A radio frequency fingerprinting method based on time scattering feature extraction

Abstract

The present application relates to the field of wireless communication and information security technology, and particularly relates to a radio frequency fingerprint identification method based on time scattering feature extraction, which can be used for physical layer identity authentication of intelligent networked vehicles, Internet of Vehicles terminals and other wireless devices. The method comprises the following steps: step S1: obtaining I / Q baseband signal samples of a terminal to be identified; step S2: pre-processing the I / Q signal samples; step S3: inputting the pre-processed I channel signal and Q channel signal into a time scattering network respectively, independently extracting multi-scale scattering coefficients, and obtaining double-channel time scattering features through scattering transformation; step S4: compressing and fusing the scattering coefficients to construct a final feature vector; and step S5: inputting the feature vector obtained in step S4 into a classification model to output a corresponding terminal identity category.

Description

Technical Field

[0001] This invention relates to the fields of wireless communication and information security technology, and in particular to a radio frequency fingerprint recognition method based on time scattering feature extraction, which can be used for physical layer identity authentication of intelligent connected vehicles, vehicle networking terminals and other wireless devices. Background Technology

[0002] Intelligent connected vehicles rely on V2X wireless communication to enable information exchange between vehicles and between vehicles and roadside facilities. However, open wireless channels are vulnerable to security threats such as identity forgery, signal replay, and unauthorized access. Existing authentication methods based on cryptography and certificate systems suffer from high overhead and insufficient real-time performance in high-dynamic, low-latency scenarios. Therefore, leveraging the inherent differences in the device's radio frequency front-end to achieve physical layer-assisted identity authentication has significant application value.

[0003] Radio frequency fingerprinting typically distinguishes different devices by analyzing the I / Q baseband features in the transmitted signals of wireless terminals. Existing technologies employ several methods. One approach directly flattens the original I / Q samples before inputting them into a classifier for identification. While simple to implement, this method is sensitive to noise, time shifts, local deformations, and channel disturbances. Another approach models the data using spectral or time-frequency transformation features, but its feature representations have limited stability in complex propagation environments. A third approach uses deep learning models for end-to-end identification. While possessing strong automatic feature learning capabilities, this approach usually relies on a large-scale training sample size, exhibiting insufficient generalization performance under conditions of small samples, domain offset, and complex channels, and also incurs high training and deployment costs.

[0004] Especially in the highly mobile scenarios of intelligent connected vehicles, the wireless propagation environment is generally characterized by additive white Gaussian noise, multipath fading, and Doppler frequency shift, which significantly change the time-domain structure and phase evolution characteristics of the I / Q baseband signal. This makes it difficult for existing radio frequency fingerprint recognition methods to simultaneously achieve feature stability, recognition accuracy, and adaptability to complex scenarios.

[0005] Therefore, there is an urgent need for a radio frequency fingerprinting method that can remain stable under time shifts and local deformations, is suitable for small sample conditions, and still has good robustness under complex channel disturbances. Summary of the Invention

[0006] This invention aims to provide a radio frequency fingerprinting method based on time scattering feature extraction, in order to solve the problems of existing radio frequency fingerprinting methods being sensitive to noise, multipath fading and Doppler frequency shift, having insufficient feature stability, and having poor generalization ability under small sample conditions in the high-mobility communication scenarios of intelligent connected vehicles.

[0007] This invention provides a radio frequency fingerprinting method based on time scattering feature extraction for physical layer identification of wireless terminals, especially the transmitting devices of intelligent connected vehicle communication terminals. The method takes the I / Q baseband sampling sequence of the terminal's transmitted signal as input, constructs a dual-channel time scattering feature extraction mechanism to extract multi-scale feature representations that are stable against time shifts, local deformations, and some channel disturbances, and then combines this with a classification model to complete terminal identification.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] A radio frequency fingerprint recognition method based on time scattering feature extraction, the method comprising the following steps:

[0010] Step S1: Obtain I / Q baseband signal samples of the terminal to be identified;

[0011] Step S2: Preprocess the I / Q signal samples;

[0012] Step S3: Input the preprocessed I-channel signal and Q-channel signal into the time scattering network respectively, extract the multi-scale scattering coefficients independently, and obtain the dual-channel time scattering characteristics through scattering transformation;

[0013] Step S4: Compress and fuse the scattering coefficients to construct the final radio frequency fingerprint feature vector;

[0014] Step S5: Input the final radio frequency fingerprint feature vector obtained in step S4 into the classification model and output the corresponding terminal identity category.

