Radiation source individual identification method and system based on unintentional phase modulation characteristics

Abstract

The invention provides a radiation source individual identification method and system based on unintentional phase modulation characteristics, and relates to the technical field of radar signal processing. The method comprises the following steps: acquiring intermediate frequency data of radar signals emitted by different radar radiation source individuals to obtain a plurality of intermediate frequency data, and extracting a plurality of monopulse signals corresponding to the plurality of intermediate frequency data from the plurality of intermediate frequency data; phase extraction is carried out on the multiple monopulse signals, and unintentional modulation phase feature maps corresponding to the multiple pieces of intermediate frequency data are obtained; and constructing a radar individual identification model and training by using the data set, and identifying the plurality of unintentional modulation phase feature maps through the trained radar individual identification model to obtain a plurality of radar radiation source individual identification results. By extracting unintentional modulation phase features of radar signals, original fingerprint information of the signals is reserved to reduce calculation complexity, and a deep residual network of a small convolution kernel is constructed as a radar individual recognition model to improve recognition accuracy.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of radar signal processing, and in particular to a method and system for identifying individual radiation sources based on unintentional phase modulation characteristics. Background Art

[0002] The subtle features of individual radar emitters are caused by the inherent non-ideal characteristics of the internal devices of the transmitter. The main factors causing the non-ideal characteristics of the emitters are the many internal devices of the emitters, such as the nonlinear characteristics of the power amplifier, the instability of the frequency source inside the emitter, and the discreteness of the device characteristics and technical indicators caused by process defects during the device processing. The existing individual radiation source identification technology extracts subtle features that can reflect the individual radiation source from the intermediate frequency signal, and uses classification and recognition technologies such as machine learning classifiers or artificial intelligence networks to identify the individual differences of subtle features to achieve the identification of individual radiation sources. However, the dimensionality and computational complexity of the subtle features of most individual radiation sources are high, and the multi-layer convolutional neural network in the existing classification and recognition technology has the disadvantages of long preprocessing time, insufficient network depth, and network degradation during training. Therefore, a high-precision and high-efficiency individual radiation source identification method is urgently needed. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method and system for identifying individual radiation sources based on unintentional phase modulation characteristics in view of the deficiencies in the prior art.

[0004] The technical solution of the present invention to solve the above technical problems is as follows:

[0005] A method for identifying individual radiation sources based on unintentional phase modulation characteristics comprises the following steps:

[0006] Collecting intermediate frequency data of radar signals emitted by different radar radiation sources to obtain intermediate frequency data of multiple different radar radiation sources, respectively performing pulse extraction on the multiple intermediate frequency data to obtain multiple single pulse signals corresponding to the multiple intermediate frequency data;

[0007] The plurality of single pulse signals are modeled respectively by the constructed radar single-channel signal model to obtain a plurality of real signals corresponding to the plurality of intermediate frequency data, and phase extraction is performed on the plurality of real signals respectively to obtain a plurality of unintentional modulation phase characteristic diagrams corresponding to the plurality of intermediate frequency data;

[0008] A radar individual recognition model is constructed, and the radar individual recognition model is trained using the constructed radiation source data set. The trained radar individual recognition model is used to recognize a plurality of the unintentional modulation phase feature maps to obtain a plurality of radar radiation source individual recognition results.

[0009] Another technical solution of the present invention to solve the above technical problems is as follows:

[0010] A radiation source individual identification system based on unintentional phase modulation characteristics, comprising:

[0011] A signal acquisition module is used to collect intermediate frequency data of radar signals emitted by different radar radiation sources, obtain intermediate frequency data of multiple different radar radiation sources, and respectively perform pulse extraction on the multiple intermediate frequency data to obtain multiple single pulse signals corresponding to the multiple intermediate frequency data;

[0012] A feature extraction module is used to respectively model multiple single pulse signals through the constructed radar single-channel signal model to obtain multiple real signals corresponding to the multiple intermediate frequency data, and respectively perform phase extraction on the multiple real signals to obtain multiple unintentional modulation phase feature diagrams corresponding to the multiple intermediate frequency data;

[0013] The target recognition module is used to construct a radar individual recognition model, train the radar individual recognition model through the constructed radiation source data set, and recognize multiple unintentional modulation phase feature maps through the trained radar individual recognition model to obtain multiple radar radiation source individual recognition results.

