CN116662783A - Deep learning-based individual identification method for radar radiation source of cross-over receiver - Google Patents
Deep learning-based individual identification method for radar radiation source of cross-over receiver Download PDFInfo
- Publication number
- CN116662783A CN116662783A CN202310655917.4A CN202310655917A CN116662783A CN 116662783 A CN116662783 A CN 116662783A CN 202310655917 A CN202310655917 A CN 202310655917A CN 116662783 A CN116662783 A CN 116662783A
- Authority
- CN
- China
- Prior art keywords
- domain
- data
- network
- receiver
- identification
- 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
- 230000005855 radiation Effects 0.000 title claims abstract description 66
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000013135 deep learning Methods 0.000 title claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 6
- 230000006870 function Effects 0.000 claims description 30
- 230000004913 activation Effects 0.000 claims description 12
- 238000011176 pooling Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000005070 sampling Methods 0.000 claims description 8
- 238000005457 optimization Methods 0.000 claims description 7
- 238000012360 testing method Methods 0.000 claims description 7
- 230000009466 transformation Effects 0.000 claims description 5
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 241000170489 Upis Species 0.000 claims description 2
- 230000010354 integration Effects 0.000 claims description 2
- 230000009467 reduction Effects 0.000 claims description 2
- 238000012935 Averaging Methods 0.000 claims 1
- 238000004088 simulation Methods 0.000 description 9
- 238000005516 engineering process Methods 0.000 description 8
- 230000000694 effects Effects 0.000 description 7
- 238000001228 spectrum Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000006835 compression Effects 0.000 description 2
- 238000007906 compression Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 230000003313 weakening effect Effects 0.000 description 1
Classifications
-
- 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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Biology (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Computer Networks & Wireless Communication (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a deep learning-based radar radiation source individual identification method, which mainly solves the problem that in the prior art, when radar radiation source individual identification is carried out across receivers, the identification accuracy is reduced due to receiver distortion pollution. The implementation scheme is as follows: processing and dividing data from different receivers; constructing a radar radiation source individual identification model based on deep learning of a double branch consisting of a feature extractor and two identification classifiers; performing iterative training on the model by adopting a back propagation method, and enabling the feature extractor to complete difference compensation of the same radiation source signal under different receiving environments through the maximized blurring of the domain label; individual identification of radar radiation source signals is performed using the trained model. The method and the device enhance the robustness to different non-ideal pollution factors, improve the individual identification accuracy of the radar radiation source under the application scene of the bridging receiver, and can be used for radar individual information reconnaissance of the bridging receiver under a complex electromagnetic environment.
Description
Technical Field
The invention belongs to the technical field of radar signal processing, and particularly relates to a radar radiation source identification method which can be applied to radar individual information reconnaissance of a bridging receiver in a complex electromagnetic environment.
Background
The radar radiation source is an electromagnetic wave emitting device for generating a detectable electromagnetic wave signal within a range so that a radar receiver can detect a target. Radar radiation sources include components such as antennas, radio frequency transmitters, modulators, etc., that typically determine target location and characteristics from transmitted electromagnetic waves and signals reflected back therefrom. Common radar radiation sources include civilian radar, military radar, which are widely used for communication, navigation, monitoring and reconnaissance.
Individual identification of radar radiation sources is a technique to match a signal to a unique transmitter based on unintentional modulation characteristics of the radiation source attached to the signal. Since the unintentional modulation signature originates from the nature of the hardware circuitry of the radiation source, it is a unique signature of each transmitter and is therefore also referred to as "fingerprint" identification. The method can acquire the most visual radar distribution condition and action strategy of radar carrying equipment by identifying radar radiation source individuals in the space, and is the basis of subsequent situation analysis and decision making.
However, with the increasing maturity of electronic information technology, the appearance of new system radars is popularized in a large scale, and the construction of a radar radiation source identification feature library is very difficult due to new waveform application and improvement of electronic device performance. Meanwhile, the coverage frequency range of radar signals is widened, the working parameter domains of different radar individuals in the same batch are highly overlapped, and the different signal forms of the multifunctional radar are agile, so that the traditional sorting and identification technology based on pulse description word parameters such as carrier frequency, pulse width, amplitude, pulse repetition frequency and the like is difficult to adapt to the current severe complex electromagnetic spectrum environment, and the identification of radar individuals depends on the radiation source individual identification technology for identifying by using the fingerprint characteristics of a transmitter. In the current application scenario of individual identification of radar radiation sources, since the known signal and the signal to be identified come from different receivers, different receivers contained in the signal are distorted, so that characteristic drift and model mismatch can be caused, and the identification accuracy is further reduced.
