Radio frequency fingerprint recognition method and device based on slice combination convolutional neural network

Abstract

The present invention relates to a field of low-carbon scheduling technology, and is a radio frequency fingerprint identification method and device based on a slice combined convolutional neural network. The method includes using variance trajectory detection to intercept the steady-state segment of a radio frequency baseband I / Q signal, and dividing it into a number of slice signals; inputting the several slice signals into a slice combined convolutional neural network model, extracting the fingerprint information of the radio frequency baseband I / Q signal based on the fusion of slice signal features, and completing radio frequency fingerprint identification. In a communication framework based on deep learning, the present invention uses slice information learning to obtain a combined convolutional neural network model, effectively guiding the deep neural network structure to learn real radio frequency fingerprint features, and after inputting several slice signals of the steady-state segment of the radio frequency baseband I / Q signal into the combined convolutional neural network model, it can extract the real radio frequency fingerprint features, and based on feature combination and fusion, extract the fingerprint information of the radio frequency baseband I / Q signal with high recognition, effectively improving the recognition of the radio frequency fingerprint.

Description

Technical Field

[0001] The present invention relates to the technical field of radio frequency fingerprint recognition, in particular to a radio frequency fingerprint recognition method and device based on a slice combined convolutional neural network. Background Art

[0002] IoT objects should have their own identities to build a trusted IoT ecological network system. In order to prevent problems such as device cloning, replay attacks, and user identity simulation, how to accurately identify and authenticate IoT objects is a key issue facing the IoT and the basis for IoT applications.

[0003] RFF identification is a non-encrypted authentication mechanism based on the hardware attributes of IoT objects, without the need to embed additional hardware or additional computing resources. Its principle is to individually identify the communication radiation source based on the subtle characteristics of the RF signal derived from SEI, where SEI refers to the ability to link the unique electromagnetic characteristics of the transmitter to the transmitter itself. In view of the utilization of physical layer resources, the existing RFF identification methods can be divided into channel characteristic-based methods and transmission signal-based methods. The channel characteristic-based methods aim to use the unique location information of the IoT device as an identity detection indicator and are always used for indoor positioning of IoT devices. The transmission signal-based methods can be divided into transient signal-based methods and steady-state signal-based methods. The RF transient signal does not involve data information, but only the hardware attributes of the mirror transmitter. Therefore, RFF was originally extracted from the RF transient signal, but the RF transient signal is difficult to capture due to its short duration. According to the different fingerprint feature extraction methods, the RFF recognition method based on the transmission signal can also be divided into the waveform domain-based method and the modulation domain-based method. Like communication signals, orthogonal modulation has been widely used today. The modulation domain method adopts the signal structure given by the modulation scheme. The RF baseband I / Q signal is used as the basic processing unit to make the specific physical characteristics of the transmitter easier to identify. The modulation domain fingerprint features include constellation traces, differential constellation traces, contour stars, differential contour stars and other features. In the identification and authentication stage, the RFF recognition method based on the transmission signal can also be divided into traditional machine learning-based methods and deep learning-based methods according to different classifiers.

[0004] From the current research status of RFF identification, extracting RF fingerprints with unique natural properties is still a challenging task. The feature extraction method plays a vital role in the recognition accuracy of fingerprint recognition technology based on traditional machine learning. Common RF fingerprint features include entropy features, Holder coefficient features, multifractal dimension features, fractal box dimension features, RF-DNA, integrated bispectral features, power spectral density features, etc. Among them, integrated bispectral and power spectral density are two relatively good methods for extracting RF fingerprint features of steady-state signals. In order to pursue higher data rates and spectrum efficiency, communication systems usually use linear modulation modes such as QPSK and 16QAM. In multi-carrier systems, the peak-to-average ratio is relatively large and the signal envelope changes, so the system should maintain linear amplification. If nonlinear amplification is used, it will cause in-band signal distortion, reduce system performance, cause out-of-band intermodulation products, and interfere with adjacent channels of the transmitter carrier frequency.

