Transmitter authentication and identification method and system based on twin network and radio frequency fingerprint

Abstract

The invention discloses a transmitter authentication and identification method and system based on a twin network and a radio frequency fingerprint, and belongs to the technical field of physical layer security protection of a wireless communication system. Training a pre-established twinborn recognition structure by using the twinborn training set to obtain a trained twinborn recognition structure; inputting an unknown receiving signal and the orthogonal sampling IQ image of each legal transmitter into the trained twin identification structure, and authenticating the unknown receiving signal; if the unknown receiving signal does not pass the authentication, judging that the corresponding unknown receiving signal comes from the unknown attack node and discarding the unknown receiving signal; and if the unknown received signal passes the authentication, identifying and classifying the unknown received signal. According to the invention, three twin identification structures are designed based on the twin network, so that the receiver has the capabilities of anomaly detection and identification classification at the same time, and the detection process of first authentication and then identification classification of the received signal is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of physical layer security protection of wireless communication systems, and specifically relates to a transmitter authentication and identification method and system based on twin networks and radio frequency fingerprints. Background Art

[0002] With the emergence of new communication scenarios such as satellite Internet, Internet of Things, and Internet of Vehicles, the scale of wireless networks and the number of devices connected have also increased dramatically. A large number of low-cost terminal devices are widely deployed in production and life, and are interconnected on a large scale through the network. The widespread deployment of terminals has facilitated production and life, but also brought more severe security challenges due to the openness of wireless channels. Therefore, the authentication of the access device identity is particularly important for protecting privacy, confidential information, and ensuring communication security.

[0003] In emerging application scenarios such as the Internet of Things, terminals have four characteristics: large number, low cost, limited computing power, and limited hardware resources. Traditional device identity authentication is usually implemented using the upper layer based on key encoding and decoding. This method ensures security by preventing attackers from completing decryption within the effective time. However, when the attacker has strong computing power, the key is easily cracked. At the same time, the distribution, management, and sharing of keys are complicated. The complex operations of the upper layer lead to higher complexity, while the sharing and management of keys lead to higher costs, such as the cost of storing too many keys and the cost of defending against eavesdropping attacks caused by frequent key exchanges. Traditional device authentication schemes are in conflict with the characteristics of current terminals. Therefore, it is necessary to seek identity authentication schemes with higher accuracy and lower complexity. Physical layer authentication is a technology that studies the access authentication security of wireless communication devices based on the physical characteristics of the wireless signal propagation link or the structural characteristics of the wireless device itself. Due to its low complexity and high accuracy, it is being widely studied as an effective supplement to the upper layer authentication scheme.

[0004] At present, there are some studies on physical layer authentication using machine learning and deep learning algorithms. For example, there are schemes that use time domain IQ samples, frequency domain fast Fourier transform results and time-frequency domain spectrograms as inputs to three deep learning networks: multilayer perceptron, long short-term memory network and convolutional neural network for feature extraction and classification; there are schemes that resample the received I / Q signals at multiple sampling rates and then send them to CNN for classification and recognition. Physical layer authentication in satellite scenarios also uses the ResNet18 network to identify and classify IQ data of low-orbit satellites. The receiver first needs to perform anomaly detection on the received signal to authenticate the legitimate device and determine whether the received signal comes from a legitimate transmitter in the training set or a new attack device. If the identity authentication is passed, identification and classification are performed to determine which legitimate transmitter the received signal comes from. However, the above existing schemes lack the steps of anomaly detection and authentication for the received signal. Summary of the invention

[0005] The purpose of the present invention is to address the problems in the above-mentioned prior art and provide a transmitter authentication and identification method and system based on twin network and radio frequency fingerprint. A twin identification structure is designed based on the twin network so that the receiver has both anomaly detection and identification and classification capabilities, thereby realizing an inspection process of first authentication and then identification and classification.

[0006] In order to achieve the above object, the present invention has the following technical solutions:

[0007] In a first aspect, a transmitter authentication and identification method based on a twin network and a radio frequency fingerprint is provided, comprising:

[0008] A twin training set is constructed based on the original baseband received signal training set, and the pre-established twin recognition structure is trained using the twin training set to obtain the trained twin recognition structure;

[0009] The unknown received signal and the orthogonal sampled IQ images of each legal transmitter are input into the trained twin recognition structure to authenticate the unknown received signal;

[0010] If the unknown received signal fails to pass the authentication, the corresponding unknown received signal is determined to come from an unknown attacking node and is discarded; if the unknown received signal passes the authentication, the unknown received signal is identified and classified.