In order to solve the above problems, an improvement is needed to be made to the design thought of the individual identification algorithm of the radar radiation source, and the robustness to the distortion of different receivers is improved so as to improve the universality under different receiving scenes.
The patent document with the application number of CN202010494523.1 discloses a radar radiation source individual identification method under the condition of differential signal-to-noise ratio, which effectively extracts the fingerprint characteristics of radiation sources with different dimensions by splicing a plurality of characteristics such as rising edge time, pulse width, bispectral waveform and the like, and has higher identification precision under the simulation environment of differential signal-to-noise ratio. However, the method only stacks the feature quantity to improve the recognition accuracy in the noise difference environment, does not aim at designing a related algorithm for weakening the pollution difference, and does not consider pollution with higher coupling degree such as receiver distortion, so that the individual recognition accuracy in the cross-over receiver application scene is low.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a cross receiver radar radiation source individual identification method based on deep learning, so as to improve the identification precision of the cross receiver on radar radiation source individuals in application occasions.
In order to achieve the above purpose, the technical scheme adopted by the invention comprises the following steps:
(1) Acquiring pulse signal data from different receivers, defining a domain label of each pulse signal, and defining source domain data and target domain data; dividing more than half of the source domain data and the target domain data into a training data set Q and the rest of test data sets T respectively;
(2) Performing bispectrum transformation on each pulse signal to obtain a dimension-reduced one-dimensional characteristic domain SIB;
(3) Constructing a radar radiation source individual identification model O based on deep learning:
(3a) Establishing a feature extractor G formed by sequentially cascading a first-stage one-dimensional convolution layer, a maximum pooling layer 1, a second-stage one-dimensional convolution layer, a maximum pooling layer 2 and a flattening layer f ;
(3b) Establishing a single identification classifier G formed by sequentially cascading a first-stage full-connection layer, a second-stage full-connection layer and a Softmax classifier y ;
(3c) Establishing a domain identification classifier G formed by sequentially cascading a gradient inversion layer R, a 1 st-level full-connection layer, a 2 nd-level full-connection layer and a Softmax classifier d The gradient inversion layer R changes the optimization direction of domain classification so as to blur the difference between different domains and inhibit the characteristic drift pollution caused by different receivers;
(3d) Will G f And G y Cascading to form individual identification branch, and G f And G d Cascading forms a domain identification branch;
(4) Iterative training is carried out on the network identification model O to obtain an optimized model O * :
(4a) Initializing iteration number I, setting the maximum iteration number as I, wherein I is more than or equal to 50, setting the loss function of network optimization as a cross entropy function, and the learning rate as mu p Initializing network parameters of O as theta;
(4b) Randomly selecting a batch of data from source domain training data, inputting the data into a network model O, and using the network prediction probability that the nth sample of the source domain of the individual identification branch belongs to the c radiation source individualObtaining individual classification predictive tag->Network prediction probability +.A.n. sample derived from the kth receiver using domain identification tributary source domain>Obtaining the domain prediction tag->Calculating individual classification prediction tags respectively>And the actual classification label Lf 1 Loss L of (2) 1 Domain prediction tag->And actual domain label Ld 1 Loss L of (2) 2 ;
(4c) Randomly selecting a batch of data from the training data of the target domain, inputting the data into a network model O, and calculating the network prediction probability of the first sample of the target domain from the kth receiver by using a domain identification branchObtaining the domain prediction tag->Calculating the domain prediction tag +.>And actual domain label Ld 2 Loss L of (2) 3 ;
(4d) Summing all losses gives the total loss of the network l=l 1 +L 2 +L 3 And carrying out back propagation according to the method, optimizing network parameters by using a gradient descent method, and completing one iteration;
(4e) Repeating the steps (4 b) - (4 d) until I is more than or equal to I, stopping iteration, and obtaining an optimized network model O * ;
(5) Inputting the test sample T into the trained network model O * Calculating by using individual identification branch to obtain identification result L p 。
Compared with the prior art, the invention has the following advantages:
firstly, the invention adopts the bispectrum surrounding line integral as the characteristic domain, so that the characteristic dimension can be reduced under the condition of not losing the characteristic information as much as possible, and compared with the existing technology for identifying through the two-dimensional bispectrum domain, the calculation amount of the device is smaller;
secondly, the domain identification classification branch comprising the gradient inversion layer is introduced to compensate individual fingerprint characteristic drift of the radar radiation source caused by different pollution factors, so that the method has a good inhibition effect on characteristic drift generated by distortion of different receivers and characteristic measurement errors caused by noise;
third, the invention builds the radar radiation source individual identification model O based on deep learning, and compared with the existing support vector machine technology, the deep learning technology adopted by the invention has high dimension feature processing capability and more excellent deep feature extraction capability, thereby improving the identification accuracy.
