Radio frequency fingerprint authentication method, device and equipment based on wavelet detail coefficient
Through the combination of wavelet decomposition and convolutional neural network, the detailed partial signals of the RF signal are extracted and two-dimensional images are generated, which solves the problems of high complexity and low accuracy of feature extraction in RF fingerprint recognition, and achieves more efficient recognition and improvement Low-energy RF fingerprint authentication.
Patent Information
- Application Number
- CN202411352365.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-26
- Publication Date
- 2024-12-10
AI Technical Summary
The existing RF fingerprint recognition methods have challenges in the problems of high complexity, low accuracy and low recognition accuracy, especially in the identity authentication of IoT devices. Traditional methods rely on complicated feature engineering and deep learning. The model converges too slowly, resulting in high computational complexity and increased energy consumption.
The detailed partial signals of the radio frequency signal are obtained through wavelet decomposition, a two-dimensional image is generated, and the trained convolutional neural network is used for identification, removing redundant approximate partial signals to extract the fingerprint features of the device, simplifying the feature extraction process and reducing the computational complexity degree and energy consumption.
It improves the accuracy and recognition accuracy of feature extraction, simplifies the feature extraction process, accelerates the convergence speed of neural network models, reduces the energy consumption in the system training stage, and improves the overall efficiency and recognition accuracy.
Smart Images
Figure CN119110299A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of communication security technology, and in particular relates to a radio frequency fingerprint authentication method, device and equipment based on wavelet detail coefficients. Background Art
[0002] With the development of 6G networks, the network environment with massive terminal access has become an important part of building an Internet of Everything communication network. In this environment, the key to intrinsic security lies in an efficient, reliable and low-power identity authentication mechanism. Traditional security mechanisms often rely on key management systems and powerful computing resources, while IoT devices have limited computing power, storage space and battery capacity, which limits their security capabilities. Therefore, a lightweight authentication mechanism is needed that can support large-scale IoT deployment and prevent attackers from maliciously operating the interconnection protocol.
[0003] Radio Frequency Fingerprint Authentication does not require any password or other identity information verification, meets the requirements of high speed, convenience, and lightweight, and is a safe technology. Radio Frequency Fingerprint (RFF) is generated during the manufacturing process of the device due to inevitable minor errors in electronic components. This signal feature information will not disappear with changes in the signal transmission scenario. It is unique and non-replicable, similar to human fingerprints. Radio Frequency Fingerprint Identification (RFFI) uses the unique and unalterable errors generated by the internal hardware manufacturing of wireless devices to achieve device identification and authentication. Therefore, radio frequency fingerprint technology is considered to be an ideal technology for IoT device access authentication.
[0004] Existing RFID fingerprint recognition research can be mainly divided into traditional methods based on manual feature selection and methods based on deep learning:
[0005] Methods based on manual feature selection: These methods mainly start from several major features such as waveform features, parameter features, transform domain features, statistical features, etc., and then use classifiers to identify devices. However, this method is highly dependent on feature extraction algorithms, requires rich knowledge and experience in signal processing, and some hardware features are related to each other and difficult to extract individually through manual methods.
[0006] Methods based on deep learning: With the development of neural network technology in recent years, many studies on RF fingerprint authentication use deep learning to automatically extract RF features, eliminating the need for complex feature engineering. RF fingerprint recognition methods based on deep learning usually use classification neural networks to process the original signal or its transformed form and classify RF signals from different devices. These existing deep learning-based methods have problems such as low recognition accuracy and slow model convergence.
[0007] That is to say, the existing radio frequency fingerprint recognition method has high complexity in extracting radio frequency features, low precision in extracting features, and low recognition accuracy. Summary of the invention
[0008] In order to solve the above problems existing in the prior art, the present invention provides a radio frequency fingerprint authentication method, device and equipment based on wavelet detail coefficients.
[0009] The technical problem to be solved by the present invention is achieved through the following technical solutions:
[0010] The present invention provides a radio frequency fingerprint authentication method based on wavelet detail coefficients, comprising:
[0011] Obtaining an original radio frequency signal to be identified;
[0012] Performing wavelet decomposition on the original radio frequency signal to be identified to obtain approximate partial signals and detail partial signals of different scales; wherein the detail partial signal contains the fingerprint feature of the radio frequency device used to generate the original radio frequency signal to be identified; and the approximate partial signal contains the transmission data of the radio frequency device;
[0013] retaining the detail signal, and generating a two-dimensional image corresponding to the original radio frequency signal to be identified according to the detail signal;
[0014] The two-dimensional image is identified using a trained convolutional neural network model to obtain an identification result that characterizes whether the signal contained in the two-dimensional image is legal, thereby completing the authentication of the radio frequency device; wherein the trained convolutional neural network model is trained using a data set containing multiple training samples, and each training sample is a two-dimensional image corresponding to an original radio frequency signal.
