Internet of vehicles broadcast frame radio frequency fingerprint identification method based on LMMSE channel estimation

Abstract

The invention provides an Internet of Vehicles broadcast frame radio frequency fingerprint identification method based on LMMSE channel estimation. The method comprises the following steps: obtaining subcarrier data in a resource grid based on an obtained signal of a physical side link broadcast channel; performing root-mean-square delay expansion and channel autocorrelation matrix construction based on the subcarrier data, and combining priori signal-to-noise ratio regularization and time domain windowing operation to obtain a channel estimation value; initial radio frequency fingerprint features are obtained through the channel equalization and the channel estimation value, and the improved neural network and the initial radio frequency fingerprint features are utilized to carry out classification identification on the Internet of Vehicles equipment. Influences of noise and channels on fingerprints are effectively considered, and the purpose of extracting fingerprints of different devices in a complex environment more accurately is achieved.

Description

Technical Field

[0001] The present invention belongs to the field of V2X communication technology, and in particular relates to a radio frequency fingerprint recognition method for vehicle-to-everything (V2X) broadcast frames based on LMMSE channel estimation. Background Art

[0002] With the rapid development of intelligent transportation systems (ITS), cellular vehicle-to-everything (C-V2X) technology has become a key enabler of smart transportation and has been widely applied in areas such as vehicle cooperative perception, emergency warning, and traffic efficiency optimization. However, in C-V2X communication environments, unauthorized access, malicious behavior, and device identity forgery are becoming increasingly prominent, posing a serious threat to network security and reliability. Radio frequency fingerprinting offers a new approach to physical layer security. This approach leverages inherent hardware defects (such as I / Q imbalance and power amplifier nonlinearity) as unclonable physical identifiers, avoiding computational complexity. Its unforgeability and uniqueness make it an excellent medium for device identification and authentication.

[0003] Complex wireless channel environments affect the robustness of RF-based systems. Frequency-selective and time-varying channels make it difficult to extract pure and stable RFF features that are decoupled from channel characteristics, resulting in poor performance in multipath and mobility scenarios. Channel interference can reduce fingerprint classification accuracy by up to 80%. Although channel state information (CSI) itself can serve as a potential medium for device authentication, physical layer authentication mechanisms based on environmental semantics face performance degradation in dynamic scenarios (such as transmitter mobility or the spatial proximity of malicious and legitimate devices). To address channel sensitivity, current research is focused on designing channel-robust RF fingerprint feature extraction methods. Mainstream technical approaches focus on extracting channel-independent intrinsic transmitter features, such as using carrier frequency offset (CFO) as a device identification feature. However, CFO suffers from limitations such as insufficient time-domain stability and low class discrimination, making it difficult to achieve reliable device classification in multipath fading and multi-device coexistence scenarios. With the widespread application of deep learning technology in RF fingerprinting, data-driven methods based on adversarial training have become a research hotspot. For example, adversarial examples are used as augmented data to constrain the feature extractor through strong regularization, but this will sacrifice model sensitivity and standard accuracy.

[0004] However, these methods are mostly designed for narrowband or low-mobility scenarios and are difficult to adapt to the dynamic demands of cellular vehicle-to-everything (CVI) networks. Some attempts have employed data augmentation and artificial noise addition to mitigate channel interference, but the former significantly increases model training overhead, while the latter, based on the linear superposition assumption of "received signal = transmitted signal + RFF," suffers from a model mismatch with the nonlinear coupling characteristics of channel distortion and hardware impairments in physical layer signal transmission. Furthermore, while systems such as Wi-Fi and LoRa can achieve channel compensation through protocol-specific frame structure features, their methodologies are limited by the heterogeneity of wireless communication standards, making direct migration to the C-V2X system architecture difficult. Furthermore, a differential constellation trajectory diagram construction method based on the transient characteristics of the physical random access channel uses a multi-channel convolutional neural network to extract RFF features during transient on / off phases. However, results show that its classification accuracy significantly degrades for persistent symbol features such as the demodulation reference signal (DMRS). Achieving dynamic channel robustness while maintaining the inherent characteristics of RFF is crucial for the reliability of C-V2X security authentication. Summary of the Invention

[0005] In response to the problems existing in the prior art, the present invention provides a radio frequency fingerprint recognition method for vehicle network broadcast frames based on LMMSE channel estimation, which at least partially solves the problems existing in the prior art of incomplete fingerprint extraction and low classification accuracy in the C-V2X system.

[0006] The present disclosure provides a method for identifying radio frequency fingerprints of broadcast frames in an Internet of Vehicles (IoV) based on LMMSE channel estimation, including: Obtaining subcarrier data in a resource grid based on the acquired signal of the physical sidelink broadcast channel; Based on the subcarrier data, the root mean square delay spread and channel autocorrelation matrix are constructed, and the channel estimation value is obtained by combining the prior signal-to-noise ratio regularization and time domain windowing operation; The initial radio frequency fingerprint features are obtained through channel equalization and channel estimation, and the improved neural network and the initial radio frequency fingerprint features are used to classify and identify Internet of Vehicles devices.

[0007] Optionally, obtaining subcarrier data in the resource grid based on the acquired signal of the physical sidelink broadcast channel, including: signal detection, frame synchronization, carrier frequency offset and resource grid demodulation; The signal detection adopts block energy gradient and threshold determination to locate the signal starting point by calculating the energy ratio of adjacent data blocks; The frame synchronization uses the periodic characteristics of the main synchronization signal to perform coarse synchronization and searches for the signal peak point through amplitude normalization cross-correlation operation; The carrier frequency offset compensation utilizes the characteristics that two consecutive primary synchronization symbols are the same and two consecutive secondary synchronization symbols are the same, and adopts the signal autocorrelation of the physical side link broadcast channel to obtain the frequency offset estimation value; The resource grid demodulation utilizes the structural characteristics of the cyclic prefix and eliminates the influence of time domain offset and subcarrier misalignment through phase and half subcarrier compensation to obtain frequency domain resources.

