Radio frequency fingerprint identification method and device, electronic equipment and storage medium

Abstract

The invention provides a radio frequency fingerprint identification method and device, electronic equipment and a storage medium, and the method comprises the steps: collecting a radio frequency signal of to-be-identified equipment, and extracting a radio frequency signal feature of the radio frequency signal; classifying the radio frequency signal features through a KAN network, replacing a full connection layer with a KANLinear layer in the KAN network, performing nonlinear mapping on the radio frequency signal features according to a B-spline primary function, and determining attention distribution of the radio frequency signal features; and determining a radio frequency fingerprint matching result output by the KAN network, and generating a radio frequency fingerprint identifier corresponding to the to-be-identified device according to the radio frequency fingerprint matching result. The feature mapping relation can be adaptively adjusted by introducing a learnable activation function and a B-spline function, nonlinear mapping of high-dimensional data is achieved in a low-dimensional space, calculation complexity is reduced, and meanwhile classification accuracy and robustness are improved.

Description

Technical Field

[0001] The present disclosure relates to the field of wireless communication technologies, and in particular, to a radio frequency fingerprint recognition method, apparatus, electronic device, and storage medium. Background Art

[0002] Radio Frequency (RF) signals, as key information carriers of wireless communication devices, play an important role in fields such as device identity authentication, Internet of Things device management, and wireless network security. With the rapid development of wireless communication technologies, more and more devices are accessing wireless networks, and the need for identity verification and secure communication between devices is becoming increasingly urgent. Therefore, how to accurately identify and verify device identities through radio frequency fingerprints has become a research hotspot in the current wireless communication field.

[0003] Radio frequency fingerprint recognition is to achieve unique identity recognition of devices by analyzing the radio frequency signal characteristics of wireless devices. The formation of radio frequency fingerprints is usually related to the hardware characteristics of devices, such as nonlinear distortion, frequency offset, power amplifier characteristics, etc. in transmitters. These hardware characteristics are somewhat unique in different devices and can thus be extracted as radio frequency fingerprints.

[0004] Existing radio frequency fingerprint recognition algorithms usually adopt traditional machine learning algorithms such as Support Vector Machine (SVM) and K-Nearest Neighbor (KNN). Due to the high-dimensional, time-varying, and non-linear characteristics of radio frequency signals, and the fact that radio frequency signals are often affected by factors such as environmental noise, device interference, and signal attenuation, resulting in reduced stability of signal characteristics. When dealing with high-dimensional non-linear data, these algorithms are prone to problems such as dimensionality disaster and overfitting, and cannot fully utilize the potential characteristics in radio frequency signals. In addition, the real-time performance and scalability of traditional algorithms in complex scenarios are also insufficient. Summary of the Invention

[0005] Embodiments of the present disclosure at least provide a radio frequency fingerprint recognition method, apparatus, electronic device, and storage medium, which can adaptively adjust the feature mapping relationship by introducing a learnable activation function and B-spline basis functions, realize non-linear mapping of high-dimensional data in a low-dimensional space, reduce the computational complexity, and improve the classification accuracy and robustness at the same time.

[0006] Embodiments of the present disclosure provide a radio frequency fingerprint recognition method, including: Collect the radio frequency signal of the device to be recognized, and extract the radio frequency signal characteristics of the radio frequency signal; Classify the radio frequency signal characteristics through a KAN network, replace the fully connected layer with a KANLinear layer in the KAN network, perform non-linear mapping on the radio frequency signal characteristics according to the B-spline basis function, and determine the attention distribution of the radio frequency signal characteristics; Determine the radio frequency fingerprint matching result output by the KAN network, and generate a radio frequency fingerprint identifier corresponding to the device to be identified according to the radio frequency fingerprint matching result.

[0007] In an optional implementation manner, after collecting the radio frequency signal of the device to be identified, the method further includes: Obtain the channel matrix corresponding to the radio frequency signal data; Determine the pseudo-inverse matrix corresponding to the channel matrix; Use the pseudo-inverse matrix as a zero-forcing equalizer, multiply the zero-forcing equalizer by the radio frequency signal data, and determine the equalized radio frequency signal data.

[0008] In an optional implementation manner, extracting the radio frequency signal features of the radio frequency signal specifically includes: Input the equalized radio frequency signal data into a deep convolutional neural network composed of multiple BasicBlock residual blocks for convolutional processing. Each BasicBlock residual block contains two convolutional layers, and batch normalization is set after each convolutional layer; In the process of extracting the radio frequency signal features, the input of the convolutional layer is passed to the output end of the convolutional layer through a residual connection, and the output of the convolutional layer is added to the input; After the convolutional processing of each BasicBlock residual block, the radio frequency signal features are obtained, and the spatial dimension of the radio frequency signal features is compressed through a downsampling layer arranged between different BasicBlock residual blocks, and the number of channels of the radio frequency signal features is adjusted; Compress the feature map corresponding to the radio frequency signal features to a preset fixed size through an adaptive average pooling layer, and provide it as an input to the KAN network.

