Lte-v2x radio frequency fingerprint extraction and identification method and system based on sidelink channel

Abstract

The application relates to the field of radio frequency fingerprint extraction and recognition, and discloses an LTE-V2X radio frequency fingerprint extraction and recognition method and system based on a sidelink channel, which adopts an adjacent double sliding window power detection algorithm to capture an LTE-V2X signal and perform pretreatment; extracts fingerprint spectrum features from a sidelink physical broadcast channel through an SC-FDM demodulation algorithm and a radio frequency fingerprint feature based on a synchronization symbol; for a sidelink physical data channel, generates a constellation diagram through an SC-FDM decoding precoding algorithm, compensates the phase of the constellation diagram through an adaptive weighted sliding window, sets different weights to generate a fingerprint brightness diagram; extracts fingerprint brightness diagram features, maps the fingerprint spectrum features to the fingerprint brightness diagram feature space; introduces a cross-attention module to calculate the cross-attention between the fingerprint spectrum features and the fingerprint brightness diagram features to perform feature fusion and obtain fused fingerprint features; and obtains a final fingerprint classification result according to the fused fingerprint features.

Description

LTE-V2X RF fingerprint extraction and identification method and system based on direct link channel Technical Field

[0001] This invention relates to the field of radio frequency fingerprint extraction and recognition, and specifically to an LTE-V2X radio frequency fingerprint extraction and recognition method and system based on a direct link channel. Background Technology

[0002] Due to the complexity and high security standards of the connected vehicle environment, secure and reliable identity authentication technologies are required. Radio frequency fingerprints, as an important physical characteristic of communication transmitters, possess uniqueness, stability, and difficulty in cloning, making them a crucial research direction for identity authentication within the field of physical layer security research.

[0003] Most existing LTE-V2X RF fingerprint extraction methods rely on physical random access channels for RF fingerprint extraction followed by neural network recognition. They neglect the need to address the larger number of broadcast and data channels. This is because LTE-V2X signals are wideband and have complex modulation schemes and transmission protocols, leading to instability and difficulty in obtaining extracted RF fingerprints. Therefore, designing an LTE-V2X RF fingerprint extraction and recognition method and system based on direct-access channels is essential to improve the stability and accuracy of RF fingerprint extraction and recognition. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes an LTE-V2X radio frequency fingerprint extraction and identification method and system based on a direct link channel. By capturing and preprocessing LTE-V2X signals, fingerprint spectral features are extracted from the direct link physical broadcast channel and fingerprint brightness maps are generated from the physical data channel. After feature mapping and cross-attention fusion, accurate classification of radio frequency fingerprints is achieved.

[0005] The technical solution to achieve the objective of this invention is as follows:

[0006] The LTE-V2X radio frequency fingerprint extraction and identification method based on the direct link channel includes the following steps:

[0007] The adjacent double sliding window power detection algorithm is used to capture LTE-V2X signals in units of subframes from the original received signal. Let the [number]th subframe be [details of subframe]. The initial LTE-V2X signal corresponding to each subframe is Subsequently, the starting position of the signal is accurately located using a symbol cross-correlation time synchronization algorithm based on CP to obtain the synchronized LTE-V2X signal. The frequency offset estimation algorithm based on synchronization symbols is used to analyze the synchronization LTE-V2X signal. Frequency offset compensation is performed, and phase compensation is also performed based on the phase deviation between the original received signal and the ideal signal to obtain the final LTE-V2X signal. ;

[0008] For the direct-link physical broadcast channel, the SC-FDM demodulation algorithm is used to demodulate the final LTE-V2X signal. The direct link physical broadcast channel signal frame is demodulated to obtain the demodulated LTE-V2X signal. Subsequently, the physical ID number of the direct link physical broadcast channel signal frame is obtained using the synchronization symbol calculation method, and the corresponding demodulation reference symbol DMRS is calculated based on this ID number. This is then used to demodulate the LTE-V2X signal. The frequency domain estimate is obtained by performing channel equalization on the symbols of the cut-through physical broadcast channel signal frame using the demodulation reference symbol (DMRS). Finally, the fingerprint spectral features are obtained using a radio frequency fingerprint feature extraction method based on synchronization symbols. ;

[0009] For the direct link physical data channel, the final LTE-V2X signal After bandwidth detection, the pass-through physical data channel signal frame is demodulated using the SC-FDM demodulation algorithm, and then precoded using the SC-FDM de-precoding algorithm to generate the standard QPSK constellation diagram of the pass-through physical data channel signal frame. Based on the trend of phase shift between different constellation points, an adaptive weighted sliding window is used to apply an algorithm to the standard QPSK constellation diagram for the phase difference sequence of constellation points. Phase compensation is performed; subsequently, during the conversion of the constellation map into a fingerprint luminance map, different weights are assigned to pixels in different constellation point density regions based on kernel density to generate the fingerprint luminance map. ;

[0010] Fingerprint brightness map is analyzed using convolutional and pooling layers in a convolutional neural network. Feature extraction is performed to obtain fingerprint brightness map features. And the fingerprint spectral features are mapped using a feature mapping matrix. The fingerprint spectral features are mapped into the fingerprint luminance map feature space; subsequently, a cross-attention module is introduced to calculate the attention of the fingerprint spectral features to the fingerprint luminance map features. Attention to fingerprint spectral features and fingerprint brightness map features The fused fingerprint features are obtained by fusing features based on the calculated cross-attention. ; will integrate fingerprint features The fingerprint is input into a predictor consisting of an output layer to obtain the final fingerprint classification result. ;

[0011] Furthermore, LTE-V2X signals are wideband signals. According to the frame structure format of the direct-link physical data channel, in standard mode, a subframe contains 14 SC-FDM symbols. The last SC-FDM symbol is used as a guard symbol and is not modulated with data. Therefore, there is still a gap between two consecutive subframes. Since the smallest unit for the same device to transmit data in the time domain is a subframe, and the same device will not transmit data continuously in the time domain, LTE-V2X signal acquisition is performed in units of subframes.

[0012] Furthermore, an adjacent double sliding window power detection algorithm is employed. Two adjacent sliding windows are constructed in the time domain. The original received signal is passed through the sliding windows in chronological order. The average signal power within the adjacent sliding windows is then calculated. The position of the subframe signal start point is determined by whether the power ratio between the two windows exceeds a threshold. For example, the time series of the original received signal is denoted as... The lengths of the two sliding windows are The average signal power within the two sliding windows and The calculation method is to average the squared signal values ​​at each position within the window, and then calculate the average power ratio of the two sliding windows; when the average power ratio is... When a certain threshold is exceeded, it indicates that data has begun to be captured, and the peak value of the signal segment exceeding the threshold is taken as the starting position of a received subframe signal;

[0013] Furthermore, regarding the first The initial LTE-V2X signal corresponding to each subframe Extraction and preprocessing are performed to obtain the final LTE-V2X signal. Includes the following steps:

[0014] Two sliding windows with a distance of one SC-FDM symbol are constructed in the time domain, and the initial LTEX-V2X signals are processed in chronological order. By using a sliding window, symbol cross-correlation based on CP is calculated across the entire signal subframe to obtain the cross-correlation coefficient between CP and each SC-FDMA symbol. The cross-correlation operations of all SC-FDMA symbols are superimposed, and the sampling point with the largest sum of cross-correlation magnitudes is used as the initial LTE X-V2X signal. The precise starting position to obtain synchronized LTE-V2X signals ;

[0015] Using a frequency offset estimation algorithm based on cyclic prefix, and on the basis of time synchronization, the frequency offset is estimated by calculating the difference in phase angle between the CP and the end segment of the symbol. The frequency offset is then estimated by calculating complex correlation values, and applied to the synchronization of LTE-V2X signals. The final frequency offset estimate is calculated by averaging multiple CPs and the end segments of symbols. Based on frequency offset estimation For synchronous LTE-V2X signals After frequency offset compensation, the LTE-V2X signal is synchronized by calculation. The initial phase deviation of the actual signal is obtained by calculating the complex correlation value between the actual signal and the ideal signal. Then, the average value of the initial phase deviation is calculated to obtain the estimated phase deviation value. Finally, phase compensation is performed on the signal after frequency offset compensation to obtain the final LTE-V2X signal. ;

[0016] Furthermore, regarding the initial LTE-V2X signal... Time synchronization is performed to obtain synchronized LTE-V2X signals. Includes the following steps:

[0017] Due to the initial LTE-V2X signal It contains 14 SC-FDM symbols. The SC-FDM symbols are designed using a CP+symbol structure, with different symbols corresponding to different CP lengths. The real and imaginary parts of the last part of each CP and SC-FDM symbol are of the same value but opposite in direction. Based on this, since a subframe contains 13 symbols and 1 guard symbol, the modulus of the cross-correlation coefficient between each symbol and its corresponding CP signal is calculated and superimposed. The sampling point with the largest sum of cross-correlation moduli is used as the initial LTE-V2X signal. The precise starting position to obtain synchronized LTE-V2X signals ;

[0018] Furthermore, regarding the synchronization of LTE-V2X signals Frequency offset compensation and phase compensation are performed to obtain the final LTE-V2X signal. Includes the following steps:

[0019] A frequency offset estimation algorithm based on cyclic prefix is ​​used to calculate the frequency offset by measuring the difference between the phase angle of the CP signal and the phase angle of the corresponding SC-FDM symbol's last segment. Since the CP signal and the corresponding SC-FDM symbol's last segment have the same ideal amplitude but opposite signs, the frequency offset is estimated by calculating the complex correlation value between them, and corresponding frequency offset compensation is performed to obtain the frequency offset compensated LTE-V2X signal. ;

[0020] Due to the initial phase difference, there is a constant phase difference between the transmitted and received signals; in frequency offset compensation LTE-V2X signals The initial phase deviation of the actual signal is obtained by calculating the complex correlation value between the actual signal and the ideal signal, and then the phase deviation is obtained by calculating the first complex correlation value between the actual signal and the ideal signal. The average initial phase deviation of all points in each subframe is used as the phase deviation estimate of the entire subframe signal, and the frequency offset compensated LTE-V2X signal is then used. and Multiplication is used for phase compensation to obtain the final LTE-V2X signal. ;

[0021] Furthermore, a radio frequency fingerprint feature extraction method based on synchronization symbols is employed to extract features from the final LTE-V2X signal. Fingerprint spectral features are obtained from the physical broadcast channel signal frames of the direct link. Includes the following steps:

[0022] The final LTE-V2X signal is demodulated using the SC-FDM algorithm. The direct link physical broadcast channel signal frame is demodulated to obtain the demodulated LTE-V2X signal. Then, the physical ID number of the direct link physical broadcast channel signal frame is obtained by using the synchronization symbol calculation method, and the corresponding demodulation reference symbol DMRS is calculated based on the ID number;

[0023] Based on demodulating LTE-V2X signals The frequency domain estimate is obtained by performing channel equalization on the symbols of the cut-through physical broadcast channel signal frame using the demodulation reference symbol (DMRS). Finally, the fingerprint spectral features are obtained using a radio frequency fingerprint feature extraction method based on synchronization symbols. ;

[0024] Furthermore, the direct link physical broadcast channel signal frame consists of the direct link physical broadcast symbol PSBCH, the primary synchronization symbol PSSS, the secondary synchronization symbol SSSS, and the demodulation reference symbol DMRS. The primary synchronization symbol is located in bits 2 and 3, the secondary synchronization symbol in bits 12 and 13, and the demodulation reference symbol in bits 5, 7, and 10. First, the final LTE-V2X signal is demodulated using the SC-FDM demodulation algorithm. Demodulate each symbol on the direct-link physical broadcast channel signal frame; for example, since each SC-FDM symbol includes a cyclic prefix (CP) and a symbol segment, the latter part of the cyclic prefix and the former part of the symbol segment are selected as the Fast Fourier Transform signal for demodulation. Therefore, the following settings are used: The value is used as a percentage when using a cyclic prefix; subsequently, the signal is half-wave shifted to avoid the effects of non-proportional high crosstalk; finally, the LTE-V2X signal... The Middle a symbol The shifted LTE-V2X signal is obtained after half-wave shifting. Subsequently, a Fast Fourier Transform (FFT) is performed on the shifted LTE-V2X signal. Since a portion of the cyclic prefix was removed during modulation, phase compensation is required on the signal obtained after the FFT. That is, the signal obtained after phase compensation... The demodulated LTE-V2X signal of each symbol is ;

[0025] In a direct-link physical broadcast channel signal frame, there are two types of primary synchronization symbols (PSSS) and 168 types of secondary synchronization symbols. The physical layer ID of the direct-link physical broadcast channel is determined by calculating the correlation coefficient between the synchronization symbols and the local synchronization symbols in the signal. According to the construction principle of the primary synchronization symbol PSSS, its sequence ID... This includes two possibilities, and the sequence ID of the secondary synchronization symbol SSSS. There are 168 possibilities; since the generation sequence of the synchronization symbol is The sequence has the same length as the number of subcarriers on the synchronization symbol, which is 62. For different sequences, the cross-correlation between the two sequences is very low. The Pearson correlation coefficient is used to calculate the correlation coefficient between the synchronization symbol in the signal and the local synchronization symbol. Based on the calculation results, the corresponding auxiliary synchronization symbol ID number and correlation coefficient diagram are drawn to obtain the final auxiliary synchronization symbol ID.

[0026] In the direct-link physical broadcast channel signal frame, the demodulation reference symbol DMRS is located at positions 5, 7, and 10. Similar to the synchronization symbol, DMRS also uses... A sequence, where each complex value in the sequence has an amplitude of 1; for example, for the same subframe, the above three symbols are based on... Different methods are used or The orthogonal sequence; channel equalization is performed using the two primary synchronization symbols with the first DMRS symbol as the 2nd and 3rd bits, and the two secondary synchronization symbols with the third DMRS symbol as the 12th and 13th bits;

[0027] Furthermore, channel equalization is achieved through channel estimation. In the frequency domain, a least squares (LS) channel estimation algorithm, which does not consider noise, is used to perform channel equalization on the synchronization symbols. For example, the first... Demodulating LTE-V2X signals in individual frames Taking the primary synchronization symbol as an example, since the primary synchronization symbol and DMRS pass through the same channel, they have the same frequency domain response, denoted as . Subsequently, based on the frequency domain LS estimation algorithm, the channel estimate of the primary synchronization symbol can be obtained. Using this channel estimate, channel equalization is performed on the primary synchronization symbol to obtain the frequency domain estimate of the channel-equalized primary synchronization symbol. Similarly, the frequency domain estimate of the secondary synchronization symbol can be obtained. ;

[0028] Subsequently, a radio frequency fingerprint extraction method based on synchronization symbols is used for fingerprint extraction. Two primary synchronization symbols and two secondary synchronization symbols are denoted as PSSS1, PSSS2, SSSS1, and SSSS2. Multiple identical synchronization symbols can be superimposed to eliminate the influence of noise on the fingerprint. The fingerprint information from both the primary and secondary synchronization symbols is obtained and used as a whole to identify the fingerprint spectral characteristics of a subframe, denoted as... ;

[0029] Furthermore, a data symbol-based radio frequency fingerprinting method is employed to extract the final LTE-V2X signal. Fingerprint luminance map obtained from the physical data channel signal frame of the direct link. Includes the following steps:

[0030] By superimposing the final LTE-V2X signal The spectral amplitudes of all demodulated reference symbols are calculated, and the bandwidth of the signal frame is determined using adjacent sliding windows and threshold decisions. Subsequently, the SC-FDM demodulation algorithm is used for demodulation, and the standard QPSK constellation diagram is generated using the SC-FDM deprecoding algorithm. ;

[0031] An adaptive weighted sliding window method is used to perform phase compensation based on the fluctuation of the phase difference between adjacent constellation points and the order of the constellation points, resulting in a corrected QPSK constellation diagram. The QPSK constellation chart will be revised. The image is converted to grayscale, and the constellation point density of each pixel is calculated using a Gaussian kernel function. Weights are then set based on the constellation point density of the pixels to generate a fingerprint brightness map. ;

[0032] Furthermore, the final LTE-V2X signal The direct-link physical data channel signal frame consists of data symbols, a demodulation reference signal, and a guard interval. The demodulation reference signal is located in bits 3, 6, 9, and 12, the guard interval is located in bit 14, and the remaining bits are data symbols. Two adjacent sliding windows are constructed in the frequency domain. The sum of the amplitude spectra within the adjacent sliding windows is calculated. The end and start positions are determined based on the ratio of the amplitude spectrum sums between the two windows, and the bandwidth of the signal frame is calculated. Subsequently, similar to the signal frames in the direct link physical broadcast channel, the final LTE-V2X signal is processed through half-wave shift, fast Fourier transform, and phase offset compensation in the direct link physical data channel. Demodulate;

[0033] In the direct link physical data channel for the final LTE-V2X signal After demodulation, the modulated signal is decoded using the SC-FDM de-precoding algorithm; subsequently, it is decoded using IFFT transform, and the resulting constellation symbol is multiplied inversely. The amplitude of the constellation is restored to obtain the first... Standard QPSK constellation diagram corresponding to each subframe ;

[0034] Furthermore, the final LTE-V2X signal totals Each subframe is generated through the above steps. A standard QPSK constellation diagram is presented. Considering that the phase offset between subframes in LTE-V2X typically changes slowly and linearly, while ordinary linear fitting uses a fixed window for global fitting, which accumulates fitting errors over long time periods and struggles to handle phase anomalies in a single frame caused by sudden interference, an adaptive weighted sliding window method is adopted to address these issues. This method weights the phase difference fluctuations between adjacent constellation points and the constellation point order to perform phase compensation, resulting in a corrected QPSK constellation diagram. ;

[0035] According to the revised QPSK constellation chart Plot the grayscale images corresponding to the real and imaginary parts of each constellation point; treat the real part data as... Axis, imaginary part data is considered The corresponding grayscale image is plotted along the axis, and the constellation point density of each pixel in the grayscale image is calculated using a Gaussian kernel function. The calculated constellation point density of each pixel is then used to calculate the brightness value of the corresponding pixel to generate a modified QPSK constellation map. Corresponding fingerprint brightness map ;

[0036] Furthermore, the first Fingerprint spectral features of each subframe fingerprint brightness map The fusion process yields the fused fingerprint features. It will integrate fingerprint features The fingerprint is input into a predictor to obtain the fingerprint classification result;

[0037] Furthermore, a convolutional neural network consisting of two convolutional layers, two pooling layers, and one fully connected layer is used to... fingerprint brightness map of each subframe The fingerprint brightness map features are obtained through processing. Its dimensions are 128;

[0038] Furthermore, based on the extraction of synchronization symbols, the first... Fingerprint spectral features of each subframe It contains 62 real-part features and 62 imaginary-part features extracted from the primary synchronization symbol PSSS, 62 real-part features and 62 imaginary-part features extracted from the secondary synchronization symbol, and 1 physical ID feature. Therefore, the first... Fingerprint spectral features of each subframe The dimension is 249; due to the fingerprint brightness map features The dimension is 128, therefore the feature mapping matrix The size is Using the feature mapping matrix The mapped fingerprint spectral features are obtained by mapping the fingerprint spectral features onto the fingerprint luminance map feature space. ;

[0039] Furthermore, a cross-attention mechanism is introduced to calculate the attention of fingerprint spectral features to fingerprint luminance map features. Attention to fingerprint spectral features and fingerprint brightness map features After aligning the fingerprint spectral features and fingerprint luminance map features to the same space using a feature mapping matrix, they are normalized to avoid the influence of dimensional differences. Subsequently, the similarity matrix is ​​obtained by calculating the similarity between the normalized features. ;

[0040] Based on similarity matrix Calculate the attention of fingerprint spectral features to fingerprint luminance map features separately. Attention to fingerprint spectral features and fingerprint brightness map features Based on these two attention points, the fingerprint brightness map features are analyzed respectively. and mapped fingerprint spectral features After weighting and fusing, the fused fingerprint features are obtained. ; will integrate fingerprint features The fingerprint is input into a predictor consisting of an output layer to obtain the final fingerprint classification result. .

[0041] Furthermore, this invention proposes an LTE-V2X radio frequency fingerprint extraction and identification system based on a direct link channel, including an acquisition and processing module, a spectrum feature module, an image feature module, and a fingerprint classification module;

[0042] The acquisition and processing module uses an adjacent dual sliding window power detection algorithm to capture LTE-V2X signals from the original received signals. Then, it uses a symbol cross-correlation time synchronization algorithm based on CP to accurately locate the starting position of the signal to obtain a synchronized LTE-V2X signal. The synchronized LTE-V2X signal is compensated for frequency offset using a frequency offset estimation algorithm based on synchronization symbols, and the final LTE-V2X signal is obtained by performing phase compensation based on the phase deviation between the original received signal and the ideal signal.