[0008] Optionally, the signal detection includes: Define the signal block length as , traverse the signal of the physical side link broadcast channel, according to the block length Divide the data into non-overlapping segments and calculate the energy block by block : , in For the Grid data block, is the sampling point index, for The transpose of is a natural number.

[0009] Optionally, the carrier frequency offset compensation includes: Extract the signal segments and calculate the conjugate cross-correlation phase difference to estimate the frequency offset:

[0010] in, is the frequency deviation, , is the number of Fourier transform points, and is the relevant segment of the main synchronization signal, and is the related segment of the auxiliary synchronization symbol, and n is a natural number.

[0011] Optionally, the resource grid demodulation includes: Determine the starting position of the Fourier transform and then extract the length The time domain sample segment of length The time domain sample segment and factor Multiply, then perform Fourier transform on the time domain signal after multiplication to convert the time domain sample segment to the frequency domain, factor Used for half-subcarrier frequency offset compensation.

[0012] Optionally, performing root mean square delay spread and channel autocorrelation matrix construction based on subcarrier data, combining a priori signal-to-noise ratio regularization and time domain windowing operations to obtain a channel estimate, includes: Estimate channel power to characterize signal energy; The channel autocorrelation matrix is ​​constructed based on the estimated channel power and RMS delay spread.

[0013] Optionally, the channel autocorrelation matrix is ​​constructed based on the estimated channel power and a delay-related phase offset, and the delay-related phase offset is obtained based on a frequency difference between subcarriers.

[0014] Optionally, a priori signal-to-noise ratio regularization and time-domain windowing operations are combined to obtain a channel estimate, including: Generate dynamically adjusted regularization terms , the regularization term It is used to reduce the effect of regularization when the prior signal-to-noise ratio is high, and to increase the regularization strength when the prior signal-to-noise ratio is low.

[0015] Optionally, performing root mean square delay spread and channel autocorrelation matrix construction based on subcarrier data, combining a priori signal-to-noise ratio regularization and time domain windowing operations to obtain a channel estimate, includes: In low signal-to-noise ratio scenarios, the default RMS delay spread value is set; Perform multi-symbol averaging on the initial frequency-domain channel response to smooth out the noise impact; The data differences of the channel autocorrelation matrix are eliminated through symmetry processing, and the main path is retained through windowing operation to suppress noise and RF fingerprint interference.

[0016] Optionally, an initial radio frequency fingerprint feature is obtained through channel equalization and channel estimation, and the improved neural network and the initial radio frequency fingerprint feature are used to classify and identify the Internet of Vehicles devices. The improved neural network is an improved ShuffleNet V2 network, which includes: The convolutional block attention module is integrated after the final convolutional layer to dynamically focus on key spectral features through joint optimization of channel attention and spatial attention. The channel attention module extracts channel features through global average pooling and global maximum pooling, and generates channel weights in combination with the fully connected layer. The spatial attention module generates spatial features through average pooling and maximum pooling along the channel dimension on the feature map after channel attention weighting, and then uses one-dimensional convolution to generate spatial weights.

[0017] The present invention provides a radio frequency fingerprint recognition method for Internet of Vehicles broadcast frames based on LMMSE channel estimation. The method obtains a channel estimation value by processing the frequency domain signal in the signal, obtains an initial radio frequency fingerprint feature by channel equalization and channel estimation, and then uses a neural network to classify and identify devices, thereby taking into account the correlation between time and frequency, maintaining the robustness of LMMSE, reducing excessive reliance on pilot data in high-noise environments, and suppressing the noise amplification effect. In response to the resource constraint requirements of C-V2X, a lightweight model is constructed based on the analysis of the complexity of the model in this field. The model is robust to different scenarios and noises and has the potential for on-site training and adjustment. The influence of noise and channel on fingerprints is effectively considered, thereby achieving the purpose of more accurately extracting fingerprints of different devices in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and other objects, features and advantages of the present disclosure will become more apparent through a more detailed description of exemplary embodiments of the present disclosure with reference to the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present disclosure.

[0019] Figure 1 A flowchart of a method for radio frequency fingerprint recognition of vehicle network broadcast frames based on LMMSE channel estimation provided by an embodiment of the present disclosure; Figure 2 A schematic diagram of a physical side link broadcast frame structure provided by an embodiment of the present disclosure; Figure 3 Schematic diagram of cross-correlation between a standard PSSS sequence and a received signal provided by an embodiment of the present disclosure; Figure 4a A schematic diagram of an unchanneled radio frequency fingerprint provided in an embodiment of the present disclosure; Figure 4b Schematic diagram of channel processing radio frequency fingerprint provided by an embodiment of the present disclosure; Figure 5 The overall architecture diagram of the improved ShuffleNet V2 module provided in the embodiments of the present disclosure; Figure 6 The inverse residual basic module of the neural network provided by the embodiment of the present disclosure; Figure 7 The inverse residual downsampling module of the neural network provided in the embodiment of the present disclosure; Figure 8 This is a comparison chart of accuracy under different signal-to-noise ratios provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0020] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0021] It should be clear that the following embodiments of the present disclosure are described through specific concrete examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other in the absence of conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0022] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on this disclosure, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this device and / or practice this method.

[0023] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The illustrations only show components related to the present disclosure and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.

[0024] Additionally, in the following description, specific details are provided to provide a thorough understanding of the examples. However, one skilled in the art will appreciate that the aspects described can be practiced without these specific details.