[0011] As a preferred solution, the step of constructing a twin training set based on the original baseband received signal training set includes:

[0012] There are IQ samples of N devices in the original baseband received signal training set. All IQ samples of length q are converted into a matrix of dimension IQ images; for each IQ image in the original baseband received signal training set, randomly select an IQ image of the same type from the device class to which it belongs to form a pair, and assign a label 1 to indicate that the two IQ images are from the same device; randomly select an IQ image from other different types of devices to form a pair, and assign a label 0 to indicate that the two IQ images are from different devices; perform the above operations on all IQ images in the original baseband received signal training set to obtain the twin training set.

[0013] As a preferred solution, there are three types of twin recognition structures, and the twin recognition structures pre-established by using the twin training set for training include:

[0014] The input of the twin recognition structure 1 is the IQ image pair in the twin training set. The feature extraction layer adopts the structure based on the deep convolutional neural network AlexNet8, retains the convolution extraction part of AlexNet8, removes the fully connected classification layer, and then connects a fully connected layer with 48 hidden nodes after global average pooling for feature compression to obtain a 1*48-dimensional feature vector;

[0015] After feature extraction, the Euclidean distance of the feature vectors corresponding to the two input IQ images is calculated to obtain the similarity measure between the features. The Euclidean distance of the similarity measure is mapped to a single neuron and converted into a probability form through sigmoid activation.

[0016] The value of the output neuron represents the probability that the input IQ images come from the same device;

[0017] The training of the twin recognition structure one is completed by calculating the loss back propagation through the binary cross entropy loss function to update the network parameters.

[0018] As a preferred solution, there are three types of twin recognition structures, and the twin recognition structures pre-established by using the twin training set for training include:

[0019] The input, output and feature extraction network of the twin recognition structure 2 are the same as those of the twin recognition structure 1. The difference lies in the similarity measurement part. The twin recognition structure 2 uses the absolute value of the difference between the feature vectors v1 and v2 |v1-v2| as the feature difference layer, and then completes the training by learning the mapping relationship from the feature difference vector to the output probability of the same class through several layers of fully connected networks. The twin recognition structure 2 replaces the Euclidean distance calculation layer of the twin recognition structure 1 with a 48-dimensional feature difference layer, and then connects a 16-dimensional fully connected layer to learn the mapping relationship of feature differences, and finally outputs it to a single neuron activated by the sigmoid function. The training of the twin recognition structure 2 is completed by calculating the loss back propagation through the binary cross entropy loss function to update the network parameters.

[0020] As a preferred solution, there are three types of twin recognition structures, and the twin recognition structures pre-established by using the twin training set for training include:

[0021] The network input of the twin recognition structure three is still the IQ image pair, and the feature extraction network is the same as that of the twin recognition structure one and the twin recognition structure two. The twin recognition structure three uses the Euclidean distance to measure the similarity between the feature vectors v1 and v2, and directly outputs the Euclidean distance. The contrast loss function is used to complete the training of the twin recognition structure three. The expression of the contrast loss function is:

[0022] Contrasive_loss=Y*d 2 +(1-Y)*max(margin-d,0) 2

[0023] Among them, Y is the label 0 or 1, d is the Euclidean distance between feature vectors, and margin is a pre-set hyperparameter.

[0024] As a preferred solution, the unknown received signal and the orthogonal sampling IQ images of each legal transmitter are input into the trained twin recognition structure. When the unknown received signal is authenticated, if the twin recognition structure 1 or the twin recognition structure 2 is used, the authentication step includes:

[0025] Randomly extract m IQ images from each category of the N legitimate devices and combine them with the transmission signal of the unknown device to form an IQ image pair and input them into the twin recognition structure 1 or the twin recognition structure 2 to obtain m output values ​​[p i,1 ,p i,2 ,…,p i,m ], where i = 1, 2, ..., N; As the confidence level for judging whether the received signal comes from the legitimate device of the i-th category, we get {p1,p2,…,p N} represents the confidence level of judging that the received signal comes from N legitimate devices; if all p i If all of them are smaller than the confidence threshold λ, the corresponding received signal is considered to come from the attacking device and the corresponding received signal is discarded. i If it is greater than the confidence threshold λ, the received signal is considered to come from the legitimate device set and the subsequent identification and classification steps are performed.