[0009] In an optional implementation manner, the KAN network includes multiple nested KAN layers; Each KAN layer includes an activation function formed by a linear combination of the B-spline basis function and the spline function.

[0010] In an optional implementation manner, the activation function is defined as:

[0011] Wherein, represents the activation function; represents the B-spline basis function; represents the spline function, and the spline function is a linear combination of the B-spline basis function; represents the function weight corresponding to the basis function; representing the function weights corresponding to the spline function.

[0012] In an alternative embodiment, a KANLinear layer is used to replace the fully connected layer in the KAN network, and a non-linear mapping is performed on the radio frequency signal features according to the B-spline basis function, specifically including: replacing the linear weight matrix with the B-spline basis function, and linearly combining the B-spline basis function with the basic activation function; determining a weight matrix that enables the B-spline basis function to approximate a specified output under specified input data, and performing a non-linear mapping on the radio frequency signal features according to the weight matrix.

[0013] In an alternative embodiment, the method further includes: defining an initial grid corresponding to the KANLinear layer according to a preset grid step, grid upper and lower boundaries, and grid size; dynamically adjusting each point in the initial grid according to the radio frequency signal features by using a uniform distribution and an adaptive generation method to determine an updated target grid; determining the position of the B-spline basis function according to the target grid.

[0014] The embodiments of the present disclosure further provide a radio frequency fingerprint recognition device, including: a feature extraction module, configured to collect radio frequency signals of a device to be recognized, and extract radio frequency signal features of the radio frequency signals; a classification module, configured to classify the radio frequency signal features through a KAN network, replace the fully connected layer with a KANLinear layer in the KAN network, perform a non-linear mapping on the radio frequency signal features according to the B-spline basis function, and determine the attention distribution of the radio frequency signal features; a fingerprint recognition module, configured to determine a radio frequency fingerprint matching result output by the KAN network, and generate a radio frequency fingerprint identifier corresponding to the device to be recognized according to the radio frequency fingerprint matching result.

[0015] The embodiments of the present disclosure further provide an electronic device, including: a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device runs, the processor communicates with the memory through the bus. When the machine-readable instructions are executed by the processor, the above-mentioned radio frequency fingerprint recognition method, or the steps in any possible implementation manner of the above-mentioned radio frequency fingerprint recognition method are executed.

[0016] An embodiment of the present disclosure also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the above-mentioned radio frequency fingerprint recognition method or the steps in any possible implementation manner of the above-mentioned radio frequency fingerprint recognition method.

[0017] An embodiment of the present disclosure also provides a computer program product, including a computer program / instructions. When the computer program / instructions are executed by a processor, they implement the above-mentioned radio frequency fingerprint recognition method or the steps in any possible implementation manner of the above-mentioned radio frequency fingerprint recognition method.

[0018] A radio frequency fingerprint recognition method, device, electronic device and storage medium provided by an embodiment of the present disclosure collect radio frequency signals of a device to be recognized and extract radio frequency signal features of the radio frequency signals; classify the radio frequency signal features through a KAN network, replace a fully connected layer with a KANLinear layer in the KAN network, perform a non-linear mapping on the radio frequency signal features according to a B-spline basis function, and determine an attention distribution of the radio frequency signal features; determine a radio frequency fingerprint matching result output by the KAN network, and generate a radio frequency fingerprint identifier corresponding to the device to be recognized according to the radio frequency fingerprint matching result. It is possible to adaptively adjust a feature mapping relationship by introducing a learnable activation function and a B-spline function, implement non-linear mapping of high-dimensional data in a low-dimensional space, reduce computational complexity, and improve classification accuracy and robustness at the same time.