[0025] The radio frequency fingerprint recognition method for broadcast frames of the Internet of Vehicles based on LMMSE channel estimation provided in this embodiment is specifically a radio frequency fingerprint recognition method for broadcast frames of the Internet of Vehicles based on LMMSE channel estimation. The goal is to optimize the channel response estimation by combining the channel covariance matrix and the statistical information of the noise. Based on this, the initial fingerprint feature is generated by removing the channel response, and by improving the ShuffleNet V2 network architecture, the network architecture is optimized for low signal-to-noise ratio and dynamic channel conditions, while having a higher feature extraction capability. During the training process, the training data collected from different scenarios are integrated, and the model generalization ability is enhanced by utilizing diversified samples. In addition, by improving the information richness of the features, the amount of calculation and the number of parameters are effectively reduced, thereby improving the calculation efficiency while maintaining the detection accuracy and achieving network lightweighting. In low signal-to-noise ratio static and mobile scenarios, more accurate device classification can be achieved.

[0026] The received signal in this embodiment is the signal of the physical side link broadcast channel.

[0027] For ease of understanding, this embodiment discloses a method for identifying radio frequency fingerprints of broadcast frames of Internet of Vehicles based on LMMSE channel estimation. The method includes obtaining subcarrier data in a resource grid based on a signal of an acquired physical sidelink broadcast channel; Based on the subcarrier data, the root mean square delay spread and channel autocorrelation matrix are constructed, and the channel estimation value is obtained by combining the prior signal-to-noise ratio regularization and time domain windowing operation; The method of performing root mean square delay spread and channel autocorrelation matrix construction based on subcarrier data, combining prior signal-to-noise ratio regularization and time domain windowing operations to obtain channel estimation includes: The estimated rms delay spread is used to construct the channel autocorrelation matrix; Introducing a priori signal-to-noise ratio regularization term to dynamically adjust the regularization strength in LMMSE estimation; The frequency domain channel response is converted to the time domain through inverse Fourier transform, and a windowing operation is performed to suppress noise and non-channel components. The frequency domain is then restored through Fourier transform to obtain the final channel estimate.

[0028] The initial RF fingerprint features are obtained through channel equalization and channel estimation, and the improved neural network and initial RF fingerprint features are used to classify and identify Internet of Vehicles devices. Therefore, channel equalization technology is used to remove the channel impact and extract the initial RF fingerprint features. The channel impact is obtained based on the final channel estimation.

[0029] Obtain subcarrier data in the resource grid based on the acquired signal of the physical sidelink broadcast channel, including: signal detection, frame synchronization, carrier frequency offset and resource grid demodulation; The signal detection adopts block energy gradient and threshold determination to locate the signal starting point by calculating the energy ratio of adjacent data blocks; The frame synchronization uses the periodic characteristics of the main synchronization signal to perform coarse synchronization and searches for the signal peak point through amplitude normalization cross-correlation operation; The carrier frequency offset compensation utilizes the characteristics that two consecutive primary synchronization symbols are the same and two consecutive secondary synchronization symbols are the same, and adopts the signal autocorrelation of the physical side link broadcast channel to obtain the frequency offset estimation value; The resource grid demodulation utilizes the structural characteristics of the cyclic prefix and eliminates the influence of time domain offset and subcarrier misalignment through phase and half subcarrier compensation to obtain frequency domain resources.

[0030] Signal detection determines the effective signal segment through dynamic interception strategy to ensure the accuracy of the signal starting point; Frame synchronization is based on the strong periodicity of PSSS and SSSS signals, combined with fine synchronization of CP, verifying the standard subframe length of 2048 and optimizing synchronization accuracy; Carrier frequency offset compensation utilizes the same characteristics of the synchronization symbols and combines the phase difference between the CP and the data tail to achieve high-precision frequency offset estimation and compensation.

[0031] like Figure 1 As shown, the process includes: Step S1: The receiver oversamples and downconverts the received signal in the time domain to obtain a baseband received signal, which is then preprocessed. Time synchronization, carrier frequency offset estimation, and compensation are performed to obtain a received signal in the time domain that carries the RF fingerprint and channel information. Resource grid demodulation is then performed to obtain the frequency domain resource. Step S2: Root mean square delay spread (RMS delay spread) and channel autocorrelation matrix construction are combined with prior signal-to-noise ratio regularization and time domain windowing to generate a high-precision frequency domain channel response, effectively suppressing noise and multipath interference. Step S3: Channel equalization is used to extract initial RF fingerprint features, and an improved lightweight neural network is used for device classification and identification. This embodiment demonstrates high robustness in low signal-to-noise ratio and dynamic scenarios, making it suitable for device authentication in connected vehicle-to-everything (C-V2X) scenarios.

[0032] The LMMSE channel estimation is constructed using the root mean square delay spread and the channel autocorrelation matrix. In low signal-to-noise ratio scenarios, the RMS delay spread value is set by default to ensure estimation stability; Perform multi-symbol averaging on the initial frequency domain channel response to smooth out the noise impact; The data differences of the channel autocorrelation matrix are eliminated through symmetry processing, and the main path is retained through windowing operation to suppress noise and RFF interference, thereby obtaining an accurate channel estimation value.