[0026] As a preferred solution, the unknown received signal and the orthogonal sampling IQ images of each legal transmitter are input into the trained twin recognition structure. When the unknown received signal is authenticated, if the twin recognition structure three is adopted, the authentication step includes:

[0027] Randomly extract m IQ images from each category of the N legitimate devices and combine them with the transmission signal of the unknown device to form an IQ image pair and input them into the twin recognition structure 3 to obtain m output values ​​[d i,1 ,d i,2 ,…,d i,m ], where i = 1, 2, ..., N; As the similarity distance between the received signal and the legitimate device of the i-th category, we get {d1,d2,…,d N} indicates the similarity of judging the received signal from N legitimate devices; if all d i If all of them are greater than the similarity threshold η, the corresponding received signal is considered to come from the attacking device and the corresponding received signal is discarded; if there is a certain d i If it is less than the similarity threshold η, the received signal is considered to come from the legitimate device set and the subsequent identification and classification steps are performed.

[0028] As a preferred solution, when the unknown received signal is identified and classified, if the twin recognition structure 1 or the twin recognition structure 2 is adopted, the steps of identification and classification are:

[0029] For the network output, the category corresponding to the value with the largest confidence is selected as the classification result, that is,

[0030] If twin recognition structure three is adopted, the steps for recognition and classification are:

[0031] For the network output, the category corresponding to the value with the smallest Euclidean distance is selected as the classification result, that is,

[0032] In the second aspect, a transmitter authentication and identification system based on a twin network and radio frequency fingerprint is provided, including:

[0033] The twin recognition structure training module is used to construct a twin training set based on the original baseband received signal training set, and use the twin training set to train the pre-established twin recognition structure to obtain the trained twin recognition structure;

[0034] The authentication module is used to input the unknown received signal and the orthogonal sampling IQ images of each legal transmitter into the trained twin recognition structure to authenticate the unknown received signal;

[0035] The identification and classification module is used to determine that the corresponding unknown received signal comes from an unknown attack node and discard it if the unknown received signal fails to pass the authentication; if the unknown received signal passes the authentication, the unknown received signal is identified and classified.

[0036] In a third aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the transmitter authentication and identification method based on a twin network and a radio frequency fingerprint is implemented.

[0037] Compared with the prior art, the present invention has at least the following beneficial effects:

[0038] Based on the twin network, a twin recognition structure is designed to enable the receiver to have both anomaly detection and identification and classification capabilities. By inputting the unknown received signal and the orthogonal sampling IQ images of each legal transmitter into the trained twin recognition structure, the unknown received signal can be authenticated. If the unknown received signal fails to pass the authentication, the corresponding unknown received signal is determined to come from an unknown attack node and is discarded. If the unknown received signal passes the authentication, the unknown received signal is further identified and classified. The transmitter authentication and identification method based on the twin network and radio frequency fingerprint of the present invention realizes the inspection process of first anomaly detection and then identification and classification. By training a single twin network, anomaly detection and identification and classification of a given number of devices can be achieved. Experiments have proved that the method proposed in the present invention can achieve similar recognition and classification performance compared to traditional convolutional neural network solutions in ground wireless communication and low-orbit satellite communication scenarios, and at the same time has the ability to detect anomalies of unknown transmitters, filling the technical gap in the industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0040] Figure 1 It is a schematic diagram of a model of a transmitter authentication and identification method based on a twin network and radio frequency fingerprint applied to a physical layer authentication system according to an embodiment of the present invention;

[0041] Figure 2 This is a schematic diagram of the architecture of the twin recognition structure 1 according to an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of the architecture of the twin recognition structure 2 of an embodiment of the present invention;

[0043] Figure 4 This is a schematic diagram of the architecture of the twin recognition structure 3 of an embodiment of the present invention;

[0044] FIG5( a ) is a diagram showing the distribution of output values ​​of the twin recognition structure 1 for the same type and different types of inputs according to an embodiment of the present invention;

[0045] FIG5( b ) is a diagram showing the output value distribution of the twin recognition structure 2 for the same type and different types of inputs according to an embodiment of the present invention;

[0046] FIG5( c ) is a diagram showing the output value distribution of the twin recognition structure three for the same and different types of inputs according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, ordinary technicians in this field can also obtain other embodiments without making creative work.

[0048] See also Figure 1 When a transmitter authentication and identification method based on twin network and radio frequency fingerprint is applied to a physical layer authentication system, there are a set of legitimate transmitting nodes and a set of attacking nodes in the network. At the same time, a signal is sent to the legitimate receiver Bob to try to establish a communication connection. Bob determines the source of the received signal based on the radio frequency fingerprint feature of the wireless device. The authentication based on radio frequency fingerprint feature is divided into two stages: anomaly detection and identification classification.