[0014] The beneficial effects of the present invention are: by extracting multiple single pulse signals from intermediate frequency data, an unintentional modulation phase feature map is obtained from multiple single pulse signals using an unintentional phase modulation feature extraction and processing method, which can retain the original fingerprint information of the signal on the basis of reducing the complexity of radar subtle feature extraction, can express the radar nonlinear characteristics, and thus improve the efficiency of the radar individual recognition model in identifying individual radiation sources. The constructed radar single-channel signal model is used for iterative training during the training process, which optimizes problems such as network degradation, improves the network depth, solves problems such as insufficient network depth and network degradation during the training process, and at the same time improves the accuracy of radar radiation source individual recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flow chart of a method for identifying individual radiation sources based on unintentional phase modulation features provided by an embodiment of the present invention;

[0016] Figure 2 A flow chart of phase extraction provided by an embodiment of the present invention;

[0017] Figure 3 A schematic diagram of unintentional phase modulation characteristics of two radar radiation sources provided by an embodiment of the present invention;

[0018] Figure 4 A structural diagram of a radar individual identification model provided by an embodiment of the present invention;

[0019] Figure 5A schematic diagram of a prediction accuracy curve based on a radiation source training set and a radiation source verification set provided by an embodiment of the present invention;

[0020] Figure 6 A schematic diagram of a confusion matrix based on a radiation source test set provided by an embodiment of the present invention;

[0021] Figure 7 A flowchart of an implementation of a method for identifying individual radiation sources based on unintentional phase modulation features provided by an embodiment of the present invention;

[0022] Figure 8 A module block diagram of a radiation source individual identification system based on unintentional phase modulation characteristics provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The principles and features of the present invention are described below in conjunction with the accompanying drawings. The examples given are only used to explain the present invention and are not used to limit the scope of the present invention.

[0024] The individual characteristics of the radiation source mainly include envelope front, bispectrum, cyclic spectrum, time-frequency spectrum, unintentional phase modulation, etc. Among them, the individual characteristics such as bispectrum, cyclic spectrum, time-frequency spectrum have the characteristics of high dimension and high computational complexity, but the unintentional phase modulation feature can express the nonlinear characteristics of the radar and has low computational complexity. At this stage, deep learning technology has driven the development of radar signal recognition technology. The use of multiple hidden layers with nonlinear logic functions can mine deeper information in feature data and amplify the differences in signal characteristics to complete the automatic identification of radiation sources. In view of the shortcomings of multi-layer convolutional neural networks such as long preprocessing time, insufficient network depth, and network degradation during training, the present invention combines the unintentional phase modulation characteristics of individual radiation sources for differential identification, and uses multiple small-sized convolution kernels and residual unit layers in the convolutional neural network to solve the problems of network depth gradient disappearance and gradient explosion.

[0025] The present invention uses radar signals as processing objects to achieve individual identification of radiation sources. Figure 1 As shown, an embodiment of the present invention provides a method for identifying individual radiation sources based on unintentional phase modulation features, comprising the following steps:

[0026] Collecting intermediate frequency data of radar signals emitted by different radar radiation sources to obtain intermediate frequency data of multiple different radar radiation sources, respectively performing pulse extraction on the multiple intermediate frequency data to obtain multiple single pulse signals corresponding to the multiple intermediate frequency data;

[0027] The plurality of single pulse signals are modeled respectively by the constructed radar single-channel signal model to obtain a plurality of real signals corresponding to the plurality of intermediate frequency data, and phase extraction is performed on the plurality of real signals respectively to obtain a plurality of unintentional modulation phase characteristic diagrams corresponding to the plurality of intermediate frequency data;

[0028] A radar individual recognition model is constructed, and the radar individual recognition model is trained using the constructed radiation source data set. The trained radar individual recognition model is used to recognize a plurality of the unintentional modulation phase feature maps to obtain a plurality of radar radiation source individual recognition results.

[0029] Specifically, the intermediate frequency data of the same type of signals emitted by 10 different radars are collected. The carrier frequency of the radar radiation source is 100MHz, the sampling rate of the radar receiver is 400MSps, and the number of pulses collected by each radar is 4000. The intermediate frequency signal data of each radiation source is pulse positioned and the pulse data is intercepted. Each radiation source corresponds to 4000 single pulse signals divided by the number of pulses.

