research based on intercepted physical layer electromagnetic signals is just started, few researches have been carried out at present, and in a complex electromagnetic environment, particularly a low signal-to-noise ratio environment, how to infer behaviors through the intercepted physical layer signals is a problem to be solved urgently.
Disclosure of Invention
The invention aims to provide a short wave communication behavior identification method based on self-correlation high-order spectrum characteristics, and the method is used for achieving the aim of reverse reasoning behavior from a physical layer.
The technical solution for realizing the purpose of the invention is as follows: a short wave communication behavior identification method based on self-correlation high-order spectrum characteristics comprises the following steps:
s1, solving an autocorrelation function of a time domain signal to obtain an autocorrelation sequence;
s2, performing bispectrum transformation on the autocorrelation sequence, and calculating third-order cumulant;
s3, sampling the obtained third-order cumulant matrix, converting the three-order cumulant matrix into a three-channel format similar to a picture, and constructing a data set;
s4, designing two paths of input convolutional neural network models, and further improving the models by increasing the depth of the models, introducing a Batch Normalization layer and improving an activation function to obtain a final network model;
and S5, inputting the data set into a final network model to complete the tasks of training and classification.
Compared with the prior art, the invention has the following remarkable advantages:
(1) Combining signal processing and deep learning, extracting fingerprint characteristics of physical layer signals of different behaviors through signal processing, and training and classifying the characteristics by using a convolutional neural network to realize the purpose of reversely reasoning behaviors from the physical layer signals and provide an idea for the behavior research of a communication radiation source;
(2) The method is suitable for the environment with larger noise, has better recognition effect under the condition of low signal to noise ratio, is fast in model training, has good real-time performance, and can carry out analysis in real time;
(3) The method can directly analyze the obtained physical layer short wave signals obtained by detection and reception, and can realize the purpose of short wave communication behavior without complex operations such as demodulation, de-interleaving, decoding and the like on the signals.
Drawings
Fig. 1 is a flowchart of the short-wave communication behavior identification method based on the autocorrelation high-order spectrum features.
Fig. 2 is a diagram of a network model architecture.
Fig. 3 is a diagram of the relationship between five types of burst waveforms and the communication behavior.
Fig. 4a is a signal forming flow chart of the short wave burst waveform BW 0.
Fig. 4b is a signal forming flow chart of the short wave burst waveform BW 1.
Fig. 4c is a signal forming flow chart of the short wave burst waveform BW 2.
Fig. 4d is a signal forming flow chart of the short wave burst waveform BW 3.
Fig. 4e is a signal forming flow chart of the short wave burst waveform BW 4.
FIG. 5 is a graph of accuracy results of different network model identifications.
Detailed Description
The invention relates to a short wave communication behavior identification method based on autocorrelation high-order spectrum characteristics, which comprises the following steps:
s1, solving an autocorrelation function of a time domain signal to obtain an autocorrelation sequence;
s2, performing bispectrum transformation on the autocorrelation sequence, and calculating third-order cumulant;
s3, sampling the obtained third-order cumulant matrix, converting the three-order cumulant matrix into a three-channel format similar to a picture, and constructing a data set;
s4, designing two paths of input convolutional neural network models, and further improving the models by increasing the depth of the models, introducing a Batch Normalization layer and improving an activation function to obtain a final network model;
and S5, inputting the data set into a final network model to complete the tasks of training and classification.