[0043] The spectrum feature module, targeting the direct-link physical broadcast channel, demodulates the direct-link physical broadcast channel signal frames in the final LTE-V2X signal using the SC-FDM demodulation algorithm to obtain the demodulated LTE-V2X signal. Subsequently, the physical ID number of the direct link physical broadcast channel signal frame is obtained by using the synchronization symbol calculation method, and the corresponding demodulation reference symbol DMRS is calculated based on the ID number. Based on the demodulated LTE-V2X signal and the demodulation reference symbol DMRS, channel equalization is performed on the symbols of the direct link physical broadcast channel signal frame to obtain the frequency domain estimate. Finally, the fingerprint spectrum features are obtained by using the radio frequency fingerprint feature extraction method based on the synchronization symbol.

[0044] The image feature module, targeting the direct-link physical data channel, performs bandwidth detection on the final LTE-V2X signal frame of the direct-link physical data channel, then demodulates it using the SC-FDM demodulation algorithm, and finally generates a standard QPSK constellation diagram of the direct-link physical data channel signal frame using the SC-FDM deprecoding algorithm. Based on the phase shift trend between different constellation points, an adaptive weighted sliding window is used to perform phase compensation on the standard QPSK constellation diagram for the constellation point phase difference sequence. Subsequently, in the process of converting the constellation diagram into a fingerprint brightness map, different weights are set for pixels in different constellation point density regions according to the kernel density to generate the fingerprint brightness map.

[0045] The fingerprint classification module extracts fingerprint luminance features from the fingerprint luminance map using convolutional and pooling layers in a convolutional neural network, and maps the fingerprint spectral features to the fingerprint luminance map feature space using a feature mapping matrix. Subsequently, a cross-attention module is introduced to calculate the attention of the fingerprint spectral features to the fingerprint luminance map features and the attention of the fingerprint luminance map features to the fingerprint spectral features, respectively. Based on the calculated cross-attention, feature fusion is performed to obtain fused fingerprint features. The fused fingerprint features are then input into a predictor consisting of an output layer to obtain the final fingerprint classification result.

[0046] Furthermore, the acquisition and processing module includes a signal acquisition unit, a signal synchronization unit, and a signal processing unit;

[0047] The signal acquisition unit uses an adjacent dual sliding window power detection algorithm to capture LTE-V2X signals in units of subframes from the original received signal;

[0048] The signal synchronization unit uses a symbol cross-correlation time synchronization algorithm based on CP to accurately locate the starting position of the signal and obtain a synchronized LTE-V2X signal.

[0049] The signal processing unit uses a frequency offset estimation algorithm based on synchronization symbols to perform frequency offset compensation on the synchronous LTE-V2X signal, and performs phase compensation based on the phase deviation between the original received signal and the ideal signal to obtain the final LTE-V2X signal.

[0050] Furthermore, an adjacent double sliding window power detection algorithm is adopted to construct two adjacent sliding windows in the time domain. The original received signal is passed through the sliding windows in chronological order. Then, the average signal power in the adjacent sliding windows is calculated, and the position of the subframe signal start point is obtained by judging whether the power ratio between the two windows exceeds the threshold.

[0051] Furthermore, the initial LTE-V2X signal is extracted and preprocessed to obtain the final LTE-V2X signal. Includes the following steps:

[0052] Two sliding windows with a distance of one SC-FDM symbol are constructed in the time domain. The initial LTEX-V2X signal is passed through the sliding windows in chronological order. Symbol cross-correlation calculation based on CP is performed on the entire signal subframe to obtain the cross-correlation coefficient between CP and each SC-FDMA symbol. The cross-correlation operations of all SC-FDMA symbols are superimposed, and the sampling point with the largest sum of cross-correlation magnitudes is calculated as the precise starting position of the initial LTEX-V2X signal to obtain the synchronized LTE-V2X signal.

[0053] Using a frequency offset estimation algorithm based on cyclic prefixes, the frequency offset is estimated by calculating the difference in phase angle between the CP and the symbol tail segment, based on time synchronization. Frequency offset estimation is performed by calculating complex correlation values. The final frequency offset estimate is calculated by averaging the frequency offset estimates of multiple CPs and symbol tail segments in the synchronized LTE-V2X signal. Based on the frequency offset estimate... After frequency offset compensation of the synchronous LTE-V2X signal, the initial phase deviation of the actual signal is obtained by calculating the complex correlation value between the synchronous LTE-V2X signal and the ideal signal. Then, the average value of the initial phase deviation is calculated to obtain the phase deviation estimate. Finally, phase compensation is performed on the signal after frequency offset compensation to obtain the final LTE-V2X signal.

[0054] Furthermore, the spectrum feature module includes a signal demodulation unit, a signal equalization unit, and a feature extraction unit;

[0055] The signal demodulation unit demodulates the through-link physical broadcast channel signal frame in the final LTE-V2X signal using the SC-FDM demodulation algorithm to obtain the demodulated LTE-V2X signal;

[0056] The signal equalization unit uses the synchronization symbol calculation method to obtain the physical ID number of the direct link physical broadcast channel signal frame and calculates the corresponding demodulation reference symbol DMRS based on the ID number. Based on the demodulated LTE-V2X signal and the demodulation reference symbol DMRS, the unit performs channel equalization on the symbols of the direct link physical broadcast channel signal frame to obtain the frequency domain estimate.

[0057] The feature extraction unit uses a radio frequency fingerprint feature extraction method based on synchronization symbols to obtain fingerprint spectral features;

[0058] Furthermore, the method of extracting fingerprint spectral features from the direct link physical broadcast channel signal frame in the final LTE-V2X signal using a synchronization symbol-based radio frequency fingerprint feature extraction method includes the following steps:

[0059] The SC-FDM demodulation algorithm is used to demodulate the direct link physical broadcast channel signal frame in the final LTE-V2X signal to obtain the demodulated LTE-V2X signal. Then, the physical ID number of the direct link physical broadcast channel signal frame is obtained by using the synchronization symbol calculation method, and the corresponding demodulation reference symbol DMRS is calculated based on the ID number.

[0060] Based on the demodulated LTE-V2X signal and demodulated reference symbol DMRS, channel equalization is performed on the symbols of the physical broadcast channel signal frame of the direct link to obtain the frequency domain estimate. Finally, the RF fingerprint feature extraction method based on synchronization symbol is used to obtain the fingerprint spectrum features.

[0061] Furthermore, the image feature module includes a decoding unit and a brightness map unit;

[0062] After bandwidth detection, the decoding unit demodulates the final LTE-V2X signal through the pass-through physical data channel signal frame using the SC-FDM demodulation algorithm, and then generates the standard QPSK constellation diagram of the pass-through physical data channel signal frame using the SC-FDM deprecoding algorithm.

[0063] The luminance map unit performs phase compensation on the standard QPSK constellation map by using an adaptive weighted sliding window for the constellation point phase difference sequence based on the trend of phase shift between different constellation points. Subsequently, in the process of converting the constellation map into a fingerprint luminance map, different weights are set for pixels in different constellation point density regions according to the kernel density to generate the fingerprint luminance map.

[0064] Furthermore, the method of extracting fingerprint brightness map features from the direct-link physical data channel signal frame in the final LTE-V2X signal using a data symbol-based radio frequency fingerprint extraction method includes the following steps:

[0065] The bandwidth of the signal frame is determined by superimposing the spectral amplitudes of all demodulated reference symbols in the final LTE-V2X signal and using adjacent sliding windows and threshold decisions. Then, the SC-FDM demodulation algorithm is used for demodulation, and the standard QPSK constellation diagram is generated by the SC-FDM deprecoding algorithm.

[0066] An adaptive weighted sliding window method is adopted to perform phase compensation based on the fluctuation of the phase difference between adjacent constellation points and the order of constellation points to obtain a corrected QPSK constellation map. The corrected QPSK constellation map is converted into a grayscale image, and the constellation point density of each pixel is calculated by using a Gaussian kernel function. The weights are set according to the constellation point density of the pixels, and fingerprint brightness map features are generated.

[0067] Furthermore, the fingerprint classification module includes a feature processing unit, a feature fusion unit, and a fingerprint classification unit;

[0068] The feature processing unit extracts fingerprint brightness map features by using convolutional and pooling layers in the convolutional neural network, and maps the fingerprint spectral features to the fingerprint brightness map feature space through a feature mapping matrix;

[0069] The feature fusion unit introduces a cross-attention module to calculate the attention of fingerprint spectral features to fingerprint brightness map features and the attention of fingerprint brightness map features to fingerprint spectral features, respectively. Based on the calculated cross-attention, feature fusion is performed to obtain fused fingerprint features.

[0070] The fingerprint classification unit inputs the fused fingerprint features into a predictor consisting of an output layer to obtain the final fingerprint classification result;

[0071] Furthermore, the fingerprint spectral features and fingerprint brightness map are fused to obtain fused fingerprint features, which are then input into a predictor to obtain fingerprint classification results.

[0072] Furthermore, a convolutional neural network consisting of two convolutional layers, two pooling layers, and one fully connected layer is used to process the fingerprint luminance map to obtain fingerprint luminance map features; the fingerprint spectral features are then mapped into the fingerprint luminance map feature space through a feature mapping matrix.

[0073] Furthermore, a cross-attention mechanism is introduced to calculate the attention of fingerprint spectral features to fingerprint brightness map features and the attention of fingerprint brightness map features to fingerprint spectral features. The fingerprint spectral features and fingerprint brightness map features are normalized to avoid the influence of dimensional differences, and then the similarity between the normalized features is calculated to obtain a similarity matrix. Based on the similarity matrix, the attention of fingerprint spectral features to fingerprint brightness map features and the attention of fingerprint brightness map features to fingerprint spectral features are calculated respectively. The fingerprint brightness map features and the mapped fingerprint spectral features are weighted according to these two attentions and then fused to obtain the fused fingerprint features. The fused fingerprint features are input into a predictor consisting of an output layer to obtain the final fingerprint classification result.