[0049] Among them, in the anomaly detection stage: First, Bob needs to perform an anomaly detection on the received signal to determine whether the received signal comes from a set of legitimate transmitting nodes or an anomaly comes from an attacking node. If it comes from a set of legitimate transmitting nodes, it is received and the next stage of identification and classification is performed. If it comes from an attacking node, it is discarded and access is denied. In the identification and classification stage: When the anomaly detection in the first stage passes and the received signal comes from a set of legitimate nodes, the source ID of the received signal is further identified and classified. Determine which specific legitimate transmitting node the signal comes from and give the final classification result.

[0050] The transmitter authentication and identification method based on twin network and radio frequency fingerprint in the embodiment of the present invention includes the following steps:

[0051] Step 1: construct a twin training set based on the original baseband received signal training set, and use the twin training set to train the pre-established twin recognition structure to obtain the trained twin recognition structure;

[0052] Step 2: Input the unknown received signal and the orthogonal sampled IQ images of each legal transmitter into the trained twin recognition structure to authenticate the unknown received signal;

[0053] Step 3: If the unknown received signal fails to pass the authentication, the corresponding unknown received signal is determined to come from an unknown attacking node and is discarded; if the unknown received signal passes the authentication, the unknown received signal is identified and classified.

[0054] In a possible implementation manner, the step of constructing a twin training set based on the original baseband received signal training set in step 1 includes:

[0055] There are IQ samples of N devices in the original baseband received signal training set. All IQ samples of length q are converted into a matrix of dimension IQ images; for each IQ image in the original baseband received signal training set, randomly select an IQ image of the same type from the device class to which it belongs to form a pair, and assign a label 1 to indicate that the two IQ images are from the same device; randomly select an IQ image from other different types of devices to form a pair, and assign a label 0 to indicate that the two IQ images are from different devices; perform the above operations on all IQ images in the original baseband received signal training set to obtain the twin training set.

[0056] In a possible implementation, there are three twin recognition structures pre-established in the embodiment of the present invention, namely, twin recognition structure 1, twin recognition structure 2, and twin recognition structure 3. Step 2 of the embodiment of the present invention uses a twin training set to train the pre-established twin recognition structure, including:

[0057] like Figure 2 As shown in Figure 1, the input of the twin recognition structure 1 is the IQ image pair in the twin training set. The feature extraction layer adopts the structure based on the deep convolutional neural network AlexNet8, retains the convolution extraction part of AlexNet8, removes the last fully connected classification layer, and connects a fully connected layer of 48 hidden nodes after global average pooling for feature compression to obtain a 1*48-dimensional feature vector. The structure of the feature extraction layer is shown in Table 1. After feature extraction, the Euclidean distance of the feature vectors corresponding to the two input IQ images is calculated to obtain the similarity measure between the features. The measured Euclidean distance is mapped to a single neuron and converted into a probability form by sigmoid activation. The value of the output neuron represents the probability that the input IQ image comes from the same device. The training of the twin recognition structure 1 is completed by calculating the loss back propagation and updating the network parameters through the binary cross entropy loss function.

[0058] Table 1 Twin recognition structure - feature extraction layer parameters

[0059]

[0060] like Figure 3As shown in the figure, the input, output and feature extraction network of the twin recognition structure 2 are exactly the same as those of the twin recognition structure 1, and the difference lies in the similarity measurement part. The twin recognition structure 1 uses the Euclidean distance to measure the similarity between two feature vectors. The twin recognition structure 2 uses the absolute value of the difference between the feature vectors v1 and v2 |v1-v2| as the feature difference layer, and then completes the training by learning the mapping relationship between the feature difference vector and the output of the same class probability through several layers of fully connected networks. The network parameters of the twin recognition structure 2 are shown in Table 2. Specifically, the Euclidean distance calculation layer is replaced with a 48-dimensional feature difference layer, and then a 16-dimensional fully connected layer is connected to learn the mapping relationship of the feature difference, and finally output to a single neuron activated by the sigmoid function. The training of the twin recognition structure 2 is completed by calculating the loss back propagation through the binary cross entropy loss function to update the network parameters.