As a specific example, the specific step of S1 is:
step S11, generating 1000 signal samples of the five burst waveforms by MATLAB simulation according to five burst waveforms BW0, BW1, BW2, BW3 and BW4 specified by the American military short wave radio rule MIL-STD-188-141B, wherein the signals are I and Q signals, and generating a time sequence xk;
Step S12, finding xkIs defined as:
for the receiving end signal skConsider additive white Gaussian noise nkThen sk=xk+nk;
Defining a receiver-side signal skThe autocorrelation function r of (a) is:
As a specific example, the specific step of S2 is:
s21, estimating a third-order spectrum of the generated autocorrelation sequence by using a nonparametric method;
signal high order spectrum skx(ω1,···,ωk-1) Is defined as follows:
wherein c iskx(τ1,···,τk-1) Is xkThe k-th order cumulant of (c);
defining the k-1 spectrum of the signal as the k-th order spectrum of the signal, so that the bispectrum s of the signal3x(ω1,ω2) The definition is as follows:
c3x(τ1,τ2) Is xkThe 3 rd order cumulant of (c);
step S22, according to the definition formula of bispectrum, the autocorrelation sequence r is solvedssOf the dual spectrum transformation matrixs;
The length of the FFT is set to 256, the length of the rao optimum window function is set to 5, the length of each segment is 250, the overlap length of each segment is 30, resulting in a 256 × 256 matrix s:
wherein, c3r(τ1,τ2) Is the 3 rd order cumulant of the autocorrelation function r, s (ω)1,ω2) Is according to c3r(τ1,τ2) The resulting bispectrum is calculated.
As a specific example, the specific step of S3 is:
step S31, sampling a matrix S obtained by double-spectrum transformation, wherein the original matrix is 256 multiplied by 256 dimensions, odd-numbered lines are sampled at 1, 3 and 5 '\ 8230255 points, even-numbered lines are sampled at 2, 4 and 6' \ 8230256 points, and two 128 multiplied by 128 matrices S are obtainedI、sQ;
Step S32, adopting maximum and minimum normalization to SI、sQPreprocessing line by line to make the value be [0, 1%]Within the scope, the specific update process is as follows:
wherein,is s isIThe (c) th row of (a),is s isQJ, i, j ∈ [1,128 ]](ii) a Obtaining a matrix s updated after preprocessingIAnd sQ;
Step S33,Will s isIAnd sQSplicing in the third dimension to obtain a 128 multiplied by 2 three-dimensional matrix S;
and step S34, performing the operations on 5000 samples one by one to construct a four-dimensional matrix Y of 5000 multiplied by 128 multiplied by 2, and manufacturing a data set.
As a specific example, the specific step of S4 is:
s41, constructing two branches for the network model according to the structures of a classical convolutional neural network LeNet-5 and AlexNet, and respectively marking as a path I and a path II;
in two branches, the activation function includes relu and leakyrelu, and the expression of the relu activation function is:
f(x)=max(0,x)
the expression for the Leakyrelu activation function is:
f(x)=max(0,x)+leak*min(0,x)
wherein x is the input of the activation function and f (x) is the output of the activation function; leak is a very small constant to preserve the value of the negative x-axis, usually around 0.01;
s42, increasing the model depth of the second path by adding a full connection layer; the last fully-connected layer output dimension of the two branches is 256;
and S43, superposing the 256-dimensional vectors output by the two full-connection layers, sequentially passing through a leakyrelu activation function, a batchnormalization layer and a dropout layer, and finally outputting a classification result through a softmax classifier.
As a specific example, the specific step of S5 is:
s51, manufacturing 5000 labels according to the data set, and performing label learning through a network; dividing the preprocessed data set Y into a training set, a verification set and a test set according to the proportion of 3;
inputting the pre-training set into a network model, and updating the weight by adopting a Momentum gradient descent algorithm, namely a Momentum algorithm, wherein the updating process of the Momentum algorithm parameters is as follows:
W←W-ανdw
b←b-ανdb
wherein x is(i)Collecting m samples { x) for a training set(1),x(2),···,x(m)I data of g is gradient, vkIs the kth update speed, θkIs the kth iteration weight; epsilon is a learning rate, alpha is a momentum parameter, the initial learning rate epsilon is set to be 0.001, and the momentum parameter alpha is set to be 0.9; w and b are weight values and offset of the neural network, vdw、νdbThe relationship of the gradient between two adjacent steps is established,in order to obtain the gradient of the loss function in the weight value and the offset, α and β are momentum parameters in the algorithm, and β is usually set to 0.9, and may be set to other values;
using a cross entropy loss function, the expression is as follows:
where M is the number of classes, yiIs a true label of the specimen, PiIs the probability that the sample belongs to the i-th class;
the sample size selected in one training is set to be 64, and the sample training time epoch is set to be 500;
and S52, after the model training is finished, saving the model weight, inputting the preprocessed test set into the trained model for testing, and outputting a classification result and the identification accuracy to perform real-time classification and identification on the communication behavior.