Drawings
FIG. 1 is a general flow chart of an implementation of the present invention;
FIG. 2 is a scenario diagram of an embodiment of the present invention;
FIG. 3 is a diagram of a network model structure in the present invention;
FIG. 4 is a sub-flowchart of training a network model in accordance with the present invention;
FIG. 5 is a graph showing the contrast of the characteristic fields of the signals of three radiation sources at 40dB passing through two receivers respectively;
FIG. 6 is a graph showing the contrast of the characteristic fields of signals from three radiation sources at 20dB through two receivers, respectively, according to the present invention;
FIG. 7 is a graph comparing classification accuracy of the present invention with prior art embodiments.
Detailed Description
Embodiments and effects of the present invention are described in further detail below with reference to the accompanying drawings.
Referring to fig. 2, the usage scenario of this example includes three radar radiation sources and two signal receivers, the three radar radiation sources transmit signals and are respectively received by two different receivers, wherein the signals of the receiver 1 have classification identification tags, the hardware parameters of the three radar radiation sources are different, the transmitted signal waveforms and pulse parameters are the same, the mixing parameters of the two receivers are the same, and the hardware parameters are different.
The invention recognizes in this context part of the signal of the receiver 1 and all signals of the receiver 2.
Referring to fig. 1, the implementation steps of this example are as follows:
step 1, acquiring pulse signal data of two receivers, and processing and dividing the pulse signal data.
(1.1) defining the domain tag of the pulse according to the source of the pulse signal, the pulse signal domain tag Ld received by the receiver 1 1 1, the pulse signal domain label Ld received by the receiver 2 2 Is 2;
(1.2) defining the signal received by the receiver 1 as source domain data and the signal received by the receiver 2 as target domain data;
(1.3) dividing more than half of the source domain data and the target domain data into training data sets respectively, and the rest is test data sets.
And step 2, performing double-spectrum transformation on each pulse signal to obtain a dimension-reduced one-dimensional characteristic domain SIB.
(2.1) for a sampling frequency f s The pulse signal of (2) is subjected to zero mean normalization processing to obtain a processed pulse sampling sequence of x (0), x (1), x (N-1), and dividing the pulse sampling sequence into K sections, wherein each section contains M sampling points, and the K section data is x (k) (0),x (k) (1),...x (k) (M-1);
(2.2) calculating discrete fourier transform, DFT, coefficients for each piece of data:
wherein λ is the argument of the fourier transform;
(2.3) calculating triple correlation of DFT coefficient to obtain double spectrum estimated value of the data
Delta in 0 =f s /N 0 ,L 1 Total sliding window length for correlation operation, N 0 Is the normalization factor of the correlation operation, N 0 And L 1 Satisfy m= (2L 1 +1)N 0 ,λ 1 、λ 2 Sliding independent variables of two different dimensions are calculated for correlation;
(2.4) taking the average of the bispectral estimates of all the K-segment signals to obtain the bispectral estimate of the whole-segment signal
In the middle ofIs two frequency dimensions of the bispectrum;
(2.5) for bispectrum estimatesPerforming line-surrounding integration to obtain a bispectrum feature domain SIB (omega) after dimension reduction
And 3, constructing a radar radiation source individual identification model O based on deep learning.