[0030] In the embodiment of the present invention, an unintentional phase modulation feature extraction and processing method is used to reduce the complexity of radar subtle feature extraction, with the purpose of improving the efficiency of individual identification of radiation sources and solving the problem of long preprocessing time of multi-layer convolutional neural networks. A deep residual network model with a small convolution kernel (i.e., a radar single-channel signal model) is constructed to solve the problems of insufficient network depth and network degradation during training, while ensuring the accuracy of individual identification of radar radiation sources.

[0031] Preferably, said respectively performing pulse extraction on the plurality of intermediate frequency data to obtain a plurality of single pulse signals corresponding to the plurality of intermediate frequency data comprises:

[0032] The intermediate frequency data is subjected to Hilbert transform processing to obtain an envelope, the envelope is normalized, the normalized envelope is differentially calculated to obtain a plurality of pulse positioning data, and signal extraction is performed on the intermediate frequency data according to the plurality of pulse positioning data. In this process, all the intermediate frequency data are processed to obtain a plurality of single pulse signals corresponding to the intermediate frequency data.

[0033] It should be understood that the differential value is obtained by calculating the normalized envelope differential, and each radar signal pulse position is located according to the maximum differential value and a single pulse signal is intercepted. Since the envelope image of the pulse signal reaches a peak value, its derivative (i.e., the differential value) will reach a maximum, so the envelope is peak-identified to obtain the signal pulse position.

[0034] In the embodiment of the present invention, the radar pulse position is located by using the differential envelope curve, so that each radar pulse data is aligned, thereby ensuring the precision and accuracy of the subsequent unintentional phase modulation feature extraction.

[0035] Preferably, the construction process of the radar single-channel signal model comprises:

[0036] A radar single-channel signal model is constructed based on the pulse amplitude, initial phase, signal carrier frequency, Gaussian white noise, number of sampling points, phase modulation function and unintentional modulation phase obtained when collecting intermediate frequency data. The radar single-channel signal model is:

[0037]

[0038] Among them, x(n) is the real part signal of the single pulse signal, A is the pulse amplitude, is the initial phase, f c is the signal carrier frequency, ω(n) has a mean of zero and a variance of σ 2 Gaussian white noise, N is the number of sampling points within the pulse duration, n∈[0,N-1], is the phase modulation function, Unintentional phase modulation.

[0039] Preferably, if Figure 2 As shown, the phase extraction is performed on multiple real part signals respectively to obtain multiple unintentional modulation phase characteristic diagrams corresponding to the intermediate frequency data, including:

[0040] The real signal is subjected to Hilbert transform processing to obtain the imaginary signal, and the real signal and the imaginary signal are calculated by the instantaneous phase calculation expression to obtain the instantaneous phase. In this process, all real signals of any intermediate frequency data are processed to obtain multiple instantaneous phases. The instantaneous phase calculation expression is:

[0041]

[0042] in, is the instantaneous phase, x(n) is the real part of the pulse signal, is the imaginary part of the pulse signal, tan -1 is the inverse trigonometric function;

[0043] Modulation processing is performed on the multiple instantaneous phases to obtain an unintentional modulation phase characteristic diagram. In this process, all real signals of all intermediate frequency data are processed to obtain unintentional modulation phase characteristic diagrams corresponding to the multiple intermediate frequency data.

[0044] Preferably, the step of performing modulation processing on the plurality of instantaneous phases to obtain an unintentional modulation phase characteristic diagram comprises:

[0045] An initial instantaneous phase curve is generated according to the multiple instantaneous phases, the initial instantaneous phase curve is de-averaged to obtain an instantaneous phase curve, the instantaneous phase curve is fitted by the least square method to obtain a fitting straight line, the difference between the instantaneous phase curve and the fitting straight line is calculated to obtain a demodulated instantaneous phase curve, and the demodulated instantaneous phase curve is denoised to obtain an unintentional modulation phase characteristic diagram.

[0046] Specifically, after performing Hilbert transform on the single-channel pulse signal (i.e., the real signal x(n) of the single pulse signal), the imaginary signal x(n) of the single pulse signal is obtained. The imaginary signal is expressed as:

[0047]

[0048] The instantaneous phase is calculated by calculating the inverse trigonometric function of the ratio of the real signal and the imaginary signal of the single pulse signal through the instantaneous phase calculation expression to obtain the instantaneous phase. The instantaneous phase calculation expression is:

[0049]

[0050] A phase observation model is constructed based on Hilbert transform processing and instantaneous phase calculation expression. The instantaneous phase of each single pulse signal is calculated through the phase observation model, and an initial instantaneous phase curve containing noise and without modulation is drawn according to the sampling order of multiple instantaneous phases.