[0019] To make the above objects, features, and advantages of the present disclosure more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, provides detailed descriptions as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] To more clearly illustrate the technical solutions of the embodiments of the present disclosure, the following briefly introduces the drawings required for the embodiments. The accompanying drawings are incorporated into the specification and form a part of this specification. These drawings show embodiments that conform to the present disclosure and, together with the specification, are used to illustrate the technical solutions of the present disclosure. It should be understood that the following drawings only show some embodiments of the present disclosure and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0021] Figure 1 Shows a flowchart of a radio frequency fingerprint recognition method provided by an embodiment of the present disclosure; Figure 2 Shows a flowchart of another radio frequency fingerprint recognition method provided by an embodiment of the present disclosure; Figure 3Shows a schematic diagram of a radio frequency fingerprint recognition device provided by an embodiment of the present disclosure; Figure 4 Shows a schematic diagram of an electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present disclosure. Apparently, the described embodiments are only some of the embodiments of the present disclosure, rather than all of the embodiments. Components of the embodiments of the present disclosure described and illustrated in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the detailed description of the embodiments of the present disclosure provided in the accompanying drawings is not intended to limit the scope of the present disclosure claimed, but merely represents selected embodiments of the present disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of the present disclosure without creative efforts fall within the scope of protection of the present disclosure.

[0023] It should be noted that: like reference numerals and letters denote like items in the following drawings, and thus, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0024] The term "and / or" in this article merely describes an association relationship and indicates that three relationships may exist. For example, A and / or B may represent: A exists alone, both A and B exist simultaneously, and B exists alone. In addition, the term "at least one" in this article means any one of a plurality or any combination of at least two of a plurality. For example, including at least one of A, B, and C may represent including any one or more elements selected from the set composed of A, B, and C.

[0025] Through research, it is found that existing radio frequency fingerprint recognition algorithms usually adopt traditional machine learning algorithms such as support vector machine (SVM) and K-nearest neighbor (KNN). Since radio frequency signals have high-dimensional, time-varying, and non-linear characteristics, and radio frequency signals are often affected by factors such as environmental noise, device interference, and signal attenuation, the stability of signal features is reduced. When these algorithms process high-dimensional non-linear data, problems such as dimensionality disaster and overfitting are likely to occur, and the potential features in radio frequency signals cannot be fully utilized. In addition, the real-time performance and scalability of traditional algorithms are also insufficient in complex scenarios.

[0026] Based on the above research, the present disclosure provides a radio frequency fingerprint recognition method, apparatus, electronic device, and storage medium. By collecting the radio frequency signals of the device to be recognized and extracting the radio frequency signal features of the radio frequency signals; classifying the radio frequency signal features through a KAN network, replacing the fully connected layer with a KANLinear layer in the KAN network, performing a non-linear mapping on the radio frequency signal features according to the B-spline basis function, and determining the attention distribution of the radio frequency signal features; determining the radio frequency fingerprint matching result output by the KAN network, and generating a radio frequency fingerprint identifier corresponding to the device to be recognized according to the radio frequency fingerprint matching result.

[0027] To facilitate the understanding of this embodiment, first, a radio frequency fingerprint recognition method disclosed in the embodiments of the present disclosure will be introduced in detail. The execution subject of the radio frequency fingerprint recognition method provided in the embodiments of the present disclosure is generally a computer device with certain computing capabilities. Such a computer device includes, for example: a terminal device, a server, or other processing devices. The terminal device may be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementation manners, the radio frequency fingerprint recognition method may be implemented by a processor invoking computer-readable instructions stored in a memory.

[0028] See Figure 1 As shown, it is a flowchart of a radio frequency fingerprint recognition method provided in an embodiment of the present disclosure. The method includes steps S101 to S103, where: S101: Collect the radio frequency signals of the device to be recognized, and extract the radio frequency signal features of the radio frequency signals.

[0029] In a specific implementation, a high-precision radio frequency receiving device (such as a spectrum analyzer, a software-defined radio, etc.) is used to capture the radio frequency signals of the device to be recognized. The frequency band range of the signals is determined by the communication frequency of the device to be recognized (such as Wi-Fi, Bluetooth, ZigBee, etc.). The received analog radio frequency signals are converted into discrete digital signals through analog-to-digital conversion (ADC) at a certain sampling rate, and signal preprocessing is performed.

[0030] Here, the preprocessing process includes frequency offset estimation and low-pass filtering. Frequency offset estimation is used to estimate the frequency offset in the signal, eliminate the frequency offset caused by different signal sources or interference, perform spectrum analysis on the input original I / Q signal, calculate the frequency offset of the signal, and then use Fourier transform (FFT) to calculate the frequency distribution. Based on spectrum peak detection, the offset value is determined and corrected. Low-pass filtering is used to eliminate high-frequency noise and retain effective low-frequency information. A low-pass filter is used to select a suitable cutoff frequency according to the signal bandwidth, filter out high-frequency components and output a denoised smooth signal.

[0031] As a possible implementation, after preprocessing, zero-forcing equalization can be performed through the following steps 1 to 3: Step 1: Obtain a channel matrix corresponding to the radio frequency signal data.

[0032] Step 2: Determine the pseudo-inverse matrix corresponding to the channel matrix.

