A method for identifying specific radiation sources based on fuzzy function graph method

Abstract

This invention discloses a method for identifying specific radiation sources based on a fuzzy function graph method, belonging to the field of radiation source identification technology. The method comprises the following steps: S1: obtaining an actual received signal and reconstructing an ideal signal; S2: obtaining a fuzzy function graph of the signal based on the actual received signal and the reconstructed ideal signal; S3: using the fuzzy function graph as a two-dimensional image to extract radio frequency fingerprint features; S4: using a neural network model to learn and classify the radio frequency fingerprint features; and S5: deploying the neural network model in an actual system for real-time radiation source identification. This method fully utilizes the information in the fuzzy function graph and the powerful feature learning capabilities of convolutional neural networks, and has great application potential.

Description

Technical Field

[0001] The present invention belongs to the technical field of radiation source identification, and in particular relates to a specific radiation source identification method based on a fuzzy function graph method. Background Art

[0002] Most existing signal RF fingerprinting methods target a specific type of signal and use raw I / Q baseband signals as input parameters. While baseband signals retain their original signal characteristics, they cannot directly reflect the characteristics of RF fingerprints. Furthermore, raw I / Q data alone contains a large amount of noise and channel environment influences, resulting in reduced recognition accuracy. Current methods lack intuitive analysis capabilities.

[0003] Traditional RF fingerprint recognition methods are usually based on (1) statistical features: amplitude, phase noise, IQ balance; (2) nonlinear features: harmonic distortion, clock deviation, ADC / DAC nonlinear error; (3) signal transformation: frequency domain analysis, phase space reconstruction, time-frequency analysis, etc. However, the performance of these methods is often affected by factors such as signal quality, environmental noise, and hardware limitations, and requires a lot of manual feature extraction and selection. The signal itself is a kind of RF fingerprint, but this fingerprint contains too much redundant information and has too large a dimension. The computational complexity in the recognition stage is huge, resulting in insufficient recognition efficiency. Therefore, it is necessary to first remove irrelevant and redundant information from the identifiable signal and retain as many device features as possible. This is essentially a process of signal dimensionality reduction. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a specific radiation source identification method based on fuzzy function graph method, which uses the convolution two-dimensional representation of the real received signal and the ideal reconstructed signal to extract the radio frequency fingerprint features; through the residual convolutional neural network design, more information exchange and cross-layer connections are introduced, making the network more stable and robust, able to handle more complex and diverse image scenes, and with better generalization ability and adaptability.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for identifying a specific radiation source based on a fuzzy function graph method comprises the following steps:

[0007] S1: Obtain the actual received signal and reconstruct the ideal signal;

[0008] S2: Obtain the fuzzy function diagram of the signal based on the actual received signal and the reconstructed ideal signal;

[0009] S3: Use the fuzzy function graph as a two-dimensional image to extract RF fingerprint features;

[0010] S4: Using neural network model to learn and classify RF fingerprint features;

[0011] S5: Deploy the neural network model into the actual system for real-time radiation source identification.

[0012] Furthermore, in step S1, the ideal baseband signal of the ideal signal transmitter is x(t)=x I (t)+j*x Q (t), modulate it to a carrier frequency f c High frequency signal:

[0013]

[0014] where x I (t), x Q (t) I / Q two-way digital baseband signal of the communication system; Considering the frequency drift f in the wireless communication system g , the actual carrier frequency is f c +f g ; For the received signal, assuming the channel is ideal, the signal received at the receiver is:

[0015]

[0016] where f cx is the frequency drift of the receiving system, fc x ≠f g For all received signals, for the same receiver, Δf=(f g +f c )-(f cx +f c ) is a fixed value; after pre-processing such as mixing and filtering, the actual received signal is obtained Reconstruct the ideal signal based on the actual received signal

[0017] Furthermore, in step S2, the fuzzy function diagram of the signal is obtained based on the actual received signal and the reconstructed ideal signal:

[0018]

[0019] where τ, f D are the delay in the time domain and the offset in the frequency domain respectively;

[0020] For the incoming signal, first make a frequency estimate. For a signal with a time length of T and a frequency of f, take τ∈[-T / N1, T / N1], f D∈[-f / N2, f / N2], where N1 and N2 are constants greater than 1; for the same transmitter, there is a same maximum point at the same position in the frequency dimension of the ambiguity function graph; in addition to the position, the image characteristics also include the nonlinear response of the signal in the time domain and the offset characteristics in the frequency domain.