[0015] In a further optimization of this technical solution, in step S1, the complex baseband signal obtained after downconversion and sampling of the wireless terminal's transmitted signal is collected or retrieved, and represented as a dual-channel I / Q discrete time series; wherein, the I channel represents the in-phase component, the Q channel represents the quadrature component, and each sample consists of time-series sampling points of a fixed length.

[0016] In a further optimization of this technical solution, the preprocessing step S2 is as follows: standardize the I and Q channels of each sample respectively, calculate the mean and standard deviation along each channel, and complete sample-level normalization.

[0017] In a further optimization of this technical solution, the time scattering network consists of a fixed wavelet filter bank, nonlinear modulus operation, and low-pass averaging operation. It does not rely on training to learn filter parameters, but obtains a stable structural representation of the input signal at different time scales through multi-layer cascaded operations.

[0018] This technical solution is further optimized, wherein the scattering coefficients include 0th-order scattering coefficients, 1st-order scattering coefficients and 2nd-order scattering coefficients; wherein, the 0th-order scattering coefficients reflect the low-frequency average characteristics of the signal; the 1st-order scattering coefficients are used to characterize the first-order local time-frequency structure of the signal after wavelet convolution; and the 2nd-order scattering coefficients further describe the higher-level local variation characteristics in the first-order coefficients.

[0019] Further optimization of this technical solution, step S4 is as follows: logarithmically compress the scattering coefficients of the I-path and Q-path respectively; then, flatten the two scattering coefficients respectively, and splice them in a preset order to form the final radio frequency fingerprint feature vector characterizing the current sample.

[0020] In a further optimization of this technical solution, the classification model is a support vector machine, random forest, or other lightweight classifier suitable for small sample classification tasks.

[0021] Further optimization of this technical solution includes introducing channel disturbance samples in step S1, wherein the channel disturbances include:

[0022] (1) Additive white Gaussian noise perturbation: By superimposing white Gaussian noise under different signal-to-noise ratio conditions onto the original I / Q samples, a low signal-to-noise ratio communication environment is simulated;

[0023] (2) Multipath fading disturbance: By constructing a multi-tap discrete time delay channel, different time delays, different amplitudes and different phases are applied to the complex baseband signal to simulate complex propagation scenarios;

[0024] (3) Doppler frequency shift disturbance: by applying a phase rotation that accumulates over time to the complex baseband signal, the frequency shift effect caused by the high-speed relative motion of the terminal is simulated.

[0025] Unlike existing technologies, the above technical solution has the following beneficial effects:

[0026] (1) By performing time scattering transformation on the I / Q dual-channel signal, the present invention extracts multi-scale stable features, which can improve the adaptability to time shift, local deformation and complex channel disturbance, thereby enhancing the stability and distinguishability of radio frequency fingerprint features.

[0027] (2) The present invention uses a fixed wavelet filter to construct the time scattering feature extraction front end, which does not rely on large-scale labeled samples for end-to-end training. Therefore, it can still achieve good recognition results under small sample conditions and is suitable for wireless terminal identity recognition scenarios with limited sample size.

[0028] (3) The present invention adopts the technical route of "time scattering feature extraction + lightweight classifier", which reduces the model training complexity and deployment difficulty while ensuring recognition performance, and is more suitable for application environments with limited resources or high real-time requirements.

[0029] (4) This invention can be used for physical layer identity authentication of intelligent connected vehicles, vehicle network terminals and other wireless devices, and has good engineering application value and promotion significance. Attached Figure Description

[0030] Figure 1 The flowchart shows the overall process of the radio frequency fingerprinting method based on time scattering features.

[0031] Figure 2 This is a schematic diagram of the time scattering feature extraction structure. Detailed Implementation

[0032] To explain in detail the technical content, structural features, objectives, and effects of the technical solution, the following description is provided in conjunction with specific embodiments and accompanying drawings.

[0033] Step S1: Obtain I / Q signal samples of the terminal to be identified.

[0034] The complex baseband signal obtained by down-converting and sampling the transmitted signal from the wireless terminal is collected or retrieved at the receiving end, and represented as a dual-channel I / Q discrete-time sequence. Here, the I channel represents the in-phase component, and the Q channel represents the quadrature component. Each sample consists of time-series sampling points of a fixed length. In this embodiment, a single sample can be represented as a dual-channel sequence matrix of length 256, with each time sampling point containing both I-channel and Q-channel sample values. The input sample data can be derived from publicly available datasets or actual collected data.

[0035] Step S2: Preprocess the I / Q signal samples.

[0036] To reduce the impact of numerical scale differences between samples on feature extraction and classification, the I and Q channels of each sample are standardized separately. Preferably, the mean and standard deviation are calculated for each channel, and sample-level normalization is performed. This ensures that the processed dual-channel samples maintain the original temporal structure while reducing training instability caused by amplitude fluctuations. This step provides a unified input basis for subsequent scattering feature extraction and classifier training.