[0005] Existing technologies include:

[0006] Prior art 1, patent publication number CN109919015A discloses a method for extracting and identifying radio frequency fingerprints based on a multi-sampling convolutional neural network, which uses slicing technology to expand the training set, validation set and test set, selects and trains the parameters of the multi-sampling convolutional neural network according to the training set and validation set, and uses the trained multi-sampling convolutional neural network to classify and identify radio frequency fingerprints for the actual received wireless device output signal. The slicing technology of the present invention is only used to expand the training set, validation set and test set, so that the multi-sampling convolutional neural network can be trained more fully, and the slicing technology and feature fusion are not used to extract subtle and real radio frequency fingerprint features and perform feature enhancement.

[0007] Prior art 2, patent publication number CN112464713A discloses a method for identifying radio frequency fingerprints of communication radiation sources based on deep learning, extracting three types of fingerprint information of I / Q two-way digital zero intermediate frequency signals, dividing the three types of fingerprint information data sets into training sets and test sets, training a multi-channel deep convolutional neural network model, and using the trained multi-channel deep convolutional neural network model to classify and identify radio frequency fingerprints. This patent uses a multi-channel deep convolutional neural network model for secondary feature extraction to improve the recognition accuracy of radio frequency fingerprints of communication radiation sources, but it cannot extract subtle features in I / Q signals, and enhances single features to guide the deep neural network structure to learn real radio frequency fingerprint features. Summary of the invention

[0008] The present invention provides a radio frequency fingerprint recognition method and device based on a slice combined convolutional neural network, which are used for radio frequency fingerprint recognition of the Internet of Things.

[0009] The first aspect of the present application provides a radio frequency fingerprint recognition method based on a slice combination convolutional neural network, comprising:

[0010] The steady-state signal of the RF baseband I / Q signal is intercepted by using variance trajectory detection, and is divided into several slice signals;

[0011] Several slice signals are input into the slice combination convolutional neural network model, and the fingerprint information of the RF baseband I / Q signal is extracted based on the slice signal feature fusion to complete the RF fingerprint recognition. The slice combination convolutional neural network model is obtained through deep learning using multiple sets of training data, and each set of training data includes: the slice signal of the steady-state signal and the identification information used to identify the RF fingerprint.

[0012] Preferably, the steady-state signal of the RF baseband I / Q signal is intercepted by using variance trajectory detection and divided into a number of slices, including:

[0013] Use the receiver to collect RF baseband I / Q signals and perform ADC sampling;

[0014] Perform variance trajectory detection on the sampled RF baseband I / Q signal to intercept the steady-state signal of the RF baseband I / Q signal;

[0015] The steady-state signal of the radio frequency baseband I / Q signal is sliced ​​to obtain a plurality of slice signals.

[0016] Preferably, the radio frequency baseband I / Q signal includes a steady-state signal intercepted by a radio frequency fingerprint obtained using a steady-state signal.

[0017] Preferably, a plurality of slice signals are input into a slice combination convolutional neural network model, and fingerprint information of the RF baseband I / Q signal is extracted based on the slice signal feature fusion to complete RF fingerprint recognition, including:

[0018] Input several slice signals into the slice combination convolutional neural network model, and use the deep convolutional neural network feature extractor to extract the features in each slice signal;

[0019] The features extracted from each slice signal are fused, and the fingerprint information of the RF baseband I / Q signal is extracted from the fused features;

[0020] The softmax function is used to identify the fingerprint information of the RF baseband I / Q signal.

[0021] In view of this, the second aspect of the present application provides a radio frequency fingerprint recognition device based on a slice combination convolutional neural network, comprising:

[0022] The slicing unit intercepts the steady-state signal of the RF baseband I / Q signal by using variance trajectory detection and divides it into a number of slice signals;

[0023] The recognition unit inputs several slice signals into the slice combination convolutional neural network model, extracts the fingerprint information of the RF baseband I / Q signal based on the slice signal feature fusion, and completes the RF fingerprint recognition, wherein the slice combination convolutional neural network model is obtained through deep learning using multiple sets of training data, and each set of training data includes: the slice signal of the steady-state signal of the RF baseband I / Q signal and the identification information used to identify the RF fingerprint.

