Individual identification method of radiation sources based on differential reconstruction of constellation diagram

Abstract

The present invention relates to the field of radiation source individual identification technology. In order to improve the recognition accuracy, effectiveness and reliability under low signal-to-noise ratio, the present invention proposes a radiation source individual identification method based on differential reconstruction constellation diagram, and the initial I / Q two-way signal data of each signal source are subjected to differential processing, and EMD, VMD and ITD decomposition are performed respectively, and the three items with the highest frequency after decomposition are added and reconstructed to generate a constellation diagram, and the radiation source individual identification is performed through a deep convolutional neural network. The present invention can combine the advantages of each decomposition through three different decomposition methods, so that individual characteristics are more likely to be highlighted, and then use the advantages of deep convolutional neural network in image processing for identification, which can improve the accuracy and reliability of identification.

Description

Technical Field

[0001] The present invention relates to the technical field of radiation source individual identification, and in particular to a radiation source individual identification method based on differential reconstruction constellation diagram. Background Art

[0002] The radiation source individual identification technology faces severe challenges in today's electromagnetic environment. The complexity of the radiation source system continues to increase, making the identification process more difficult. This is due to the diversity and complexity of modern communication systems. The types and structures of radiation sources are becoming more and more diverse, requiring the identification technology to constantly change and adapt.

[0003] The core of radiation source fingerprint feature extraction is to accurately characterize the hardware differences of radiation sources. The unintentional modulation information hidden in the signals of different radiation sources is the key to distinguishing individual radiation sources. Compared with intentional modulation, unintentional modulation is subtle and difficult to detect. For individual radiation sources of the same model and the same modulation method, their hardware differences are very small, resulting in only very weak fingerprint information in their signals. Especially under low signal-to-noise ratio conditions, the fingerprint features of the signal are easily submerged in the noise. If the features cannot be separated from the noise, the subsequent classification and identification work will be affected. The performance of existing technologies under low signal-to-noise ratios is still not ideal, and there are still many problems to be solved in terms of the robustness of radiation source fingerprint features and resistance to channel interference. Summary of the invention

[0004] Based on the technical problems of robustness and channel interference resistance of the fingerprint features of radiation sources in the above-mentioned prior art, the purpose of the present invention is to provide an individual identification method based on differential reconstruction of constellation diagram, which transforms the I / Q two-way signal data into the image domain of the constellation diagram through differential, decomposition and reconstruction, and inputs the constellation diagram as a fingerprint feature into the deep convolutional neural network to complete the individual identification and classification of radiation sources. It can improve the recognition accuracy, effectiveness and reliability under low signal-to-noise ratio.

[0005] The technical solution of the present invention is:

[0006] Step 1: Collect RF baseband signals from multiple radiation sources to obtain initial I / Q two-way signal data;

[0007] Step 2: Perform differential processing on the initial I / Q signals to form differentially processed I and Q signals;

[0008] Step 3: Perform EMD, VMD, and ITD decomposition on the I signal after differential processing to obtain the first IMF set, the second IMF set, and the third IMF set respectively;

[0009] The Q signal after the difference processing is also decomposed by EMD, VMD, and ITD respectively to obtain the fourth IMF set, the fifth IMF set, and the sixth IMF set respectively;

[0010] Step 4: reconstructing by adding the three items with the highest frequencies in each of the first IMF set, the second IMF set, the third IMF set, the fourth IMF set, the fifth IMF set, and the sixth IMF set, respectively obtaining a first reconstructed signal, a second reconstructed signal, a third reconstructed signal, a fourth reconstructed signal, a fifth reconstructed signal, and a sixth reconstructed signal;

[0011] Step 5: Normalize the first reconstructed signal and the fourth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, normalize the second reconstructed signal and the fifth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, and normalize the third reconstructed signal and the sixth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, to obtain three constellation diagrams;

[0012] Step 6: Using the three constellation diagrams as radiation source fingerprints, individual radiation source identification is performed through a convolutional neural network. The convolutional neural network may be a network of the Resnet series.