As a specific example, step S52 is followed by:
step S53, setting different signal-to-noise ratio conditions of 0dB, 5dB, 8dB, 10dB and 15dB respectively; and (3) respectively adding corresponding noise to the 5000 samples generated in the step S1, making a data set according to the steps of the step S2 and the step S3, repeating the experiment, and verifying the reliability of the algorithm under the noise condition.
The technical solution of the present invention will be further described in detail with reference to the accompanying drawings and examples.
Examples
With reference to fig. 1, the invention provides a short-wave communication behavior identification method based on autocorrelation high-order spectrum characteristics, which achieves the purpose of physical layer signal classification to behavior identification, and comprises the following steps:
s1, solving an autocorrelation function of a time domain signal to obtain an autocorrelation sequence;
step S11, generating 1000 signal samples of the five burst waveforms by using MATLAB simulation according to five burst waveforms BW0, BW1, BW2, BW3 and BW4 specified by the American military shortwave radio regulation MIL-STD-188-141B, wherein the signals are I and Q signals, and generating a time sequence xk。
Step S1.2, solving xkIs defined as:
for the receiving end signal skConsider additive white Gaussian noise nkThen sk=xk+nk。
Defining a receiver-side signal skHas an autocorrelation function r of
WhereinIs the variance of Gaussian white noise, rxx(m) is xkThe autocorrelation function of. It can be seen that the autocorrelation function can effectively eliminate noise. r is xkFromA sequence of correlation functions.
S2, performing bispectrum transformation on the autocorrelation sequence, and calculating third-order cumulant;
and S21, estimating a third-order spectrum of the generated autocorrelation sequence by using a nonparametric method.
Signal high order spectrum skx(ω1,···,ωk-1) Is defined as follows:
wherein c iskx(τ1,···,τk-1) Is xkThe k-th order cumulant of (c);
defining the k-1 spectrum of the signal as the k-th order spectrum of the signal, so that the bispectrum s of the signal3x(ω1,ω2) The definition is as follows:
step S22, according to the definition formula of bispectrum, the autocorrelation sequence r is solvedssThe bispectral transformation matrix s.
The length of the FFT is set to 256, the length of the rao optimum window function is set to 5, the length of each segment is 250, the overlap length of each segment is 30, resulting in a 256 × 256 matrix s:
wherein, c3r(τ1,τ2) Is the 3 rd order cumulant of the autocorrelation function r, s (ω)1,ω2) Is according to c3r(τ1,τ2) The resulting bispectrum was calculated.
S3, sampling the obtained third-order cumulant matrix, converting the three-order cumulant matrix into a three-channel format similar to a picture, and constructing a data set;
step S31, sampling the matrix S obtained by double spectrum transformation, the original matrix SThe matrix is 256 x 256 dimension, odd number line samples are 1, 3, 5 \823030 '; 255 points, even number line samples are 2, 4, 6 \8230'; 256 points, two 128 x 128 matrixes are obtainedI、sQ。
Step S32, adopting maximum and minimum normalization to SI、sQPreprocessing line by line to make the value be in [0,1 ]]Within the scope, the specific update process is as follows:
wherein,is s isIThe (c) th row of (a),is s isQJ, i, j e [1,128 ]](ii) a Obtaining a matrix s updated after preprocessingIAnd sQ。
Step S33, adding SIAnd sQAnd (4) splicing in the third dimension to obtain a 128 multiplied by 2 three-dimensional matrix S.