[0024] Preferably, the slicing unit comprises:

[0025] The sampling module uses the receiver to collect the RF baseband I / Q signal and performs ADC sampling;

[0026] The interception module performs variance trajectory detection on the sampled RF baseband I / Q signal and intercepts the steady-state signal of the RF baseband I / Q signal;

[0027] The slicing module slices the steady-state signal of the RF baseband I / Q signal to obtain a number of sliced ​​signals.

[0028] Preferably, the identification unit comprises:

[0029] The feature extraction module inputs several slice signals into the slice combination convolutional neural network model and uses the deep convolutional neural network feature extractor to extract the features in each slice signal;

[0030] The fingerprint extraction module fuses the features extracted from each slice signal and extracts the fingerprint information of the RF baseband I / Q signal from the fused features;

[0031] The fingerprint recognition module uses the softmax function to identify the fingerprint information of the RF baseband I / Q signal.

[0032] Beneficial effect: Under the communication framework based on deep learning, the present invention uses slice information learning to obtain a combined convolutional neural network model, effectively guides the deep neural network structure to learn the real RF fingerprint features, and after inputting several slice signals of the steady-state signal of the RF baseband I / Q signal into the combined convolutional neural network model, extracts subtle features on different slices, that is, extracts the real RF fingerprint features, and based on feature combination and fusion, increases the amount of information on the technical differences between RF signals, thereby enhancing a single feature, thereby effectively improving the recognition of RF fingerprints. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Attached Figure 1 A schematic diagram of a recognition method flow chart according to an embodiment of the present invention;

[0034] Attached Figure 2 A schematic diagram of a process for steady-state signal slicing of a radio frequency baseband I / Q signal in one embodiment of the present invention;

[0035] Attached Figure 3 A schematic flow chart of a method for identifying a slice signal by inputting a slice combination convolutional neural network model in one embodiment of the present invention;

[0036] Attached Figure 4 A schematic diagram of a process of obtaining an optimal solution using a genetic algorithm in one embodiment of the present invention;

[0037] Attached Figure 5 This is a recognition accuracy diagram of 21 aircraft slices before combination and feature fusion in one embodiment of the present invention;

[0038] Attached Figure 6 This is a recognition accuracy diagram of 21 aircraft slice combinations and feature fusions in one embodiment of the present invention;

[0039] Attached Figure 7 is the recognition accuracy of 198 aircraft slices before combination and feature fusion in one embodiment of the present invention;

[0040] Attached Figure 8 is the recognition accuracy of 198 aircraft slices before combination and feature fusion in one embodiment of the present invention;

[0041] Attached Fig. 9 It is a training convergence curve diagram of slice combination and feature fusion before and after under the same deep convolutional neural network feature extractor structure in one embodiment of the present invention;

[0042] Attached Fig.10 This is a graph of the recognition accuracy of 21 aircraft based on PSD fingerprint and GRA classifier in one embodiment of the present invention;

[0043] Attached Fig.11 This is a graph of the recognition accuracy of 21 aircraft based on PSD fingerprint and SVM classifier in one embodiment of the present invention;

[0044] Attached Fig.12 This is a graph of the recognition accuracy of 35 aircraft based on PSD fingerprint and SVM classifier in one embodiment of the present invention;

[0045] Attached Fig.13 This is a graph of the recognition accuracy of 100 aircraft based on PSD fingerprint and SVM classifier in one embodiment of the present invention;

[0046] Attached Fig.14 This is a graph of the recognition accuracy of 198 aircraft based on PSD fingerprint and SVM classifier in one embodiment of the present invention;

[0047] Attached Fig.15 A diagram showing the recognition accuracy of different numbers of aircraft in one embodiment of the present invention;