[0033] Step 3: Using the pseudo inverse matrix as a zero-forcing equalizer, multiplying the zero-forcing equalizer by the radio frequency signal data to determine equalized radio frequency signal data.

[0034] In specific implementations, channel equalization is to process the signal that arrives at the receiving end after being transmitted through the channel in a manner opposite to the channel characteristics, similar to reverse filtering, and is used to resist channel distortion caused by an imperfect channel, especially inter-symbol interference (ISI), so as to accurately restore the received signal. Unlike the minimum mean square error (MMSE) equalization that relies on channel and noise knowledge, zero-forcing equalization (ZF) can completely eliminate the inter-symbol interference (ISI) caused by multipath effects without considering noise, and the algorithm is relatively simple and the computational complexity is low.

[0035] Here, the RF signal transmission signal Affected by many factors such as the channel, the received signal The core idea of ​​zero-forcing equilibrium is to find a matrix , so that the equalized signal As close to the original signal as possible ,Right now:

[0036] in, Represents the equalized RF signal data; a received signal representing a radio frequency signal; Represents the equalized signal As close to the original signal as possible The matrix of .

[0037] In order to eliminate the channel effect, it is necessary to (where is the identity matrix, represents the channel matrix), the zero-forcing equalizer selects as the pseudo-inverse matrix of the channel matrix , that is .

[0038] Further, by multiplying the received signal of the radio frequency signal with the zero-forcing equalizer , the equalized radio frequency signal data after the radio frequency signal passes through the zero-forcing equalization can be obtained. In an ideal situation (i.e., when there is no noise), since , the original signal can be completely restored, thereby eliminating inter-symbol interference (ISI). Therefore, zero-forcing equalization can effectively eliminate the interference of the multipath effect and provide clearer data for the deep learning model.

[0039] In a specific implementation, refer to Figure 2 shown in the flowchart of another radio frequency fingerprint recognition method provided by the embodiments of the present disclosure. The method includes steps S1011 to S1014, where: S1011. Input the equalized radio frequency signal data into a deep convolutional neural network composed of multiple BasicBlock residual blocks for convolutional processing. Each of the BasicBlock residual blocks contains two convolutional layers, and batch normalization is set after each convolutional layer.

[0040] S1012. During the extraction of the radio frequency signal features, the input of the convolutional layer is passed to the output end of the convolutional layer through a residual connection, and the output of the convolutional layer is added to the input.

[0041] S1013. After the convolutional processing of each BasicBlock residual block, the radio frequency signal features are obtained, and the spatial dimension of the radio frequency signal features is compressed through a downsampling layer arranged between different BasicBlock residual blocks, and the number of channels of the radio frequency signal features is adjusted.

[0042] S1014. Compress the feature map corresponding to the radio frequency signal features to a preset fixed size through an adaptive average pooling layer and provide it as an input to the KAN network.

[0043] In a specific implementation, the core of the feature extraction stage consists of multiple BasicBlock residual blocks, aiming to extract high-level features of the RF signal through a deep network and effectively reduce the training difficulty of the deep network. These residual blocks overcome the vanishing gradient problem of the deep network by introducing a residual connection design (Skip Connection), while retaining the original information of the input signal.

[0044] Here, each residual block contains two convolutional layers, followed by batch normalization (BatchNormalization) and ReLU activation function. Each convolutional layer uses a 3×3 convolutional kernel, with a stride of 1 and a padding of 1. This configuration can extract local features without changing the size of the feature map, thus avoiding the loss of spatial information of the feature map.

[0045] Among them, introducing batch normalization after each convolutional layer can stabilize the distribution of the convolutional output and effectively reduce the vanishing gradient or exploding gradient problem during training. The ReLU activation function is adopted after each convolutional layer to introduce non-linearity into the model and enhance the expressive ability of the network.

[0046] Furthermore, the residual block realizes the residual connection by directly passing the input to the output end of the block and adding it to the output after being processed by two convolutional layers, which can make the network easier to learn the changes between the input and output and retain the original features of the input signal in the deep structure, enabling the network to train deeper structures more effectively.

[0047] Among them, in order to adjust the size of the feature map between different residual blocks to match the input dimension of the next convolutional layer, a downsampling layer is introduced. The downsampling layer is usually implemented by a 1×1 convolutional operation with a stride of 2 to reduce the spatial dimension of the feature map and adjust the number of channels. This operation not only compresses the size of the feature map, but also improves the computational efficiency while retaining important feature information.

[0048] Preferably, in the final stage of feature extraction, an adaptive average pooling layer is used to compress the feature map to a fixed size (usually 1×1), which not only realizes the dimensionality reduction operation of the features and prepares for the input of the KAN network, but also retains the global feature information.