[0015] The present invention also provides a radio frequency fingerprint authentication device based on wavelet detail coefficients, comprising:
[0016] An acquisition module, used for acquiring an original radio frequency signal to be identified;
[0017] A decomposition module, used for performing wavelet decomposition on the original radio frequency signal to be identified to obtain approximate partial signals and detail partial signals of different scales; wherein the detail partial signal contains the fingerprint feature of the radio frequency device used to generate the original radio frequency signal to be identified; and the approximate partial signal contains the transmission data of the radio frequency device;
[0018] A feature extraction and conversion module, used to retain the detail signal and generate a two-dimensional image corresponding to the original radio frequency signal to be identified according to the detail signal;
[0019] An authentication module is used to use a trained convolutional neural network model to identify the two-dimensional image, obtain an identification result used to characterize whether the signal contained in the two-dimensional image is legal, and complete the authentication of the radio frequency signal; wherein the trained convolutional neural network model is trained using a data set containing multiple training samples, and each training sample is a two-dimensional image corresponding to an original radio frequency signal.
[0020] The present invention also provides a radio frequency fingerprint authentication device based on wavelet detail coefficients, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus;
[0021] The memory is used to store computer programs;
[0022] The processor is used to implement the steps of the above-mentioned radio frequency fingerprint authentication method based on wavelet detail coefficients when executing the program stored in the memory.
[0023] Compared with the prior art, the present invention has the following beneficial effects:
[0024] The present invention removes the approximate part signal after wavelet decomposition and only retains the detail part signal as the basis for subsequent feature extraction, thereby filtering out redundant information containing a large amount of transmission data, retaining rich information highly related to the hardware characteristics of the device (fingerprint characteristics of the device), highlighting the key characteristics of the hardware characteristics of the device, not only reducing the complexity of feature extraction, simplifying the feature extraction process, but also improving the accuracy of feature extraction, thereby improving the accuracy of subsequent identification of the extracted features. On this basis, the present invention generates a two-dimensional image based on the retained detail part signal, and uses the obtained two-dimensional image as the input of the convolutional neural network, which can give full play to the advantages of the convolutional neural network in the field of image processing and further improve the accuracy of identification. In addition, by simplifying the feature extraction process, the present invention also reduces the computational complexity in radio frequency fingerprint recognition, accelerates the convergence speed of the neural network model, and reduces the energy consumption in the system training stage, thereby reducing the cost of the server and improving the overall efficiency of the system.
[0025] The present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods. BRIEF DESCRIPTION OF THE DRAWINGS
[0026] Figure 1 It is a flow chart of a radio frequency fingerprint authentication method based on wavelet detail coefficients provided by an embodiment of the present invention;
[0027] Figure 2 is a schematic diagram of a wavelet decomposition and signal extraction provided by an embodiment of the present invention;
[0028] Figure 3 It is a flow chart of generating an I / Q characteristic scatter plot according to a detail signal provided by an embodiment of the present invention;
[0029] Figure 4 is a schematic diagram of the model structure of a convolutional neural network model for identification provided by an embodiment of the present invention;
[0030] Figure 5 is a schematic diagram of an exemplary data set for model training provided by an embodiment of the present invention;
[0031] Figure 6 is an exemplary flowchart of constructing a data set and using the constructed data set to train and test a model provided by an embodiment of the present invention;
[0032] Figure 7 is a comparison chart of the accuracy of a solution using detail coefficients after wavelet decomposition provided by an embodiment of the present invention (i.e., the present invention) and a solution not using detail coefficients after wavelet decomposition;
[0033] Figure 8 It is a comparison chart of the loss values output by two models of the solutions using and not using the wavelet decomposition detail coefficients provided by the embodiments of the present invention. DETAILED DESCRIPTION
[0034] The present invention is further described in detail below with reference to specific embodiments, but the embodiments of the present invention are not limited thereto.
[0035] Figure 1 FIG. 1 is a flow chart of a radio frequency fingerprint authentication method based on wavelet detail coefficients provided by an embodiment of the present invention. Figure 1 As shown, the method includes:
[0036] S101: Acquire an original radio frequency signal to be identified.