[0048] Attached Fig.16 This is a map of 21 aircraft identification accuracy based on the contour star method in one embodiment of the present invention;

[0049] Attached Fig.17 This is a diagram of the recognition accuracy of 21 aircraft based on a complex convolutional neural network in one embodiment of the present invention;

[0050] Attached Fig.18 This is a diagram of the recognition accuracy of 21 aircraft based on a deep residual network in one embodiment of the present invention;

[0051] Attached Fig.19 This is a diagram of the recognition accuracy of 21 aircraft based on a deep complex residual network in one embodiment of the present invention;

[0052] Attached Fig. 20 The figure is a block diagram of an identification device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0054] It should be understood that in various embodiments of the present invention, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0055] It should be understood that in the present invention, "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0056] The present application provides a radio frequency fingerprint recognition method and device based on a slice combination convolutional neural network, which is used to solve the problem of how to extract subtle features of radio frequency fingerprints and improve the radio frequency fingerprint recognition rate.

[0057] The following is a detailed description of the embodiments of the present application.

[0058] Embodiment 1: As attached Figure 1 As shown, this embodiment discloses a radio frequency fingerprint recognition method based on a slice combination convolutional neural network, including:

[0059] Step S101, using variance trajectory detection to intercept the steady-state signal of the RF baseband I / Q signal, the steady-state signal is the signal contained in the stable signal segment after the transient segment is removed, and it is cut into several slice signals according to the length of equal data points;

[0060] Step S102, inputting several slice signals into a slice combination convolutional neural network model, extracting fingerprint information of the RF baseband I / Q signal based on slice signal feature fusion, and completing RF fingerprint recognition, wherein the slice combination convolutional neural network model is obtained through deep learning using multiple sets of training data, and each set of training data includes: a slice signal of a steady-state signal and identification information used to identify the RF fingerprint.

[0061] The present invention uses slice information learning to obtain a combined convolutional neural network model under a communication framework based on deep learning, effectively guides the deep neural network structure to learn the real RF fingerprint features, and after inputting several slice signals of the steady-state signal of the RF baseband I / Q signal into the combined convolutional neural network model, extracts the subtle features on different slices, that is, extracts the real RF fingerprint features, and based on feature combination and fusion, increases the amount of information about the technical differences between RF signals, thereby enhancing a single feature, thereby effectively improving the recognition of RF fingerprints. Furthermore, the slicing technology and the slice combination convolutional neural network model in the present invention, starting from the two aspects of signal slice combination, feature combination and fusion, can be fully used for large-scale individual identification of communication radiation sources.

[0062] Embodiment 2: This embodiment discloses a radio frequency fingerprint recognition method based on a slice combination convolutional neural network, comprising:

[0063] Step S201, using variance trajectory detection to capture the steady-state signal of the RF baseband I / Q signal, and divide it into a number of slice signals, as shown in the attached Figure 2 As shown, specifically including:

[0064] Step S2011, using a receiver to collect RF baseband I / Q signals and perform ADC sampling;

[0065] Step S2012, performing variance trajectory detection on the sampled RF baseband I / Q signal to intercept the steady-state signal of the RF baseband I / Q signal;

[0066] In this step, variance trajectory detection is performed on the sampled RF baseband I / Q signal to intercept the steady-state signal of the RF baseband I / Q signal, including:

[0067] When a signal segment containing a target signal is detected, how to accurately detect the starting point and the end point of the target signal plays an important role in subsequent fingerprint feature extraction.

[0068] First, for the input data sequence, the transient starting point detection can be performed by calculating its variance trajectory sequence and setting the threshold. The variance trajectory sequence of the input data can be calculated by the following formula:

[0069] ,

[0070] ,

[0071] where ;

[0072] in, Indicates rounding down; Indicates the length of the sliding window used for variance calculation; Indicates the displacement of the rear window function after each calculation; Represents the average value of the window function sequence; when the receiver receives a transient signal, The amplitude will increase significantly. After the variance trajectory calculation is completed, the starting point detection problem of the transient signal can be simplified to the common change point detection problem. The change point position where the change trajectory sequence begins to rise is the starting point of the transient signal. Similarly, when the variance trajectory begins to remain unchanged, it is the position where the transient signal ends, and the signal power is basically stable at this time. The variance trajectory detection method has low computational complexity and is suitable for real-time detection of signal change points and interception of identifiable signals.