[0061] Table 2 Network parameters of twin recognition structure II

[0062]

[0063] like Figure 4 As shown in the figure, the twin recognition structure three only intercepts the part of the network that measures similarity. The main difference is that it uses the contrast loss function for training. The input of the network is still the IQ image pair, and the feature extraction network is exactly the same as the previous two. The twin recognition structure three uses the Euclidean distance to measure the similarity between the feature vectors v1 and v2, and directly outputs the Euclidean distance. The contrast loss function is used to complete the training of the twin recognition structure three. The expression of the contrast loss function is:

[0064] Contrasive_loss=Y*d 2 +(1-Y)*max(margin-d,0) 2

[0065] Among them, Y is the label 0 or 1, d is the Euclidean distance between feature vectors, and margin is a pre-set hyperparameter.

[0066] In a possible implementation manner, in step 2 of the embodiment of the present invention, the unknown received signal and the orthogonal sampling IQ images of each legal transmitter are input into the trained twin recognition structure. When authenticating the unknown received signal, if the twin recognition structure 1 or the twin recognition structure 2 is used, the authentication step includes:

[0067] First, m IQ images are randomly selected from the IQ image set of each category of N legitimate devices to form IQ image pairs, which are input into the twin recognition structure 1 or the twin recognition structure 2 to obtain m output values ​​[p i,1 ,p i,2 ,…,p i,m ], where i = 1, 2, ..., N. As the confidence level for judging that the received signal comes from the legitimate device of the i-th category. Finally, we get {p1,p2,…,p N} represents the confidence level of judging that the received signal comes from N legitimate devices. i If all of them are smaller than the confidence threshold λ, the received signal is considered to be from the attacking device and is discarded. i If it is greater than the confidence threshold λ, the received signal is considered to come from the legitimate device set and the subsequent identification and classification steps are performed.

[0068] If twin recognition structure 3 is adopted, the authentication steps include:

[0069] First, m IQ images are randomly selected from the IQ image set of each category of N legitimate devices to form IQ image pairs and input into the twin recognition structure 3 to obtain m output values ​​[d i,1 ,d i,2 ,…,d i,m ], where i = 1, 2, ..., N. As the similarity distance between the received signal and the legitimate device of the i-th category. Finally, {d1, d2, …, d N} indicates the similarity of judging that the received signal comes from N legitimate devices. i If all of them are greater than the similarity threshold η, the received signal is considered to be from the attacking device and is discarded. i If the similarity is less than the threshold η, the received signal is considered to come from the legitimate device set and the subsequent identification and classification steps are performed.

[0070] In a possible implementation, when step 3 of the embodiment of the present invention identifies and classifies the unknown received signal, if the twin recognition structure 1 or the twin recognition structure 2 is used, the steps of identification and classification are: for the network output, the category corresponding to the value with the largest confidence value is selected as the classification result, that is,

[0071] If the twin recognition structure 3 is adopted, the steps for recognition and classification are as follows: for the network output, the category corresponding to the value with the smallest Euclidean distance is selected as the classification result, that is,

[0072] Experimental results show that the method of the embodiment of the present invention can effectively detect anomalies and identify and classify received signals in ground and low-orbit satellite scenarios, realizing a detection process of first detecting anomalies and then identifying and classifying.

[0073] In order to verify the performance of the transmitter authentication and identification method proposed in the embodiment of the present invention, the following experiments were conducted:

[0074] Experimental conditions

[0075] (1) Ground wireless communication scenario

[0076] In the ground scenario, the ZigBee CC2530 module is used as the transmitter, and the modulation mode is OQPSK modulation, which complies with the IEEE 802.15.4 physical layer transmission standard. In the experiment, the preamble sequence of each frame of data is collected as the IQ sample. The preamble contains 8 symbols, corresponding to 1024 sampling points. The legal receiver uses the USRP B210 software radio device. In the experiment, the signal is sent and received in the 2.4GHz frequency band. The B210 is connected to the PC and the signal is received and processed by the GNU Radio software. The receiver sampling rate is 8MHz and the receiving gain is 35dB. In the office scenario, wireless data is collected. Each ZigBee module transmits a signal at a distance of 1m from the receiver B210, and the transmitting and receiving antennas are placed vertically on the desktop.