[0074] Compared with existing technologies, this invention employs an adjacent dual sliding window power detection algorithm to capture LTE-V2X signals and perform preprocessing; it extracts fingerprint spectral features from the direct-link physical broadcast channel using an SC-FDM demodulation algorithm and RF fingerprint features based on synchronization symbols; for the direct-link physical data channel, it generates a constellation diagram using an SC-FDM deprecoding algorithm, performs phase compensation on the constellation diagram using an adaptive weighted sliding window, and sets different weights according to pixel kernel density to generate a fingerprint brightness map; it extracts fingerprint brightness map features and maps the fingerprint spectral features to the fingerprint brightness map feature space using a feature mapping matrix; subsequently, it introduces a cross-attention module to calculate the cross-attention between the fingerprint spectral features and the fingerprint brightness map features respectively, and performs feature fusion to obtain fused fingerprint features; and finally, it obtains the fingerprint classification result based on the fused fingerprint features. Attached Figure Description

[0075] Figure 1 is a flowchart of the LTE-V2X radio frequency fingerprint extraction and identification method based on the direct link channel;

[0076] Figure 2 is a flowchart of LTE-V2X signal extraction and preprocessing;

[0077] Figure 3 is a flowchart of fingerprint spectral feature extraction;

[0078] Figure 4 is a flowchart of fingerprint brightness map extraction;

[0079] Figure 5 is a block diagram of an LTE-V2X radio frequency fingerprint extraction and identification system based on a direct link channel. Detailed Implementation

[0080] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

[0081] As shown in Figure 1, a specific embodiment of the present invention discloses an LTE-V2X radio frequency fingerprint extraction and identification method based on a direct link channel, comprising the following steps:

[0082] The adjacent double sliding window power detection algorithm is used to capture LTE-V2X signals in units of subframes from the original received signal. Let the [number]th subframe be [details of subframe]. The initial LTE-V2X signal corresponding to each subframe is Subsequently, the starting position of the signal is accurately located using a symbol cross-correlation time synchronization algorithm based on CP to obtain the synchronized LTE-V2X signal. The frequency offset estimation algorithm based on the cyclic prefix is ​​used to estimate the synchronous LTE-V2X signal. Frequency offset compensation is performed, and phase compensation is also performed based on the phase deviation between the original received signal and the ideal signal to obtain the final LTE-V2X signal. ;

[0083] For the direct-link physical broadcast channel, the SC-FDM demodulation algorithm is used to demodulate the final LTE-V2X signal. The direct link physical broadcast channel signal frame is demodulated to obtain the demodulated LTE-V2X signal. Subsequently, the physical ID number of the direct link physical broadcast channel signal frame is obtained using the synchronization symbol calculation method, and the corresponding demodulation reference symbol DMRS is calculated based on this ID number. This is then used to demodulate the LTE-V2X signal. The frequency domain estimate is obtained by performing channel equalization on the symbols of the cut-through physical broadcast channel signal frame using the demodulation reference symbol (DMRS). Finally, the fingerprint spectral features are obtained using a radio frequency fingerprint feature extraction method based on synchronization symbols. ;

[0084] For the direct link physical data channel, the final LTE-V2X signal After bandwidth detection, the pass-through physical data channel signal frame is demodulated using the SC-FDM demodulation algorithm, and then precoded using the SC-FDM de-precoding algorithm to generate the standard QPSK constellation diagram of the pass-through physical data channel signal frame. Based on the trend of phase shift between different constellation points, an adaptive weighted sliding window is used to apply an algorithm to the standard QPSK constellation diagram for the phase difference sequence of constellation points. Phase compensation is performed; subsequently, during the conversion of the constellation map into a fingerprint luminance map, different weights are assigned to pixels in different constellation point density regions based on kernel density to generate the fingerprint luminance map. ;

[0085] Fingerprint brightness map is analyzed using convolutional and pooling layers in a convolutional neural network. Feature extraction is performed to obtain fingerprint brightness map features. And the fingerprint spectral features are mapped using a feature mapping matrix. The fingerprint spectral features are mapped into the fingerprint luminance map feature space; subsequently, a cross-attention module is introduced to calculate the attention of the fingerprint spectral features to the fingerprint luminance map features. Attention to fingerprint spectral features and fingerprint brightness map features The fused fingerprint features are obtained by fusing features based on the calculated cross-attention. ; will integrate fingerprint features The fingerprint is input into a predictor consisting of an output layer to obtain the final fingerprint classification result. ;

[0086] Furthermore, LTE-V2X signals are wideband signals. According to the frame structure format of the direct-link physical data channel, in standard mode, a subframe contains 14 SC-FDM symbols. The last SC-FDM symbol is used as a guard symbol and is not modulated with data. Therefore, there is still a gap between two consecutive subframes. Since the smallest unit for the same device to transmit data in the time domain is a subframe, and the same device will not transmit data continuously in the time domain, LTE-V2X signal acquisition is performed in units of subframes.

[0087] Furthermore, an adjacent double sliding window power detection algorithm is employed. Two adjacent sliding windows are constructed in the time domain. The original received signal is passed through the sliding windows in chronological order. The average signal power within the adjacent sliding windows is then calculated. The position of the subframe signal start point is determined by whether the power ratio between the two windows exceeds a threshold. For example, the time series of the original received signal is denoted as... , This indicates the signal position on the original received signal, and the lengths of the two sliding windows are... The average signal power within the two sliding windows and The calculation method is to average the squares of the signal values ​​at each position within the window, and the specific formula is as follows:

[0088] ,

[0089] ,

[0090] After calculating the average signal power of the two sliding windows, the ratio of the average power of the two sliding windows is expressed as:

[0091] ,

[0092] When the average power ratio When a certain threshold is exceeded, it indicates that data has begun to be captured, and the peak value of the signal segment exceeding the threshold is taken as the starting position of a received subframe signal;

[0093] Furthermore, as shown in Figure 2, for the first... The initial LTE-V2X signal corresponding to each subframe Extraction and preprocessing are performed to obtain the final LTE-V2X signal. Includes the following steps:

[0094] Two sliding windows with a distance of one SC-FDM symbol are constructed in the time domain, and the initial LTEX-V2X signals are processed in chronological order. By using a sliding window, symbol cross-correlation based on CP is calculated across the entire signal subframe to obtain the cross-correlation coefficient between CP and each SC-FDMA symbol. The cross-correlation operations of all SC-FDMA symbols are superimposed, and the sampling point with the largest sum of cross-correlation magnitudes is used as the initial LTE X-V2X signal. The precise starting position to obtain synchronized LTE-V2X signals ;

[0095] Using a frequency offset estimation algorithm based on cyclic prefix, and on the basis of time synchronization, the frequency offset is estimated by calculating the difference in phase angle between the CP and the end segment of the symbol. The frequency offset is then estimated by calculating complex correlation values, and applied to the synchronization of LTE-V2X signals. The final frequency offset estimate is calculated by averaging multiple CPs and the end segments of symbols. Based on frequency offset estimation For synchronous LTE-V2X signals After frequency offset compensation, the LTE-V2X signal is synchronized by calculation. The initial phase deviation of the actual signal is obtained by calculating the complex correlation value between the actual signal and the ideal signal. Then, the average value of the initial phase deviation is calculated to obtain the estimated phase deviation value. Finally, phase compensation is performed on the signal after frequency offset compensation to obtain the final LTE-V2X signal. ;

[0096] Furthermore, regarding the initial LTE-V2X signal... Time synchronization is performed to obtain synchronized LTE-V2X signals. Includes the following steps:

[0097] Due to the initial LTE-V2X signal It contains 14 SC-FDM symbols. The SC-FDM symbols are designed using a CP+ symbol structure, with different symbols corresponding to different CP lengths, specifically:

[0098] ,

[0099] in Let be the index of the symbol in the subframe; denoted as _____. The CP signal corresponding to each symbol is:

[0100] ,

[0101] in, express road signal, express road signal, Indicates the first The length of the CP signal corresponding to each symbol If the symbol is an imaginary part, then the signal representation of the last segment of the corresponding SC-FDM symbol is:

[0102] ;

[0103] The real and imaginary parts of the last part of each CP and SC-FMD symbol are of the same value but opposite in direction, therefore the... The cross-correlation coefficient between a symbol and its corresponding CP signal is calculated as follows:

[0104] ;

[0105] Based on this, since a subframe contains 13 symbols and 1 guard symbol, the modulus of the cross-correlation coefficient between each symbol and its corresponding CP signal is calculated and summed. The sampling point with the largest sum of cross-correlation moduli is taken as the initial LTE-V2X signal. The precise starting position to obtain synchronized LTE-V2X signals ;

[0106] Furthermore, regarding the synchronization of LTE-V2X signals Frequency offset compensation and phase compensation are performed to obtain the final LTE-V2X signal. Includes the following steps:

[0107] Because LTE-V2X has a small frequency offset, only the LTE-V2X signal needs to be synchronized. Perform fractional frequency offset estimation to synchronize LTE-V2X signals. The Middle The CP signal corresponding to each symbol is represented as:

[0108] ,

[0109] in, express Signal amplitude, express signal frequency, Indicates the fractional frequency offset. Indicates the signal sampling frequency. express Initial phase, express The phase deviation is calculated using a frequency offset estimation algorithm based on a cyclic prefix. The difference between the phase angle of the CP signal and the last segment of the corresponding SC-FDM symbol is used to estimate the frequency offset. Since the ideal amplitude of the CP signal and the last segment of the corresponding SC-FDM symbol are the same but their signs are opposite, the frequency offset is estimated by calculating the complex correlation between them. The derivation formula is as follows:

[0110] ,

[0111] ,

[0112] in, Indicates the length of an SC-FDM symbol. and These represent the signals at the end of the CP signal and the SC-FDM symbol, respectively. The function representing the calculation of the phase angle difference. For the calculated first The frequency offset estimate at the nth symbol, for the nth symbol Subframe The synchronous LTE-V2X signal is obtained by averaging all CP signals using the frequency offset estimate calculated from the end segment of the SC-FDM symbol. Frequency offset estimation And perform corresponding frequency offset compensation to obtain a frequency offset compensated LTE-V2X signal. ;

[0113] Due to the initial phase difference, there is a constant phase difference between the transmitted and received signals; in frequency offset compensation LTE-V2X signals The initial phase deviation of the actual signal is obtained by calculating the complex correlation value between the actual signal and the ideal signal. The calculation formula is as follows:

[0114] ,

[0115] in Indicates the first The ideal signal corresponding to the first subframe; then, by calculating the first... The average initial phase deviation of all points in each subframe is used as the phase deviation estimate of the entire subframe signal, and the frequency offset compensated LTE-V2X signal is then used. and Multiplication is used for phase compensation to obtain the final LTE-V2X signal. ;

[0116] By employing the aforementioned LTE-V2X signal extraction and preprocessing methods, the effects of frequency offset and phase offset are eliminated, resulting in the final LTE-V2X signal for the subframe. This provides a foundation for the subsequent extraction of highly stable radio frequency fingerprint features;

[0117] Furthermore, as shown in Figure 3, a radio frequency fingerprint feature extraction method based on synchronization symbols is used to extract features from the final LTE-V2X signal. Fingerprint spectral features are obtained from the physical broadcast channel signal frames of the direct link. Includes the following steps:

[0118] The final LTE-V2X signal is demodulated using the SC-FDM algorithm. The direct link physical broadcast channel signal frame is demodulated to obtain the demodulated LTE-V2X signal. Then, the physical ID number of the direct link physical broadcast channel signal frame is obtained by using the synchronization symbol calculation method, and the corresponding demodulation reference symbol DMRS is calculated based on the ID number;

[0119] Based on demodulating LTE-V2X signals The frequency domain estimate is obtained by performing channel equalization on the symbols of the cut-through physical broadcast channel signal frame using the demodulation reference symbol (DMRS). Finally, the fingerprint spectral features are obtained using a radio frequency fingerprint feature extraction method based on synchronization symbols. ;

[0120] Furthermore, the direct link physical broadcast channel signal frame consists of the direct link physical broadcast symbol PSBCH, the primary synchronization symbol PSSS, the secondary synchronization symbol SSSS, and the demodulation reference symbol DMRS. The primary synchronization symbol is located in bits 2 and 3, the secondary synchronization symbol in bits 12 and 13, and the demodulation reference symbol in bits 5, 7, and 10. First, the final LTE-V2X signal is demodulated using the SC-FDM demodulation algorithm. Demodulate each symbol on the direct-link physical broadcast channel signal frame; for example, since each SC-FDM symbol includes a cyclic prefix (CP) and a symbol segment, the latter part of the cyclic prefix and the former part of the symbol segment are selected as the Fast Fourier Transform signal for demodulation. Therefore, the following settings are used: The value is used as a percentage when using the cyclic prefix; subsequently, the signal is half-wave shifted to avoid the effects of non-proportional high crosstalk, specifically the half-wave shift formula is as follows:

[0121] ,

[0122] in, This is a simplified expression for the exponential function. This is the starting point for performing a Fast Fourier Transform on the symbols. Indicates the first Each symbol corresponds to the length of the CP. Indicates the first The length of each symbol; Record the final LTE-V2X signal. The Middle a symbol The shifted LTE-V2X signal is obtained after half-wave shifting. Subsequently, a Fast Fourier Transform (FFT) is performed on the shifted LTE-V2X signal. Since a portion of the cyclic prefix was removed during modulation, phase compensation is required for the signal obtained after the FFT. The specific phase compensation formula is as follows:

[0123] ;

[0124] Let the first result obtained after phase compensation be... The demodulated LTE-V2X signal of each symbol is ;

[0125] In a direct-link physical broadcast channel signal frame, there are two types of primary synchronization symbols (PSSS) and 168 types of secondary synchronization symbols. The physical layer ID of the direct-link physical broadcast channel is determined by calculating the correlation coefficient between the synchronization symbols and the local synchronization symbols in the signal. According to the construction principle of the primary synchronization symbol PSSS, its sequence ID... This includes two possibilities, and the sequence ID of the secondary synchronization symbol SSSS. There are 168 possibilities; since the generation sequence of the synchronization symbol is The sequence has the same length as the number of subcarriers on the synchronization symbol, which is 62. For different sequences, the cross-correlation between the two sequences is very low. The Pearson correlation coefficient is used to calculate the correlation coefficient between the synchronization symbol in the signal and the local synchronization symbol. Based on the calculation results, the corresponding auxiliary synchronization symbol ID number and correlation coefficient diagram are drawn to obtain the final auxiliary synchronization symbol ID.

[0126] In the direct-link physical broadcast channel signal frame, the demodulation reference symbol DMRS is located at positions 5, 7, and 10. Similar to the synchronization symbol, DMRS also uses... A sequence, where each complex value in the sequence has an amplitude of 1; for example, for the same subframe, the above three symbols are based on... Different methods are used or The orthogonal sequence; channel equalization is performed using the two primary synchronization symbols with the first DMRS symbol as the 2nd and 3rd bits, and the two secondary synchronization symbols with the third DMRS symbol as the 12th and 13th bits;

[0127] Furthermore, channel equalization is achieved through channel estimation, denoted as the transmitted signal. The received signal after passing through the physical broadcast channel of the direct link is Then the relationship between the two in the frequency domain satisfies:

[0128] ,

[0129] in, For the channel frequency domain response, For broadband white noise that follows a Gaussian distribution, channel estimation is performed in the frequency domain using the least squares (LS) channel estimation algorithm that does not consider noise, as described below:

[0130] ;

[0131] Channel frequency response obtained using the above frequency domain LS estimation algorithm Channel equalization is performed on the synchronization symbols, for example, using the first... Demodulating LTE-V2X signals in individual frames Taking the primary synchronization symbol as an example, since the primary synchronization symbol and DMRS pass through the same channel, they have the same frequency domain response, denoted as . Then we have the following formula:

[0132] ,

[0133] in, The received spectrum of the main synchronization symbol. For the received spectrum of DMRS symbols, The ideal symbol spectrum for the master synchronization symbol, For the ideal symbol spectrum of DMRS symbols, The fingerprint information loaded on the main synchronization symbol. Fingerprint information loaded onto the DMRS symbol. The fingerprint causes distortion to the signal; subsequently, based on the frequency domain LS estimation algorithm, the estimated value of the channel for the main synchronization symbol can be obtained, calculated as follows:

[0134] ,

[0135] in, This represents the channel estimate corresponding to the DMRS symbol obtained through the LS estimation algorithm. The obtained channel estimate is used to perform channel equalization on the primary synchronization symbol. The calculation formula is as follows:

[0136] ,

[0137] in, This is the frequency domain estimate of the primary synchronization symbol after channel equalization; similarly, the frequency domain estimate of the secondary synchronization symbol can be obtained. ;

[0138] Subsequently, a radio frequency fingerprint extraction method based on synchronization symbols was used for fingerprint extraction. The two primary synchronization symbols and two secondary synchronization symbols were denoted as PSSS1, PSSS2, SSSS1, and SSSS2. The received... Demodulating LTE-V2X signals in individual frames The signals corresponding to the four synchronization symbols above are represented as follows:

[0139] ,

[0140] in, The received signals corresponding to the four synchronization symbols. , , , These are the transmission signals corresponding to the four synchronization symbols. For the loaded fingerprint information, The four synchronization symbols correspond to Gaussian white noise. Since the mean of Gaussian white noise is 0, the influence of noise on the fingerprint can be eliminated by superimposing multiple identical synchronization symbols. The primary synchronization symbols PSSS1 and PSSS2 are superimposed separately, and the secondary synchronization symbols SSSS1 and SSSS2 are superimposed separately. The calculation formula is as follows:

[0141] ,

[0142] ,

[0143] in, This indicates the calculation of mathematical expectation. The radio frequency fingerprint representing the master synchronization symbol, The RF fingerprint represents the secondary synchronization symbol; it can be seen from the calculation formula that the two RF fingerprints are only related by... , and Related; therefore, by accumulating the fingerprint information of the primary synchronization symbol and the secondary synchronization symbol from a subframe, and treating them as a whole, the fingerprint frequency offset feature of a subframe is denoted as... ;

[0144] Furthermore, as shown in Figure 4, a data symbol-based radio frequency fingerprint extraction method is used to extract the final LTE-V2X signal. Fingerprint luminance map obtained from the physical data channel signal frame of the direct link. Includes the following steps:

[0145] By superimposing the final LTE-V2X signal The spectral amplitudes of all demodulated reference symbols are calculated, and the bandwidth of the signal frame is determined using adjacent sliding windows and threshold decisions. Subsequently, the SC-FDM demodulation algorithm is used for demodulation, and the standard QPSK constellation diagram is generated using the SC-FDM deprecoding algorithm. ;

[0146] An adaptive weighted sliding window method is used to perform phase compensation based on the fluctuation of the phase difference between adjacent constellation points and the order of the constellation points, resulting in a corrected QPSK constellation diagram. The QPSK constellation chart will be revised. The image is converted to grayscale, and the constellation point density of each pixel is calculated using a Gaussian kernel function. Weights are then set based on the constellation point density of the pixels to generate a fingerprint brightness map. ;

[0147] Furthermore, the final LTE-V2X signal The direct-link physical data channel signal frame consists of data symbols, demodulation reference signals, and guard intervals. The demodulation reference signals are located in bits 3, 6, 9, and 12, the guard interval is located in bit 14, and the remaining bits are data symbols. Since the direct-link physical data channel signal frame contains four demodulation reference symbols, and each subcarrier of the demodulation reference symbols is modulated with a complex signal of amplitude 1, the spectral amplitude difference between the signal band and the non-signal band can be amplified by superimposing the spectral amplitude of the reference signals, thus reducing the impact of noise. Since SC-FDM uses a resource grid modulation method in the frequency domain, and the minimum unit is 12 subcarriers, a sliding window of size 12 is set. In addition, since the position of each resource grid in the spectrum is fixed, the sliding unit of the sliding window is also set to 12 to reduce the computational complexity. Two adjacent sliding windows are constructed in the frequency domain, and the sum of the amplitude spectra within the adjacent sliding windows is calculated. The threshold judgment is based on the ratio of the sum of the amplitude spectra between the two windows. Specifically, when both sliding windows are in the noise band at the same time, the ratio of the sum of the amplitude spectra between the two windows is close to 1. When the first resource cell of the effective frequency band enters the first sliding window, the second sliding window is still in the noise frequency band. At this time, the ratio of the two windows will be much greater than 1, and the current position is the starting position of the effective signal frequency band. When both sliding windows are in the effective frequency band at the same time, the ratio of the amplitude spectrum sum between the two windows returns to around 1. When the noise frequency band enters the first sliding window, the second sliding window is still in the effective signal frequency band. At this time, the ratio of the amplitude spectrum sum between the two windows will drop instantaneously and approach 0, and the current position is the ending position of the effective signal frequency band. Subsequently, the bandwidth of the signal frame is calculated using the ending position and the starting position. Subsequently, similar to the signal frames in the direct link physical broadcast channel, the final LTE-V2X signal is processed through half-wave shift, fast Fourier transform, and phase offset compensation in the direct link physical data channel. Demodulate;