Referring to fig. 3, this step is specifically implemented as follows:
(3.1) establishing a feature extractor G formed by sequentially cascading a first-stage one-dimensional convolution layer, a first maximum pooling layer, a second-stage one-dimensional convolution layer, a second maximum pooling layer and a flattening layer f Wherein:
the number of channels of the first convolution layer is 16, the convolution kernel size is 1x32, and the activation function is a ReLU function;
the number of channels of the second convolution layer is 16, the convolution kernel size is 1x25, and the activation function is a ReLU function;
the size of the pooling cores of the two largest pooling layers is 1x2;
the node number of the flattening layer is 192;
(3.2) establishing a single identification classifier G formed by sequentially cascading a first-stage full-connection layer, a second-stage full-connection layer and a Softmax classifier y Wherein:
the node number of the first-stage full-connection layer is 192, and the activation function is a Sigmoid function;
the node number of the second-stage full-connection layer is C, the activation function is a Sigmoid function, and the node number C is the individual number of radar radiation sources contained in a data set;
(3.3) establishing a domain identification classifier G formed by sequentially cascading a gradient inversion layer R, a 1 st stage full-connection layer, a 2 nd stage full-connection layer and a Softmax classifier d Wherein:
the node number of the 1 st-level full-connection layer is 192, and the activation function is a Sigmoid function;
the node number of the 2 nd-level full-connection layer is N, the activation function is a Sigmoid function, wherein N is the number of receivers for receiving data;
the gradient inversion layer R, when the network forward calculates, its output is equal to the input x, i.e., R (x) =x; in the network back propagation optimization, R inverts the input loss gradient, i.eWherein p is the ratio of the current iteration number I to the total iteration number I, and represents the iteration process of the network, loss is network Loss in the process of counter propagation, and gamma=10 is a super parameter;
(3.4) G f And G y Cascading to form individual identification branch, and G f And G d The concatenation constitutes a domain identification branch.
Step 4, performing iterative training on the network identification model O to obtain an optimized modelO * 。
(4.1) initializing iteration number i=1, setting the maximum iteration number as i=50, setting the loss function of network optimization as a cross entropy function, and learning rate as mu p Initializing network parameters of O, wherein learning rate mu p The calculation formula of (2) is as follows:
mu in the middle p Initial value mu 0 =0.01, the multiplication super-parameter α=10, β=0.75 is an exponential super-parameter;
(4.2) inputting training set data into the network, and updating network parameters in a back propagation mode:
referring to fig. 4, the implementation of this step is as follows:
(4.2.1) randomly selecting 32 pulse characteristic domain data from the source domain training data, inputting the 32 pulse characteristic domain data into the network model O, and calculating the network prediction probability of the nth sample of the batch of source domain data belonging to the c radiation source individual by using an individual identification branchThe specific calculation steps are as follows:
first, the SIB passing parameter of the input feature field is theta 1 Feature extractor G of f The output feature vector f:
f=G f (SIB;θ 1 );
then, the characteristic vector f is input into the parameter theta 2 Individual identification classifier G of (2) y Performing classification prediction to obtain a classification prediction vector with a length of C
Wherein C is the number of radar radiation sources,representing the size of the likelihood that the nth source domain input sample belongs to each individual radar radiation source;
finally, set upIs z at the c-th point c Prediction vector +.>Calculating the probability of the nth source domain input sample belonging to each class label>
p n c I.e. the network prediction probability of the nth sample of the source domain belonging to the c-th label, F is the total number of labels, wherein(4.2.2) calculating the individual classification prediction result and the true classification label Lf 1 Loss L of (2) 1 :
Wherein U is the number of radar radiation sources, N is the total number of samples,network prediction probability for the nth sample of the source domain belonging to the c-th tag, +.>True probability of whether the nth sample of the source domain belongs to the c label, belonging to the cRadiation source->1, not belonging to the c-th radiation source->Is 0;
(4.2.3) calculating a network prediction probability that an mth sample of the batch source domain data originates from a kth receiver using a domain identification branchThe calculation mode is the same as (4.2.1);
(4.2.4) calculating the domain classification prediction result and the real domain classification label Ld 1 Loss L of (2) 2 ;
Where H is the number of individual receivers, M is the total number of samples,network prediction probability derived from kth receiver for nth sample of source domain, +.>True probability of whether the nth sample of the source domain belongs to the kth receiver or not, belonging to the kth receiver1, not belonging to the kth receiver +.>Is 0;
(4.2.5) randomly selecting 32 pulse characteristic domain data from the target domain training data, inputting the 32 pulse characteristic domain data into the network model O, and calculating the network prediction probability of the first sample of the target domain data from the kth receiver by using a domain identification branchThe calculation mode is the same as (4.2.1);
(4.2.6) calculating the loss L of the domain classification prediction result and the real domain classification label 3 :
Where H is the number of individual receivers, L is the total number of samples,network prediction probability derived from kth receiver for the first sample of the target domain, +.>For the true probability of whether the ith sample belongs to the kth receiver, belonging to the kth receiver +.>1, not belonging to the kth receiver +.>Is 0;
(4.3) summing all losses to give the total loss of the network l=l 1 +L 2 +L 3 And carrying out back propagation according to the method, optimizing network parameters by using a gradient descent method, and completing one iteration;
(4.4) repeating the steps (4.2) - (4.3) until the I is more than or equal to I, stopping iteration, and obtaining the optimized network model O * 。
Step 5, inputting the test sample T into the trained network model O * Obtaining a recognition result L through calculation of an individual recognition branch p And (5) completing identification of the individual cross-receiver radar radiation source.
