Unmanned aerial vehicle identity verification method based on active radio frequency fingerprint injection

Abstract

The invention discloses an unmanned aerial vehicle identity verification method based on active radio frequency fingerprint injection, and belongs to the technical field of unmanned aerial vehicle identification and identity authentication. Each unmanned aerial vehicle with a legal identity firstly generates baseband IQ signal data which does not contain a special field or a lead code sequence of an identification identity, then carries out random attenuation on the amplitudes of I-path data and Q-path data, injects radio frequency fingerprint features, and then transmits the data through an antenna; s2, receiving and storing wireless signals transmitted by each unmanned aerial vehicle; s3, preprocessing: dividing the sample into a training sample set and a verification sample set; s4, constructing a radio frequency fingerprint identification network model; s5, training and verifying the radio frequency fingerprint identification network model; s6, the radio frequency fingerprint identification network model identifies a wireless signal transmitted by the unmanned aerial vehicle, and verifies the identity legality of the unmanned aerial vehicle; the method does not need to send a characteristic message embedded with identity ID information, is difficult to counterfeit, and can effectively avoid counterfeit attacks.

Description

Technical Field

[0001] This invention belongs to the field of drone identification and authentication technology, specifically relating to a drone authentication method based on active radio frequency fingerprint injection. Background Technology

[0002] With the development of drone technology, its application in logistics delivery, agricultural and forestry inspection and emergency rescue is becoming increasingly widespread.

[0003] With the increasing number of drone applications, security incidents caused by issues such as drone identity forgery are occurring frequently, seriously threatening airspace safety and public order. Therefore, drone identity verification is particularly important.

[0004] Current drone authentication methods typically require drones to transmit authentication request messages containing their own identity ID during flight. This method is vulnerable to spoofing attacks, meaning that unauthorized drones can intercept and parse other drones' authentication request messages to obtain their identity IDs and use those IDs to authenticate with the drone management system.

[0005] In view of this, a drone authentication method based on active radio frequency fingerprint injection is designed to solve the above problems. Summary of the Invention

[0006] To address the problems mentioned in the background section, this invention provides a drone authentication method based on active radio frequency fingerprint injection. This method is characterized by its difficulty in counterfeiting, effective avoidance of counterfeiting attacks, ease of detection, resistance to environmental changes, and improved accuracy of authentication.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a drone authentication method based on active radio frequency fingerprint injection, comprising the following steps: S1: Under the premise of cooperative communication mode, each drone with a legitimate identity first generates baseband IQ signal data without special fields or preamble sequences that do not contain identification information. Then, the amplitude of the I and Q data is randomly attenuated, radio frequency fingerprint features are injected, and then the data is transmitted through the antenna. S2: Receive and store the wireless signals transmitted by each UAV; S3: After preprocessing the stored drone wireless signals, they are divided into training sample sets and validation sample sets; S4: Construct an RFID fingerprint recognition network model; S5: Train the radio frequency fingerprint recognition network model using the training sample set, and verify the radio frequency fingerprint recognition network model using the verification sample set; S6: After successful verification, the radio frequency fingerprinting network model identifies the wireless signals emitted by the drone to verify the drone's identity and legitimacy.

[0008] Furthermore, in step S1, the specific steps for randomly attenuating the amplitudes of the I-channel and Q-channel data and injecting radio frequency fingerprint features include: Attenuation of sampling points for I-channel data Attenuation of sampling points for Q-channel data : ; ; In the formula: ; ; ; ; Indicates from and Random selection; Indicates from and Random selection.

[0009] Furthermore, in step S1, the IQ imbalance of the injected radio frequency fingerprint feature is 1dB.

[0010] Furthermore, the specific steps of step S3 include: S3-1: Add random-amplitude Gaussian white noise to the wireless signal of each stored drone; S3-2: Divide the wireless signal with added Gaussian white noise into equal-length signal segments; S3-3: Normalize signal segments of equal length; S3-4: Divide the normalized signal segments into training sample sets and validation sample sets.