[0037] Step S3: Construct dual-channel time-scattering features

[0038] The preprocessed I-channel and Q-channel signals are input into a one-dimensional time-scattering network to independently extract multi-scale scattering coefficients. The time-scattering network consists of a fixed wavelet filter bank, nonlinear modulus arithmetic, and low-pass averaging. It does not rely on training to learn filter parameters but instead obtains a stable structural representation of the input signal at different time scales through multi-layer cascaded operations. The scattering coefficients include at least 0th-order, 1st-order, and 2nd-order scattering coefficients.

[0039] Among them, the 0th-order scattering coefficient reflects the low-frequency average characteristics of the signal; the 1st-order scattering coefficient is used to characterize the first-order local time-frequency structure of the signal after wavelet convolution; and the 2nd-order scattering coefficient further describes the higher-level local variation characteristics in the first-order coefficients. By performing the above scattering transformation on the I and Q channels respectively, dual-channel time scattering features with multi-scale characterization capabilities and a certain time-shift stability can be obtained while preserving the difference information between in-phase and quadrature components. The dual-channel time scattering features serve as intermediate feature representations for subsequent compression, fusion, and construction of the final RF fingerprint feature vector.

[0040] Step S4: Compress and fuse the scattering coefficients to construct the final radio frequency fingerprint feature vector.

[0041] Because the dynamic range of the scattering output may be large, logarithmic compression is performed on the I-path and Q-path scattering coefficients to improve the stability of subsequent classification model training. Then, the two scattering coefficients are flattened and concatenated in a preset order to form the final RF fingerprint feature vector representing the current sample. This feature vector contains multi-scale temporal structure information from both the I-path and Q-path, and serves as input to the subsequent classification model for terminal identification.

[0042] Step S5: Perform terminal identity classification and recognition based on the feature vector.

[0043] The feature vector obtained in step S4 is input into the classification model, which outputs the corresponding terminal identity category. The classification model can be a support vector machine, random forest, or other lightweight classifier suitable for small-sample classification tasks. In a preferred embodiment, a support vector machine is used to model the temporal scattering features to obtain higher recognition accuracy and better engineering deployability. After training the model using the training set, inference is performed on the test samples to output the category to which the target terminal belongs, thereby achieving radio frequency fingerprint identification.

[0044] Step S6: Output the recognition results and complete the performance evaluation.

[0045] Steps S1 to S5 are executed on the sample to be identified, and the terminal identity recognition result is output. The recognition performance and robustness of the method can be tested and evaluated under different signal-to-noise ratios, multipath conditions and Doppler conditions.

[0046] Preferably, the time scattering network is a one-dimensional time scattering network; preferably, the input sample is a fixed-length dual-channel I / Q time-series signal; preferably, the classification model is a support vector machine model.

[0047] To enhance the adaptability of this invention in high-mobility communication scenarios for intelligent connected vehicles, when acquiring I / Q baseband signal samples in step S1, channel perturbation samples can be further constructed based on the original acquired samples. These channel perturbation samples can be used for data augmentation during the training phase and / or robustness evaluation during the testing phase. The channel perturbations include:

[0048] (1) Additive white Gaussian noise perturbation: By superimposing white Gaussian noise under different signal-to-noise ratio conditions onto the original I / Q samples, a low signal-to-noise ratio communication environment is simulated;

[0049] (2) Multipath fading disturbance: By constructing a multi-tap discrete time delay channel, different time delays, different amplitudes and different phases are applied to the complex baseband signal to simulate complex propagation scenarios;

[0050] (3) Doppler frequency shift disturbance: by applying a phase rotation that accumulates over time to the complex baseband signal, the frequency shift effect caused by the high-speed relative motion of the terminal is simulated.

[0051] By introducing the aforementioned perturbation samples during the sample acquisition and construction stages, the classification model's adaptability to channel variations such as noise, multipath, and Doppler can be enhanced, thereby improving its recognition stability under domain offset conditions.

[0052] Example 1: A radio frequency fingerprint recognition method based on time scattering feature extraction

[0053] like Figure 1 and Figure 2 As shown, this embodiment provides a radio frequency fingerprint recognition method based on time scattering feature extraction, used for physical layer identity recognition of wireless terminals, especially vehicle-to-everything (V2X) communication terminals in intelligent connected vehicles. Among them, Figure 1 The overall flow of the method of the present invention is shown, including steps such as I / Q baseband signal sample acquisition, sample preprocessing, dual-channel time scattering feature extraction, feature compression and fusion, and classification and recognition. Figure 2 The dual-channel temporal scattering feature extraction and feature fusion structure in steps S3 and S4 is further demonstrated. This method includes the following steps:

[0054] Step 1: Obtain I / Q signal samples.