[0013] Furthermore, the features extracted in the above steps 2 and 3 include:

[0014] The differential processing of the I / Q signals is as follows:

[0015]

[0016] Where Y(t) = X(t)e j2πθt+φ is the baseband signal obtained by down-conversion of the receiver, Y * To take the conjugate value, n is the interval of difference, is the phase error. The differentially processed signal D(t) contains a constant phase rotation factor e -j2πθn .

[0017] In summary, the present invention has the following beneficial effects:

[0018] By differentially reconstructing the constellation diagram, the signal after differential processing only contains a phase rotation factor of a constant value. Without estimating and compensating the carrier frequency deviation and phase deviation of the receiver, a stable radio frequency fingerprint of the communication radiation source can also be obtained. Since individual features will be drowned by noise and difficult to extract, and individual features are often hidden in high-frequency components, decomposing and reconstructing the signal can separate the high-frequency features from the original signal. The three different decomposition methods can combine the advantages of each decomposition, making it more likely that individual features will be highlighted. Relying on the advantages of deep convolutional neural networks in image processing for recognition can improve the accuracy and reliability of recognition.

[0019] The technical solution of the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A flowchart of the method proposed for the present invention;

[0021] Figure 2 This is the constellation trajectory diagram in the comparative experiment when the signal-to-noise ratio is 10dB;

[0022] Figure 3 This is the constellation trajectory diagram in the comparative experiment when the signal-to-noise ratio is -2dB;

[0023] Figure 4 This is the differential constellation diagram in the comparison experiment when the signal-to-noise ratio is -2dB;

[0024] Figure 5 This is the differential reconstruction constellation diagram using the VMD decomposition method at a signal-to-noise ratio of -2dB in the comparative experiment;

[0025] Figure 6 It is a comparison diagram between the radiation source individual identification method based on differential reconstruction constellation diagram of the present invention and the traditional feature extraction method. DETAILED DESCRIPTION

[0026] Figure 1 The flowchart of the method proposed in the present invention is shown in FIG. Figure 1 As shown, the present invention provides a method for identifying individual radiation sources based on a differentially reconstructed constellation diagram, and the method for identifying individual radiation sources based on a differentially reconstructed constellation diagram comprises the following steps:

[0027] Step 1: Collect RF baseband signals of multiple radiation sources of the same model to obtain initial I / Q two-way signal data;

[0028] Step 2: Perform differential processing on the initial I / Q signals to form differentially processed I and Q signals;

[0029] Step 3: Perform EMD, VMD, and ITD decomposition on the I signal after differential processing to obtain the first IMF set, the second IMF set, and the third IMF set respectively;

[0030] The Q signal after the difference processing is also decomposed by EMD, VMD, and ITD respectively to obtain the fourth IMF set, the fifth IMF set, and the sixth IMF set respectively;

[0031] Step 4: reconstructing by adding the three items with the highest frequencies in each of the first IMF set, the second IMF set, the third IMF set, the fourth IMF set, the fifth IMF set, and the sixth IMF set, respectively obtaining a first reconstructed signal, a second reconstructed signal, a third reconstructed signal, a fourth reconstructed signal, a fifth reconstructed signal, and a sixth reconstructed signal;

[0032] Step 5: Normalize the first reconstructed signal and the fourth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, normalize the second reconstructed signal and the fifth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, and normalize the third reconstructed signal and the sixth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, to obtain three constellation diagrams;

[0033] Step 6: Use the three constellation diagrams as radiation source fingerprints and perform individual radiation source identification through a deep convolutional neural network.

[0034] Furthermore, the features extracted in the above steps 2 and 3 include:

[0035] The difference processing described in step 2 is specifically as follows:

[0036]

[0037] Where Y(t) = X(t)e j2πθt+φ is the baseband signal obtained by down-conversion of the receiver, Y * To obtain the conjugate value, n is the interval of the difference, n is taken as 40, is the phase error. The differentially processed signal D(t) contains a constant phase rotation factor e -j2πθn .