[0011] Furthermore, in step S4, the constructed radio frequency fingerprint recognition network model adopts a deep convolutional neural network, consisting of a large kernel convolutional layer, K layers of residual convolutional blocks, pooling layers, fully connected layers, and softmax layers, wherein: After extracting features from the normalized signal segment, the large kernel convolutional layer inputs the signal into a K-layer residual convolutional block. After extracting features from the output of the large kernel convolutional layer again, the K-layer residual convolutional block is input into the pooling layer; The pooling layer compresses the output of the K residual convolutional blocks into a one-dimensional fixed-length feature code, which is then output to a classifier consisting of fully connected layers and softmax layers. ; ; In the formula: This represents the feature value at the i-th position in the first to C-th channels of the output of the K-layer residual convolutional block; This represents the result of processing the output of the K-layer residual convolutional block after applying local maxima; Indicates the feature aggregation range parameter; This represents the offset of elements within the local neighborhood during feature aggregation. This represents the i-th element in the final feature representation obtained after classifier processing; A classifier consisting of fully connected layers and softmax layers classifies the feature encodings output by the pooling layers to obtain the classification results.

[0012] Furthermore, in step S5, the training objective of the radio frequency fingerprint recognition network model is to minimize the cross-entropy loss function, which is: ; In the formula: B represents the number of samples in each training batch; This represents the probability that the current sample output by the RFID fingerprint classifier belongs to the i-th type of drone.

[0013] Furthermore, the specific steps of step S6 include: S6-1: Normalize the received wireless signal from the drone; S6-2: Divide the normalized UAV wireless signal into a set of signal segments of equal length; S6-3: Input each signal segment in the set of equal-length signal segments into the radio frequency fingerprinting network model to predict the drone to which the signal segment belongs; S6-4: Count the drones to which all signal segments belong; S6-5: The drone that appears most frequently is the drone category to which the current wireless signal to be identified belongs; S6-6: If the drone is a registered drone, then it is determined to be a legitimate user.

[0014] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention constructs an RF fingerprinting network model, injecting RF fingerprints into wireless signals such as image transmission signals and status feedback signals transmitted by the UAV. The RF fingerprinting network model identifies the injected RF fingerprints to verify the UAV's identity. Compared with existing technologies, it does not require sending special messages with embedded identity ID information, making it difficult to counterfeit and effectively avoiding counterfeiting attacks.

[0015] 2. The radio frequency fingerprint injection of the present invention randomly attenuates the amplitude of the I-channel data and Q-channel data, which is not only easy to detect, but also less affected by environmental changes, thus improving the accuracy of identity verification. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a structural diagram of the radio frequency fingerprint recognition network model of the present invention; Figure 3 This is a diagram of the experimental scenario for this invention; Figure 4 This is a comparison diagram of the signal constellation before and after the active injection of IQ imbalance features in this invention; Figure 5 This is a comparison chart of the accuracy of drone identity verification under different environments when the present invention does not actively inject radio frequency fingerprints; Figure 6 This is a comparison chart showing the accuracy of drone identity verification under different environments after actively injecting radio frequency fingerprints according to the present invention. Figure 7 This diagram illustrates the impact of actively injected radio frequency fingerprints on the communication bit error rate in this invention. Detailed Implementation

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0018] Example 1 See appendix Figure 1 The present invention provides the following technical solution: a drone authentication method based on active radio frequency fingerprint injection, comprising the following steps: S1: Under the premise of cooperative communication mode, each drone with a legitimate identity first generates baseband IQ signal data without special fields or preamble sequences that do not contain identification information. Then, the amplitude of the I and Q data is randomly attenuated, radio frequency fingerprint features are injected, and then the data is transmitted through the antenna. S2: Receive and store the wireless signals transmitted by each UAV; S3: After preprocessing the stored drone wireless signals, they are divided into training sample sets and validation sample sets; S4: Construct an RFID fingerprint recognition network model; S5: Train the radio frequency fingerprint recognition network model using the training sample set, and verify the radio frequency fingerprint recognition network model using the verification sample set; S6: After successful verification, the radio frequency fingerprinting network model identifies the wireless signals emitted by the drone to verify the drone's identity and legitimacy.

[0019] Specifically, in step S1, the amplitudes of the I-channel data and Q-channel data are randomly attenuated, and the specific steps for injecting RF fingerprint features include: Attenuation of sampling points for I-channel data Attenuation of sampling points for Q-channel data : ; ; In the formula: ; ; ; ; Indicates from and Random selection; Indicates from and Random selection.

[0020] Specifically, in step S1, the IQ imbalance of the injected radio frequency fingerprint feature is 1dB.

[0021] Specifically, step S3 includes the following steps: S3-1: Add random-amplitude Gaussian white noise to the stored wireless signal of each drone: ;

[0022] In the formula: This represents the stored wireless signal of each drone; Indicates Gaussian white noise; S3-2: Wireless signal with added Gaussian white noise The signal is divided into equal-length L segments; S3-3: Normalize the signal segments of equal length L; S3-4: Divide the normalized signal segments into training sample sets. and validation sample set In the formula: This represents a signal segment after normalization. The label represents the individual label of the communication radiation source to which the signal segment belongs; T represents the number of training samples; N represents the total number of samples.