[0055] The complex baseband data of the transmitted signal from the wireless terminal to be identified, after down-conversion and sampling at the receiving end, is acquired and represented as a dual-channel I / Q discrete-time sequence, where I represents the in-phase component and Q represents the quadrature component. In this embodiment, a single sample is preferably a fixed-length dual-channel time-series signal, and each sampling point simultaneously contains I-channel and Q-channel sample values. Preferably, the input sample length can be set to 256. The input data can be derived from a publicly available dataset or from an actual wireless signal acquisition system.

[0056] Step 2: Perform sample preprocessing.

[0057] To reduce the impact of amplitude scale differences between samples on subsequent feature extraction and classification, the I and Q channels of each I / Q sample are standardized separately. Preferably, the mean and standard deviation of each channel are calculated separately, and sample-level normalization is performed, thereby improving the stability of subsequent feature modeling while preserving the original temporal structure.

[0058] Step 3: Extract dual-channel time scattering features.

[0059] The preprocessed I-channel and Q-channel signals are input into a one-dimensional time-scattering network, and multi-scale feature extraction is performed independently on the two channels. The time-scattering network consists of a fixed wavelet filter bank, modulus nonlinear operation, and low-pass averaging operation, and its output includes 0th-order scattering coefficients, 1st-order scattering coefficients, and 2nd-order scattering coefficients.

[0060] Among them, the 0th-order scattering coefficient reflects the low-frequency average characteristics of the signal; the 1st-order scattering coefficient reflects the main time-frequency structure of the signal at the local scale; and the 2nd-order scattering coefficient reflects the local variation characteristics at a higher level. Since the hardware non-ideals of different transmitting devices will cause subtle differences in the I / Q baseband signals, by performing time scattering transformation on the I / Q dual channels respectively, it is possible to extract radio frequency fingerprint features with multi-scale stability.

[0061] Step 4: Construct feature vectors.

[0062] Because the dynamic range of the time-scattering output is large, logarithmic compression is performed on the I-path and Q-path scattering coefficients to improve the stability of subsequent classification model training. Then, the compressed scattering coefficients are flattened and concatenated in a preset order to form the final RF fingerprint feature vector. This feature vector simultaneously contains multi-scale time-frequency structure information from both the I-path and Q-path.

[0063] Step 5: Complete the terminal identity classification and recognition.

[0064] The feature vector obtained in step 4 is input into the classification model, which outputs the corresponding terminal identity category. The classification model is preferably a lightweight classifier such as a support vector machine or random forest. Preferably, a support vector machine model is used for classification and recognition to balance recognition accuracy and deployment complexity. After training the classification model with a training set, inference can be performed on the test samples to output the category to which the target terminal belongs, thereby achieving radio frequency fingerprint identification.

[0065] In this embodiment, the technical approach of "temporal scattering feature extraction + lightweight classification model" can obtain a relatively stable radio frequency fingerprint feature representation under small sample conditions and improve the terminal recognition effect.

[0066] Example 2: Robust Identification Method under Complex Channel Perturbations

[0067] Based on Example 1, in order to improve the adaptability of the present invention in the highly mobile communication environment of intelligent connected vehicles, this example further constructs a robust identification process under complex channel disturbances.

[0068] Step 1: Construct noise perturbation samples.

[0069] Additive white Gaussian noise under different signal-to-noise ratios is superimposed on the original I / Q samples to simulate low signal-to-noise ratio communication environments. By using the noise-perturbed samples for model testing or reinforcement training, the model's adaptability to random noise can be analyzed and improved.

[0070] Step 2: Construct multipath fading perturbation samples.

[0071] After converting the dual-channel I / Q signal into complex baseband form, a multi-tap discrete-time-delay channel is constructed. Multiple propagation paths with different delays, amplitudes, and phases are superimposed on the original signal to simulate the multipath effect in a real wireless propagation environment. After perturbation, the resulting complex baseband signal is converted back into I / Q dual-channel form for use by the subsequent identification module.

[0072] Step 3: Construct Doppler frequency shift perturbation samples.

[0073] By applying a phase rotation that accumulates linearly over time to the complex baseband signal, the Doppler frequency shift effect caused by the high-speed relative motion of the terminal is simulated. This Doppler perturbation alters the phase evolution and local timing structure of the received signal, thus serving to test the robustness of the method in highly maneuverable scenarios.

