Signal fingerprint feature extraction and recognition method of the same radiation source in multiple environments

Abstract

The present invention discloses a method for extracting and identifying fingerprint features of the same radiation source signal in multiple environments, which belongs to the field of non-radiation source identification. The main idea of ​​the method is: first, the radiation source signal is received in environment one, saved and the signal features are extracted, and a radio frequency fingerprint library is established through machine learning for subsequent radiation source identification. Then, the radiation source signal is received in environment two, the radiation source signal is input into an adaptive filter as an input signal, the radiation source signal received in environment one is input into the adaptive filter as the expected signal, and the calibration coefficient is obtained through iterative adaptive algorithm. Finally, the radiation source signal obtained in environment two is processed by the adaptive filter according to the calibration coefficient, the processed signal feature is extracted and combined with the radio frequency fingerprint library to classify and identify the radiation source in environment two.

Description

Technical Field

[0001] The present invention belongs to the field of radiation source identification, and specifically proposes a method for extracting and identifying fingerprint features of signals of the same radiation source under various environments. Background Art

[0002] With the rapid development of wireless communications and the Internet of Things (IoT), more and more devices and objects are communicating and interacting via radio frequency (RF) signals. However, this also raises security and privacy concerns. Malicious users could exploit wireless signals to gain unauthorized access, steal data, or engage in other malicious activities. Identifying malicious users has become a pressing issue. RF fingerprinting technology uses RF signal characteristics to identify and authenticate objects or devices. It leverages signal characteristics within the radio spectrum, such as frequency, amplitude, phase, and time domain characteristics, to distinguish and identify RF signals between different devices or objects. RF fingerprinting technology has applications in a variety of fields, including communications, the IoT, security, military, and industry. However, RF signal characteristics are easily affected by the surrounding environment. Even RF signals emitted by the same source can exhibit different signal characteristics in different environments. This makes RF fingerprint libraries built in one environment unusable across multiple environments.

[0003] In practical applications, it is essential to achieve uniform RF fingerprinting across multiple environments. Nowadays, mobile phones, drones, and other devices travel over large distances, making RF fingerprint libraries established for a single environment incapable of identifying target signals across a wide range. Extracting features from the same radiation source in multiple environments and supplementing the fingerprint library are labor-intensive. Therefore, a signal fingerprint feature extraction and identification method is needed that enables the fingerprint features of the same radiation source to be universally recognized across different environments.

[0004] The method for extracting and identifying fingerprint features of signals of the same radiation source in multiple environments proposed by the present invention uses simple equipment and can unify signal features in multiple environments, thus solving the problem that the radio frequency fingerprint library cannot be used in multiple environments. Summary of the Invention

[0005] The purpose of the present invention is to provide a method for extracting and identifying fingerprint features of signals from the same radiation source in multiple environments, which can effectively solve the problem that the current radio frequency fingerprint library cannot be used in multiple environments.

[0006] The present invention provides the following technical solutions:

[0007] A method for extracting and identifying fingerprint features of signals from the same radiation source in multiple environments. This technical solution includes:

[0008] First, the receiver collects and saves signals from the radiation source in environment 1, extracts the RF fingerprint and establishes an RF fingerprint library; then, it collects signals from the unified radiation source in environment 2, inputs the collected signals into the adaptive filter as input signals, and inputs the signals collected in environment 1 into the adaptive filter as reference signals, and obtains the calibration coefficients through iterative adaptive algorithms; finally, the adaptive filter is used as a calibrator to calibrate the signals collected in environment 2, and the radiation sources in environment 2 are classified and identified in combination with the RF fingerprint library established in environment 1.

[0009] A radiation source signal correction method based on adaptive filtering comprises the following steps:

[0010] Step 1: In the following environment, use a superheterodyne receiver to receive and save the radiation source signal.

[0011] Step 2: Based on step 1, feature extraction is performed on the collected radiation source signal and a radio frequency fingerprint library is established using the KNN algorithm.

[0012] Step 3: Under environment 2, use a superheterodyne receiver to receive and save the radiation source signal.

[0013] Step 4: Based on steps 1 and 2, the radiation source signal received in environment 1 is input into the adaptive filter as a reference signal, and the radiation source signal received in environment 2 is input into the adaptive filter as an input signal, and the calibration coefficient is obtained through iterative adaptive algorithm.

[0014] Step 5: Based on step 4, the radiation source signal collected from the second environment is processed using the adaptive filter with the calibration coefficient obtained.

[0015] Step 6: Based on step 5 and step 2, feature extraction is performed on the processed signal, and the radiation source in environment 2 is classified and identified using the KNN algorithm combined with the radio frequency fingerprint library established in step 2. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of the process of obtaining the calibration coefficient of the calibrator in the present invention;

[0017] Figure 2 This is a schematic diagram of the working process under environment 2 of the present invention;

[0018] Figure 3 It is the iterative flow chart of the adaptive filter; DETAILED DESCRIPTION

[0019] The present invention will be further described below in conjunction with the accompanying drawings:

[0020] The technical solution adopted by the present invention is: a method for extracting and identifying fingerprint features of signals from the same radiation source under various environments, specifically comprising the following steps:

[0021] Step 1: In the following environment, use an X310 and a B210 USRP software radio platform as the radiation source, and transmit the QPSK signal. Use a superheterodyne receiver to receive and save the transmission signals of the two radiation sources.