[0051] The initial instantaneous phase curve is de-averaged to obtain the instantaneous phase curve, which is expressed as:

[0052]

[0053] Among them, mean(·) is the mean removal process;

[0054] Demodulation of the instantaneous phase curve is performed, including: using the least square method to transform the instantaneous phase curve Fitting to a straight line The least squares method minimizes the sum of square errors from all phase data points to the fitting line, thereby modulating the instantaneous phase curve. is the instantaneous phase curve With the fitting straight line The difference is expressed as:

[0055]

[0056] Demodulation instantaneous phase curve Perform time domain windowing and smoothing operation to reduce noise and obtain unintentional modulation phase characteristic diagram;

[0057] Repeat the above steps to extract the unintentional phase modulation features of the intermediate frequency data of 10 individual radiation sources in turn. Figure 3 Shown is the unintentional phase modulation characteristics (i.e., unintentional modulation phase characteristic diagram) of two radar radiation sources radar1 and radar2.

[0058] In an embodiment of the present invention, the Hilbert transform method is used to estimate the instantaneous phase curve in the pulse, and then the curve is subjected to processing such as de-averaging and noise reduction to extract individual unintentional phase modulation features, and the unintentional modulation phase features of multiple pulse signals from the same radar radiation source are accumulated, and stable individual features are obtained through filtering.

[0059] Preferably, if Figure 4 As shown, the construction of the radar individual recognition model includes:

[0060] Construct an input layer, multiple convolutional layers, multiple batch normalization layers, activation function layers, multiple pooling layers, fully connected layers, and normalization layers;

[0061] Connecting the residual input interface to the input of the first convolutional layer, connecting the output of the first convolutional layer to the input of the second convolutional layer, connecting the residual input interface and the output of the second convolutional layer to the input of the first batch of normalization layers, connecting the output of the first batch of normalization layers to the input of the first pooling layer, and using the output of the first pooling layer as the residual output interface to construct multiple residual unit layers;

[0062] The output of the input layer is connected to the input of the third convolutional layer, the output of the third convolutional layer is connected to the input of the second batch of normalization layers, the output of the second batch of normalization layers is connected to the input of the activation function layer, the output of the activation function layer is connected to the input of multiple residual unit layers connected in series, the output of the last residual unit layer is connected to the input of the second pooling layer, the output of the second pooling layer is connected to the input of the fully connected layer, and the output of the fully connected layer is connected to the input of the normalization layer to obtain a radar individual recognition model.

[0063] It should be understood that an individual recognition deep residual network model (i.e., radar individual recognition model) is constructed based on the pytorch deep learning framework.

[0064] In an embodiment of the present invention, each residual unit layer is composed of two convolutional layers, a batch normalization layer and a pooling layer. At the same time, a short connection is added across the two convolutional layers to achieve cross-layer flow of data. Multiple small-size convolution kernels and residual unit layers are added to the network model to solve problems such as network depth gradient disappearance and gradient explosion.

[0065] Preferably, the trained radar individual identification model is used to identify the multiple unintentional modulation phase feature graphs to obtain multiple radar radiation source individual identification results, including:

[0066] The unintentional modulation phase feature map is convolved by the third convolution layer to obtain a convolution feature map, the convolution feature map is normalized by the second batch normalization layer to obtain a normalized feature map, the normalized feature map is nonlinearly mapped by the activation function layer to obtain a nonlinear feature map, the nonlinear feature map is residually mapped by a plurality of residual unit layers connected in series one by one to obtain a residual feature map, the residual feature map is downsampled by the second pooling layer to obtain a downsampled feature map, the downsampled feature map is classified by the fully connected layer to obtain an identification probability feature map, the identification probability feature map is converted into a probability distribution by the normalization layer to obtain an individual identification result of a radar radiation source, and all unintentional modulation phase feature maps are processed in this way to obtain individual identification results of multiple radar radiation sources.

[0067] Specifically, the radar individual recognition model includes eight feature processing layers. The feature data is forward propagated through the first to eighth layers. The operation of each layer of the network on the data is as follows:

[0068] The first layer is the convolution layer, which uses a small-sized convolution kernel to slide on the input feature map. The convolution kernel size is 3, and the convolution kernel parameters are convolved with the feature map data. As the number of network layers increases, the convolution network extracts higher-dimensional features, which can be expressed as:

[0069] X 1 =X*W+b,

[0070] Among them, X 1 represents the output of the first convolutional layer (i.e., the convolution feature map), X represents the input feature set (i.e., multiple unintentional modulation phase feature maps), W represents the convolution kernel, and b represents the bias term.