[0038] The following is a brief description of the basic decomposition models of EMD, VMD, and ITD:

[0039] The EMD decomposition process is to continuously obtain the intrinsic mode function components of each layer. Suppose the original signal is s(t) and the signal to be decomposed is r(t). The main steps of decomposition are as follows:

[0040] (1) Initially, let the signal to be decomposed r(t) = s(t), and fit r(t) to obtain the upper and lower envelopes with an average value of m(t). The middle component h(t) is defined as the difference between s(t) and m(t), that is:

[0041] h(t)=s(t)-m(t)

[0042] (2) Judge the intermediate component h(t). If two conditions are met: first, the number of extreme points and zero-crossing points of the upper and lower envelopes cannot differ by more than one; second, the mean of the upper and lower envelopes is 0, then h(t) is the first IMF component c1(t), and r(t) = s(t) - c1(t). If both conditions are not met, r(t) = h(t), and the iterative decomposition is continued until the conditions are met.

[0043] VMD decomposition assumes that the original signal f is decomposed into k components, ensuring that the decomposition sequence is a modal component with a limited bandwidth and a center frequency. At the same time, the sum of the estimated bandwidths of each mode is minimized. The constraint condition is that the sum of all modes is equal to the original signal. The VMD constrained variational model is as follows:

[0044]

[0045]

[0046] Among them, u k ={u1,u2,...,u k} are the modal functions, ω k ={ω1,ω2,...,ω k ,} is the center frequency of each mode, is the partial derivative operator.

[0047] VMD decomposition can set the number of decomposed modes of a given signal. During the search and solution process, it can adaptively match the optimal center frequency and finite bandwidth of each mode, achieve effective separation of inherent modal components and frequency domain division of the signal, and then obtain the effective decomposition components of the given signal, and finally obtain the optimal solution to the variational problem.

[0048] The basic model of ITD decomposition is as follows:

[0049] For signal X t , define a piecewise linear baseline extraction operator ζ, so that the original signal X t Decompose into a baseline signal and make the remaining components into appropriate rotation components. t It can be expressed as:

[0050] X t =ζX t +(1-ζ)X t =Lt +H t

[0051] Where: L t =ζX t Represents the baseline signal, which represents the relatively high-frequency part of the input signal; H t =(1-ζ)X t It represents the rotation component, which represents the low-frequency part of the input signal. After the input signal is decomposed into the baseline signal and the rotation signal, the baseline signal can be further decomposed until the obtained baseline signal is a monotonic signal.

[0052] In step 6, the present invention does not limit the type of the convolutional neural network, as long as it can realize image recognition and classification. Preferably, in the present invention, the convolutional neural network ResNet18 is used to automatically extract features and classify the differential reconstruction constellation diagram. Since the ResNet18 is a public network, the specific process in step 6 is not repeated here.

[0053] The above is only a preferred embodiment of the present invention and does not limit the present invention in any way. Any simple modification, change and equivalent change made to the above embodiment according to the technical essence of the invention still falls within the protection scope of the technical solution of the present invention.

[0054] Figure 2 This is the constellation trajectory diagram for the comparison experiment under a signal-to-noise ratio of 10dB. Figure 3 To compare the constellation trajectory diagram in the experiment with a signal-to-noise ratio of -2dB, Figure 2 , Figure 3 It can be seen that the reduction of signal-to-noise ratio has a greater impact on the constellation trajectory diagram, which also directly affects the final recognition accuracy.

[0055] Figure 4 To compare the differential constellation diagram in the experiment with a signal-to-noise ratio of -2dB, Figure 5 This is the differential reconstruction constellation diagram using the VMD decomposition method at a signal-to-noise ratio of -2dB in the comparative experiment. Figure 4 , Figure 5 It can be seen that the differential reconstruction constellation diagram can filter out some useless information in the differential constellation diagram and highlight the high-frequency characteristics of the signal. The present invention uses three decomposition methods, EMD, VMD, and ITD, to obtain three differential reconstruction constellation diagrams. First, the three constellation diagrams meet the input requirements of the neural network, and there is no need to convert the constellation diagram into RGB three-channel form; secondly, the three decomposition methods have different effects under different signal-to-noise ratios, and the neural network can extract different features to better complete the classification task.