[0049] S102. Classify the RF signal features through the KAN network. In the KAN network, replace the fully connected layer with the KANLinear layer, perform a non-linear mapping on the RF signal features according to the B-spline basis function, and determine the attention distribution of the RF signal features.

[0050] In a specific implementation, the KAN network (Kernel Attention Network) is a neural network embedded with an attention mechanism, aiming to capture the high-dimensional feature relationships of input features through non-linear mapping. It consists of two KANLinear layers. Different from traditional neural networks, the KAN network introduces a KANLinear layer based on B-spline basis functions in the fully connected layer to enhance the network's ability to express complex features. Its purpose is to perform complex feature combinations through learnable non-linear functions, rather than relying on simple linear functions, thereby improving the accuracy and generalization ability of classification tasks.

[0051] Here, an N-layer KAN network can be described as a nesting of multiple KAN layers, and its mathematical expression is:

[0052] Among them, represents the KAN network; Z represents the features of the input radio frequency signal; represents the i th layer of the KAN network. Each KAN layer in the KAN network is composed of a learnable activation function , and the activation function has dimensional input and dimensional output: , all values of the (n + 1)th layer can be regarded as a matrix about all activation values of the n th layer, as shown in the following formula:

[0053] Here, for a network composed of layer grids and neurons per layer, ([[]] n, i ) represents the n th neuron of the i th layer, and the activation value is expressed as . Then there are n * n +1 activation functions between the th layer and the th layer. The activation function connecting the )th neuron can be defined as: where the value before activation of the activation function

[0054] is , and the value after activation is , which is always equal to , the th The activation value of a neuron can be simplified to the sum of all activated values, as shown in the following formula:

[0055] That is, all values in the (n + 1)-th layer can be regarded as a matrix of all activation values in the n-th layer.

[0056] Preferably, the activation function is composed of a basis function and a spline function spline ( x ), and the activation function is defined as:

[0057] where represents the activation function; represents the B-spline basis function; represents the spline function, and the spline function is a linear combination of the B-spline basis functions; represents the function weight corresponding to the basis function; represents the function weight corresponding to the spline function.

[0058] Optionally, the basis function is usually set to the SiLU activation function, and the spline function can be parameterized as a linear combination of B-spline basis functions: , where is the corresponding coefficient and is trainable, is the i -th B-spline basis function.

[0059] Here, and can be redundant and can be absorbed into and . However, these factors are still included (trainable by default) to better control the overall amplitude of the activation function. When initializing the activation function, is set to 1, is set to 0.

[0060] Furthermore, replace the linear weight matrix with a B-spline basis function, and linearly combine the B-spline basis function with the basic activation function; determine the weight matrix that enables the B-spline basis function to approximate the specified output under the specified input data, and perform a non-linear mapping on the radio frequency signal features according to the weight matrix.

[0061] In a specific implementation, the traditional fully connected layer plays an important role in the neural network, but its linear structure and fixed activation function have limitations in processing complex data. To enhance the computational efficiency and expressive power of the network, the KANLinear layer is adopted in this application to replace the fully connected layer. The KANLinear layer combines the B-spline basis function and the traditional activation function to achieve a non-linear mapping of complex functions.

[0062] Here, the core idea of the KANLinear layer is to replace the traditional linear weight matrix with the B-spline basis function, allowing the neural network to apply a learnable non-linear transformation to each input feature, significantly enhancing the flexibility of the model. In traditional implementations, all intermediate variables must be expanded to apply different activation functions, resulting in a large memory consumption. The KANLinear layer significantly reduces the memory requirements by introducing the B-spline basis function and linearly combining it with the basic activation function.

[0063] Among them, the B-spline basis function is an important tool for function approximation and interpolation, with advantages such as locality, smoothness, and numerical stability. In the KANLinear layer, the B-spline basis function is used to interpolate and approximate the input radio frequency signal feature tensor on a given grid to achieve complex non-linear transformations. The higher-order B-spline basis function is obtained through a recurrence relation, and the k-th order B-spline basis function can be linearly combined by two (k - 1)-th order B-spline basis functions.

[0064] Furthermore, to achieve the non-linear combination of data, the model interpolates by calculating the weights of the spline basis functions. This process involves solving a system of linear equations to find a set of coefficients such that the B-spline basis function can approximate the output at the given input points, that is, given the input data X and the output data Y , it is necessary to find the weight matrix such that , where is the B-spline basis function matrix. Solving this formula by the least squares method can obtain the weight matrix .