[0077] (2) Low-orbit satellite communication scenario

[0078] The authentication performance of low-orbit satellites for ground terminal devices is tested in the low-orbit satellite scenario. The wireless signals collected in the above ground scenario are simulated by the low-orbit satellite channel generated by the Quasi Deterministic Radio Channel Generator (QuaDRiGa) channel generation software. The low-orbit satellite altitude is set to 1400km, the elevation angle is 31.6°, the azimuth angle is 180°, and the scenario is QuaDRiGa_NTN_Urban_LOS. The satellite antenna uses a dish parabolic antenna with a diameter of 3m, a gain of 44dBi, and LHCP polarization. The terminal antenna uses a patch antenna with LHCP polarization, pointing vertically to the sky. The antenna operating frequency is 2.2GHz. The terminal position is 20 devices randomly placed within a radius of 50m.

[0079] The Adam optimizer is used for network training, with a learning rate of 1e-3, a batch_size of 64, and training stops after 100 epochs. The cross entropy loss function is used for AlexNet8 training. The baseband IQ data is collected as the baseband received signal data set, and the preamble sequence of each frame is used as a sample (1024 sampling points). Each device in the training set has 800 frames of data, and each device in the test set has 100 frames of data. In the experiment, IQ samples of 20 ZigBee devices are continuously collected to form the training set and the test set. ZigBee devices No. 1 to 15 constitute the legal transmitter set, and ZigBee devices No. 16 to 20 constitute the attack node set.

[0080] Simulation content and result analysis

[0081] Table 3 shows the parameter amounts of the AlexNet8 network and the three twin recognition structures, as well as the classification accuracy and anomaly detection capabilities in the ground and low-orbit satellite scenarios. As can be seen from Table 3, the AlexNet8 network has the largest number of parameters, the three twin recognition structures have similar parameter amounts, about 1 / 4 of the AlexNet8 network, and the twin recognition structure three has the smallest number of parameters. For the recognition and classification effects of 15 ZigBee devices, they are almost on par with the AlexNet8 network in both the ground and low-orbit satellite scenarios, and are even better than the AlexNet8 network in the ground scenario. In terms of anomaly detection capabilities, the AlexNet8 network does not have anomaly detection capabilities and can only perform intra-class classification. The three twin recognition structures have the ability to detect anomalies of unseen categories while having classification capabilities. Figures 5(a), 5(b), and 5(c) show the output value distribution of the three twin recognition structures for the same and different types of input IQ image pairs. The output values ​​of the twin recognition structures one and two represent the probability that the input is from the same type, so the output values ​​for inputs from the same device are close to 1, and the output values ​​for inputs from different devices are close to 0. The output value of the twin recognition structure three represents the Euclidean distance between the input features. Therefore, the Euclidean distance from the same device is smaller, and the Euclidean distance from different devices is larger. It can be seen that the outputs of the same and different types of the three twin recognition structures are clearly distinguishable. By selecting appropriate thresholds λ and η, the function of anomaly detection can be achieved.

[0082] Table 4 compares the anomaly detection accuracy of 15 legal devices against 5 unknown new devices in the ground and low-orbit satellite scenarios of the three twin recognition structures. During the detection, the new device and the original legal set device signals are mixed for anomaly detection. It can be seen that the twin recognition structure three has the best anomaly detection capability. The accuracy of anomaly detection of the twin recognition structure three in the ground scenario can reach 89.26%, and the accuracy of anomaly detection in the low-orbit satellite scenario can reach 81.43%.

[0083] Table 3 Parameters, classification accuracy and anomaly detection capabilities of three twin recognition structures

[0084]

[0085] Table 4 Authentication accuracy of three twin recognition structures

[0086]

[0087] Therefore, in summary, the transmitter identity authentication method based on the twin network proposed in the present invention can have good anomaly detection and identification and classification capabilities in ground and low-orbit satellite scenarios, and has anomaly detection capabilities that traditional convolutional neural networks do not have. By training a single twin network, anomaly detection and identification and classification of a given number of devices can be achieved.

[0088] The embodiment of the present invention further proposes a transmitter authentication and identification system based on a twin network and a radio frequency fingerprint, comprising:

[0089] The twin recognition structure training module is used to construct a twin training set based on the original baseband received signal training set, and use the twin training set to train the pre-established twin recognition structure to obtain the trained twin recognition structure;

[0090] The authentication module is used to input the unknown received signal and the orthogonal sampling IQ images of each legal transmitter into the trained twin recognition structure to authenticate the unknown received signal;

[0091] The identification and classification module is used to determine that the corresponding unknown received signal comes from an unknown attack node and discard it if the unknown received signal fails to pass the authentication; if the unknown received signal passes the authentication, the unknown received signal is identified and classified.

[0092] Another embodiment of the present invention further provides an electronic device, comprising:

[0093] A memory storing at least one instruction; and a processor executing the instructions stored in the memory to implement the transmitter authentication and identification method based on the twin network and radio frequency fingerprint.