[0056] Figure 6This is a comparison chart between the radiation source individual identification method based on differential reconstruction constellation diagram of the present invention and the existing feature extraction method. Figure 6 In this paper, the method of using differential reconstruction of constellation diagram is compared with the existing constellation trajectory diagram method and differential constellation diagram method through convolutional neural network ResNet18. The data set used in this comparative experiment is as follows: The output data of eight power amplifiers of the same model are collected by baseband signal receiver as the research object. The center frequency of the amplifier module is 355MHz. Each individual is studied using 160 samples, and each sample has 1200 sample points. Figure 6 It can be seen that the differential reconstruction constellation diagram method used in the present invention is generally superior to the existing constellation trajectory diagram and differential constellation diagram methods. Especially in the case of low signal-to-noise ratio, since the present method extracts high-frequency feature signals and adds and reconstructs them, it has a certain filtering effect, so individual features are easier to show in the constellation diagram. In the case of high signal-to-noise ratio, the accuracy of the three methods is not much different.

[0057] In summary, the present invention transforms the I / Q two-way signal data into the graph domain of the constellation diagram after differentiation, decomposition and reconstruction, and then automatically and effectively extracts the deep-level features of the individual radiation source carried by the signal through a convolutional neural network, and corresponds each differentially reconstructed constellation diagram to the individual radiation source, thereby completing the individual radiation source identification and classification more safely and effectively.

Claims

1. A method for identifying individual radiation sources based on differential reconstruction of constellation diagrams, characterized in that: The following steps are involved: Step 1: Collect radio frequency baseband signals from a plurality of radiation sources to obtain initial I / Q two-way signal data; the plurality of radiation sources are radiation sources of the same model; Step 2: Perform differential processing on the initial I / Q signals to form differentially processed I and Q signals; Step 3: Perform EMD, VMD, and ITD decomposition on the I signal after differential processing to obtain the first IMF set, the second IMF set, and the third IMF set respectively; The Q signal after the difference processing is also decomposed by EMD, VMD, and ITD respectively to obtain the fourth IMF set, the fifth IMF set, and the sixth IMF set respectively; Step 4: reconstructing by adding the three items with the highest frequencies in each of the first IMF set, the second IMF set, the third IMF set, the fourth IMF set, the fifth IMF set, and the sixth IMF set, respectively obtaining a first reconstructed signal, a second reconstructed signal, a third reconstructed signal, a fourth reconstructed signal, a fifth reconstructed signal, and a sixth reconstructed signal; Step 5: Normalize the first reconstructed signal and the fourth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, normalize the second reconstructed signal and the fifth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, and normalize the third reconstructed signal and the sixth reconstructed signal respectively and use them as the horizontal and vertical axes to generate a constellation diagram, to obtain three constellation diagrams; Step 6: Use the three constellation diagrams as radiation source fingerprints and perform individual radiation source identification through a convolutional neural network.

2. The method for identifying individual radiation sources based on differential reconstruction of constellation diagrams according to claim 1, characterized in that: The differential processing in step 2 includes: in, is the baseband signal obtained by down-conversion of the receiver, Y * To take the conjugate value, n is the interval of difference, is the phase error. The differentially processed signal D(t) contains a constant phase rotation factor e -j2πθn .

3. The method for identifying individual radiation sources based on differential reconstruction of constellation diagrams according to claim 2, characterized in that: In step 3, the maximum number of decompositions of EMD, VMD, and ITD is limited to 5.

4. The method for identifying individual radiation sources based on differential reconstruction of constellation diagrams according to claim 3, characterized in that: In step 2, the value of the difference interval n is 40.

5. The method for identifying individual radiation sources based on differential reconstruction of constellation diagrams according to claim 3, characterized in that: In step 6, the convolutional neural network is ResNet18.

Images