[0073] Step S2013, slicing the steady-state signal of the RF baseband I / Q signal to obtain a plurality of sliced ​​signals.

[0074] Step S202, inputting a plurality of slice signals into a slice combination convolutional neural network model, extracting fingerprint information of the RF baseband I / Q signal based on the fusion of slice signal features, and completing RF fingerprint recognition, wherein the slice combination convolutional neural network model is obtained by deep learning using multiple sets of training data, and each set of training data includes: a slice signal of a steady-state signal of the RF baseband I / Q signal and identification information used to identify the RF fingerprint; as shown in the attached Figure 3 As shown, specifically including:

[0075] Step S2021, inputting a plurality of slice signals into a slice combination convolutional neural network model, and using a deep convolutional neural network feature extractor to extract features from each slice signal;

[0076] Step S2022, fusing the features extracted from each slice signal, and extracting fingerprint information of the RF baseband I / Q signal from the fused features;

[0077] Step S2023: Use the softmax function to identify the fingerprint information of the RF baseband I / Q signal.

[0078] The above slice combination convolutional neural network model is shown in the attached Figure 4 As shown; the structure of the deep convolutional neural network feature extractor is shown in the following table:

[0079] Table 1 Structure of deep convolutional neural network feature extractor

[0080] Network Layer Parameter structure Input Layer 1×2×N 2D Convolutional Layer 1×2×48 Normalization layer 1×2×48 2D Max Pooling Layer 1×1×48 Dropout Layer 1×1×48 2D Convolutional Layer 3×3×64 Normalization layer 3×3×64 2D Average Pooling Layer 3×1×64 Fully connected layer 198

[0081] Among them, the activation function of the two-dimensional convolutional layer is ReLU.

[0082] Using normalization layers can effectively reduce the dependence of deep networks on weight initialization.

[0083] The activation function of the output layer is Softmax.

[0084] At the same time, the loss function of the deep network in the combined convolutional neural network model is sparse category cross entropy. Since a lot of sparsity is introduced into the deep network, such as ReLU, Dropout, Pooling, etc., this may lead to a decrease in the computational performance of the deep network. In order to maintain the sparsity of the network structure and utilize the high computational performance of dense matrices, the design of the sliced ​​combined convolutional neural network can cluster sparse matrices into denser sub-matrices to improve computational performance.

[0085] In Example 3, ADS-B signals are selected as test signals for large-scale real-world communication radiation source personal identification. The ADS-B system installed on the aircraft uses GPS to determine its position and transmits its position, identity, altitude, speed and other data to the outside through a transmitter at very short time intervals. A dedicated ADS-B ground station can receive the transmitted data and forward it to air traffic controllers to accurately track the aircraft.

[0086] Three test cases are set for analysis. Test case 1 tests the change in recognition accuracy before and after slice combination and feature fusion under the same deep convolutional neural network feature extractor structure; Test case 2 compares the method proposed in the present invention with the existing fingerprint recognition technology based on traditional machine learning, and analyzes the difference in recognition accuracy; Test case 3 compares the method proposed in the present invention with the existing fingerprint recognition technology based on deep learning, and analyzes the difference in recognition accuracy.