[0021] Furthermore, the radio frequency fingerprint features in step S3 include phase noise, Dochler shift, and time delay.

[0022] Furthermore, the neural network model described in step S4 is based on the ResNet neural network, and the residual block is introduced so that each layer is directly connected to the previous layer, so that the gradient is directly back-propagated to any layer, thereby making the network easier to optimize. At the same time, the cross-layer connection of ResNet is used to propagate the features of the previous layer to the subsequent layer; the processed signal fuzzy function graph is used as the input of the neural network, trained in the model, and the results are verified with the data of the validation set.

[0023] Furthermore, the neural network model described in step S4 is a ResNet-based neural network, and a residual block is introduced so that each layer is directly connected to the previous layer, so that the gradient is directly back-propagated to any layer, thereby making the network easier to optimize. At the same time, the cross-layer connection of ResNet is used to propagate the features of the previous layer to the subsequent layer; the processed signal fuzzy function graph is used as the input of the neural network, trained in the model, and the results are verified with the data of the validation set.

[0024] The beneficial effects of the present invention are as follows: the fuzzy function graph of the present invention can provide a relatively more comprehensive time-domain analysis of linear characteristics and frequency deviations. The fuzzy function graph is a universal image format. The classic convolutional neural network model designed for computer vision can be directly applied to process fuzzy function graph data, reducing the requirements for professional knowledge of designing specific convolutional neural network models for wireless signals and ensuring the stability of the results. The present invention overcomes the difficulties of constructing and executing matching filters due to the uncertainty of the form and content of the received signal. An ideal reference signal is created through prior information, and the signal processing part is completed by generating a fuzzy function graph through processing with the actual received signal. This specific radiation source identification technical solution based on fuzzy function graph and convolutional neural network makes full use of the information of the fuzzy function graph and the powerful feature learning ability of the convolutional neural network, and has great application potential.

[0025] Other advantages, objectives and features of the present invention will be described in the following description and will be apparent to those skilled in the art to some extent, or those skilled in the art can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to make the purpose, technical solutions and beneficial effects of the present invention more clear, the present invention provides the following drawings for illustration:

[0027] Figure 1 It is a signal transmission and reception model;

[0028] Figure 2 is the fuzzy function diagram, where (a) and (b) are different viewing angles;

[0029] Figure 3 This is the ResNet neural network architecture diagram. DETAILED DESCRIPTION

[0030] Aiming at the problems existing in the current specific radiation source algorithm, the present invention proposes a specific radiation source identification method based on a fuzzy function method.

[0031] First, the system block diagram of signal transmission and reception is as follows Figure 1 As shown, the ideal baseband signal of the ideal signal transmitter is x(t)=x I (t)+j*x Q (t), modulate it to a carrier frequency f c High frequency signal:

[0032]

[0033] where x I (t), x Q (t) I / Q two-way digital baseband signal of the communication system. Considering the wireless communication system, due to various reasons (such as drift in the circuit, temperature change in the environment, aging of the equipment, etc.), the carrier frequency may deviate from its original or preset value, that is, the frequency drift f g , the actual carrier frequency is f c +f g The differences in RF signals between devices are due to differences in transmitter circuit design and manufacturing tolerances of electronic components during the production process. In other words, differences in hardware will lead to differences in frequency drift. For the received signal, assuming the channel is ideal, the signal received by the receiver is

[0034]