The technical effects of the invention are further described below in conjunction with simulation experiments:
1. experimental conditions:
hardware environment: CPU is Inter (R) core (TM) i7-10700F, main frequency is 2.90GHz, memory is 16.0GB,64 bit operating system; the display card is an English-to-Chinese RTX2060 display card; software environment: microsoft windows 10 professional edition, python3.8, systemVue10.0.
Setting 3 radar radiation sources and 2 signal receivers, wherein the signals emitted by the radar radiation sources are single-frequency pulses with the same amplitude, the pulse width of 2 mu s and the carrier frequency of 250MHz, the hardware performance parameters of different radar radiation sources and the hardware performance parameters of different receivers are different, the mixing parameters of the devices are shown in a table 1, the hardware performance parameters of the transmitters are shown in a table 2, and the hardware performance parameters of the receivers are shown in a table 3:
table 1 mixing parameter settings
Transmitter intermediate frequency f EI | 50MHz |
Radio frequency f R | 200MHz |
Intermediate frequency f of receiver RI | 120MHz |
Receiver bandwidth | 50MHz |
Table 2 hardware performance parameter settings for transmitters
Table 3 hardware performance parameter settings for receiver
Receiver 1 | Receiver 2 | |
Three-order cut-off point/dBm of amplifier 1 | 45 | 35 |
1dB compression point/dBm of amplifier 1 | 30 | 20 |
Mixer second order cut-off point/dBm | 40 | 30 |
Mixer third order cut-off point/dBm | 30 | 20 |
Three-order cut-off point/dBm of amplifier 2 | 45 | 35 |
1dB compression Point of Amplifier 2/dBm | 30 | 20 |
Referring to the data set dividing method studied in the past, 200 samples are generated for each radiation source, 120 samples are used as training data, 80 samples are used as test data, two receivers are arranged, the data of the receiver 1 is used as source domain data, and the data of the receiver 2 is used as target domain data.
The recognition effect is evaluated by the recognition accuracy, and the calculation formula of the recognition accuracy is as follows:
second, simulation content
Simulation 1: under the condition that the signal-to-noise ratio is 40dB, signals generated by radar radiation sources 1, 2 and 3 arranged in the table 2 are respectively used for obtaining received signals through a receiver 1 and a receiver 2, the method is used for carrying out double-spectrum line-surrounding integral transformation on all the received signals to obtain a pulse characteristic domain, the characteristic difference among different transmitters and the influence of the receiver on the characteristics of the radiation sources are simulated, the result is shown in a figure 5, wherein the figure 5 (a) sequentially shows that the signals generated by the radar radiation sources 1, 2 and 3 pass through the characteristic domain of the receiver 1 from left to right; fig. 5 (b) shows, from left to right, the characteristics of the signals generated by the radar radiation sources 1, 2, 3 passing through the receiver 2.
Simulation 2: under the condition that the signal-to-noise ratio is 20dB, signals generated by using the radar radiation sources 1, 2 and 3 arranged in the table 2 are respectively transmitted to the receiver 1 and the receiver 2 to obtain received signals, the dual-spectrum line-surrounding integral transformation is carried out on all the received signals to obtain pulse characteristic fields, the influence of noise and the receiver on the characteristics of the radiation sources is simulated, the result is shown in fig. 6, fig. 6 (a) shows the characteristic fields of the signals generated by the radar radiation sources 1, 2 and 3 transmitted through the receiver 1, and fig. 6 (b) shows the characteristic fields of the signals generated by the radar radiation sources 1, 2 and 3 transmitted through the receiver 2.