[0071] The second layer is the batch normalization layer, which calculates the batch data X 1 The mean μ and variance σ of 2 , normalize the feature data to improve data stability, expressed as:

[0072]

[0073] Among them, ε represents a very small constant, μ represents the mean of the input data of this layer, σ represents the variance of the input data of this layer, and X 2 represents the output of the batch normalization layer, X 1 Represents the output of the first convolutional layer.

[0074] The third layer is a nonlinear transformation layer (i.e., activation function layer), which enables the model to learn more complex nonlinear mappings. The activation function is the ReLU function, which is expressed as:

[0075] RELU(x)=max(0,X 2 ),

[0076] Among them, x represents the input of the nonlinear transformation layer, and RELU(x) represents the output of the nonlinear transformation layer.

[0077] The fourth and fifth layers are residual unit layers. The residual block structure can propagate gradients losslessly, solving the problem of gradient vanishing due to gradient multiplication during the back-propagation process of the convolutional neural network without increasing the amount of additional computation.

[0078] The sixth layer is the pooling layer, which is used to reduce the dimension of the feature vector output by the convolutional layer, while improving the generalization performance of the network and reducing the overfitting phenomenon of the network.

[0079] The seventh layer is a fully connected layer, which reduces the dimension of the input feature map and purifies it, outputs a feature vector with a dimension equal to the number of categories, and obtains the value of each dimension of the feature vector, which is the recognition probability of each radar individual.

[0080] The eighth layer is the Softmax normalization layer, which converts the output of the network into a probability value and distributes its value between 0 and 1.

[0081] Preferably, the training of the radar individual recognition model by using the constructed radiation source data set includes:

[0082] Constructing a radiation source data set, and dividing the radiation source data set into a radiation source training set, a radiation source verification set, and a radiation source test set according to a set ratio;

[0083] The radiation source training set is learned by the radar individual recognition model to identify the identification label of each radiation source individual in the radiation source training set, and the loss of the multiple identification labels is calculated by the cross entropy loss function to obtain the cross entropy loss. The cross entropy loss function is:

[0084]

[0085] Among them, E is the cross entropy loss, y i is the true label, p i is the identification tag, n is the number of radiation source individuals;

[0086] The radar individual recognition model is optimized by the cross entropy loss, and the optimized radar individual recognition model is verified by the radiation source verification set. When the verification passes, the radiation source test set is predicted by the optimized radar individual recognition model to obtain the predicted label of each radiation source individual in the radiation source test set, and a confusion matrix is ​​constructed based on the multiple predicted labels and the corresponding true predicted labels in the radiation source test set. The confusion matrix is ​​ratio-calculated to obtain the recognition rate, and the training is terminated when the recognition rate reaches the set accuracy.

[0087] Specifically, the training and testing of radar emitter individual recognition are completed on the pytorch operating platform under the Phytium FT2000+64-bit processor, NVIDIA-RTX 2080Ti-GPU and Linux system. The feature data is forward propagated from the input layer to the output layer. First, the emitter training set is input into the radar individual recognition model. The radar individual recognition model is used to learn the emitter training set, including: capturing the edge and texture features of multiple unintentional modulation phase feature maps in the emitter training set through a convolution layer with multiple layers of small-size convolution kernels, and then passing through a batch data normalization layer and a nonlinear transformation layer (i.e., a ReLU activation function layer) to improve data stability and nonlinear mapping capabilities, passing through multiple residual unit layers to reduce overfitting and accelerate convergence, and then inputting the pooling layer and the fully connected layer to obtain the category probability. The Softmax normalization layer is used to convert the network's output category probability into a probability value distributed between 0 and 1, and the label with the largest probability value is used as the output label. Use the loss function to calculate the loss value between the output label and the annotated true label, and use the gradient descent method to input the loss value from the output layer of the model to back-propagate from back to front until it propagates to the input layer. During the back-propagation process, update the various parameter values ​​of the model according to the loss value and the optimizer; continuously iterate the above process until the model parameters converge. The initial learning rate is 0.01, and the optimizer uses the radam optimizer. Repeat the forward propagation (i.e., the process of learning the radiation source training set) and the back-propagation process 60 times, that is, the training rounds are 60, the batch size is 128, and the accuracy curves of the radiation source training set and the radiation source verification set are recorded, as shown in Figure 5 As shown, it can be seen that the accuracy of the radiation source training set S1 and the radiation source verification set S2 has reached a convergence state, and the network model parameters are saved.