[0037] S102, performing wavelet decomposition on the original RF signal to be identified to obtain approximate partial signals and detail partial signals of different scales; wherein the detail partial signal contains the fingerprint feature of the RF device used to generate the original RF signal to be identified; and the approximate partial signal contains the transmission data of the RF device.
[0038] Here, the approximate part is the low-frequency part of the RF signal, which contains a large amount of transmission data, and the detail part, also known as the detail coefficient, is the high-frequency part of the RF signal, which contains rich RF fingerprint information of the RF device.
[0039] Wavelet decomposition is usually implemented by discrete wavelet transform (DWT). The present invention can use any specific wavelet decomposition method to decompose the original radio frequency signal, and the present invention is not limited to this. Exemplarily, the original radio frequency signal can be decomposed by wavelet first-level decomposition to obtain approximate partial signals and detailed partial signals of different scales.
[0040] S103: retain the detail signal, and generate a two-dimensional image corresponding to the original radio frequency signal to be identified according to the detail signal.
[0041] By performing wavelet decomposition and signal retention on the original RF signal to be identified, the effective signal can be accurately extracted from the original RF signal to be identified. Figure 2 This is a schematic diagram of wavelet decomposition and signal extraction.
[0042] S104. Use the trained convolutional neural network model to identify the two-dimensional image, obtain an identification result used to characterize whether the signal contained in the two-dimensional image is legal, and complete the authentication of the radio frequency device; wherein the trained convolutional neural network model is trained using a data set containing multiple training samples, and each training sample is a two-dimensional image corresponding to an original radio frequency signal.
[0043] In some embodiments, the step of “generating a two-dimensional image corresponding to the original radio frequency signal to be identified according to the detail signal” in S103 is implemented by the following steps:
[0044] S1031. Perform transformation processing on the detail signal to obtain a transformed signal.
[0045] Specifically, the Hilbert transform may be performed on the detail signal to obtain a transformed signal.
[0046] S1032. Extract I / Q features from the transformed signal; wherein the extracted I / Q features are a two-dimensional feature matrix.
[0047] S1033. Perform a visualization operation on the two-dimensional feature matrix to convert the two-dimensional feature matrix into a two-dimensional image to obtain an I / Q feature scatter plot, and use the I / Q feature scatter plot as the two-dimensional image corresponding to the original RF signal to be identified.
[0048] Here, by performing a visualization operation on the two-dimensional feature matrix, each element (data point) in the two-dimensional feature matrix can be mapped to a coordinate graph, thereby obtaining an I / Q feature scatter plot corresponding to the two-dimensional feature matrix. For example, Figure 3 It is a flow chart for generating an I / Q characteristic scatter plot based on a detail signal (detail coefficient).
[0049] Please refer to Figure 4 The convolutional neural network model of the present invention includes: two convolutional layers (Convolution), two maximum pooling layers (Max-Pool) and two fully connected layers (Fully-Connected). The image input to the convolutional neural network model is processed by the first convolutional layer, the first maximum pooling layer, the second convolutional layer, the second maximum pooling layer and the two fully connected layers in sequence to obtain the output result.
[0050] In some embodiments, the training samples in the above-mentioned data set for training the convolutional neural network model are obtained by the following method:
[0051] S001. Acquire multiple different original radio frequency signals with labels; the label of each original radio frequency signal is used to indicate whether the original radio frequency signal is a legal signal.
[0052] S002. Perform wavelet decomposition on each original radio frequency signal to obtain approximate partial signals and detail partial signals of different scales of the original radio frequency signal.
[0053] S003. Retain the detail signal of the original RF signal, and generate a two-dimensional image corresponding to the original RF signal based on the detail signal, use the two-dimensional image corresponding to the original RF signal as a training sample, and use the label of the original RF signal as the label of the training sample, to obtain multiple training samples that correspond to multiple different original RF signals with labels.
[0054] It should be noted that the specific implementation principle of "generating a two-dimensional image corresponding to the original RF signal according to the detail signal" in S003 is the same as the specific implementation principle of "generating a two-dimensional image corresponding to the original RF signal to be identified according to the detail signal" in the above S102, and will not be repeated here.
[0055] In some embodiments, the above data set is composed of a training set and a test set. The training set contains multiple training samples, and the test set contains multiple test samples. The principle of obtaining multiple test samples is the same as the principle of obtaining multiple training samples, which will not be repeated here. Figure 5 Schematic diagram of the data set obtained for model training. Since each sample is an image composed of multiple two-dimensional scattering points, the data set is a two-dimensional point set.