Simulation 3: under the conditions that the signal to noise ratio is 5dB,10dB,20dB,30dB and 40dB respectively, signals generated by using the radar radiation sources 1, 2 and 3 arranged in the table 2 are respectively received through the receiver 1 and the receiver 2, the signals are respectively subjected to simulation recognition by using the method and the existing radar radiation source individual recognition technology based on the one-dimensional convolutional neural network, the recognition accuracy of the two technologies under different signal to noise ratios is obtained, the result is shown in fig. 7, 1D-DANN in the diagram is the method, and 1D-CNN is the prior art.
3. Simulation result analysis:
as can be seen from fig. 4 and 5, in the case of higher signal-to-noise ratio, the same radiation source signal, after passing through different receivers, has different degrees of distortion in the characteristic field, and as the performance of the receiver hardware decreases, the degree of distortion increases. When noise is present and the noise power is greater than the unintentional contaminated power generated by the receiver, the noise becomes a major factor in contaminating the feature field.
As can be seen from fig. 6, when the signal-to-noise ratio is higher than 20dB, the recognition accuracy of the prior art is between 80% and 90%, and the recognition accuracy of the present invention is higher than 95%; at a signal-to-noise ratio of 5dB, the recognition accuracy of the prior art is reduced to 50%, while the recognition accuracy of the invention is still 79%. Simulation results show that: under different signal-to-noise ratio conditions, the identification effect of the invention is higher than that of the existing algorithm by more than 8%, and the invention has better optimization effect on hardware characteristics and characteristic domain drift caused by noise, and has better identification effect on radar radiation source signals of a bridging receiver.
Claims (11)
1. A cross receiver radar radiation source individual identification method based on deep learning is characterized by comprising the following steps:
(1) Acquiring pulse signal data from different receivers, defining a domain label of each pulse signal, and defining source domain data and target domain data; dividing more than half of the source domain data and the target domain data into a training data set Q and the rest of test data sets T respectively;
(2) Performing bispectrum transformation on each pulse signal to obtain a dimension-reduced one-dimensional characteristic domain SIB;
(3) Constructing a radar radiation source individual identification model O based on deep learning:
(3a) Establishing a feature extractor G formed by sequentially cascading a first-stage one-dimensional convolution layer, a maximum pooling layer 1, a second-stage one-dimensional convolution layer, a maximum pooling layer 2 and a flattening layer f ;
(3b) Establishing a single identification classifier G formed by sequentially cascading a first-stage full-connection layer, a second-stage full-connection layer and a Softmax classifier y ;
(3c) Establishing a domain identification classifier G formed by sequentially cascading a gradient inversion layer R, a 1 st-level full-connection layer, a 2 nd-level full-connection layer and a Softmax classifier d ;
(3d) Will G f And G y Cascading to form individual identification branch, and G f And G d Cascading forms a domain identification branch;
(4) Iterative training is carried out on the network identification model O to obtain an optimized model O * :
(4a) Initializing iteration number I, setting the maximum iteration number as I, wherein I is more than or equal to 50, setting the loss function of network optimization as a cross entropy function, and the learning rate as mu p Initializing network parameters of O as theta;
(4b) Randomly selecting a batch of data from source domain training data, inputting the data into a network model O, and using the network prediction probability that the nth sample of the source domain of the individual identification branch belongs to the c radiation source individualObtaining individual classification prediction labels Lf 1 * Network prediction probability of being derived from kth receiver using domain identification tributary source domain nth sample +.>Obtaining the Domain prediction tag->Calculating individual classification prediction tags Lf 1 * And the actual onesInter-classification label Lf 1 Loss L of (2) 1 Domain prediction tag->And actual domain label Ld 1 Loss L of (2) 2 ;
(4c) Randomly selecting a batch of data from the training data of the target domain, inputting the data into a network model O, and calculating the network prediction probability of the first sample of the target domain from the kth receiver by using a domain identification branchObtaining the Domain prediction tag->Calculating the domain prediction tag +.>And actual domain label Ld 2 Loss L of (2) 3 ;
(4d) Summing all losses gives the total loss of the network l=l 1 +L 2 +L 3 And carrying out back propagation according to the method, optimizing network parameters by using a gradient descent method, and completing one iteration;
(4e) Repeating the steps (4 b) - (4 d) until I is more than or equal to I, stopping iteration, and obtaining an optimized network model O * ;
(5) Inputting the test sample T into the trained network model O * Calculating by using individual identification branch to obtain identification result L p 。
2. The method of claim 1, wherein the defining of the domain label for each pulse signal in (1) and defining the source domain data and the destination domain data is accomplished by:
(1a) Defining the domain label of the pulse according to the source of the pulse signal, namely, endowing the pulse signal received by the same receiver with the same domain label and endowing signals from different receivers with different domain labels;
(1b) Defining source domain data and target domain data according to the presence or absence of the individual identification tag of the pulse signal:
if the pulse has an individual identification tag, defining the pulse as source domain data;
and otherwise, defining the target domain data to be identified.