[0023] Specifically, in step S4, the constructed radio frequency fingerprint recognition network model adopts a deep convolutional neural network, which consists of a large kernel convolutional layer, K residual convolutional blocks, pooling layers, fully connected layers, and softmax layers, wherein: The kernel size of the large kernel convolutional layer is 2×15. After extracting features from the normalized signal segment, it is input into the K-layer residual convolutional block. K-layer residual convolutional blocks extract features again from the output of the large-kernel convolutional layer. Where C represents the number of convolution kernels in the output convolutional layer of the residual convolutional block, and M represents the output dimension, which is input to the pooling layer; Pooling layers compress the output of K residual convolutional blocks into one-dimensional fixed-length feature codes. The output is fed into a classifier consisting of fully connected layers and softmax layers: ; ;

[0024] In the formula: This represents the feature value at the i-th position in the first to C-th channels of the output of the K-layer residual convolutional block; This represents the result of processing the output of the K-layer residual convolutional block after applying local maxima; Indicates the feature aggregation range parameter; This represents the offset of elements within the local neighborhood during feature aggregation. This represents the i-th element in the final feature representation obtained after classifier processing; The classifier, consisting of fully connected layers and softmax layers, encodes the features output by the pooling layer. Classify the data to obtain the classification results; See appendix Figure 2 In this embodiment, K=5 for the K-layer residual convolutional block.

[0025] Specifically, in step S5, the training objective of the RFID fingerprint recognition network model is to minimize the cross-entropy loss function. for: ;

[0026] In the formula: B represents the number of samples in each training batch; This represents the probability that the current sample output by the RFID fingerprint classifier belongs to the i-th type of drone.

[0027] Specifically, step S6 includes the following steps: S6-1: Normalize the received wireless signal from the drone; S6-2: Divide the normalized UAV wireless signal into a set of equal-length signal segments. Where L represents the segment length; S6-3: Combine each signal segment within the set of equal-length signal segments. The signal is input into an RFID fingerprinting network model to predict the drone to which the signal segment belongs. S6-4: Count the drones to which all signal segments belong; S6-5: The drone that appears most frequently is the drone category to which the current wireless signal to be identified belongs; S6-6: If the drone is a registered drone, then it is determined to be a legitimate user.

[0028] Example 2 See appendix Figure 3 In the specific experiment of this application, the experimental scenario includes one UAV network management center 100 and four UAVs 101; The drone 101 uses the DJI Matrice M100, and is equipped with an Ettus B200mini SDR board and antenna for transmitting and receiving radio signals. It is also equipped with an NVIDIA Jetson TX2 board, which is responsible for the drone's control, baseband IQ output generation, and IQ imbalance fingerprint feature injection. See appendix Figure 4 The changes in the signal constellation diagram before and after IQ imbalance injection are shown. The gray dots represent the constellation points before IQ imbalance injection and the gray dots represent the constellation points after IQ imbalance injection. The UAV network management center 100 consists of a GPU server, an Ettus B200mini SDR board, and an antenna. The GPU server controls the Ettus B200mini to collect the wireless radio frequency signals of the UAV 101 at a sampling rate of 10MHz, and uses the radio frequency fingerprinting network model to identify and verify the UAV's identity. ① Before transmitting signals, the UAV 101 does not actively inject radio frequency fingerprints. The accuracy of UAV 101's identity verification was tested in both static and dynamic environments. The results are attached. Figure 5 As shown, it can be seen that in a static environment, that is, the electromagnetic environment during the training phase is the same as that during the testing phase, a high accuracy rate of identity verification can be achieved without actively injecting radio frequency fingerprints. However, when the process is moved to a dynamic environment, that is, the electromagnetic environment during the testing phase is different from that during the training phase, the accuracy rate of drone identity verification drops significantly. ② Before transmitting signals, the UAV 101 actively injects a 1dB IQ imbalance as an RF fingerprint. The accuracy of UAV 101's identity verification is tested in both static and dynamic environments. The results are attached. Figure 6 As shown, it can be seen that when the drone 101 actively injects radio frequency fingerprints, the identity verification accuracy rate reaches 100% in a static environment. Even in a dynamic environment, the identity verification accuracy rate does not drop significantly and remains above 85%. ③ Before transmitting signals, the UAV 101 actively injects different degrees of IQ imbalance as radio frequency fingerprints, and tests the received bit error rate under different signal-to-noise ratios. The results are attached. Figure 7 As shown, it can be seen that the IQ imbalance feature injected actively should not be too large and should be controlled within 1dB so as not to affect normal communication.