[0148] In the direct link physical data channel for the final LTE-V2X signal After demodulation, the modulated signal is decoded using the SC-FDM de-precoding algorithm; for example, the precoding formula for data symbol transmission in a direct-link physical data channel is as follows:

[0149] ,

[0150] in, This is a pre-encoded complex-valued signal, i.e., a signal modulated onto the resource cell. This represents the QPSK constellation symbol before precoding. Indicates the final LTE-V2X signal Upper The bandwidth of each symbol; then decoded by IFFT transformation, and the decoded constellation symbol is inversely multiplied by [missing information]. The amplitude of the constellation is restored to obtain the first... Standard QPSK constellation diagram corresponding to each subframe ;

[0151] Furthermore, the final LTE-V2X signal totals Each subframe is generated through the above steps. A standard QPSK constellation diagram is presented. Considering that the phase offset between subframes in LTE-V2X typically changes slowly and linearly, while ordinary linear fitting uses a fixed window for global fitting, which accumulates fitting errors over long time periods and struggles to handle phase anomalies in a single frame caused by sudden interference, an adaptive weighted sliding window method is adopted to address these issues. This method weights the phase difference fluctuations between adjacent constellation points and the constellation point order to perform phase compensation, resulting in a corrected QPSK constellation diagram. ;

[0152] For example, for the first Standard QPSK constellation diagram corresponding to each subframe In Calculate the standard QPSK constellation chart using constellation points. The average phase of all constellation points is calculated using the following formula:

[0153] ,

[0154] in, Indicates the first Standard QPSK constellation diagram corresponding to each subframe The average phase of all constellation points in the middle, Represents the standard QPSK constellation diagram The Middle The phase of each constellation point is calculated; then the phase difference between constellation points in adjacent subframes is calculated. The calculation formula is as follows:

[0155] ,

[0156] Obtain the phase difference sequence ;

[0157] For the Standard QPSK constellation diagram corresponding to each subframe Take the nearest Phase difference sequence of frames Then the variance of the phase difference sequence is calculated. The calculation formula is as follows:

[0158] ,

[0159] ,

[0160] in, Represents the phase difference sequence The mean, Subframe The corresponding weights; based on the most recent Frame weights and phase difference sequences The mean was calculated to obtain the first The number of subframes compared to the first The offset of each subframe is calculated using the following formula:

[0161] ;

[0162] Subsequently, based on the calculated offset, the first Standard QPSK constellation diagram corresponding to each subframe The corrected QPSK constellation chart is obtained by performing phase compensation on all constellation points. ;

[0163] According to the revised QPSK constellation chart Plot the grayscale images corresponding to the real and imaginary parts of each constellation point; treat the real part data as... Axis, imaginary part data is considered The corresponding grayscale image is plotted along the axis; then, the constellation point density of each pixel in the grayscale image is calculated using a Gaussian kernel function, and corresponding weights are set for coloring to generate a fingerprint brightness map; for example, for pixels in the grayscale image... The constellation density at the corresponding pixel position is calculated based on the Gaussian kernel function, using the following formula:

[0164] ,

[0165] in, Represents pixels constellation density at that location Indicates the number of constellation points. Indicates the pixel size of the grayscale image. Indicates the first The coordinates of the constellation points This indicates the selected Gaussian kernel function; subsequently, the brightness value of the corresponding pixel is calculated based on the calculated constellation point density of each pixel to generate a modified QPSK constellation map. Corresponding fingerprint brightness map The calculation formula is as follows:

[0166] ,

[0167] in, Represents the pixels on the fingerprint brightness map The brightness value at the location; by reducing the brightness contribution of high-density pixels to avoid saturation of the fingerprint brightness map, and increasing the brightness contribution of low-density pixels to highlight the subtle offsets at the fingerprint edges.

[0168] Furthermore, the first Fingerprint spectral features of each subframe fingerprint brightness map The fusion process yields the fused fingerprint features. It will integrate fingerprint features The fingerprint is input into a predictor to obtain the fingerprint classification result;

[0169] Furthermore, a convolutional neural network consisting of two convolutional layers, two pooling layers, and one fully connected layer is used to... fingerprint brightness map of each subframe Processing is required; for example, the first... fingerprint brightness map of each subframe The size is The kernel size in the first convolutional layer is 5×5, and the output dimension remains the same after processing by the first convolutional layer. The output dimension after dimensionality reduction by the first max pooling layer with a pooling size of 2 and a step size of 2 is: The kernel size in the second convolutional layer is 3×3, and the output dimension after processing by the second convolutional layer is... The output dimension after dimensionality reduction via a second max-pooling layer with a pooling size of 2 and a step size of 2 is: Ultimately through a Fingerprint luminance map extracted using fully connected layers and ReLU activation function Fingerprint brightness map features Its dimensions are 128;

[0170] Furthermore, based on the extraction of synchronization symbols, the first... Fingerprint spectral features of each subframe It contains 62 real-part features and 62 imaginary-part features extracted from the primary synchronization symbol PSSS, 62 real-part features and 62 imaginary-part features extracted from the secondary synchronization symbol, and 1 physical ID feature. Therefore, the first... Fingerprint spectral features of each subframe The dimension is 249; due to the fingerprint brightness map features The dimension is 128, therefore the feature mapping matrix The size is The specific feature mapping calculation formula is as follows:

[0171] ,

[0172] in, The bias term is 128-dimensional, and ReLU is the activation function. To map fingerprint spectral features;

[0173] Furthermore, a cross-attention mechanism is introduced to calculate the attention of fingerprint spectral features to fingerprint luminance map features. Attention to fingerprint spectral features and fingerprint brightness map features After aligning the fingerprint spectral features and fingerprint luminance map features to the same space using a feature mapping matrix, they are normalized to avoid the influence of dimensional differences. The calculation formula is as follows:

[0174] ,

[0175] ,

[0176] in, For normalized fingerprint brightness map features, The fingerprint spectral features are normalized; then the similarity between the normalized features is calculated to obtain the similarity matrix. The calculation formula is as follows:

[0177] ,

[0178] in, The first feature in the normalized fingerprint luminance map One dimension, The first characteristic in the normalized fingerprint spectral features One dimension, The first feature in the normalized fingerprint luminance map The first dimension and normalized fingerprint spectral features Relevance of each dimension;

[0179] Based on similarity matrix Calculate the attention of fingerprint spectral features to fingerprint luminance map features separately. Attention to fingerprint spectral features and fingerprint brightness map features The specific calculation formula is as follows:

[0180] ,

[0181] ;

[0182] The attention of fingerprint spectral features to fingerprint luminance map features is obtained through the two formulas above. Attention to fingerprint spectral features and fingerprint brightness map features And based on these two attentions, the fingerprint brightness map features are analyzed respectively. and mapped fingerprint spectral features After weighting and fusing, the fused fingerprint features are obtained. The specific calculation formula is as follows:

[0183] ;

[0184] Fusing fingerprint features The input is fed into a predictor consisting of an output layer to obtain the final fingerprint classification result. This output layer is a fully connected layer with dimensions of . , The formula for calculating the classification result, representing the number of devices, is as follows:

[0185] ,

[0186] in, The output fingerprint classification results, This is the weight matrix. For bias terms;

[0187] Furthermore, as shown in Figure 5, this invention proposes an LTE-V2X radio frequency fingerprint extraction and identification system based on a direct link channel, including an acquisition and processing module, a spectrum feature module, an image feature module, and a fingerprint classification module;

[0188] The acquisition and processing module uses an adjacent dual sliding window power detection algorithm to capture LTE-V2X signals in subframe units from the original received signal. Then, it uses a symbol cross-correlation time synchronization algorithm based on CP to accurately locate the starting position of the signal to obtain the synchronized LTE-V2X signal. The frequency offset estimation algorithm based on synchronization symbols is used to compensate for the frequency offset of the synchronized LTE-V2X signal, and phase compensation is performed according to the phase deviation between the original received signal and the ideal signal to obtain the final LTE-V2X signal.

[0189] The spectrum feature module targets the direct-link physical broadcast channel. It demodulates the direct-link physical broadcast channel signal frame in the final LTE-V2X signal using the SC-FDM demodulation algorithm to obtain the demodulated LTE-V2X signal. Then, it uses the synchronization symbol calculation method to obtain the physical ID number of the direct-link physical broadcast channel signal frame and calculates the corresponding demodulation reference symbol DMRS based on the ID number. Based on the demodulated LTE-V2X signal and the demodulation reference symbol DMRS, it performs channel equalization on the symbols of the direct-link physical broadcast channel signal frame to obtain the frequency domain estimate. Finally, it uses the radio frequency fingerprint feature extraction method based on synchronization symbols to obtain the fingerprint spectrum features.

[0190] The image feature module, targeting the direct-link physical data channel, performs bandwidth detection on the final LTE-V2X signal frame of the direct-link physical data channel, then demodulates it using the SC-FDM demodulation algorithm, and finally generates a standard QPSK constellation diagram of the direct-link physical data channel signal frame using the SC-FDM deprecoding algorithm. Based on the phase shift trend between different constellation points, an adaptive weighted sliding window is used to perform phase compensation on the standard QPSK constellation diagram for the constellation point phase difference sequence. Subsequently, in the process of converting the constellation diagram into a fingerprint brightness map, different weights are set for pixels in different constellation point density regions according to the kernel density to generate the fingerprint brightness map.

[0191] The fingerprint classification module extracts fingerprint luminance features from the fingerprint luminance map using convolutional and pooling layers in a convolutional neural network, and maps the fingerprint spectral features to the fingerprint luminance map feature space using a feature mapping matrix. Subsequently, a cross-attention module is introduced to calculate the attention of the fingerprint spectral features to the fingerprint luminance map features and the attention of the fingerprint luminance map features to the fingerprint spectral features, respectively. Based on the calculated cross-attention, feature fusion is performed to obtain fused fingerprint features. The fused fingerprint features are then input into a predictor consisting of an output layer to obtain the final fingerprint classification result.