And S34, performing the operation on 5000 samples one by one to construct a four-dimensional matrix Y of 5000 multiplied by 128 multiplied by 2, and making a data set.
S4, designing two paths of input convolutional neural network models, and further improving the models in modes of increasing the depth of the models, introducing a BatchNormalization layer, improving an activation function and the like, as shown in FIG. 2.
And S41, constructing two branches for the network model according to the structures of the classical convolutional neural network LeNet-5 and AlexNet, and respectively recording the two branches as a path I and a path II.
In two branches, the activation function includes relu and leakyrelu, and the expression of the relu activation function is as follows:
f(x)=max(0,x)
the expression for the Leakyrelu activation function is:
f(x)=max(0,x)+leak*min(0,x)
wherein x is the input of the activation function and f (x) is the output of the activation function; leak is a very small constant to preserve the value of the negative x-axis, typically around 0.01.
And S42, increasing the model depth of the second path by adding a full connection layer. The last fully-connected layer output dimension of both branches is 256.
And S43, overlapping the 256-dimensional vectors output by the two fully-connected layers, sequentially passing through a leakyrelu activation function, a batchnormalization layer and a dropout layer, and finally outputting a classification result through a softmax classifier.
And S5, inputting the data set into a network model to complete the tasks of training and classification.
And S51, manufacturing 5000 labels according to the data set, and performing label learning through a network. The preprocessed data set Y is divided into a training set, a verification set and a test set according to the proportion of 3. Inputting the pre-training set into a network model, and updating the weight by adopting a Momentum gradient descent algorithm (Momentum). The Momentum algorithm is specifically as follows:
W←W-ανdw
b←b-ανdb
wherein x(i)Collecting m samples { x) for a training set(1),x(2),···,x(m)I data of g is gradient, vkIs the k-th update speed, θkIs the kth iteration weight; epsilon is a learning rate, alpha is a momentum parameter, the initial learning rate epsilon is set to be 0.001, and the momentum parameter alpha is set to be 0.9; w, b are weights of neural networkWeight and offset, vdw、νdbThe relationship of the gradient between two adjacent steps is established,in order to obtain the gradient of the loss function in the weight value and the offset, α and β are momentum parameters in the algorithm, and β is usually set to 0.9, and may be set to other values;
a cross entropy loss function is used, which is expressed as follows:
where M is the number of classes, yiIs a true label of the specimen, PiIs the probability that the sample belongs to class i.
In this experiment, the batch size was set to 64 and the epoch was set to 500.
And S52, after model training is finished, saving the model weight, inputting the preprocessed test set into the trained model for testing, and outputting a classification result and the real-time classification and identification of the communication behavior by the identification accuracy.
And S53, setting different signal-to-noise ratio conditions of 0dB, 5dB, 8dB, 10dB and 15dB respectively. And (3) respectively adding corresponding noise to the 5000 samples generated in the step S1, making a data set according to the steps of the step S2 and the step S3, repeating the experiment, and verifying the reliability of the algorithm under the noise condition.
The superiority of the present invention is demonstrated by experiments 1 to 3 below.
The experiment MATLAB simulates physical layer burst waveforms and extracts characteristic making data sets. The deep learning environment is as follows: windows 10 operating system, 11th Gen Intel (R) Core (TM) I5-11260H CPU, NVIDIA GeForce RTX3050 graphics card, python 3.7, tensorFlow 2.5.0, and Keras 2.8.0.