3. The method of claim 1, wherein the performing a bispectrum transform on each pulse signal received by a different receiver in (2) is implemented as follows:
(2a) For a sampling frequency f s The pulse signal of (2) is subjected to zero mean normalization processing to obtain a processed pulse sampling sequence of x (0), x (1), x (N-1), and dividing the pulse sampling sequence into K sections, wherein each section contains M sampling points, and the K section data is x (k) (0),x (k) (1),...x (k) (M-1);
(2b) Calculating Discrete Fourier Transform (DFT) coefficients of each piece of data:
(2c) Calculating triple correlation of DFT coefficients to obtain a bispectrum estimated value of the segment of data
Delta in 0 =f s /N 0 ,N 0 And L 1 Satisfy m= (2L 1 +1)N 0 ,λ 1 、λ 2 Sliding independent variables of two different dimensions are calculated for correlation;
(2d) Averaging the bispectral estimates of all K-segment signals to obtain bispectral estimates of the whole-segment signals
In the middle of
(2e) For bispectrum estimationPerforming line-surrounding integration to obtain a bispectrum feature domain SIB (omega) after dimension reduction
4. The method of claim 1, wherein the feature extractor G constructed in step (3 a) f The parameters of each layer are as follows:
the number of channels of the first convolution layer is 16, the convolution kernel size is 1x32, and the activation function is a ReLU function;
the number of channels of the second convolution layer is 16, the convolution kernel size is 1x25, and the activation function is a ReLU function;
the size of the pooling cores of the two largest pooling layers is 1x2;
the number of nodes in the flattened layer is 192.
5. The method of claim 1, wherein the individual recognition classifier G constructed in step (3 b) y The parameters of each layer are as follows:
the node number of the first-stage full-connection layer is 192, and the activation function is a Sigmoid function;
the number of nodes of the second-stage full-connection layer is C, the activation function is a Sigmoid function, and the number of nodes C is the number of radar radiation source individuals contained in the data set.
6. The method of claim 1, wherein the domain identification classifier G constructed in step (3 c) d The structure and parameters of each layer are as follows:
the node number of the 1 st-level full-connection layer is 192, and the activation function is a Sigmoid function;
the node number of the 2 nd-level full-connection layer is N, the activation function is a Sigmoid function, wherein N is the number of receivers for receiving data;
the gradient inversion layer R has an output equal to the input x, i.e. R (x) =x,
in the network back propagation optimization, R inverts the input loss gradient:
where p is the ratio of the current iteration number I to the total iteration number I, and represents the iteration process of the network, loss is network Loss during back propagation, and γ is a constant 10.
7. The method according to claim 1, wherein the learning rate μ in step (4 a) p The calculation formula is as follows:
mu in the middle 0 Mu is p Alpha is a multiplication super parameter and beta is an index super parameter.
8. The method of claim 1, wherein the individual classification prediction tags Lf are calculated in step (4 b) 1 * And the actual classification label Lf 1 Loss L of (2) 1 The formula is as follows:
wherein U is the number of radar radiation sources, N is the total number of samples,network prediction probability for the nth sample of the source domain belonging to the c-th tag, +.>True probability of whether the nth sample of the source domain belongs to the c-th tag, belonging to the c-th radiation source +.>1, not belonging to the c-th radiation source->Is 0.
9. The method of claim 1, wherein in step (4 b) domain classification predictive labels are calculatedAnd the actual classification label Ld 1 Loss L of (2) 2 The formula is as follows:
where H is the number of individual receivers, M is the total number of samples,network prediction probability derived from kth receiver for nth sample of source domain, +.>Is the source domainWhether n samples belong to the true probability of the kth receiver, belonging to the kth receiver +.>1, not belonging to the kth receiver +.>Is 0.