[0033] In the RF fingerprint recognition method for Internet of Vehicles broadcast frames based on LMMSE channel estimation in this embodiment, Figure 2 and Figure 3 As shown in the figure, the Primary Sidelink Synchronization Signal (PSSS) and Secondary Sidelink Synchronization Signal (SSSS) are configured in symbols 2-3 and 12-13, respectively, to achieve fast time-frequency synchronization; the DMRS cluster (symbols 5, 7, and 10) provides high-resolution channel state information (CSI); the GP is deployed in symbol 14 to absorb multipath energy; the remaining symbols are PSBCH symbols. The physical sidelink broadcast channel (PSBCH) signal is received, and the subcarrier data in the resource grid is obtained through signal detection, frame synchronization, carrier frequency offset (CFO), and resource grid demodulation. Specifically, the following steps are involved: Signal detection uses a method based on block energy gradient and threshold judgment to locate the signal starting point, and a dynamic interception strategy to determine the effective signal segment. Specifically: First, define the signal block length as , traverse the received signal ,in, is the baseband signal length. Divide the data into non-overlapping segments and calculate the energy block by block : , in For the When the energy ratio of adjacent blocks exceeds the threshold T, it can be determined that the signal starting point is located at Department, is the sampling point index, for The transposition of , thereby achieving the positioning and detection of the target signal segment; Then, frame synchronization, due to the special characteristics of PSBCH, the number of Fast Fourier Transform (FFT) points is set in the initial stage. , and define the cyclic prefix length vector , where the CP length of the first and eighth symbols is 160 samples, and the rest are 144 samples, which is consistent with the normal CP configuration of the side chain. The coarse synchronization is performed by using the periodic characteristics of the primary synchronization signal (PSSS) and loading the locally generated standardized PSSS sequence. Perform amplitude normalization cross-correlation operation on the received signal to search for the peak point. The formula is: , in is the correlation window length, is the sampling point index, is the correlation coefficient. By searching for the peak point of the correlation curve and , verifying that its interval meets the standard subframe length of 2048. Subsequently, fine synchronization based on cyclic prefix (CP) further optimizes the synchronization accuracy, defines a local search range near the coarse synchronization point, and accumulates the CP cross-correlation of consecutive M grid OFDM symbols. Select the offset with the largest cumulative value as the final synchronization point : , For the The starting point of an OFDM symbol, is the total number of symbols.

[0034] Furthermore, carrier frequency offset compensation uses the characteristics of two consecutive primary synchronization symbols being identical and two consecutive secondary synchronization symbols being identical to obtain a frequency offset estimate using the received signal autocorrelation method, without the need to regenerate the reference sequence. The frequency offset is estimated by extracting the signal segment and calculating the conjugate cross-correlation phase difference: , in , that is, the time interval between two repeated synchronization symbols (measured in sampling points), , and , They are the relevant segments of PSSS and secondary synchronization symbol (SSSS), Secondly, based on the fine compensation of the cyclic prefix, we traverse 13 symbols, extract the CP segment and the data tail segment, calculate the phase difference and average it to obtain a high-precision frequency offset. , and then the signal after phase rotation compensation : , Represents the time interval between the CP and the end of its symbol. The starting position of FFT is , after time domain synchronization, the signal is , where the total frequency offset estimate is , is the sum of the coarse estimate and the fine estimate. And the sampling rate .

[0035] Finally, the resource grid is demodulated. For each symbol, the demodulation process first determines the FFT starting position. , using Calculation, where the factor 0.55 ensures that the FFT starting point is located in the middle of the cyclic prefix, minimizing inter-symbol interference. When demodulating, first extract the length of The time domain sample segment and the factor used for half subcarrier frequency offset compensation Multiply, and then perform FFT transformation on the compensated time domain signal to convert it to the frequency domain. The phase distortion caused by the offset, the transformed frequency domain signal will be further compared with the phase compensation factor Multiply them together and eventually move the zero-frequency component to the center. and Defined as: , is an index vector from 0 to 2047. The method utilizes the structural characteristics of the cyclic prefix and eliminates the influence of time domain offset and subcarrier misalignment through phase and half subcarrier compensation to obtain frequency domain resources.

[0036] Channel estimation is performed on the pre-processed PSBCH signal based on the optimized LMMSE technology, specifically including: First, estimate the channel power To characterize the signal energy, the formula is: , is the number of subcarriers, where for The flattened vector of the channel estimate, if C is 2 or 3, is the PSSS symbol at the corresponding position, containing two columns of 62 subcarriers, with a total of 124 elements. The noise power of the input is estimated by calculating the average energy of the noise segment. Then take a non-negative value to ensure It is not negative and provides channel energy parameters for the subsequent correlation matrix construction. The symbol dimension is averaged, and the least squares estimate of multiple symbols is averaged to smooth the noise and enhance the reliability of the channel response. The inverse fast Fourier transform after zero padding converts the average channel response in the frequency domain into the channel impulse response (CIR) in the time domain, which represents the amplitude and delay information of each path in the multipath channel. The zero frequency position of the CIR is adjusted to ensure that the delay is aligned. The power delay spectrum is calculated to provide the energy distribution of the multipath components. ,but: , It reflects the channel delay characteristics and keeps the delay resolution consistent with the frequency domain by excluding the redundant data introduced by zero padding. Setting a dynamic threshold , to focus on the main path: , Convert the dB value to a linear scale by multiplying The maximum value of gets the threshold, where Ensure a minimum attenuation of 5dB to prevent the threshold from being too high when the signal-to-noise ratio is low, thereby retaining a large amount of noise components. Then normalize the power delay distribution to make it a probability distribution to facilitate statistical delay characteristics. Assume that the delay vector is ,but: , in is the time domain sampling period, is the subcarrier spacing, RMS delay spread It is defined as the root mean square deviation of the delay, and its calculation formula is: , is the average delay, which is the expected delay value of the power delay distribution. If the calculated result is invalid, the default value is used. , suitable for low signal-to-noise ratio scenarios; Then, the channel autocorrelation matrix The construction of is based on the root mean square delay spread, where is the effective channel power, which is obtained by deducting the noise power from the power expectation value of the least squares estimation channel response, and is finally defined as: , is the frequency difference between the subcarriers, multiplied by The phase shift related to the time delay is obtained. It simulates the coherent bandwidth effect caused by the time delay and is applicable to multipath channels. In order to eliminate the error introduced by the numerical calculation, the matrix After adjusting the Hermitian symmetric form, the original matrix and its conjugate transposed matrix are added and divided by 2 to ensure that the matrix proof remains positive. Considering that the channel estimation process in the Internet of Vehicles environment is greatly affected by noise and multipath effects, especially under low signal-to-noise ratio or complex channel conditions, the estimation may drop significantly. Therefore, the correlation matrix based The obtained prior signal-to-noise ratio To optimize the LMMSE estimate, is the trace of the matrix, It combines channel correlation and delay information to reflect more accurate channel statistical characteristics: , And generate a dynamic adjustment regularization term ,when When high, the regularization effect is reduced and channel details are preserved. When is low, increase the regularization strength to enhance the matrix stability. Then, the frequency domain channel estimate is calculated by the following formula : , is the identity matrix, The LMMSE weight matrix is ​​used to correct for noise and systematic errors in the LS estimate, improving estimation accuracy. In multipath channels, the CIR is typically concentrated in a small number of samples in the time domain. This is related to the RMS delay spread. In C-V2X, typical values ​​range from 0.1 to 5 microseconds, reflecting the transient nature of multipath paths in urban or highway environments. In contrast, since noise and RF interference are more widely distributed in the time domain and lack the CIR concentration, the frequency domain channel estimate can be further converted to the time domain using the IDFT and windowed. This removes wide-ranging noise and RFF, making the channel estimate more accurate. This suppresses the noise component while preserving the channel's primary multipath paths. Finally, the time domain estimate is converted to the frequency domain using the DFT to obtain a frequency domain estimate. In low- and medium-speed vehicle scenarios, the channel transformation experienced by adjacent symbols is minimal, so the frequency domain estimates for adjacent symbols can be approximately equal. The frequency domain estimates for adjacent symbols are averaged, leveraging the redundancy of the reference signal to smooth out the noise and interference, resulting in the final frequency domain channel estimate.