[0022] Step 2: Based on step 1, the box dimension and radial integral bispectrum of the two radiation source signals are extracted as fingerprint features, and the KNN algorithm is used for machine learning to establish a radio frequency fingerprint library.

[0023] Step 3: In environment 2, use the receiver to receive and save signals from the two radiation sources.

[0024] Step 4: Based on steps 1 and 2, the radiation source signal received in environment 1 and the radiation source signal received in environment 2 are input into the adaptive filter, and the calibration coefficient is obtained through the adaptive algorithm iteration. The specific steps are as follows:

[0025] 4.1 The radiation source signal obtained in environment 1 is the desired signal d(n), and the radiation source signal obtained in environment 2 is the input signal X(n) = [x(n), x(n-1), ..., x(n-M+1)] T , the tap weight vector of the filter, i.e. the calibration coefficient, is w(n)=[w0(n),w1(n),...,w M-1 (n)] T , where M is the filter order, the filter output signal is:

[0026]

[0027] 4.2 Subtracting the output signal from the expected signal, we can get the error signal e(n):

[0028] e(n)=d(n)-y(n)=d(n)-w T (n)x(n)

[0029] 4.3 Using mean square error to construct the cost function is:

[0030] ξ(n)=E[d 2 (n)]-2E[d(n)w T (n)X(n)]+E[w T (n)X(n)X T (n)w(n)]

[0031] 4.4 When the filter weights tend to be stable

[0032] p=E[d(n)X T (n)]

[0033] R=E[X(n)X T (n)]

[0034] Among them, p is the cross-correlation between the expected signal and the input signal, R is the autocorrelation matrix of the input signal, and the cost function is:

[0035] ζ(n)=E[d 2 (n)]-2w T p+w T R

[0036] 4.5 The cost function is a quadratic function. When the cost function is minimum, the filter weight coefficient is the optimal solution. By taking the derivative of the weight vector, we can get:

[0037]

[0038] when When , the optimal solution can be obtained as:

[0039] W opt =R -1 p

[0040] 4.6 According to the LMS algorithm ζ(n)≈e 2 (n), then the instantaneous value of the gradient is:

[0041]

[0042] 4.7 According to the instantaneous value of the gradient, the LMS weight update formula can be obtained as:

[0043]

[0044] Step 5: Based on step 4, the radiation source signal collected in the second environment is filtered using the adaptive filter with the calibration coefficient obtained.

[0045] Step 6: Based on step 5 and step 2, the box dimension and radial integral bispectrum of the processed signal are extracted, and the radiation source in environment 2 is classified and identified using the KNN algorithm combined with the RF fingerprint library established in step 2.

Claims

1. A method for extracting and identifying fingerprint features of signals from the same radiation source under various environments, specifically comprising the following steps: Step 1: In the following environment, use an X310 and a B210 USRP software radio platform as the radiation source, the transmission signal is a QPSK signal, and a superheterodyne receiver is used to receive and save the transmission signals of the two radiation sources; Step 2: Based on step 1, the box dimension and radial integral bispectrum of the two radiation source signals are extracted as fingerprint features, and the KNN algorithm is used for machine learning to establish a radio frequency fingerprint library; Step 3: In environment 2, use the receiver to receive and save the signals from the two radiation sources; Step 4: Based on steps 1 and 2, the radiation source signal received in environment 1 and the radiation source signal received in environment 2 are input into the adaptive filter, and the calibration coefficient is obtained through the adaptive algorithm iteration. The specific steps are as follows: 4.1 The radiation source signal obtained in environment 1 is the desired signal d(n), and the radiation source signal obtained in environment 2 is the input signal X(n) = [x(n), x(n-1), ..., x(n-M+1)] T , the tap weight vector of the filter, i.e. the calibration coefficient, is w(n)=[w0(n),w1(n),...,w M-1 (n)] T , where M is the filter order, the filter output signal is: 4.2 Subtracting the output signal from the expected signal, we can get the error signal e(n): e(n)=d(n)-y(n)=d(n)-w T (n)x(n) 4.3 Using mean square error to construct the cost function is: ξ(n)=E[d 2 (n)]-2E[d(n)w T (n)X(n)]+E[w T (n)X(n)X T (n)w(n)] 4.4 When the filter weights tend to be stable p=E[d(n)X T (n)] R=E[X(n)X T (n)] in, p is the cross-correlation between the expected signal and the input signal, R is the autocorrelation matrix of the input signal, and the cost function is: ζ(n)=E[d 2 (n)]-2w T p+w T Rw 4.5 The cost function is a quadratic function. When the cost function is minimum, the filter weight coefficient is the optimal solution. By taking the derivative of the weight vector, we can get: when When , the optimal solution can be obtained as: W opt =R -1 p 4.6 According to the LMS algorithm ζ(n)≈e 2 (n), then the instantaneous value of the gradient is: 4.7 According to the instantaneous value of the gradient, the LMS weight update formula can be obtained as: Step 5: Based on step 4, the adaptive filter with the calibration coefficient obtained is used to filter the radiation source signal collected from the second environment; Step 6: Based on step 5 and step 2, the box dimension and radial integral bispectrum of the processed signal are extracted, and the radiation sources in environment 2 are classified and identified using the KNN algorithm combined with the RF fingerprint library established in step 2.

Images