[0065] Here, during the non-linear mapping process, the B-spline basis function is defined according to the step size, boundary, and size of the preset grid (grid). The initial grid is dynamically adjusted to ensure that the B-spline basis function better fits the distribution of the input features. During the training process, by optimizing the weight matrix, the B-spline basis function can approximate the target output under specific input data. Utilizing the non-linear characteristics of the B-spline, the radio frequency signal features form a richer high-dimensional representation after passing through the KANLinear layer.

[0066] Among them, different weights are assigned to each part of the radio frequency signal features through attention calculation to highlight key information. This process calculates the attention weights of each feature through the output features of the KANLinear layer, and uses the calculated attention weights to weight the radio frequency signal features to form a global feature representation. The features after attention weighting are sent to the next layer for classification processing.

[0067] Optionally, the output features of the KAN network pass through the Softmax layer to generate the classification probability of the radio frequency fingerprint. The radio frequency fingerprint matching result is generated according to the classification probability, and finally the classification label of the radio frequency signal features is determined.

[0068] As a possible implementation, an initial grid corresponding to the KANLinear layer is defined according to the preset grid step size, grid upper and lower boundaries, and grid size; each point in the initial grid is dynamically adjusted according to the radio frequency signal features by using a uniform distribution and an adaptive generation method to determine the updated target grid; the position of the B-spline basis function is determined according to the target grid.

[0069] Here, the grid determines the position of the B-spline basis function. In the KANLinear layer, the generation and update of the grid directly affect the shape and coverage of the spline basis function. The initial grid can be defined as:

[0070]

[0071] Among them, represents the grid, represents the step size of the grid, and respectively represent the lower and upper boundaries of the grid, represents the grid size, is the i th point in the grid, n represents the spline order.

[0072] After that, the positions of the grid points are dynamically adjusted according to the input data, and the updated grid is obtained by combining the uniform grid and the adaptive grid:

[0073] Among them, represents the i th grid point generated by the uniform distribution, represents the i th grid point generated adaptively, and the parameter Control the weighted ratio of the uniform grid and the adaptive grid. The adjusted grid can capture the local features of the input data more precisely, thereby improving the fitting ability of the model.

[0074] As another possible implementation, the L1 regularization needs to perform non-linear operations on tensors and is not applicable to the redesigned computing structure. In the embodiments of the present application, the L1 regularization of the weights is used to replace the traditional regularization method, which is divided into two parts: activation regularization and entropy regularization.

[0075] Here, the activation regularization calculates the sum of the average absolute values of the function weights corresponding to all spline functions: , the entropy regularization calculates the probability of each weight relative to the sum , and then calculates the negative entropy: , and the final regularization loss is the weighted sum of the activation regularization and the entropy regularization: , where and are hyperparameters of the regularization. To optimize the performance of the network on the Wi-Fi signal dataset, the KANLinear layer uses the Kaiming initialization method to ensure the stability of the weight distribution.

[0076] S103. Determine the radio frequency fingerprint matching result output by the KAN network, and generate a radio frequency fingerprint identifier corresponding to the device to be identified according to the radio frequency fingerprint matching result.

[0077] In a specific implementation, a probability distribution belonging to a specific device category is output through the last layer of the KAN network. The classification result is directly used as the unique identifier of the device, or for a device not in the training set, the KAN network can output a feature embedding vector, which represents the high-dimensional expression of the radio frequency feature. By calculating the cosine similarity or Euclidean distance between the feature embedding vector and the radio frequency fingerprint embedding of the existing device, the matching result of the device is determined.

[0078] Here, the output result of the KAN network is compared with the known radio frequency fingerprint library, and the radio frequency feature embeddings of the known devices are stored in the device fingerprint library. The features output by the KAN network are compared with all the features in the fingerprint library and the similarity is calculated to find the radio frequency fingerprint closest to the features output by the KAN network, and the corresponding device is the device to be identified.

[0079] Among them, if the similarity is higher than a certain threshold (such as 0.95), it is determined that the input radio frequency signal belongs to the corresponding device; if the similarity is lower than the threshold, it may be a new device and needs to be saved in the fingerprint library as the radio frequency fingerprint of the new device.

[0080] A radio frequency fingerprint recognition method provided by an embodiment of the present disclosure collects radio frequency signals of a device to be recognized and extracts radio frequency signal features of the radio frequency signals; classifies the radio frequency signal features through a KAN network, replaces a fully connected layer with a KANLinear layer in the KAN network, performs a non-linear mapping on the radio frequency signal features according to a B-spline basis function, and determines an attention distribution of the radio frequency signal features; determines a radio frequency fingerprint matching result output by the KAN network, and generates a radio frequency fingerprint identifier corresponding to the device to be recognized according to the radio frequency fingerprint matching result. A learnable activation function and a B-spline function can be introduced to adaptively adjust a feature mapping relationship, realize non-linear mapping of high-dimensional data in a low-dimensional space, reduce computational complexity, and improve classification accuracy and robustness at the same time.