Experiment 1: comparative experiments under different signal-to-noise ratios
The autocorrelation high-order spectral feature extraction method designed by the invention can effectively reduce mailboxes caused by noise, in order to verify the effectiveness of the method, the experiment identifies burst waveforms corresponding to five kinds of communication behaviors under the conditions of different signal-to-noise ratios, and the test results are shown in table 1. FIG. 3 is a diagram showing the relationship between five burst waveforms and the communication behavior, and FIGS. 4a to 4e are signal shaping flow charts of the short-wave burst waveforms BW0, BW1, BW2, BW3, BW4
TABLE 1 recognition results for different SNR conditions
Experimental results show that the method has a good identification effect, when the signal-to-noise ratio is 0dB, the identification rates of other communication behaviors except the flow management behavior (BW 1) are more than 80%, and the identification rates of BW0, BW3 and BW4 are more than 90%. The feature extraction method provided by the invention can effectively reduce the interference of noise and improve the identification accuracy under the condition of low signal-to-noise ratio.
Experiment 2: comparative experiments with different algorithms
TABLE 2 recognition results of different algorithms
Experimental results show that the denoising method provided by the invention effectively solves the problem of low recognition accuracy under the condition of low signal-to-noise ratio through the autocorrelation method. When the signal-to-noise ratio is 15dB, the algorithm of the invention has lower recognition rate than a bispectrum amplitude phase algorithm, because not only noise is removed during the noise removal of the autocorrelation method, but also some useful characteristic information is included, because the noise is smaller, the removed characteristic information is relatively more, the retained characteristic is relatively reduced, and the improvement brought by the noise removal cannot be matched with the loss of the characteristic. Therefore, the invention is more suitable for the environment with certain noise.
Experiment 3: comparative experiments with different network models
The invention designs a double-input convolution neural network model, which respectively extracts characteristics through two branches, outputs through a full connection layer, then superposes the characteristics and further processes the characteristics. In order to verify the improvement of the model on the recognition rate, four network models are selected for comparison in the experiment, and the results are shown in table 3 and fig. 5.
TABLE 3 recognition results of different network models
The experimental results of table 3 and fig. 5 prove that the dual-input network model used in the invention can further extract features and improve the recognition rate.
In conclusion, the invention is optimized in the aspects of the feature extraction method and the neural network model, can further improve the recognition effect under the condition of low signal-to-noise ratio, and has practicability.
Claims (7)
1. A short wave communication behavior identification method based on self-correlation high-order spectrum features is characterized by comprising the following steps:
s1, solving an autocorrelation function of a time domain signal to obtain an autocorrelation sequence;
s2, performing bispectrum transformation on the autocorrelation sequence, and calculating third-order cumulant;
s3, sampling the obtained third-order cumulant matrix, converting the three-order cumulant matrix into a three-channel format similar to a picture, and constructing a data set;
s4, designing two paths of input convolutional neural network models, and further improving the models by increasing the depth of the models, introducing a Batch Normalization layer and improving an activation function to obtain a final network model;
and S5, inputting the data set into a final network model to complete the tasks of training and classification.
2. The short communication behavior identification method based on the autocorrelation high-order spectral features as claimed in claim 1, wherein the specific steps of S1 are:
step S11, five burst waveforms BW0, BW1, BW2, BW3 and BW4 specified by the American military short wave radio procedure MIL-STD-188-141B,using MATLAB simulation to generate 1000 signal samples of five burst waveforms, wherein the signals are I and Q signals, and generating a time sequence xk;
Step S12, finding xkIs defined as:
for the receiving end signal skConsider additive white Gaussian noise nkThen sk=xk+nk;
Defining a receiver-side signal skThe autocorrelation function r of (a) is:
3. The short-wave communication behavior identification method based on the autocorrelation high-order spectrum features as claimed in claim 2, wherein the specific steps of S2 are as follows:
s21, estimating a third-order spectrum by using a nonparametric method for the generated autocorrelation sequence;
signal high order spectrum skx(ω1,···,ωk-1) Is defined as follows:
wherein c iskx(τ1,···,τk-1) Is xkThe k-th order cumulant of (c);
define the k-1 spectrum of the signal as the k-th order spectrum of the signal, soBispectrum s of signals3x(ω1,ω2) The definition is as follows:
c3x(τ1,τ2) Is xkThe 3 rd order cumulant of (c);
step S22, according to the definition formula of bispectrum, the autocorrelation sequence r is solvedssThe bispectrum transformation matrix s;
the length of the FFT is set to 256, the length of the rao optimum window function is set to 5, the length of each segment is 250, the overlap length of each segment is 30, resulting in a 256 × 256 matrix s:
wherein, c3r(τ1,τ2) Is the 3 rd order cumulant of the autocorrelation function r, s (ω)1,ω2) Is according to c3r(τ1,τ2) The resulting bispectrum is calculated.