[0192] Furthermore, the acquisition and processing module includes a signal acquisition unit, a signal synchronization unit, and a signal processing unit;

[0193] The signal acquisition unit uses an adjacent dual sliding window power detection algorithm to capture LTE-V2X signals in units of subframes from the original received signal;

[0194] The signal synchronization unit uses a symbol cross-correlation time synchronization algorithm based on CP to accurately locate the starting position of the signal and obtain a synchronized LTE-V2X signal.

[0195] The signal processing unit uses a frequency offset estimation algorithm based on synchronization symbols to perform frequency offset compensation on the synchronous LTE-V2X signal, and performs phase compensation based on the phase deviation between the original received signal and the ideal signal to obtain the final LTE-V2X signal.

[0196] Furthermore, an adjacent double sliding window power detection algorithm is adopted to construct two adjacent sliding windows in the time domain. The original received signal is passed through the sliding windows in chronological order. Then, the average signal power in the adjacent sliding windows is calculated, and the position of the subframe signal start point is obtained by judging whether the power ratio between the two windows exceeds the threshold.

[0197] Furthermore, the initial LTE-V2X signal is extracted and preprocessed to obtain the final LTE-V2X signal. Includes the following steps:

[0198] Two sliding windows with a distance of one SC-FDM symbol are constructed in the time domain. The initial LTEX-V2X signal is passed through the sliding windows in chronological order. Symbol cross-correlation calculation based on CP is performed on the entire signal subframe to obtain the cross-correlation coefficient between CP and each SC-FDMA symbol. The cross-correlation operations of all SC-FDMA symbols are superimposed, and the sampling point with the largest sum of cross-correlation magnitudes is calculated as the precise starting position of the initial LTEX-V2X signal to obtain the synchronized LTE-V2X signal.

[0199] Using a frequency offset estimation algorithm based on cyclic prefixes, the frequency offset is estimated by calculating the difference in phase angle between the CP and the symbol tail segment, based on time synchronization. Frequency offset estimation is performed by calculating complex correlation values. The final frequency offset estimate is calculated by averaging the frequency offset estimates of multiple CPs and symbol tail segments in the synchronized LTE-V2X signal. Based on the frequency offset estimate... After frequency offset compensation of the synchronous LTE-V2X signal, the initial phase deviation of the actual signal is obtained by calculating the complex correlation value between the synchronous LTE-V2X signal and the ideal signal. Then, the average value of the initial phase deviation is calculated to obtain the phase deviation estimate. Finally, phase compensation is performed on the signal after frequency offset compensation to obtain the final LTE-V2X signal.

[0200] Furthermore, the spectrum feature module includes a signal demodulation unit, a signal equalization unit, and a feature extraction unit;

[0201] The signal demodulation unit demodulates the through-link physical broadcast channel signal frame in the final LTE-V2X signal using the SC-FDM demodulation algorithm to obtain the demodulated LTE-V2X signal;

[0202] The signal equalization unit uses the synchronization symbol calculation method to obtain the physical ID number of the direct link physical broadcast channel signal frame and calculates the corresponding demodulation reference symbol DMRS based on the ID number. Based on the demodulated LTE-V2X signal and the demodulation reference symbol DMRS, the unit performs channel equalization on the symbols of the direct link physical broadcast channel signal frame to obtain the frequency domain estimate.

[0203] The feature extraction unit uses a radio frequency fingerprint feature extraction method based on synchronization symbols to obtain fingerprint spectral features;

[0204] Furthermore, the method of extracting fingerprint spectral features from the direct link physical broadcast channel signal frame in the final LTE-V2X signal using a synchronization symbol-based radio frequency fingerprint feature extraction method includes the following steps:

[0205] The SC-FDM demodulation algorithm is used to demodulate the direct link physical broadcast channel signal frame in the final LTE-V2X signal to obtain the demodulated LTE-V2X signal. Then, the physical ID number of the direct link physical broadcast channel signal frame is obtained by using the synchronization symbol calculation method, and the corresponding demodulation reference symbol DMRS is calculated based on the ID number.

[0206] Based on the demodulated LTE-V2X signal and demodulated reference symbol DMRS, channel equalization is performed on the symbols of the physical broadcast channel signal frame of the direct link to obtain the frequency domain estimate. Finally, the RF fingerprint feature extraction method based on synchronization symbol is used to obtain the fingerprint spectrum features.

[0207] Furthermore, the direct link physical broadcast channel signal frame consists of the direct link physical broadcast symbol PSBCH, the primary synchronization symbol PSSS, the secondary synchronization symbol SSSS, and the demodulation reference symbol DMRS. The primary synchronization symbol is located in bits 2 and 3, the secondary synchronization symbol in bits 12 and 13, and the demodulation reference symbol in bits 5, 7, and 10. First, the final LTE-V2X signal is demodulated using the SC-FDM demodulation algorithm. Each symbol on the physical broadcast channel signal frame of the direct link is demodulated; then, the signal is half-wave shifted to avoid the influence of non-proportional high crosstalk; then, the shifted LTE-V2X signal is subjected to fast Fourier transform. Since part of the cyclic prefix is ​​removed during modulation, phase compensation is required on the signal obtained after fast Fourier transform to obtain the demodulated LTE-V2X signal.

[0208] In the direct link physical broadcast channel signal frame, there are two types of primary synchronization symbols (PSSS) and 168 types of secondary synchronization symbols. The physical layer ID of the direct link physical broadcast channel is determined by calculating the correlation coefficient between the synchronization symbols and the local synchronization symbols in the signal.

[0209] Furthermore, channel equalization is achieved through channel estimation. In the frequency domain, the least squares (LS) channel estimation algorithm without considering noise is used to perform channel equalization on the synchronization symbols; the frequency domain estimates of the primary and secondary synchronization symbols after channel equalization are obtained.

[0210] Subsequently, a radio frequency fingerprint extraction method based on synchronization symbols was used for fingerprint extraction. Multiple identical synchronization symbols were superimposed to eliminate the influence of noise on the fingerprint. The fingerprint information of the main synchronization symbol and the auxiliary synchronization symbol was obtained and used as a whole to identify the fingerprint frequency offset feature of a subframe.

[0211] Furthermore, the image feature module includes a decoding unit and a brightness map unit;

[0212] After bandwidth detection, the decoding unit demodulates the final LTE-V2X signal through the pass-through physical data channel signal frame using the SC-FDM demodulation algorithm, and then generates the standard QPSK constellation diagram of the pass-through physical data channel signal frame using the SC-FDM deprecoding algorithm.

[0213] The luminance map unit performs phase compensation on the standard QPSK constellation map by using an adaptive weighted sliding window for the constellation point phase difference sequence based on the trend of phase shift between different constellation points. Subsequently, in the process of converting the constellation map into a fingerprint luminance map, different weights are set for pixels in different constellation point density regions according to the kernel density to generate the fingerprint luminance map.

[0214] Furthermore, the method of obtaining a fingerprint brightness map from the direct-link physical data channel signal frame in the final LTE-V2X signal using a data symbol-based radio frequency fingerprint extraction method includes the following steps:

[0215] The bandwidth of the signal frame is determined by superimposing the spectral amplitudes of all demodulated reference symbols in the final LTE-V2X signal and using adjacent sliding windows and threshold decisions. Then, the SC-FDM demodulation algorithm is used for demodulation, and the standard QPSK constellation diagram is generated by the SC-FDM deprecoding algorithm.

[0216] An adaptive weighted sliding window method is used to perform phase compensation based on the fluctuation of the phase difference between adjacent constellation points and the order of constellation points to obtain a corrected QPSK constellation map. The corrected QPSK constellation map is converted into a grayscale image, and the constellation point density of each pixel is calculated using a Gaussian kernel function. Weights are set according to the constellation point density of the pixels, and a fingerprint brightness map is generated.

[0217] Furthermore, the final LTE-V2X signal frame for the direct-link physical data channel consists of data symbols, demodulation reference signals, and guard intervals. The demodulation reference signals are located in bits 3, 6, 9, and 12, the guard interval is located in bit 14, and the remaining bits are data symbols. Since the direct-link physical data channel signal frame contains four demodulation reference symbols, and each subcarrier of the demodulation reference symbols is modulated with a complex signal of amplitude 1, the spectral amplitude difference between the signal band and non-signal band can be amplified by superimposing the spectral amplitude of the reference signals, thus reducing the impact of noise. Two adjacent sliding windows are constructed in the frequency domain. The sum of the amplitude spectra within the adjacent sliding windows is calculated using the sliding windows. The ratio of the sum of the amplitude spectra between the two windows is used to determine the threshold and end positions. Subsequently, the bandwidth of the signal frame is calculated using the end and start positions. Then, the final LTE-V2X signal is demodulated in the direct-link physical data channel through half-wave shift, fast Fourier transform, and phase offset compensation.

[0218] After demodulation, the modulated signal is decoded using the SC-FDM de-precoding algorithm; subsequently, it is decoded using IFFT transform, and the amplitude of the constellation diagram is restored by inversely multiplying the decoded constellation symbols by the bandwidth to obtain the first constellation symbol. The standard QPSK constellation diagram corresponding to each subframe;

[0219] Furthermore, an adaptive weighted sliding window method is adopted to perform phase compensation based on the fluctuation of the phase difference between adjacent constellation points and the order of constellation points to obtain a corrected QPSK constellation diagram;

[0220] Based on the real and imaginary parts of each constellation point in the modified QPSK constellation diagram, a corresponding grayscale image is drawn; the real part data is considered as... Axis, imaginary part data is considered The corresponding grayscale image is drawn along the axis; then, the constellation point density of each pixel on the grayscale image is calculated using the Gaussian kernel function, and corresponding weights are set for coloring to generate a fingerprint brightness map;

[0221] Furthermore, the fingerprint classification module includes a feature processing unit, a feature fusion unit, and a fingerprint classification unit;

[0222] The feature processing unit extracts fingerprint brightness map features by using convolutional and pooling layers in the convolutional neural network, and maps the fingerprint spectral features to the fingerprint brightness map feature space through a feature mapping matrix;

[0223] The feature fusion unit introduces a cross-attention module to calculate the attention of fingerprint spectral features to fingerprint brightness map features and the attention of fingerprint brightness map features to fingerprint spectral features, respectively. Based on the calculated cross-attention, feature fusion is performed to obtain fused fingerprint features.