[0087] For test case 1, under the same deep convolutional neural network feature extractor structure, the recognition accuracy of 21 aircraft and 198 aircraft before and after slice combination and feature fusion can be obtained, as shown in Figure 5 , Figure 6 , Figure 7 and Figure 8 As shown. Figure 5 and Figure 6It can be seen that the recognition accuracy of the 21 aircraft before slice combination and feature fusion is 92.68%, and under the same deep convolutional neural network feature extractor structure, the recognition accuracy of the 21 aircraft after slice combination and feature fusion is 99.70%. Figure 7 and Figure 8 It can be seen that under the same deep convolutional neural network feature extractor structure, the recognition accuracy of 198 aircraft before slice combination and feature fusion is only 74.60%, and the recognition accuracy of 198 aircraft after slice merging and feature fusion can reach 99.03%.

[0088] Fig. 9 The training convergence curves of 198 aircraft before and after slice combination and feature fusion under the same deep convolutional neural network feature extractor structure are shown. The proposed method of using only a single slice is a deep network without slice combination and feature fusion; the proposed method of using two slices is a deep network with two slices combined with feature fusion under the same deep convolutional neural network feature extractor structure. Each signal slice contains 1500 steady-state data points.

[0089] from Fig. 9 It can be seen that for the deep network before slice combination and feature fusion, after about 350 training iterations, the training curve basically converged, and the final training accuracy (acc CNN) was 79.88%, and the prediction accuracy (val acc CNN) was 74.60%. For the deep network after slice combination and feature fusion under the same deep convolutional neural network feature extractor structure, we can see that after only about 50 training iterations, the training curve has basically converged, and the final training accuracy (accSCCNN) is 99.55%, and the prediction accuracy (val acc SCCNN) is 99.03%.

[0090] Therefore, test case 1 shows that the present invention can effectively guide the deep neural network structure to learn the real RF fingerprint features through signal slice combination, feature combination and fusion.

[0091] For test case 2, the recognition accuracy of 21 aircraft, 35 aircraft, 100 aircraft, and 198 aircraft can be obtained using PSD fingerprint and intelligent classifier, respectively. Figures 10 to 15 As shown. Fig.10 and Fig.11 It can be seen that the recognition accuracy of the 21 aircraft using the PSD fingerprint and GRA classifier is 82.09%, while the recognition accuracy of the 21 aircraft using the PSD and SVM classifier is 91.23%. Fig.12 It can be seen that the recognition accuracy of 35 aircraft using PSD and SVM classifiers is reduced to 83.35%. Fig.13It can be seen that the recognition accuracy of 100 aircraft using PSD and SVM classifiers is reduced to 74.23%. Fig.14 It can be seen that the recognition accuracy of 198 aircraft using PSD and SVM classifiers is further reduced to 68.31%. Fig.15 It can be seen that when the number of aircraft increases, the recognition accuracy of this method will be significantly reduced, so the traditional machine learning-based RFF recognition technology is not suitable for large-scale real-world RF fingerprint recognition in the Internet of Things.

[0092] For test case 3, in order to test the improvement of the recognition accuracy of the method of the present invention compared with the existing fingerprint recognition technology based on deep learning, the method based on contour stars, the method based on deep complex residual network, and the method based on complex convolutional neural network are selected as comparison algorithms. The confusion matrix of the comparative recognition results of these methods is as follows: Figures 16 to 19 The comparative recognition results of these methods are shown in Table 2.

[0093] from Fig.16 It can be seen that the recognition accuracy of the 21 aircraft using the method based on contour stars is only 16.53%. Fig.17 It can be seen that the recognition accuracy of the method based on complex convolutional neural network for 21 aircraft is only 67.53%. Fig.18 It can be seen that the recognition accuracy of the deep residual network-based method for 21 aircraft is only 4.34%.

[0094] The comparison of the recognition results of the methods used in test case 3 is shown in Table 2. Figures 16 to 19 It can also be seen that the RF fingerprint recognition method combining slice combination and feature fusion proposed by the method of the present invention has improved recognition accuracy compared with the existing fingerprint recognition technology based on traditional machine learning and the existing fingerprint recognition technology based on deep learning. When the number of aircraft increases, the recognition accuracy of the method of the present invention hardly decreases and remains above 99%.