[0088] The number of feature map samples of the radiation source test set is 1000. The radiation source test set is output by the model with the predicted probability of each class. The class with the largest predicted probability is the predicted label. The values ​​of the true labels are used as the columns of the matrix, and the values ​​of the predicted labels are used as the rows. The confusion matrix of the predicted labels based on the test set output is drawn. The confusion matrix of the test sample set is as follows: Figure 6As shown, the ratio of the number of predicted labels to the number of true labels is calculated according to the confusion matrix as the recognition rate of individual identification classification, and the average recognition rate is calculated to be about 87.6% according to the recognition rate of each radiation source individual.

[0089] like Figure 7 As shown, an embodiment of the present invention provides a method for identifying individual radiation sources based on unintentional phase modulation features, comprising the steps of:

[0090] Step 1: Collect radar intermediate frequency data of the same signal emitted by different radars (i.e., individual emitters), and annotate them with real labels to generate radar emitter intermediate frequency datasets (i.e., emitter datasets);

[0091] Step 2: locate the intermediate frequency pulse data of each radar and intercept multiple pulse data of each radar;

[0092] Step 3: extract the unintentional phase modulation features of multiple pulse data respectively, and after summarizing the unintentional phase modulation features of multiple pulse data, average and smooth them to generate an individual identification feature data set (i.e., a data set constructed by multiple unintentional modulation phase feature maps);

[0093] Step 4: Normalize the individual identification feature dataset and divide it into training set, validation set and test set in a ratio of 6:2:2;

[0094] Step 5: Build a small convolution kernel residual network model for radiation source individual identification based on the pytorch deep learning framework to obtain the radar individual identification model. Input the training set into the radar individual identification model for training, and verify the trained radar individual identification model through the validation set.

[0095] Step 6: Load the deep learning network model (i.e., radar individual recognition model), input the feature test set (i.e., test set) into the deep learning network model for recognition, generate a confusion matrix based on the recognition results, and calculate the recognition accuracy based on the confusion matrix.

[0096] It should be understood that since the individual identification feature data set is obtained from the radar radiation source intermediate frequency data set, the individual identification feature data set has a corresponding real label. In step 4, it can also be divided into a training set and a validation set in a ratio of 8:2, and steps 1 to 3 are repeated to obtain a test set.

[0097] like Figure 8 As shown, an embodiment of the present invention provides a radiation source individual identification system based on unintentional phase modulation characteristics, including:

[0098] A signal acquisition module is used to collect intermediate frequency data of radar signals emitted by different radar radiation sources, obtain intermediate frequency data of multiple different radar radiation sources, and respectively perform pulse extraction on the multiple intermediate frequency data to obtain multiple single pulse signals corresponding to the multiple intermediate frequency data;

[0099] A feature extraction module is used to respectively model multiple single pulse signals through the constructed radar single-channel signal model to obtain multiple real signals corresponding to the multiple intermediate frequency data, and respectively perform phase extraction on the multiple real signals to obtain multiple unintentional modulation phase feature diagrams corresponding to the multiple intermediate frequency data;

[0100] The target recognition module is used to construct a radar individual recognition model, train the radar individual recognition model through the constructed radiation source data set, and recognize multiple unintentional modulation phase feature maps through the trained radar individual recognition model to obtain multiple radar radiation source individual recognition results.

[0101] The advantage of the present invention is that it locates the radar pulse, that is, it uses the differential envelope to locate the position of the radar pulse, so that each radar pulse data is aligned, which ensures the precision and accuracy of the subsequent unintentional phase modulation feature extraction. The unintentional phase modulation feature is extracted and processed, the instantaneous phase characteristics of the radar signal are extracted and further de-modulated, denoised, etc., which retains the original fingerprint information of the radar signal (i.e., the unintentional phase modulation feature) to a greater extent, can express the nonlinear characteristics of the radar signal and has low computational complexity, which helps to improve the efficiency of recognition. A deep residual network model with a small convolution kernel (i.e., a radar individual recognition model) is constructed to perform individual recognition of radar radiation sources. The small convolution kernel can automatically extract more features with richer types, ensuring that the original information of the signal is retained. The deep residual network solves the problems of insufficient network depth and network degradation during training, and has a high recognition accuracy rate, with an average recognition rate of about 87.6%, which has certain engineering application value.