[0074] Step 4: Perform reinforcement training or robustness testing.

[0075] Noise disturbance samples, multipath fading disturbance samples, and Doppler frequency shift disturbance samples are used to expand the model training set and / or construct the test set. Then, the feature extraction and classification recognition steps in Example 1 are repeated to obtain terminal recognition results under complex channel conditions.

[0076] By introducing perturbation samples corresponding to the target scene, the generalization ability and recognition stability of the classification model under domain offset conditions can be improved.

[0077] This embodiment demonstrates that the present invention can not only complete terminal identification under controlled conditions, but also improve its adaptability under noise, multipath and Doppler conditions through channel disturbance modeling and enhancement training.

[0078] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Unless otherwise specified, an element defined by the phrase "comprising..." or "including..." does not exclude the presence of additional elements in the process, method, article, or terminal device that includes said element. Additionally, in this document, "greater than," "less than," "exceeding," etc., are understood to exclude the stated number; "above," "below," "within," etc., are understood to include the stated number.

[0079] Although the above embodiments have been described, those skilled in the art, once they understand the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the above descriptions are merely embodiments of the present invention and do not limit the scope of patent protection of the present invention. Any equivalent structural or procedural transformations made using the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A radio frequency fingerprint recognition method based on time scattering feature extraction, characterized in that, The method includes the following steps: Step S1: Obtain I / Q baseband signal samples of the terminal to be identified; Step S2: Preprocess the I / Q signal samples; Step S3: Input the preprocessed I-channel signal and Q-channel signal into the time scattering network respectively, extract the multi-scale scattering coefficients independently, and obtain the dual-channel time scattering characteristics through scattering transformation; Step S4: Compress and fuse the scattering coefficients to construct the final radio frequency fingerprint feature vector; Step S5: Input the final radio frequency fingerprint feature vector obtained in step S4 into the classification model and output the corresponding terminal identity category.

2. The radio frequency fingerprint recognition method based on time scattering feature extraction as described in claim 1, characterized in that, In step S1, the complex baseband signal obtained by downconverting and sampling the signal transmitted by the wireless terminal is collected or retrieved, and represented as a dual-channel I / Q discrete time series; wherein, the I channel represents the in-phase component, the Q channel represents the quadrature component, and each sample consists of time-series sampling points of a fixed length.

3. The time scatter feature extraction based radio frequency fingerprinting method of claim 1, wherein, The preprocessing step S2 is as follows: standardize the I and Q channels of each sample respectively, calculate the mean and standard deviation along each channel, and complete sample-level normalization.

4. The time scatter feature extraction based radio frequency fingerprinting method of claim 1, wherein, The time scattering network consists of a fixed wavelet filter bank, nonlinear modulus operation, and low-pass averaging operation. It does not rely on training to learn filter parameters, but obtains a stable structural representation of the input signal at different time scales through multi-layer cascaded operations.

5. The time scatter feature extraction based radio frequency fingerprinting method of claim 4, wherein, The scattering coefficients include 0th-order scattering coefficients, 1st-order scattering coefficients, and 2nd-order scattering coefficients; wherein, the 0th-order scattering coefficients reflect the low-frequency average characteristics of the signal; the 1st-order scattering coefficients are used to characterize the first-order local time-frequency structure of the signal after wavelet convolution; and the 2nd-order scattering coefficients further describe the higher-level local variation characteristics in the first-order coefficients.

6. The time scatter feature extraction based radio frequency fingerprinting method of claim 1, wherein, Step S4 is as follows: Logarithmically compress the scattering coefficients of the I-path and Q-path respectively; then, flatten the two scattering coefficients respectively and splice them in a preset order to form the final radio frequency fingerprint feature vector representing the current sample.

7. The time scatter feature extraction based radio frequency fingerprinting method of claim 1, wherein, The classification model is a support vector machine, random forest, or other lightweight classifier suitable for small sample classification tasks.

8. The time scatter feature extraction based radio frequency fingerprinting method of claim 1, wherein, It also includes introducing channel disturbance samples in step S1, wherein the channel disturbance includes: (1) Additive white Gaussian noise perturbation: By superimposing white Gaussian noise under different signal-to-noise ratio conditions onto the original I / Q samples, a low signal-to-noise ratio communication environment is simulated; (2) Multipath fading disturbance: By constructing a multi-tap discrete time delay channel, different time delays, different amplitudes and different phases are applied to the complex baseband signal to simulate complex propagation scenarios; (3) Doppler frequency shift disturbance: by applying a phase rotation that accumulates over time to the complex baseband signal, the frequency shift effect caused by the high-speed relative motion of the terminal is simulated.

Images