[0095] Table 2 Comparison of recognition results of these methods

[0096] Comparison Method Number of aircraft Training Accuracy Prediction accuracy Power Spectral Density and Grey Relation Classifier 21 / 82.09% Power Spectral Density and Support Vector Machines 21 / 91.23% Equipotential Planet Map 21 35.96% 16.53% Residual Network 21 22.60% 4.34% Complex Residual Network 21 45.71% 6.48% Complex Convolutional Neural Networks 21 80.79 67.53% Method of the present invention 21 99.89% 99.70% Power Spectral Density and Support Vector Machines 198 / 68.31% Method of the present invention 198 99.55% 99.03%

[0097] Embodiment 4, as attached Fig. 20 As shown, this embodiment discloses a radio frequency fingerprint recognition device based on a slice combination convolutional neural network, including:

[0098] The slicing unit uses variance trajectory detection to intercept the steady-state signal of the radio frequency baseband I / Q signal and divides it into a number of slice signals; the slicing unit includes:

[0099] The sampling module uses the receiver to collect the RF baseband I / Q signal and performs ADC sampling;

[0100] The interception module performs variance trajectory detection on the sampled RF baseband I / Q signal and intercepts the steady-state signal of the RF baseband I / Q signal;

[0101] The slicing module slices the steady-state signal of the RF baseband I / Q signal to obtain a number of sliced ​​signals.

[0102] The identification unit inputs a plurality of slice signals into a slice combination convolutional neural network model, extracts fingerprint information of the RF baseband I / Q signal based on the fusion of slice signal features, and completes RF fingerprint identification, wherein the slice combination convolutional neural network model is obtained through deep learning using multiple sets of training data, and each set of training data includes: a slice signal of a steady-state signal of the RF baseband I / Q signal and identification information used to identify the RF fingerprint. The identification unit includes:

[0103] The feature extraction module inputs several slice signals into the slice combination convolutional neural network model and uses the deep convolutional neural network feature extractor to extract the features in each slice signal;

[0104] The fingerprint extraction module fuses the features extracted from each slice signal and extracts the fingerprint information of the RF baseband I / Q signal from the fused features;

[0105] The fingerprint recognition module uses the softmax function to identify the fingerprint information of the RF baseband I / Q signal.

[0106] Embodiment 5. This embodiment discloses an electronic device, including a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement a radio frequency fingerprint recognition method based on a slice combination convolutional neural network.

[0107] Embodiment 6. This embodiment discloses a computer-readable storage medium, on which a computer program that can be read by a computer is stored. The computer program is configured to execute a radio frequency fingerprint recognition method based on a slice combination convolutional neural network when running.

[0108] In the above embodiments, the processor may be a central processing unit (CPU), a general purpose processor, a digital signal processor (DSP), an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. It may also be a combination that implements a computing function, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, and the like. The memory may include, but is not limited to, various media that can store computer programs, such as a USB flash drive, a read-only memory, a mobile hard disk, a magnetic disk or an optical disk.

[0109] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present application may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0110] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0111] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0112] The above technical features constitute the best implementation example of the present application, which has strong adaptability and optimal implementation effect. Non-essential technical features can be added or reduced according to actual needs to meet the needs of different situations.

[0113] In the implementation mode of this application, the collection of user identity information involved needs to obtain the user's consent, and the collection, processing and use of this information needs to comply with relevant national and regional laws, regulations and standards.