[0081] Those skilled in the art can understand that in the above method of the specific embodiment, the writing order of each step does not mean a strict execution order and constitutes any limitation on the implementation process. The specific execution order of each step should be determined according to its function and possible internal logic.

[0082] Based on the same inventive concept, an embodiment of the present disclosure also provides a radio frequency fingerprint recognition device corresponding to the radio frequency fingerprint recognition method. Since the principle of solving problems by the device in the embodiment of the present disclosure is similar to the above radio frequency fingerprint recognition method in the embodiment of the present disclosure, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described in detail.

[0083] Please refer to Figure 3 , Figure 3 which is a schematic diagram of a radio frequency fingerprint recognition device provided by an embodiment of the present disclosure. As Figure 3 shown in, the radio frequency fingerprint recognition device 300 provided by an embodiment of the present disclosure includes: A feature extraction module 310, configured to collect radio frequency signals of a device to be recognized and extract radio frequency signal features of the radio frequency signals.

[0084] A classification module 320, configured to classify the radio frequency signal features through a KAN network, replace a fully connected layer with a KANLinear layer in the KAN network, perform a non-linear mapping on the radio frequency signal features according to a B-spline basis function, and determine an attention distribution of the radio frequency signal features; A fingerprint recognition module 330, configured to determine a radio frequency fingerprint matching result output by the KAN network, and generate a radio frequency fingerprint identifier corresponding to the device to be recognized according to the radio frequency fingerprint matching result.

[0085] Descriptions of the processing flow of each module in the device and the interaction flow between the modules can refer to the relevant descriptions in the above method embodiment, and will not be elaborated here.

[0086] A radio frequency fingerprint recognition device provided by an embodiment of the present disclosure collects radio frequency signals of a device to be recognized and extracts radio frequency signal features of the radio frequency signals; classifies the radio frequency signal features through a KAN network, replaces a fully connected layer with a KANLinear layer in the KAN network, performs a non-linear mapping on the radio frequency signal features according to a B-spline basis function, and determines an attention distribution of the radio frequency signal features; determines a radio frequency fingerprint matching result output by the KAN network, and generates a radio frequency fingerprint identifier corresponding to the device to be recognized according to the radio frequency fingerprint matching result. By introducing a learnable activation function and a B-spline function, the feature mapping relationship can be adaptively adjusted, a non-linear mapping of high-dimensional data can be realized in a low-dimensional space, the calculation complexity can be reduced, and at the same time, the classification accuracy and robustness can be improved.

[0087] Corresponding to Figure 1 the radio frequency fingerprint recognition method in Figure 4 As shown in FIG. 400 is a schematic structural diagram of an electronic device 400 provided by an embodiment of the present disclosure, including: a processor 41, a memory 42, and a bus 43; the memory 42 is used to store execution instructions, including an internal memory 421 and an external memory 422; the internal memory 421 here is also called an internal storage, and is used to temporarily store operation data in the processor 41 and data exchanged with an external storage such as a hard disk 422. The processor 41 exchanges data with the external storage 422 through the internal memory 421. When the electronic device 400 runs, the processor 41 communicates with the memory 42 through the bus 43, so that the processor 41 executes Figure 1 the steps of the radio frequency fingerprint recognition method in

[0088] An embodiment of the present disclosure also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the steps of the radio frequency fingerprint recognition method described in the above method embodiment. Among them, the storage medium may be a volatile or non-volatile computer-readable storage medium.

[0089] An embodiment of the present disclosure also provides a computer program product, which includes computer instructions. When the computer instructions are executed by a processor, they can execute the steps of the radio frequency fingerprint recognition method described in the above method embodiment. For details, please refer to the above method embodiment, which will not be repeated here.

[0090] Among them, the above computer program product can be specifically implemented in the form of hardware, software, or a combination thereof. In an alternative embodiment, the computer program product is specifically embodied as a computer storage medium. In another alternative embodiment, the computer program product is specifically embodied as a software product, such as a Software Development Kit (SDK), etc.

[0091] Those skilled in the art can clearly understand that for the convenience and conciseness of description, the specific working process of the above-described device can refer to the corresponding process in the foregoing method embodiments, which will not be elaborated herein. In several embodiments provided by the present disclosure, it should be understood that the disclosed device and method can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical functional division, and there may be other division methods in actual implementation. For another example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces. The indirect coupling or communication connection of the device or unit can be in an electrical, mechanical, or other form.