[0037] Finally, the influence of the wireless channel is removed, the channel information is removed through channel equalization technology, and the initial RFF is extracted. After preprocessing, CP removal and DFT, the received signal of the Cth frequency domain SC-FDMA symbol is expressed as , the channel equalization operation is specifically as follows: , The initial RFFs corresponding to the same data sequence can be further averaged and subsequently used for identification in the proposed network; Extracted fingerprints, such as Figure 4a and Figure 4b As shown; The obtained RF fingerprint data is trained using the improved ShuffleNet V2 network for cross-classification in different scenarios; Among them, the network is improved in a targeted manner, including the expansion of channel and layer structure, the introduction of attention mechanism, frequency domain signal optimization strategy, etc. The overall architecture is as follows Figure 5 As shown; The Inverted Residual module is the core component of ShuffleNet V2 and is used to build each stage of the network. Figure 6 As shown in Figure 1, when stride=1, channel splitting is used to reduce the amount of computation while retaining the identity mapping to support residual links, as shown in Figure 1. Figure 7 As shown in the figure, when stride=2, the feature size is processed through two branches and the number of channels is increased. And considering that the extracted RFF signal is one-dimensional frequency domain data, in order to adapt to this characteristic, the network computing unit is adjusted to a single-dimensional configuration to achieve feature extraction along the frequency domain axis, so that the correlation between frequency points can be directly learned and modeled. For the adjustment of channel and layer structure, as shown in the figure, Figure 5 As shown, the number of blocks in Stage 2 is adjusted to 4, resulting in 116 output channels; Stage 3 is adjusted to 8, resulting in 232 output channels. The deepened network structure enhances the ability to extract deep patterns in frequency domain signals, making it particularly suitable for capturing subtle spectral differences between devices. Channel expansion improves feature diversity, enabling the network to represent more potential frequency domain features. Stage 4 maintains 1024 channels to strike a balance between expressiveness and computational cost.

[0038] The convolution block attention module is integrated after the final convolution layer. Through the joint optimization of channel attention and spatial attention, key spectral features are dynamically focused. The channel and spatial attention formulas are: , The channel attention module extracts channel features through global average pooling and global maximum pooling, and generates channel weights in combination with the fully connected layer. The spatial attention module generates spatial features through average pooling and maximum pooling along the channel dimension on the feature map after channel attention weighting, and then uses one-dimensional convolution to generate spatial weights.

[0039] The channel shuffle operation of ShuffleNet V2 is retained and alleviated by efficient feature fusion. batch size, is the number of groups, is the number of channels, is the signal length, and the dimension realization process during the channel dimension disruption is: , Finally, cosine annealing learning rate scheduling is used to optimize the training process. The formula is: , initial learning rate, Minimum learning rate, is the total number of training rounds, is the current training round, For the The specific learning rate value used for round training model; During the training process, training data sets from different scenarios are combined to enhance the generalization ability of the model through diversified samples.

[0040] The improved ShuffleNet V2 network uses channel shuffling operations and combines it with a convolutional neural network framework with an attention mechanism to build a fingerprint classification model. It further includes: Channel shuffling operation, optimizing computational efficiency through feature fusion, and is the feature after channel segmentation, is the channel splicing operation, defined as: , The convolutional block attention module enhances the ability to extract key frequency offset features in RFF signals by jointly optimizing the channel and spatial dimensions.

[0041] The classification accuracy of the method in this embodiment reaches 96.76% and 91.05% in static scenes and dynamic scenes with low signal-to-noise ratio, respectively.

[0042] This embodiment is suitable for C-V2X communication scenarios. It uses 12 identical modules for transmission, with a total bandwidth of 20 MHz. The receiving device is equipped with a universal software radio peripheral B205 with a 5.9 GHz carrier frequency and a sampling rate of 30.72 Msps. Physical sidelink broadcast data with different signal-to-noise ratios in different scenarios is collected to form a dataset, which is then divided into training and test sets in appropriate proportions.