Claims

1. A radio frequency fingerprint recognition method based on slice combination convolutional neural network, characterized in that: include: The steady-state signal of the RF baseband I / Q signal is intercepted by using variance trajectory detection, and is divided into several slice signals; Input several slice signals into the slice combination convolutional neural network model, extract the fingerprint information of the RF baseband I / Q signal based on the slice signal feature fusion, and complete the RF fingerprint recognition; The slice combination convolutional neural network model is obtained through deep learning using multiple sets of training data, each set of training data includes: a slice signal of a steady-state signal and identification information used to identify the radio frequency fingerprint; The method of using variance trajectory detection to intercept the steady-state signal of the radio frequency baseband I / Q signal and divide it into several slices includes: Use the receiver to collect RF baseband I / Q signals and perform ADC sampling; Perform variance trajectory detection on the sampled RF baseband I / Q signal to intercept the steady-state signal of the RF baseband I / Q signal; Slicing a steady-state signal of a radio frequency baseband I / Q signal to obtain a plurality of slice signals; The variance trajectory detection is performed on the sampled RF baseband I / Q signal to intercept the steady-state signal of the RF baseband I / Q signal, including: First, the transient starting point detection is performed by calculating the variance trajectory sequence of the input data sequence and setting the threshold. The variance trajectory sequence of the input data is calculated by the following formula: , , where ; in, Indicates rounding down; Indicates the length of the sliding window used for variance calculation; Indicates the displacement of the rear window function after each calculation; Represents the average value of the window function sequence; when the receiver receives a transient signal, The magnitude will increase significantly; After the variance trajectory is calculated, the starting point detection problem of the transient signal is simplified to the common change point detection problem; the change point position where the change trajectory sequence begins to rise is the starting point of the transient signal; when the variance trajectory begins to remain unchanged, it is the end position of the transient signal.

2. The radio frequency fingerprint recognition method based on slice combination convolutional neural network according to claim 1 is characterized in that: The method of inputting a plurality of slice signals into a slice combination convolutional neural network model, extracting fingerprint information of the RF baseband I / Q signal based on the fusion of slice signal features, and completing RF fingerprint recognition includes: Input several slice signals into the slice combination convolutional neural network model, and use the deep convolutional neural network feature extractor to extract the features in each slice signal; The features extracted from each slice signal are fused, and the fingerprint information of the RF baseband I / Q signal is extracted from the fused features; The softmax function is used to identify the fingerprint information of the RF baseband I / Q signal.

3. A radio frequency fingerprint recognition device based on a slice combination convolutional neural network, characterized in that: The radio frequency fingerprint recognition device is based on the radio frequency fingerprint recognition method shown in claim 1 or 2, and the radio frequency fingerprint recognition device includes: The slicing unit intercepts the steady-state signal of the RF baseband I / Q signal by using variance trajectory detection and divides it into a number of slice signals; The recognition unit inputs several slice signals into the slice combination convolutional neural network model, extracts the fingerprint information of the RF baseband I / Q signal based on the slice signal feature fusion, and completes the RF fingerprint recognition; The slice combination convolutional neural network model is obtained through deep learning using multiple sets of training data, and each set of training data includes: a slice signal of the steady-state signal of the RF baseband I / Q signal and identification information used to identify the RF fingerprint.

4. The radio frequency fingerprint recognition device based on slice combination convolutional neural network according to claim 3 is characterized in that: The slicing unit comprises: The sampling module uses the receiver to collect the RF baseband I / Q signal and performs ADC sampling; The interception module performs variance trajectory detection on the sampled RF baseband I / Q signal and intercepts the steady-state signal of the RF baseband I / Q signal; The slicing module slices the steady-state signal of the RF baseband I / Q signal to obtain a number of sliced ​​signals.

5. The radio frequency fingerprint recognition device based on slice combination convolutional neural network according to claim 3 or 4, characterized in that: The identification unit comprises: The feature extraction module inputs several slice signals into the slice combination convolutional neural network model and uses the deep convolutional neural network feature extractor to extract the features in each slice signal; The fingerprint extraction module fuses the features extracted from each slice signal and extracts the fingerprint information of the RF baseband I / Q signal from the fused features; The fingerprint recognition module uses the softmax function to identify the fingerprint information of the RF baseband I / Q signal.

6. An electronic device, characterized in that: It includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the radio frequency fingerprint recognition method based on the slice combination convolutional neural network as claimed in claim 1 or 2.

7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program that can be read by a computer, and the computer program is configured to execute the radio frequency fingerprint recognition method based on a slice combination convolutional neural network as claimed in claim 1 or 2 when running.

Images