[0056] The trained convolutional neural network model of the present invention is obtained by the following training method: obtaining an initial convolutional neural network model; using a data set to iteratively train the initial convolutional neural network model, and in each iterative training, calculating the current loss according to the labels of the training samples used and the output results of the convolutional neural network model, performing reverse gradient propagation according to the calculated current loss to adjust the network parameters of the convolutional neural network model, and stopping the training when a preset termination condition is reached to obtain a trained convolutional neural network model. Exemplarily, the loss during training is a cross entropy loss.
[0057] In some embodiments, after the training is completed, the obtained model can also be tested using a test set in the data set to determine the recognition effect of the model. Figure 6 Flowchart for constructing a dataset and using the constructed dataset to train and test the model.
[0058] The present invention has the following advantages:
[0059] 1) The wireless signal contains a large amount of transmitted data, which is not critical in RF fingerprint recognition and authentication. Retaining this information may interfere with the extraction of RF fingerprint features, increase the complexity of feature extraction, slow down the convergence of the neural network, and reduce the accuracy of recognition. The present invention removes the redundant information containing a large amount of transmitted data by removing the low-frequency components after wavelet decomposition and retains only the detail coefficients as the basis for feature extraction, thereby highlighting the key features of the device hardware characteristics, thereby improving the accuracy of recognition and reducing the complexity of feature extraction.
[0060] 2) The present invention uses Hilbert transform to extract the I / Q features of the signal after wavelet decomposition processing, constructs a two-dimensional feature matrix of the signal, and maps the data points of the two-dimensional matrix to a coordinate graph through visualization operations to generate a data set in the form of an image, giving full play to the advantages of convolutional neural networks in image processing, making the recognition accuracy of convolutional neural networks higher.
[0061] 3) The access of large-scale IoT terminals significantly increases the energy consumption and cost of the server. Even a lightweight authentication scheme based on radio frequency fingerprint consumes a lot of energy during the system training phase. The present invention filters out a large amount of redundant information through wavelet decomposition, retaining only the detail coefficients with high purity of fingerprint features, simplifying feature extraction, making the neural network model converge faster, and reducing the cost of the server.
[0062] The technical effects of the present invention are further illustrated by experiments below.
[0063] Figure 7 The accuracy comparison diagram of the solution using the detail coefficient after wavelet decomposition (i.e., the present invention) and the solution not using the detail coefficient after wavelet decomposition is shown, as shown in FIG. Figure 7 As shown, the accuracy of the scheme without using the detail coefficient after wavelet decomposition mostly hovers between 90% and 95% with the increase of the number of training times, and the highest accuracy is 95.06%; the accuracy of the scheme using the detail coefficient after wavelet decomposition is close to 95% when the number of training rounds is 10, and reaches 98.07% when the number of training rounds is 80, and then hovers around 98%, with very small fluctuations, and the highest reaches 98.87%. This shows that the scheme using the detail coefficient after wavelet decomposition has higher accuracy and stronger stability. The scheme using the detail coefficient after wavelet decomposition can achieve a very high accuracy when the number of training rounds is very small, and the model is basically stable when the number of training rounds is only 80, indicating that the present invention does simplify feature extraction, removes redundant information containing a large amount of transmission data, reduces the computational complexity in radio frequency fingerprint recognition, and improves the accuracy of recognition.
[0064] Figure 8 A comparison chart of the loss values of the two model outputs with and without the wavelet decomposition detail coefficient scheme is shown. It can be seen that the scheme using the wavelet decomposition detail coefficient speeds up the convergence of the neural network model, and the degree of convergence is better, indicating that the system does not need to spend too much cost when training the model to achieve better stability and reliability, thereby improving the overall efficiency of the system.
[0065] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0066] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification.
[0067] In the specification, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude a plurality of components. Certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good effects.
[0068] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.
Claims
1. A radio frequency fingerprint authentication method based on wavelet detail coefficients, characterized in that: include: Obtaining an original radio frequency signal to be identified; Performing wavelet decomposition on the original radio frequency signal to be identified to obtain approximate partial signals and detail partial signals of different scales; wherein the detail partial signal contains the fingerprint feature of the radio frequency device used to generate the original radio frequency signal to be identified; and the approximate partial signal contains the transmission data of the radio frequency device; retaining the detail signal, and generating a two-dimensional image corresponding to the original radio frequency signal to be identified according to the detail signal; The two-dimensional image is identified using a trained convolutional neural network model to obtain an identification result that characterizes whether the signal contained in the two-dimensional image is legal, thereby completing the authentication of the radio frequency device; wherein the trained convolutional neural network model is trained using a data set containing multiple training samples, and each training sample is a two-dimensional image corresponding to an original radio frequency signal.