[0094] Another embodiment of the present invention further proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the transmitter authentication and identification method based on the twin network and radio frequency fingerprint.

[0095] Exemplarily, the instructions stored in the memory may be divided into one or more modules / units, which are stored in a computer-readable storage medium and executed by the processor to complete the transmitter authentication and identification method based on twin network and radio frequency fingerprint of the present invention. The one or more modules / units may be a series of computer-readable instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the server.

[0096] The electronic device may be a computing device such as a smart phone, a notebook, a PDA, and a cloud server. The electronic device may include, but is not limited to, a processor and a memory. Those skilled in the art will appreciate that the electronic device may also include more or fewer components, or a combination of certain components, or different components, for example, the electronic device may also include an input / output device, a network access device, a bus, etc.

[0097] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0098] The memory may be an internal storage unit of the server, such as a hard disk or memory of the server. The memory may also be an external storage device of the server, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the server. Furthermore, the memory may include both an internal storage unit of the server and an external storage device. The memory is used to store the computer-readable instructions and other programs and data required by the server. The memory may also be used to temporarily store data that has been output or is to be output.

[0099] It should be noted that the information interaction, execution process and other contents between the above-mentioned module units are based on the same concept as the method embodiment. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0100] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0101] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device that can carry the computer program code to the camera device / terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium. For example, a USB flash drive, a mobile hard disk, a disk or an optical disk.

[0102] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0103] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A transmitter authentication and identification method based on twin network and radio frequency fingerprint, characterized in that: include: A twin training set is constructed based on the original baseband received signal training set, and the pre-established twin recognition structure is trained using the twin training set to obtain the trained twin recognition structure; The unknown received signal and the orthogonal sampled IQ images of each legal transmitter are input into the trained twin recognition structure to authenticate the unknown received signal; If the unknown received signal fails the authentication, it is determined that the corresponding unknown received signal comes from an unknown attacking node and is discarded; If the unknown received signal passes the authentication, the unknown received signal is identified and classified.

2. According to claim 1, the transmitter authentication and identification method based on twin network and radio frequency fingerprint is characterized in that: The step of constructing a twin training set based on the original baseband received signal training set includes: There are IQ samples of N devices in the original baseband received signal training set. All IQ samples of length q are converted into a matrix of dimension IQ images; for each IQ image in the original baseband received signal training set, randomly select an IQ image of the same type from the device class to which it belongs to form a pair, and assign a label 1 to indicate that the two IQ images are from the same device; randomly select an IQ image from other different types of devices to form a pair, and assign a label 0 to indicate that the two IQ images are from different devices; perform the above operations on all IQ images in the original baseband received signal training set to obtain the twin training set.

3. According to claim 2, the transmitter authentication and identification method based on twin network and radio frequency fingerprint is characterized in that: There are three types of twin recognition structures, and the twin recognition structures pre-established by using the twin training set for training include: The input of the twin recognition structure 1 is the IQ image pair in the twin training set. The feature extraction layer adopts the structure based on the deep convolutional neural network AlexNet8, retains the convolution extraction part of AlexNet8, removes the fully connected classification layer, and then connects a fully connected layer with 48 hidden nodes after global average pooling for feature compression to obtain a 1*48-dimensional feature vector; After feature extraction, the Euclidean distance of the feature vectors corresponding to the two input IQ images is calculated to obtain the similarity measure between the features. The Euclidean distance of the similarity measure is mapped to a single neuron and converted into a probability form through sigmoid activation. The value of the output neuron represents the probability that the input IQ images come from the same device; The training of the twin recognition structure one is completed by calculating the loss back propagation through the binary cross entropy loss function to update the network parameters.

4. According to claim 3, the transmitter authentication and identification method based on twin network and radio frequency fingerprint is characterized in that: There are three types of twin recognition structures, and the twin recognition structures pre-established by using the twin training set for training include: The input, output and feature extraction network of the twin recognition structure 2 are the same as those of the twin recognition structure 1. The difference lies in the similarity measurement part. The twin recognition structure 2 uses the absolute value of the difference between the feature vectors v1 and v2 |v1-v2| as the feature difference layer, and then completes the training by learning the mapping relationship from the feature difference vector to the output probability of the same class through several layers of fully connected networks. The twin recognition structure 2 replaces the Euclidean distance calculation layer of the twin recognition structure 1 with a 48-dimensional feature difference layer, and then connects a 16-dimensional fully connected layer to learn the mapping relationship of feature differences, and finally outputs it to a single neuron activated by the sigmoid function. The training of the twin recognition structure 2 is completed by calculating the loss back propagation through the binary cross entropy loss function to update the network parameters.