[0043] This embodiment also discloses a radio frequency fingerprint recognition system, including: A signal receiving module, used to receive the PSBCH signal of the C-V2X system; A signal preprocessing module for performing signal detection based on block energy gradient, coarse synchronization based on PSSS and fine synchronization based on CP, and CFO compensation based on synchronization signal and CP; The channel estimation module uses a sampling-optimized LMMSE algorithm to generate frequency domain channel responses and extract RFF features through multi-symbol LS estimation, RMS delay spread calculation, channel autocorrelation matrix construction, and time domain lengthening operations. The classification module uses an improved ShuffleNet V2 network to perform device identification on processed RFF features through expanded channels, convolutional fast attention mechanism, and learning rate scheduling. The dataset integration module is used to combine training data from visual and non-visual, static and dynamic scenes to optimize the generalization ability of the model.

[0044] The method of this embodiment obtains the statistical characteristics of the channel response through the channel covariance matrix, takes into account the correlation between time and frequency, maintains the robustness of LMMSE, reduces excessive reliance on pilot data in high-noise environments, and suppresses the noise amplification effect. In response to the resource constraints of C-V2X, based on the analysis of the complexity of the model in this field, a lightweight recognition network is constructed. It is robust to different scenarios and noise and has the potential for on-site training and adjustment. It effectively considers the impact of noise and channel on fingerprints and can more accurately extract fingerprints of different devices in complex environments. It overcomes the limitations of traditional methods and effectively solves problems such as incomplete traditional fingerprint extraction and poor classification accuracy in C-V2X systems.

[0045] This embodiment provides a method for identifying radio frequency fingerprints of broadcast frames of an Internet of Vehicles (IoV) based on LMMSE channel estimation, including the following steps: Collect physical sidelink broadcast data with different signal-to-noise ratios in different scenarios to form a dataset, which is then divided into a training set and a test set in proportion. Use the trained fingerprint model to identify devices in other scenarios; In order to verify the radio frequency fingerprint and model construction method, the identification method of this embodiment is compared with various traditional identification methods: The data sources used are disclosed as follows: For the experiment, 12 identical C-V2X modules were used to transmit PSBCH subframes with a total bandwidth of 20 MHz. Signals were received using a general-purpose software radio peripheral (B205) with a carrier frequency of 5.9 GHz and a sampling rate of 30.72 Msps. Data was collected in three different environments: direct connection, stationary LOS and NLOS scenarios, and mobile scenarios (MOV1 (LOS), MOV2 (NLOS), and MOV3 (LOS+NLOS). The PSBCH transmissions were randomized, resulting in approximately 1000 subframes per device per scenario. After preprocessing and fingerprint extraction, training and test datasets were constructed. All models were implemented in Python 3.12, with each model trained for approximately 100 epochs. The best performing model was selected for testing.

[0046] In addition, in view of the diversity of data collection scenarios and the balanced number of data packets for each device, this embodiment uses Overall Accuracy as the main evaluation indicator to measure the proportion of correctly classified samples of device fingerprints in each scenario. When the distribution of categories is uniform, it is equivalent to Macro-Recall. In addition, the confusion matrix provides detailed prediction performance for each category, and the Macro-Precision score is the harmonic mean of precision and recall to evaluate the balanced performance of the model across categories, as shown in the following formula: , In the above formula, Indicates that for a specific category i, it actually belongs to the category and is correctly classified by the model as category A sample of It does not actually belong to the category But it is misclassified by the model as category A sample of That is, it actually belongs to the category But the samples that are misclassified as other categories by the model, Indicates that it does not actually belong to the category The total number of samples that are correctly predicted by the model as other categories.

[0047] The various recognition methods compared with the fingerprint recognition method of the present invention are divided into three categories: comparison of channel processing under different signal-to-noise ratios; classification under different scenarios; comparison of different deep learning models.

[0048] (1) Classification experiments under different signal-to-noise ratios: Figure 8 The variation trend of recognition accuracy in the range of SNR from 0 dB to 30 dB is shown. Figure 8 As can be seen, the accuracy of both methods increases with increasing SNR. In the low SNR region (0-10 dB), the LS method's accuracy is lower, with a low initial value and slow growth, indicating limited noise suppression capabilities. In contrast, the LMMSE method effectively suppresses noise interference at low SNRs by leveraging channel autocorrelation and noise variance information. When the SNR is below 5 dB, the accuracy remains above 90%, with a more stable growth rate, demonstrating its robustness advantage at low SNRs. When the SNR reaches above 20 dB, the accuracy of both methods approaches 100%, but LMMSE maintains a slight advantage, highlighting its robustness in complex channel conditions.

[0049] Conduct cross-experiments in different scenarios: Classification performance in different scenarios when the training and test sets are at the same location and route. Table 1 shows the cross-certification results of the fingerprint extracted after the proposed LMMSE channel estimation in a multi-scenario test set, achieving high classification accuracy, with the highest accuracy of 100% and the lowest accuracy of 92.64%. Experimental results show that different training set combinations significantly affect the classification performance of the test set. Taking the direct connection scenario as an example, when the training set contains only direct connection data, the cross-certification accuracy is 99.75%, 98.82%, and 99.01%, 98.09%, and 99.59 in LOS, NLOS, and mobile scenarios 1-3, respectively. However, when the training set is expanded to include a combination of direct connection and NLOS, the accuracy in the NLOS scenario reaches 100%, while the accuracy in the direct connection scenario drops to 95.64%. This may be attributed to the multipath effect in the NLOS scenario, which enhances the discriminability of the fingerprint, while the channel equalization in the direct connection scenario weakens the feature diversity. Furthermore, cross-experiments with some mobile scenarios performed better than those with static scenarios. This is because static scenarios, due to the same acquisition location and consistent channels, result in a lack of discriminability in the extracted fingerprints after channel equalization. Overall, however, the combination of different training sets significantly impacts classification performance.