[0102] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0103] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed.

[0105] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of the present invention.

[0106] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for identifying individual radiation sources based on unintentional phase modulation features, characterized in that: The steps include: Collecting intermediate frequency data of radar signals emitted by different radar radiation sources to obtain intermediate frequency data of multiple different radar radiation sources, respectively performing pulse extraction on the multiple intermediate frequency data to obtain multiple single pulse signals corresponding to the multiple intermediate frequency data; The plurality of single pulse signals are modeled respectively by the constructed radar single-channel signal model to obtain a plurality of real signals corresponding to the plurality of intermediate frequency data, and phase extraction is performed on the plurality of real signals respectively to obtain a plurality of unintentional modulation phase characteristic diagrams corresponding to the plurality of intermediate frequency data; A radar individual recognition model is constructed, and the radar individual recognition model is trained using the constructed radiation source data set. The trained radar individual recognition model is used to recognize a plurality of the unintentional modulation phase feature maps to obtain a plurality of radar radiation source individual recognition results.

2. The method for identifying individual radiation sources according to claim 1, characterized in that: The step of respectively performing pulse extraction on the plurality of intermediate frequency data to obtain a plurality of single pulse signals corresponding to the plurality of intermediate frequency data comprises: The intermediate frequency data is subjected to Hilbert transform processing to obtain an envelope, the envelope is normalized, the normalized envelope is differentially calculated to obtain a plurality of pulse positioning data, and signal extraction is performed on the intermediate frequency data according to the plurality of pulse positioning data. In this process, all the intermediate frequency data are processed to obtain a plurality of single pulse signals corresponding to the intermediate frequency data.

3. The method for identifying individual radiation sources according to claim 1, characterized in that: The construction process of the radar single-channel signal model comprises: A radar single-channel signal model is constructed according to pulse amplitude, initial phase, signal carrier frequency, Gaussian white noise, number of sampling points, phase modulation function and unintentional modulation phase. The radar single-channel signal model is: Among them, x(n) is the real part signal of the single pulse signal, A is the pulse amplitude, is the initial phase, f c is the signal carrier frequency, ω(n) is Gaussian white noise, n is the number of sampling points within the pulse duration, is the phase modulation function, Unintentional phase modulation.

4. The method for identifying individual radiation sources according to claim 1, characterized in that: The phase extraction of the multiple real part signals is performed respectively to obtain multiple unintentional modulation phase characteristic diagrams corresponding to the intermediate frequency data, including: The real signal is subjected to Hilbert transform processing to obtain the imaginary signal, and the real signal and the imaginary signal are calculated by the instantaneous phase calculation expression to obtain the instantaneous phase. In this process, all real signals of any intermediate frequency data are processed to obtain multiple instantaneous phases. The instantaneous phase calculation expression is: in, is the instantaneous phase, x(n) is the real part of the pulse signal, is the imaginary part of the pulse signal, tan -1 is the inverse trigonometric function; Modulation processing is performed on the multiple instantaneous phases to obtain an unintentional modulation phase characteristic diagram. In this process, all real signals of all intermediate frequency data are processed to obtain unintentional modulation phase characteristic diagrams corresponding to the multiple intermediate frequency data.

5. The method for identifying individual radiation sources according to claim 4, characterized in that: The step of performing modulation processing on the plurality of instantaneous phases to obtain an unintentional modulation phase characteristic diagram comprises: An initial instantaneous phase curve is generated according to the multiple instantaneous phases, the initial instantaneous phase curve is de-averaged to obtain an instantaneous phase curve, the instantaneous phase curve is fitted by the least square method to obtain a fitting straight line, the difference between the instantaneous phase curve and the fitting straight line is calculated to obtain a demodulated instantaneous phase curve, and the demodulated instantaneous phase curve is denoised to obtain an unintentional modulation phase characteristic diagram.