[0029] In conclusion, actively injecting 1dB of radio frequency fingerprints can improve the accuracy of 101 authentication for drones.

[0030] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A drone authentication method based on active radio frequency fingerprint injection, characterized in that, Includes the following steps: S1: Under the premise of cooperative communication mode, each drone with a legitimate identity first generates baseband IQ signal data without special fields or preamble sequences that do not contain identification information. Then, the amplitude of the I and Q data is randomly attenuated, radio frequency fingerprint features are injected, and then the data is transmitted through the antenna. S2: Receive and store the wireless signals transmitted by each UAV; S3: After preprocessing the stored drone wireless signals, they are divided into training sample sets and validation sample sets; S4: Construct an RFID fingerprint recognition network model; S5: Train the radio frequency fingerprint recognition network model using the training sample set, and verify the radio frequency fingerprint recognition network model using the verification sample set; S6: After successful verification, the radio frequency fingerprinting network model identifies the wireless signals emitted by the drone to verify the drone's identity and legitimacy.

2. The drone authentication method based on active radio frequency fingerprint injection according to claim 1, characterized in that: In step S1, the specific steps for randomly attenuating the amplitudes of the I-channel and Q-channel data and injecting RF fingerprint features include: Attenuation of sampling points for I-channel data Attenuation of sampling points for Q-channel data : ； ； In the formula: ; ; ; ; Indicates from and Random selection; Indicates from and Randomly selected from the intervals.

3. The drone authentication method based on active radio frequency fingerprint injection according to claim 2, characterized in that: In step S1, the IQ imbalance of the injected radio frequency fingerprint feature is 1dB.

4. The drone authentication method based on active radio frequency fingerprint injection according to claim 3, characterized in that: The specific steps of step S3 include: S3-1: Add random-amplitude Gaussian white noise to the wireless signal of each stored drone; S3-2: Divide the wireless signal with added Gaussian white noise into equal-length signal segments; S3-3: Normalize signal segments of equal length; S3-4: Divide the normalized signal segments into training sample sets and validation sample sets.

5. The drone authentication method based on active radio frequency fingerprint injection according to claim 4, characterized in that: In step S4, the constructed radio frequency fingerprint recognition network model adopts a deep convolutional neural network, which consists of a large kernel convolutional layer, K residual convolutional blocks, pooling layers, fully connected layers, and softmax layers, wherein: After extracting features from the normalized signal segment, the large kernel convolutional layer inputs the signal into a K-layer residual convolutional block. After extracting features from the output of the large kernel convolutional layer again, the K-layer residual convolutional block is input into the pooling layer; The pooling layer compresses the output of the K residual convolutional blocks into a one-dimensional fixed-length feature code, which is then output to a classifier consisting of fully connected layers and softmax layers. ； ； In the formula: This represents the feature value at the i-th position in the first to C-th channels of the output of the K-layer residual convolutional block; This represents the result of processing the output of the K-layer residual convolutional block after applying local maxima; Indicates the feature aggregation range parameter; This represents the offset of elements within the local neighborhood during feature aggregation. This represents the i-th element in the final feature representation obtained after classifier processing; A classifier consisting of fully connected layers and softmax layers classifies the feature encodings output by the pooling layers to obtain the classification results.

6. The drone authentication method based on active radio frequency fingerprint injection according to claim 5, characterized in that: In step S5, the training objective of the radio frequency fingerprint recognition network model is to minimize the cross-entropy loss function, which is: ； In the formula: B represents the number of samples in each training batch; This represents the probability that the current sample output by the RFID fingerprint classifier belongs to the i-th type of drone.

7. The drone authentication method based on active radio frequency fingerprint injection according to claim 6, characterized in that: The specific steps of step S6 include: S6-1: Normalize the received wireless signal from the drone; S6-2: Divide the normalized UAV wireless signal into a set of signal segments of equal length; S6-3: Input each signal segment in the set of equal-length signal segments into the radio frequency fingerprinting network model to predict the drone to which the signal segment belongs; S6-4: Count the drones to which all signal segments belong; S6-5: The drone that appears most frequently is the drone category to which the current wireless signal to be identified belongs; S6-6: If the drone is a registered drone, then it is determined to be a legitimate user.

Images