[0092] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0093] In addition, in each embodiment of the present disclosure, the functional units can be integrated in one processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit.

[0094] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium executable by a processor. Based on this understanding, the technical solution of the present disclosure, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0095] Finally, it should be noted that the above-described embodiments are only specific implementation manners of the present disclosure, used to illustrate the technical solutions of the present disclosure, rather than limiting them. The protection scope of the present disclosure is not limited thereto. Although the present disclosure has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present disclosure can still modify the technical solutions recorded in the foregoing embodiments or can easily think of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes, or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present disclosure and should all be covered within the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. A radio frequency fingerprint recognition method, characterized in that: include: Collecting radio frequency signals of the device to be identified and extracting radio frequency signal features of the radio frequency signals; The RF signal features are classified by a KAN network, a fully connected layer is replaced by a KANLinear layer in the KAN network, the RF signal features are nonlinearly mapped according to a B-spline basis function, and an attention distribution of the RF signal features is determined; Determine the radio frequency fingerprint matching result output by the KAN network, and generate a radio frequency fingerprint identification corresponding to the device to be identified according to the radio frequency fingerprint matching result.

2. The method according to claim 1, characterized in that After collecting the radio frequency signal of the device to be identified, the method further includes: Obtaining a channel matrix corresponding to the radio frequency signal; Determine a pseudo-inverse matrix corresponding to the channel matrix; The pseudo inverse matrix is ​​used as a zero-forcing equalizer, and the zero-forcing equalizer is multiplied by the radio frequency signal to determine equalized radio frequency signal data.

3. The method according to claim 2, characterized in that Extracting the radio frequency signal feature of the radio frequency signal specifically includes: Inputting the equalized RF signal data into a deep convolutional neural network composed of a plurality of BasicBlock residual blocks for convolution processing, each of the BasicBlock residual blocks comprises two convolutional layers, and batch normalization is set after each of the convolutional layers; In the process of extracting the radio frequency signal feature, the input of the convolution layer is transmitted to the output end of the convolution layer through a residual connection, and the output of the convolution layer is added to the input; The RF signal feature is obtained after convolution processing of each BasicBlock residual block, and the spatial dimension of the RF signal feature is compressed by a downsampling layer arranged between different BasicBlock residual blocks, and the number of channels of the RF signal feature is adjusted; The feature map corresponding to the RF signal feature is compressed to a preset fixed size through an adaptive average pooling layer and provided as input to the KAN network.

4. The method according to claim 1, characterized in that: The KAN network includes multiple nested KAN layers; Each of the KAN layers includes an activation function formed by a linear combination of the B-spline basis function and a spline function.

5. The method according to claim 4, characterized in that The activation function is defined as: in, represents the activation function; represents the B-spline basis function; represents the spline function, which is a linear combination of the B-spline basis functions; Represents the function weight corresponding to the basis function; Represents the function weight corresponding to the spline function.

6. The method according to claim 5, characterized in that In the KAN network, the fully connected layer is replaced by the KANLinear layer, and the RF signal features are nonlinearly mapped according to the B-spline basis function, specifically including: Replacing a linear weight matrix with the B-spline basis function, and linearly combining the B-spline basis function with a basic activation function; A weight matrix is ​​determined that enables the B-spline basis function to approximate a specified output under specified input data, and nonlinear mapping is performed on the radio frequency signal characteristics according to the weight matrix.

7. The method according to claim 1, characterized in that The method further comprises: According to the preset grid step size, grid upper and lower boundaries, and grid size, define the initial grid corresponding to the KANLinear layer; Dynamically adjusting each point in the initial grid according to the radio frequency signal characteristics by using a uniform distribution and adaptive generation method to determine an updated target grid; The position of the B-spline basis function is determined according to the target grid.

8. A radio frequency fingerprint recognition device, characterized in that: include: A feature extraction module, used to collect the radio frequency signal of the device to be identified and extract the radio frequency signal features of the radio frequency signal; A classification module, used to classify the radio frequency signal features through a KAN network, replace the fully connected layer with a KANLinear layer in the KAN network, perform nonlinear mapping on the radio frequency signal features according to a B-spline basis function, and determine the attention distribution of the radio frequency signal features; The fingerprint recognition module is used to determine the radio frequency fingerprint matching result output by the KAN network, and generate a radio frequency fingerprint identification corresponding to the device to be identified according to the radio frequency fingerprint matching result.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the radio frequency fingerprint recognition method according to any one of claims 1 to 7 are performed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the radio frequency fingerprint identification method according to any one of claims 1 to 7 are executed.

Images