[0050] Table 1 Performance comparison of fingerprint classification in different scenarios

[0051] Model comparison experiment: To evaluate the generalizability of our improved approach to other deep learning models, we conducted comparative analysis with other deep learning models, including DenseNet, MobileNet, EfficientNet, MobileBit, and ConvNeXt. Table 2 summarizes the performance of each model in mobile scenarios. Our model achieves an overall accuracy of 99.01%, and also performs exceptionally well in terms of macro-precision and macro-F1 scores, outperforming other models. The improved channel estimation provides more precise device fingerprint features for the model input, and the introduction of a computationally inexpensive convolutional block attention mechanism enhances the model's ability to focus on discriminative features, particularly in the complex environments of mobile scenarios. In contrast, despite their computational efficiency advantages, DenseNet and MobileNet exhibit lower precision and F1 scores, demonstrating their limitations in processing multi-scenario fingerprint datasets. Notably, most existing models rely on offline training. Differences between offline and field data, such as device aging, can lead to poor performance in real-world applications. The lightweight nature of our model facilitates on-site training and tuning.

[0052] Table 2 Comparison of fingerprint classification performance of different models Deep Learning Models Overall accuracy(%) Macro Precision(%) Macro F1(%) DenseNet 95.19 93.15 95.58 MobileNet 92.7 88.93 88.12 EfficientNet 96.63 97.02 96.24 MobileVit 86.47 85.08 84.22 ConvNeXt 96.72 94 95.42 ShuffleNet V2 96.74 95.02 95.09 This embodiment model 99.01 97.44 98.05 After a series of verification and analysis, the results show that the fingerprint recognition method in this embodiment has significant advantages in improving classification accuracy and coping with complex environments. The fingerprint can still maintain a high recognition accuracy when facing a low signal-to-noise ratio.

[0053] An electronic device according to an embodiment of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), a hard disk, flash memory, etc.

[0054] The processor can be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and can control other components in the electronic device to perform desired functions. In one embodiment of the present disclosure, the processor is used to execute the computer-readable instructions stored in the memory, causing the electronic device to perform all or part of the steps of the aforementioned LMMSE channel estimation-based radio frequency fingerprinting method for vehicle network broadcast frames in various embodiments of the present disclosure.

[0055] Those skilled in the art should understand that in order to solve the technical problem of how to obtain a good user experience, this embodiment may also include well-known structures such as a communication bus and an interface, and these well-known structures should also be included in the scope of protection of this disclosure.

[0056] The electronic device provided by the embodiments of the present disclosure may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes based on a program stored in a read-only memory (ROM) or a program loaded from a storage device into a random access memory (RAM). The RAM also stores various programs and data required for the operation of the electronic device. The processing device, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0057] Typically, the following devices can be connected to an I / O interface: input devices such as sensors or visual information acquisition devices; output devices such as display screens; storage devices such as magnetic tapes and hard drives; and communication devices. Communication devices allow electronic devices to communicate with other devices (such as edge computing devices) wirelessly or wired to exchange data.

[0058] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication device, or installed from a storage device, or installed from a ROM. When the computer program is executed by the processing device, all or part of the steps of the radio frequency fingerprint recognition method for the Internet of Vehicles broadcast frame based on LMMSE channel estimation of the embodiment of the present disclosure are executed.

[0059] For detailed description of this embodiment, please refer to the corresponding description in the aforementioned embodiments, which will not be repeated here.

[0060] According to an embodiment of the present disclosure, a computer-readable storage medium stores non-transitory computer-readable instructions. When executed by a processor, the non-transitory computer-readable instructions execute all or part of the steps of the aforementioned LMMSE channel estimation-based RF fingerprinting method for broadcast frames in an Internet of Vehicles (IoV) according to each embodiment of the present disclosure.

[0061] The above-mentioned computer-readable storage media include, but are not limited to, optical storage media (e.g., CD-ROMs and DVDs), magneto-optical storage media (e.g., MOs), magnetic storage media (e.g., magnetic tapes or mobile hard disks), media with built-in rewritable non-volatile memory (e.g., memory cards), and media with built-in ROM (e.g., ROM cartridges).

[0062] For detailed description of this embodiment, please refer to the corresponding description in the aforementioned embodiments, which will not be repeated here.

[0063] The basic principles of the present disclosure have been described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, and effects mentioned in this disclosure are merely illustrative and not restrictive, and should not be construed as necessarily possessed by each embodiment of the present disclosure. Furthermore, the specific details disclosed above are provided for illustrative purposes and to facilitate understanding, rather than as limitations. These details do not limit the present disclosure to necessarily being implemented using these specific details.

[0064] In the present disclosure, relational terms such as first and second, etc. are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. The block diagrams of the devices, devices, equipment, and systems involved in the present disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As will be appreciated by those skilled in the art, these devices, devices, equipment, and systems can be connected, arranged, or configured in any manner. Words such as "including," "comprising," "having," and the like are open-ended words, meaning "including but not limited to," and can be used interchangeably therewith. The words "or" and "and" used herein refer to the words "and / or" and can be used interchangeably therewith, unless the context clearly indicates otherwise. The word "such as" used herein refers to the phrase "such as but not limited to," and can be used interchangeably therewith.

[0065] Additionally, as used herein, "or" used in a list of items beginning with "at least one" indicates a separate list, so that, for example, a list of "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not mean that the example described is preferred or better than other examples.

[0066] It should also be noted that in the system and method of the present disclosure, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present disclosure.

[0067] Various changes, substitutions, and modifications may be made to the technology described herein without departing from the teachings defined by the appended claims. Moreover, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, compositions of things, means, methods, and actions described above. Currently existing or later developed processes, machines, manufactures, compositions of things, means, methods, or actions that perform substantially the same function or achieve substantially the same results as the corresponding aspects described herein may be utilized. Accordingly, the appended claims include within their scope such processes, machines, manufactures, compositions of things, means, methods, or actions.