5. The transmitter authentication and identification method based on twin network and radio frequency fingerprint according to claim 4 is characterized in that: There are three types of twin recognition structures, and the twin recognition structures pre-established by using the twin training set for training include: The network input of the twin recognition structure three is still the IQ image pair, and the feature extraction network is the same as that of the twin recognition structure one and the twin recognition structure two. The twin recognition structure three uses the Euclidean distance to measure the similarity between the feature vectors v1 and v2, and directly outputs the Euclidean distance. The contrast loss function is used to complete the training of the twin recognition structure three. The expression of the contrast loss function is: Contrasive_loss=Y*d 2 +(1-Y)*max(margin-d,0) 2 Among them, Y is the label 0 or 1, d is the Euclidean distance between feature vectors, and margin is a pre-set hyperparameter.

6. The transmitter authentication and identification method based on twin network and radio frequency fingerprint according to claim 5 is characterized in that: The unknown received signal and the orthogonal sampling IQ images of each legal transmitter are input into the trained twin recognition structure. When the unknown received signal is authenticated, if the twin recognition structure 1 or the twin recognition structure 2 is used, the authentication step includes: Randomly extract m IQ images from each category of the N legitimate devices and combine them with the transmission signal of the unknown device to form an IQ image pair and input them into the twin recognition structure 1 or the twin recognition structure 2 to obtain m output values ​​[p i,1 ,p i,2 ,…,p i,m ], where i = 1, 2, ..., N; As the confidence level for judging whether the received signal comes from the legitimate device of the i-th category, we get {p1,p2,…,p N } represents the confidence level of judging that the received signal comes from N legitimate devices; if all p i If all of them are smaller than the confidence threshold λ, the corresponding received signal is considered to come from the attacking device and the corresponding received signal is discarded. i If it is greater than the confidence threshold λ, the received signal is considered to come from the legitimate device set and the subsequent identification and classification steps are performed.

7. The transmitter authentication and identification method based on twin network and radio frequency fingerprint according to claim 5 is characterized in that: The unknown received signal and the orthogonal sampling IQ images of each legal transmitter are input into the trained twin recognition structure. When the unknown received signal is authenticated, if the twin recognition structure three is adopted, the authentication step includes: Randomly extract m IQ images from each category of the N legitimate devices and combine them with the transmission signal of the unknown device to form an IQ image pair and input them into the twin recognition structure 3 to obtain m output values ​​[d i,1 ,d i,2 ,…,d i,m ], where i = 1, 2, ..., N; As the similarity distance between the received signal and the legitimate device of the i-th category, we get {d1,d2,…,d N } indicates the similarity of judging the received signal from N legitimate devices; if all d i If all of them are greater than the similarity threshold η, the corresponding received signal is considered to come from the attacking device and the corresponding received signal is discarded; if there is a certain d i If it is less than the similarity threshold η, the received signal is considered to come from the legitimate device set and the subsequent identification and classification steps are performed.

8. The transmitter authentication and identification method based on twin network and radio frequency fingerprint according to claim 5 is characterized in that: When the unknown received signal is identified and classified, if the twin identification structure 1 or the twin identification structure 2 is adopted, the steps of identification and classification are as follows: For the network output, the category corresponding to the value with the largest confidence is selected as the classification result, that is, If twin recognition structure three is adopted, the steps for recognition and classification are: For the network output, the category corresponding to the value with the smallest Euclidean distance is selected as the classification result, that is, 9. A transmitter authentication and identification system based on twin network and radio frequency fingerprint, characterized in that: include: The twin recognition structure training module is used to construct a twin training set based on the original baseband received signal training set, and use the twin training set to train the pre-established twin recognition structure to obtain the trained twin recognition structure; The authentication module is used to input the unknown received signal and the orthogonal sampling IQ images of each legal transmitter into the trained twin recognition structure to authenticate the unknown received signal; The identification and classification module is used to determine that the corresponding unknown received signal comes from an unknown attack node and discard it if the unknown received signal fails to pass the authentication; If the unknown received signal passes the authentication, the unknown received signal is identified and classified.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the transmitter authentication and identification method based on a twin network and radio frequency fingerprint is implemented as described in any one of claims 1 to 8.

Images