6. The method for identifying individual radiation sources according to claim 1, characterized in that: The step of constructing a radar individual recognition model comprises: Construct an input layer, multiple convolutional layers, multiple batch normalization layers, activation function layers, multiple pooling layers, fully connected layers, and normalization layers; Connecting the residual input interface to the input of the first convolutional layer, connecting the output of the first convolutional layer to the input of the second convolutional layer, connecting the residual input interface and the output of the second convolutional layer to the input of the first batch of normalization layers, connecting the output of the first batch of normalization layers to the input of the first pooling layer, and using the output of the first pooling layer as the residual output interface to construct multiple residual unit layers; The output of the input layer is connected to the input of the third convolutional layer, the output of the third convolutional layer is connected to the input of the second batch of normalization layers, the output of the second batch of normalization layers is connected to the input of the activation function layer, the output of the activation function layer is connected to the input of multiple residual unit layers connected in series, the output of the last residual unit layer is connected to the input of the second pooling layer, the output of the second pooling layer is connected to the input of the fully connected layer, and the output of the fully connected layer is connected to the input of the normalization layer to obtain a radar individual recognition model.

7. The method for identifying individual radiation sources according to claim 6, characterized in that: The trained radar individual recognition model is used to recognize the multiple unintentional modulation phase feature graphs to obtain multiple radar radiation source individual recognition results, including: The unintentional modulation phase feature map is convolved by the third convolution layer to obtain a convolution feature map, the convolution feature map is normalized by the second batch normalization layer to obtain a normalized feature map, the normalized feature map is nonlinearly mapped by the activation function layer to obtain a nonlinear feature map, the nonlinear feature map is residually mapped by a plurality of residual unit layers connected in series one by one to obtain a residual feature map, the residual feature map is downsampled by the second pooling layer to obtain a downsampled feature map, the downsampled feature map is classified by the fully connected layer to obtain an identification probability feature map, the identification probability feature map is converted into a probability distribution by the normalization layer to obtain an individual identification result of a radar radiation source, and all unintentional modulation phase feature maps are processed in this way to obtain individual identification results of multiple radar radiation sources.

8. The method for identifying individual radiation sources according to claim 1, characterized in that: The step of training the radar individual recognition model by using the constructed radiation source data set includes: Constructing a radiation source data set, and dividing the radiation source data set into a radiation source training set, a radiation source verification set, and a radiation source test set according to a set ratio; The radiation source training set is learned by the radar individual recognition model to identify the identification label of each radiation source individual in the radiation source training set, and the loss of the multiple identification labels is calculated by the cross entropy loss function to obtain the cross entropy loss. The cross entropy loss function is: Among them, E is the cross entropy loss, y i is the true label, p i is the identification tag, n is the number of radiation source individuals; The radar individual recognition model is optimized by the cross entropy loss, and the optimized radar individual recognition model is verified by the radiation source verification set. When the verification passes, the radiation source test set is predicted by the optimized radar individual recognition model to obtain the predicted label of each radiation source individual in the radiation source test set, and a confusion matrix is ​​constructed based on the multiple predicted labels and the corresponding true predicted labels in the radiation source test set. The confusion matrix is ​​ratio-calculated to obtain the recognition rate, and the training is terminated when the recognition rate reaches the set accuracy.

9. A radiation source individual identification system based on unintentional phase modulation characteristics, characterized in that: include: A signal acquisition module is used to collect intermediate frequency data of radar signals emitted by different radar radiation sources, obtain intermediate frequency data of multiple different radar radiation sources, and respectively perform pulse extraction on the multiple intermediate frequency data to obtain multiple single pulse signals corresponding to the multiple intermediate frequency data; A feature extraction module is used to respectively model multiple single pulse signals through the constructed radar single-channel signal model to obtain multiple real signals corresponding to the multiple intermediate frequency data, and respectively perform phase extraction on the multiple real signals to obtain multiple unintentional modulation phase feature diagrams corresponding to the multiple intermediate frequency data; The target recognition module is used to construct a radar individual recognition model, train the radar individual recognition model through the constructed radiation source data set, and recognize multiple unintentional modulation phase feature maps through the trained radar individual recognition model to obtain multiple radar radiation source individual recognition results.

10. The radiation source individual identification system according to claim 9, characterized in that: In the signal acquisition module, pulse extraction is performed on the plurality of intermediate frequency data respectively to obtain a plurality of single pulse signals corresponding to the plurality of intermediate frequency data, including: The intermediate frequency data is subjected to Hilbert transform processing to obtain an envelope, the envelope is normalized, the normalized envelope is differentially calculated to obtain a plurality of pulse positioning data, and signal extraction is performed on the intermediate frequency data according to the plurality of pulse positioning data. In this process, all the intermediate frequency data are processed to obtain a plurality of single pulse signals corresponding to the intermediate frequency data.

Images