[0224] The fingerprint classification unit inputs the fused fingerprint features into a predictor consisting of an output layer to obtain the final fingerprint classification result;

[0225] Furthermore, the fingerprint spectral features and fingerprint brightness map are fused to obtain fused fingerprint features, and a predictor is used to output the fingerprint classification result based on the fused fingerprint features;

[0226] Furthermore, a convolutional neural network consisting of two convolutional layers, two pooling layers, and one fully connected layer is used to process the fingerprint luminance map to obtain fingerprint luminance map features; the fingerprint spectral features are then mapped into the fingerprint luminance map feature space through a feature mapping matrix.

[0227] Furthermore, a cross-attention mechanism is introduced to calculate the attention of fingerprint spectral features to fingerprint brightness map features and the attention of fingerprint brightness map features to fingerprint spectral features. The fingerprint spectral features and fingerprint brightness map features are normalized to avoid the influence of dimensional differences, and then the similarity between the normalized features is calculated to obtain a similarity matrix. Based on the similarity matrix, the attention of fingerprint spectral features to fingerprint brightness map features and the attention of fingerprint brightness map features to fingerprint spectral features are calculated respectively. The fingerprint brightness map features and the mapped fingerprint spectral features are weighted according to these two attentions and then fused to obtain the fused fingerprint features. The fused fingerprint features are input into a predictor consisting of an output layer to obtain the final fingerprint classification result.

[0228] This invention discloses an LTE-V2X radio frequency fingerprint extraction and identification method and system based on a direct-link channel. The method employs an adjacent dual-sliding window power detection algorithm to capture and preprocess LTE-V2X signals. Fingerprint spectral features are extracted from the direct-link physical broadcast channel using an SC-FDM demodulation algorithm and radio frequency fingerprint features based on synchronization symbols. For the direct-link physical data channel, a constellation diagram is generated using an SC-FDM deprecoding algorithm. An adaptive weighted sliding window is used to perform phase compensation on the constellation diagram, and different weights are set according to pixel kernel density to generate a fingerprint brightness map. Fingerprint brightness map features are extracted, and the fingerprint spectral features are mapped to the fingerprint brightness map feature space using a feature mapping matrix. Subsequently, a cross-attention module is introduced to calculate the cross-attention between the fingerprint spectral features and the fingerprint brightness map features, and feature fusion is performed to obtain fused fingerprint features. The final fingerprint classification result is obtained based on the fused fingerprint features.

[0229] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for LTE-V2X radio frequency fingerprint extraction and identification based on a direct-link channel, characterized in that, Includes the following steps: The adjacent dual sliding window power detection algorithm is used to capture LTE-V2X signals and preprocess them by time synchronization, frequency offset compensation and phase compensation; fingerprint spectrum features are extracted after demodulation and channel equalization of the physical broadcast channel signal frames of the direct link. After demodulating the physical data channel signal frame of the direct link, deprecoding is performed to generate a constellation map. Phase compensation and weight setting are performed on the constellation points on the constellation map to generate a fingerprint brightness map. The fingerprint brightness map features are extracted, the fingerprint spectral features are mapped to the fingerprint brightness map feature space, and the fused fingerprint features are obtained through the cross-attention module. The fingerprint classification result is obtained based on the fused fingerprint features. The step of generating a fingerprint brightness map by performing phase compensation and weight setting on constellation points in the constellation map includes: using an adaptive weighted sliding window method to perform phase compensation based on the fluctuation of the phase difference between adjacent constellation points and the order of constellation points in the constellation map to obtain a corrected QPSK constellation map; and considering the real part data of each constellation point in the corrected QPSK constellation map as... Axis, imaginary part data is considered The grayscale image is plotted on the axis, and the constellation point density of each pixel on the grayscale image is calculated using a Gaussian kernel function. The calculated constellation point density of each pixel is used to calculate the brightness value of the corresponding pixel to generate a fingerprint brightness map corresponding to the corrected QPSK constellation map. The preprocessing, including time synchronization, frequency offset compensation, and phase compensation, includes: constructing two sliding windows in the time domain; passing the LTEX-V2X signal through the sliding windows in chronological order; calculating the cross-correlation coefficient between the CP signal on the LTEX-V2X signal and each SC-FDMA symbol; superimposing the cross-correlation operations of all SC-FDMA symbols; and calculating the cross-correlation coefficient with the largest sum of cross-correlation magnitudes. The sampling point serves as the starting position of the LTE-V2X signal. Using a frequency offset estimation algorithm based on a cyclic prefix, the frequency offset is calculated by the difference in phase angle between the CP and the end segment of the symbol, based on time synchronization. Frequency offset estimation is performed by calculating complex correlation values. Multiple CPs and the end segment of the symbol in the LTE-V2X signal are used to calculate the frequency offset estimate, and the average value is taken to obtain the final frequency offset estimate, which is then compensated for. The initial phase deviation of the signal is obtained by calculating the complex correlation value between the LTE-V2X signal and the ideal signal. The average value of the initial phase deviation is then calculated to obtain the phase deviation estimate, and phase compensation is performed on the signal after frequency offset compensation.

2. The LTE-V2X radio frequency fingerprint extraction and identification method based on a direct link channel as described in claim 1, characterized in that, The process of obtaining fused fingerprint features through the cross-attention module includes: aligning fingerprint spectral features and fingerprint luminance map features by mapping them to the same space using a feature mapping matrix, and then normalizing both; subsequently calculating the similarity between the normalized features to obtain a similarity matrix; calculating the attention of fingerprint spectral features to fingerprint luminance map features and the attention of fingerprint luminance map features to fingerprint spectral features based on the similarity matrix; and then weighting the fingerprint luminance map features and the mapped fingerprint spectral features according to these two attentions and fusing them to obtain the fused fingerprint features.

3. The LTE-V2X radio frequency fingerprint extraction and identification method based on a direct link channel as described in claim 2, characterized in that, The demodulation of the direct link physical broadcast channel signal frame includes: performing a half-wave shift on the LTE-V2X signal using the SC-FDM demodulation algorithm to obtain a shifted LTE-V2X signal; then performing a fast Fourier transform on the shifted LTE-V2X signal and performing phase compensation to obtain a demodulated LTE-V2X signal; subsequently, using the synchronization symbol calculation method to obtain the physical ID number of the direct link physical broadcast channel signal frame and calculating the corresponding demodulation reference symbol DMRS based on the ID number.

4. The LTE-V2X radio frequency fingerprint extraction and identification method based on a direct link channel as described in claim 3, characterized in that, The extraction of fingerprint spectrum features includes: performing channel equalization on the symbols of the direct link physical broadcast channel signal frame based on the demodulated LTE-V2X signal and demodulated reference symbol DMRS to obtain frequency domain estimates; and finally, using a radio frequency fingerprint feature extraction method based on synchronization symbols, the fingerprint information of the main synchronization symbol and the auxiliary synchronization symbol is obtained by superimposing multiple identical synchronization symbols and used as a whole to identify the fingerprint spectrum features of a subframe.

5. The LTE-V2X radio frequency fingerprint extraction and identification method based on a direct link channel as described in claim 4, characterized in that, The process of mapping fingerprint spectral features to a fingerprint luminance map feature space includes: processing the fingerprint luminance map using a convolutional neural network containing two convolutional layers, two pooling layers, and one fully connected layer to obtain fingerprint luminance map features with a dimension of 128; the fingerprint spectral features extracted based on synchronization symbols include 62 real and 62 imaginary features extracted from the primary synchronization symbol PSSS, 62 real and 62 imaginary features extracted from the secondary synchronization symbol, and one physical ID feature, resulting in a fingerprint spectral feature dimension of 249; using a size of The feature mapping matrix maps the fingerprint spectral features to the fingerprint brightness map feature space to obtain the mapped fingerprint spectral features.

6. The LTE-V2X radio frequency fingerprint extraction and identification method based on a direct link channel as described in claim 5, characterized in that, The method for capturing LTE-V2X signals includes: employing an adjacent dual sliding window power detection algorithm to construct two adjacent sliding windows in the time domain, passing the original received signal through the sliding windows in chronological order, then calculating the average signal power within the adjacent sliding windows, and determining whether the power ratio between the two windows exceeds a threshold to obtain the position of the subframe signal start point.

7. The LTE-V2X radio frequency fingerprint extraction and identification method based on a direct link channel as described in claim 6, characterized in that, The demodulation of the direct-link physical data channel signal frame includes: constructing two adjacent sliding windows in the frequency domain; calculating the sum of the amplitude spectra within the adjacent sliding windows; determining the end position and start position based on the ratio of the sum of the amplitude spectra between the two windows; calculating the bandwidth of the direct-link physical data channel signal frame based on the start position and end position; and then demodulating the direct-link physical data channel signal frame through half-wave shift, fast Fourier transform, and phase offset compensation.

8. An LTE-V2X radio frequency fingerprint extraction and identification system based on a direct-link channel, characterized in that, The LTE-V2X radio frequency fingerprint extraction and identification method based on a direct link channel as described in any one of claims 1-6 includes an acquisition and processing module, a spectrum feature module, an image feature module, and a fingerprint classification module; the acquisition and processing module uses an adjacent dual sliding window power detection algorithm to capture LTE-V2X signals from the original received signals and performs preprocessing through time synchronization, frequency offset compensation, and phase compensation; The spectrum feature module demodulates the through-link physical broadcast channel signal frame in the LTE-V2X signal using the SC-FDM demodulation algorithm. Then, it uses the synchronization symbol calculation method to obtain the physical ID number of the through-link physical broadcast channel signal frame and calculates the corresponding demodulation reference symbol DMRS based on the ID number. Subsequently, channel equalization is performed to obtain the frequency domain estimate. Finally, the RF fingerprint feature extraction method based on synchronization symbols is used to obtain the fingerprint spectrum features. The image feature module uses the SC-FDM demodulation algorithm for demodulation and then the SC-FDM de-precoding algorithm to generate a constellation map. An adaptive weighted sliding window is used to perform phase compensation on the constellation map. Then, different weights are set according to the pixel kernel density to generate a fingerprint brightness map. The fingerprint classification module introduces a cross-attention module to perform feature fusion to obtain fused fingerprint features. The fused fingerprint features are input into a predictor consisting of an output layer to obtain the final fingerprint classification result.

Images