4. The short-wave communication behavior identification method based on the autocorrelation high-order spectrum features as claimed in claim 3, wherein the specific steps of S3 are as follows:
step S31, sampling a matrix S obtained by double-spectrum transformation, wherein the original matrix is 256 multiplied by 256 dimensions, odd-numbered lines are sampled at 1, 3 and 5 '\ 8230255 points, even-numbered lines are sampled at 2, 4 and 6' \ 8230256 points, and two 128 multiplied by 128 matrices S are obtainedI、sQ;
Step S32, adopting maximum and minimum normalization to SI、sQPreprocessing line by line to make the value be in [0,1 ]]Within the scope, the specific update process is as follows:
wherein,is as sIThe number of the ith row of (a),is s isQJ, i, j e [1,128 ]](ii) a Obtaining a matrix s updated after preprocessingIAnd sQ;
Step S33, addingIAnd sQSplicing in the third dimension to obtain a 128 multiplied by 2 three-dimensional matrix S;
and step S34, performing the operations on 5000 samples one by one to construct a four-dimensional matrix Y of 5000 multiplied by 128 multiplied by 2, and manufacturing a data set.
5. The short-wave communication behavior identification method based on the autocorrelation high-order spectrum features as claimed in claim 4, wherein the specific steps of S4 are as follows:
s41, according to the structures of a classical convolutional neural network LeNet-5 and AlexNet, constructing two branches for a network model, and respectively recording the two branches as a path I and a path II;
in two branches, the activation function includes relu and leakyrelu, and the expression of the relu activation function is:
f(x)=max(0,x)
the expression for the Leakyrelu activation function is:
f(x)=max(0,x)+leak*min(0,x)
wherein x is the input of the activation function and f (x) is the output of the activation function; leak is a constant;
s42, increasing the model depth of the second path by adding a full connection layer; the last fully-connected layer output dimension of the two branches is 256;
and S43, overlapping the 256-dimensional vectors output by the two full-connection layers, sequentially passing through a leakyrelu activation function, a batchnormalization layer and a dropout layer, and finally outputting a classification result through a softmax classifier.
6. The short-wave communication behavior identification method based on the autocorrelation high-order spectrum features as claimed in claim 5, wherein the specific steps of S5 are as follows:
s51, manufacturing 5000 labels according to the data set, and performing label learning through a network; dividing the preprocessed data set Y into a training set, a verification set and a test set according to the proportion of 3;
inputting the pre-training set into a network model, and updating the weight by adopting a Momentum gradient descent algorithm, namely a Momentum algorithm, wherein the updating process of the Momentum algorithm parameters is as follows:
W←W-ανdw
b←b-ανdb
wherein x: (i) Collecting m samples { x) for a training set(1),x(2),···,x(m)I data of g is gradient, vkIs the k-th update speed, θkIs the k iteration weight; epsilon is a learning rate, alpha is a momentum parameter, the initial learning rate epsilon is set to be 0.001, and the momentum parameter alpha is set to be 0.9; w and b are weight values and offset of the neural network, vdw、νdbThe relationship of the gradient between two adjacent steps is established,the gradient of the loss function in the weight value and the offset is shown, and alpha and beta are momentum parameters in the algorithm;
using a cross entropy loss function, the expression is as follows:
where M is the number of classes, yiIs a true label of the specimen, PiIs the probability that the sample belongs to the i-th class;
the sample size selected in one training is set to be 64, and the sample training time epoch is set to be 500;
and S52, after model training is completed, saving the model weight, inputting the preprocessed test set into the trained model for testing, and outputting a classification result and the real-time classification and identification of the communication behavior by the identification accuracy.