[0035] Because the transmitter and receiver have different carrier frequencies, f cx ≠f g , where f cx is the frequency drift of the receiving system. However, for the same receiver, there is the same f cx , so for all received signals, for the same receiver, there is Δf=(fg +f c )-(f cx +f c ) is a fixed value. After pre-processing such as mixing and filtering, the actual received signal will be obtained. According to the actual received signal, the ideal signal can be reconstructed According to the actual received signal and the reconstructed ideal signal, the fuzzy function diagram of the signal can be obtained:

[0036]

[0037] where τ, f D are the delay in the time domain and the offset in the frequency domain respectively. The fuzzy function diagram is as follows Figure 2 As shown in (a) and (b) in the figure. For the incoming signal, the frequency is estimated first. Considering that the value of the frequency drift will not be too large, in general, for a signal with a time length of T and a frequency of f, τ∈[-T / N1, T / N1], f D ∈[-f / N², f / N²]. Since Δf is a constant, the same maximum value should exist for the same transmitter at the same location in the frequency domain of the ambiguity function graph. In addition to location, the image characteristics include the nonlinear response of the signal in the time domain and the offset characteristics in the frequency domain.

[0038] At the same time, a neural network based on ResNet is constructed, and the network structure is as follows Figure 3 The introduction of residual blocks allows each layer to be directly connected to the previous layer, allowing gradients to be directly backpropagated to any layer, making the network easier to optimize. Furthermore, ResNet's skip connections facilitate the propagation of features from previous layers to subsequent layers, enhancing feature utilization.

[0039] The processed signal fuzzy function graph is used as input to the neural network for model training, and the results are verified against validation data. After model testing and verification, the model can be deployed in a real-world system for real-time radiation source identification. This specific radiation source identification technology solution based on the fuzzy function graph and convolutional neural network fully utilizes the information in the fuzzy function graph and the powerful feature learning capabilities of convolutional neural networks, and has great application potential.

[0040] Compared with traditional specific radiation source identification algorithms, this method overcomes the difficulty of constructing and executing matched filtering due to the uncertainty of the received signal form and content, and constructs a universal specific RF radiation source identification method.

[0041] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A method for identifying a specific radiation source based on a fuzzy function graph method, characterized by: The following steps are involved: S1: Get the actual received signal and reconstruct the ideal signal; the ideal baseband signal of the ideal signal transmitter is , modulate it to a carrier frequency of High frequency signal: in , I / Q two-way digital baseband signal of the communication system; considering the frequency drift in the wireless communication system , the actual carrier frequency is ; For the received signal, assuming the channel is ideal, the signal received at the receiver is: in is the frequency drift of the receiving system, , for all received signals, for the same receiver, there is is a fixed value; After mixing and filtering preprocessing, the actual received signal is obtained ; Reconstruct the ideal signal based on the actual received signal ; S2: Based on the actual received signal and the reconstructed ideal signal, the fuzzy function diagram of the signal is obtained; Based on the actual received signal and the reconstructed ideal signal, the fuzzy function diagram of the signal is obtained: in are the delay in the time domain and the offset in the frequency domain respectively; For the incoming signal, first make a frequency estimate, and for the signal time length , the frequency is The signal, take , ,in is a constant greater than 1; for the same transmitter, there is a same maximum point at the same position in the frequency domain of the ambiguity function graph; in addition to the position, the image characteristics also include the nonlinear response of the signal in the time domain and the offset characteristics in the frequency domain; S3: Using the fuzzy function graph as a two-dimensional image to extract radio frequency fingerprint features; the radio frequency fingerprint features include phase noise, Doppler frequency shift, and time delay; S4: Using neural network model to learn and classify RF fingerprint features; S5: Deploy the neural network model to an actual system for real-time radiation source identification; the neural network model is based on ResNet, and introduces residual blocks so that each layer is directly connected to the previous layer, so that the gradient is directly back-propagated to any layer, and at the same time, the cross-layer connection of ResNet is used to propagate the features of the previous layer to the subsequent layer; the processed signal fuzzy function graph is used as the input of the neural network, trained in the model, and the results are verified with the data of the validation set.

Images