10. The method of claim 1, wherein the domain prediction tag is calculated in step (4 c)Loss L from actual Domain Label 3 The formula is as follows
Where H is the number of individual receivers, L is the total number of samples,network prediction probability, r, for the first sample of the target domain originating from the kth receiver l k For the true probability of whether the ith sample belongs to the kth receiver, r belonging to the kth receiver l k 1, r not belonging to the kth receiver l k Is 0.
11. The method of claim 8, wherein L is 1 Network prediction probability in a formulaThe calculation is as follows:
first, the SIB passing parameter of the input feature field is theta 1 Feature extractor G of f The output feature vector f:
f=G f (SIB;θ 1 );
then, the characteristic vector f is input into the parameter theta 2 Individual identification classifier G of (2) y Performing classification prediction to obtain a classification prediction vector with a length of C
Wherein C is the number of radar radiation sources,representing the size of the likelihood that the nth source domain input sample belongs to each individual radar radiation source;
finally, set upIs z at the c-th point c Prediction vector +.>Calculating the probability of the nth source domain input sample belonging to each class label>
I.e. the network prediction probability of the nth sample of the source domain belonging to the c-th label, F is the total number of labels, wherein +.>
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310655917.4A CN116662783A (en) | 2023-06-05 | 2023-06-05 | Deep learning-based individual identification method for radar radiation source of cross-over receiver |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310655917.4A CN116662783A (en) | 2023-06-05 | 2023-06-05 | Deep learning-based individual identification method for radar radiation source of cross-over receiver |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116662783A true CN116662783A (en) | 2023-08-29 |
Family
ID=87714925
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310655917.4A Pending CN116662783A (en) | 2023-06-05 | 2023-06-05 | Deep learning-based individual identification method for radar radiation source of cross-over receiver |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116662783A (en) |
-
2023
- 2023-06-05 CN CN202310655917.4A patent/CN116662783A/en active Pending
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107220606B (en) | Radar radiation source signal identification method based on one-dimensional convolutional neural network | |
CN110222748B (en) | OFDM radar signal identification method based on 1D-CNN multi-domain feature fusion | |
US7668699B2 (en) | Optimized stochastic resonance method for signal detection and image processing | |
CN112039820B (en) | Communication signal modulation and identification method for quantum image group mechanism evolution BP neural network | |
CN110133599B (en) | Intelligent radar radiation source signal classification method based on long-time and short-time memory model | |
CN112949387B (en) | Intelligent anti-interference target detection method based on transfer learning | |
CN111126134A (en) | Radar radiation source deep learning identification method based on non-fingerprint signal eliminator | |
CN104794264B (en) | A kind of radar communication waveform design method based on sparse frequency | |
CN113156376A (en) | Radar radiation source signal identification method based on SACNN | |
CN110163040B (en) | Radar radiation source signal identification technology in non-Gaussian clutter | |
CN114528876A (en) | Small sample signal automatic modulation identification method based on improved prototype network | |
CN112749633A (en) | Separate and reconstructed individual radiation source identification method | |
Bhatti et al. | Radar signals intrapulse modulation recognition using phase-based stft and bilstm | |
CN112882010B (en) | High-resolution range profile target identification method based on signal-to-noise ratio field knowledge network | |
CN112162255B (en) | Pulse splicing method based on intra-pulse characteristics | |
CN116540250B (en) | Laser ranging method and system | |
CN116662783A (en) | Deep learning-based individual identification method for radar radiation source of cross-over receiver | |
CN116243248A (en) | Multi-component interference signal identification method based on multi-label classification network | |
CN115169406B (en) | Instantaneous phase fingerprint feature enhancement method based on empirical mode decomposition | |
CN116760491A (en) | Signal-to-noise ratio estimation method based on deep learning | |
CN107682119B (en) | MIMO space-time code identification method based on grouping extreme value model | |
CN115980689A (en) | Point cloud detection-based radiation source signal sorting method, device, equipment and medium | |
Wang et al. | Radar emitter intrapulse signal blind sorting under modified wavelet denoising | |
CN113702901A (en) | Cross-correlation delay estimation method based on signal correlation enhancement | |
EP2684073B1 (en) | Radio frequency digital receiver system and method |
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 |