7. The short-wave communication behavior identification method based on the autocorrelation high-order spectral features as claimed in claim 6, further comprising, after step S52:
s53, setting different signal-to-noise ratio conditions of 0dB, 5dB, 8dB, 10dB and 15dB respectively; and (3) adding corresponding noise to the 5000 samples generated in the step S1, manufacturing a data set according to the steps of the step S2 and the step S3, repeating the experiment, and verifying the reliability of the algorithm under the noise condition.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210867401.1A CN115270866A (en) | 2022-07-22 | 2022-07-22 | Short-wave communication behavior identification method based on self-correlation high-order spectrum features |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210867401.1A CN115270866A (en) | 2022-07-22 | 2022-07-22 | Short-wave communication behavior identification method based on self-correlation high-order spectrum features |
Publications (1)
Publication Number | Publication Date |
---|---|
CN115270866A true CN115270866A (en) | 2022-11-01 |
Family
ID=83766762
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210867401.1A Pending CN115270866A (en) | 2022-07-22 | 2022-07-22 | Short-wave communication behavior identification method based on self-correlation high-order spectrum features |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115270866A (en) |
-
2022
- 2022-07-22 CN CN202210867401.1A patent/CN115270866A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107220606B (en) | Radar radiation source signal identification method based on one-dimensional convolutional neural network | |
CN109890043B (en) | Wireless signal noise reduction method based on generative countermeasure network | |
CN108509911B (en) | Interference signal identification method based on convolutional neural network | |
CN109471074B (en) | Radar radiation source identification method based on singular value decomposition and one-dimensional CNN network | |
CN112818891B (en) | Intelligent identification method for communication interference signal type | |
CN109711314B (en) | Radar radiation source signal classification method based on feature fusion and SAE | |
CN111428817A (en) | Defense method for resisting attack by radio signal identification | |
CN112115821B (en) | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy | |
CN114636975A (en) | LPI radar signal identification method based on spectrogram fusion and attention mechanism | |
CN114726692B (en) | SERESESESENet-LSTM-based radiation source modulation mode identification method | |
Jeong et al. | Spectrogram-based automatic modulation recognition using convolutional neural network | |
CN113673312B (en) | Deep learning-based radar signal intra-pulse modulation identification method | |
CN111310700A (en) | Intermediate frequency sampling sequence processing method for radiation source fingerprint feature identification | |
CN111222442A (en) | Electromagnetic signal classification method and device | |
CN114239657A (en) | Time sequence signal identification method based on complex value interference neural network | |
Chen et al. | Automatic modulation classification of radar signals utilizing X-net | |
CN114943245A (en) | Automatic modulation recognition method and device based on data enhancement and feature embedding | |
CN112801003A (en) | Unmanned aerial vehicle radiation source modulation pattern recognition method and device | |
CN114218984A (en) | Radio frequency fingerprint identification method based on sample multi-view learning | |
Kong et al. | A transformer-based CTDNN structure for automatic modulation recognition | |
CN114520758A (en) | Signal modulation identification method based on instantaneous characteristics | |
Feng et al. | FCGCN: Feature Correlation Graph Convolution Network for Few-Shot Individual Identification | |
CN115270866A (en) | Short-wave communication behavior identification method based on self-correlation high-order spectrum features | |
CN115267713B (en) | Intermittent sampling interference identification and suppression method based on semantic segmentation | |
CN114598577B (en) | Multi-band signal fusion filtering method for 5G communication system |
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 |