2. The radio frequency fingerprint authentication method based on wavelet detail coefficients according to claim 1 is characterized in that: The step of generating a two-dimensional image corresponding to the original radio frequency signal to be identified according to the detail portion signal includes: Performing transformation processing on the detail part signal to obtain a transformed signal; Extracting I / Q features from the transformed signal; wherein the extracted I / Q features are a two-dimensional feature matrix; A visualization operation is performed on the two-dimensional feature matrix to convert the two-dimensional feature matrix into a two-dimensional image to obtain an I / Q feature scatter plot, and the I / Q feature scatter plot is used as the two-dimensional image corresponding to the original radio frequency signal to be identified.
3. The radio frequency fingerprint authentication method based on wavelet detail coefficients according to claim 1 is characterized in that: The transforming process of the detail signal to obtain a transformed signal includes: Performing Hilbert transform on the detail signal to obtain a transformed signal.
4. The radio frequency fingerprint authentication method based on wavelet detail coefficients according to claim 1 is characterized in that: The step of performing wavelet decomposition on the original radio frequency signal to be identified to obtain approximate partial signals and detail partial signals of different scales includes: The original radio frequency signal to be identified is subjected to first-level wavelet decomposition to obtain approximate partial signals and detailed partial signals of different scales.
5. The radio frequency fingerprint authentication method based on wavelet detail coefficients according to claim 1, characterized in that: The multiple training samples are obtained by the following method: Acquire multiple different original radio frequency signals with labels; the label of each original radio frequency signal is used to indicate whether the original radio frequency signal is a legal signal; Performing wavelet decomposition on each original radio frequency signal to obtain approximate partial signals and detailed partial signals of different scales of the original radio frequency signal; The detail signal of the original RF signal is retained, and a two-dimensional image corresponding to the original RF signal is generated according to the detail signal, the two-dimensional image corresponding to the original RF signal is used as a training sample, and the label of the original RF signal is used as the label of the training sample, so as to obtain the multiple training samples that correspond one to one to the multiple different original RF signals with labels.
6. The radio frequency fingerprint authentication method based on wavelet detail coefficients according to claim 1, characterized in that: The trained convolutional neural network model is obtained by the following training method: Get the initial convolutional neural network model; The initial convolutional neural network model is iteratively trained using the data set, and in each iterative training, the current loss is calculated according to the labels of the training samples used and the output results of the convolutional neural network model, and reverse gradient propagation is performed according to the calculated current loss to adjust the network parameters of the convolutional neural network model, and the training is stopped when a preset termination condition is reached to obtain the trained convolutional neural network model.
7. The radio frequency fingerprint authentication method based on wavelet detail coefficients according to claim 6 is characterized in that: The loss is the cross entropy loss.
8. The radio frequency fingerprint authentication method based on wavelet detail coefficients according to claim 1, characterized in that: The convolutional neural network model includes, in sequence: a first convolutional layer, a first maximum pooling layer, a second convolutional layer, a second maximum pooling layer and two fully connected layers.
9. A radio frequency fingerprint authentication device based on wavelet detail coefficients, characterized in that: include: An acquisition module, used for acquiring an original radio frequency signal to be identified; A decomposition module, used for performing wavelet decomposition on the original radio frequency signal to be identified to obtain approximate partial signals and detail partial signals of different scales; wherein the detail partial signal contains the fingerprint feature of the radio frequency device used to generate the original radio frequency signal to be identified; and the approximate partial signal contains the transmission data of the radio frequency device; A feature extraction and conversion module, used to retain the detail signal and generate a two-dimensional image corresponding to the original radio frequency signal to be identified according to the detail signal; An authentication module is used to use a trained convolutional neural network model to identify the two-dimensional image, obtain an identification result used to characterize whether the signal contained in the two-dimensional image is legal, and complete the authentication of the radio frequency signal; wherein the trained convolutional neural network model is trained using a data set containing multiple training samples, and each training sample is a two-dimensional image corresponding to an original radio frequency signal.
10. A radio frequency fingerprint authentication device based on wavelet detail coefficients, comprising a processor, a communication interface, a memory and a communication bus, characterized in that: The processor, the communication interface and the memory communicate with each other via the communication bus; The memory is used to store computer programs; The processor is used to implement the method steps described in any one of claims 1-8 when executing the program stored in the memory.