[0068] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present disclosure. Therefore, the present disclosure is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0069] The above description has been provided for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present disclosure to the forms disclosed herein. Although a number of example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A radio frequency fingerprint recognition method for Internet of Vehicles broadcast frames based on LMMSE channel estimation, characterized in that: include: Obtaining subcarrier data in a resource grid based on the acquired signal of the physical sidelink broadcast channel; Based on the subcarrier data, the root mean square delay spread and channel autocorrelation matrix are constructed, and the channel estimation value is obtained by combining the prior signal-to-noise ratio regularization and time domain windowing operation; The initial radio frequency fingerprint features are obtained through channel equalization and channel estimation, and the improved neural network and the initial radio frequency fingerprint features are used to classify and identify Internet of Vehicles devices.

2. The radio frequency fingerprint recognition method for Internet of Vehicles broadcast frames based on LMMSE channel estimation according to claim 1 is characterized in that: Obtain subcarrier data in the resource grid based on the acquired signal of the physical sidelink broadcast channel, including: signal detection, frame synchronization, carrier frequency offset and resource grid demodulation; The signal detection adopts block energy gradient and threshold determination to locate the signal starting point by calculating the energy ratio of adjacent data blocks; The frame synchronization uses the periodic characteristics of the main synchronization signal to perform coarse synchronization and searches for the signal peak point through amplitude normalization cross-correlation operation; The carrier frequency offset compensation utilizes the characteristics that two consecutive primary synchronization symbols are the same and two consecutive secondary synchronization symbols are the same, and adopts the signal autocorrelation of the physical side link broadcast channel to obtain the frequency offset estimation value; The resource grid demodulation utilizes the structural characteristics of the cyclic prefix and eliminates the influence of time domain offset and subcarrier misalignment through phase and half subcarrier compensation to obtain frequency domain resources.

3. The radio frequency fingerprint recognition method for Internet of Vehicles broadcast frames based on LMMSE channel estimation according to claim 2 is characterized in that: The signal detection comprises: Define the signal block length as , traverse the signal of the physical side link broadcast channel, according to the block length Divide the data into non-overlapping segments and calculate the energy block by block : ， in For the Grid data block, is the sampling point index, for The transpose of is a natural number.

4. The radio frequency fingerprint recognition method for Internet of Vehicles broadcast frames based on LMMSE channel estimation according to claim 3 is characterized in that: The carrier frequency offset compensation includes: Extract the signal segments and calculate the conjugate cross-correlation phase difference to estimate the frequency offset: ， in, is the frequency deviation, , is the number of Fourier transform points, and is the relevant segment of the main synchronization signal, and is the related segment of the auxiliary synchronization symbol, and n is a natural number.

5. The radio frequency fingerprint recognition method for Internet of Vehicles broadcast frames based on LMMSE channel estimation according to claim 4 is characterized in that: The resource grid demodulation includes: Determine the starting position of the Fourier transform and then extract the length The time domain sample segment of length The time domain sample segment and factor Multiply, then perform Fourier transform on the time domain signal after multiplication to convert the time domain sample segment to the frequency domain, factor Used for half-subcarrier frequency offset compensation.

6. The method for radio frequency fingerprint recognition of Internet of Vehicles broadcast frames based on LMMSE channel estimation according to claim 5 is characterized in that: The root mean square delay spread and channel autocorrelation matrix construction based on subcarrier data, combined with prior signal-to-noise ratio regularization and time domain windowing operations, to obtain a channel estimate value, includes: Estimate channel power to characterize signal energy; The channel autocorrelation matrix is ​​constructed based on the estimated channel power and RMS delay spread.

7. The method for radio frequency fingerprint recognition of vehicle network broadcast frames based on LMMSE channel estimation according to claim 6 is characterized in that: The channel autocorrelation matrix is ​​constructed based on the estimated channel power and a delay-dependent phase offset, wherein the delay-dependent phase offset is obtained based on a frequency difference between subcarriers.

8. The method for radio frequency fingerprint recognition of Internet of Vehicles broadcast frames based on LMMSE channel estimation according to claim 7 is characterized in that: Combining the prior signal-to-noise ratio regularization and time-domain windowing operations, the channel estimation value is obtained, including: Generate dynamically adjusted regularization terms , the regularization term It is used to reduce the effect of regularization when the prior signal-to-noise ratio is high, and to increase the regularization strength when the prior signal-to-noise ratio is low.

9. The method for radio frequency fingerprint recognition of Internet of Vehicles broadcast frames based on LMMSE channel estimation according to claim 8, characterized in that: The root mean square delay spread and channel autocorrelation matrix construction based on subcarrier data, combined with prior signal-to-noise ratio regularization and time domain windowing operations, to obtain a channel estimate value, includes: In low signal-to-noise ratio scenarios, the default RMS delay spread value is set; Perform multi-symbol averaging on the initial frequency-domain channel response to smooth out the noise impact; The data differences of the channel autocorrelation matrix are eliminated through symmetry processing, and the main path is retained through windowing operation to suppress noise and RF fingerprint interference.

10. The method for radio frequency fingerprint recognition of Internet of Vehicles broadcast frames based on LMMSE channel estimation according to claim 9, characterized in that: The improved neural network is used to obtain the initial RF fingerprint features through channel equalization and channel estimation, and the improved neural network and the initial RF fingerprint features are used to classify and identify Internet of Vehicles devices. The improved neural network is the improved ShuffleNet V2 network, which includes: The convolutional block attention module is integrated after the final convolutional layer to dynamically focus on key spectral features through joint optimization of channel attention and spatial attention. The channel attention module extracts channel features through global average pooling and global maximum pooling, and generates channel weights in combination with the fully connected layer. The spatial attention module generates spatial features through average pooling and maximum pooling along the channel dimension on the feature map after channel attention weighting, and then uses one-dimensional convolution to generate spatial weights.

Images