An ADS-B signal radio frequency fingerprint open set identification method

Abstract

The application provides an ADS-B signal radio frequency fingerprint open set recognition method, which comprises the following steps: constructing a Gaussian mixture model according to an electromagnetic wave signal and reconstructing an ideal signal; performing time shift and frequency shift on the ideal signal, and transforming the electromagnetic wave signal and the ideal signal subjected to time shift and frequency shift to obtain a time-frequency transformation characteristic matrix; establishing a time-frequency characteristic map according to the time-frequency transformation characteristic matrix; inputting the time-frequency characteristic map into a radio frequency fingerprint recognition model to obtain a radio frequency fingerprint recognition result, and the radio frequency fingerprint recognition model comprises a frequency offset layer of a frequency offset probability matrix of a transmitter. The radio frequency fingerprint characteristics possibly existing in the signal are directly represented by the time-frequency characteristic map, the frequency offset layer structure is introduced into the Swin Transformer radio frequency fingerprint recognition model to improve the recognition accuracy, the characteristics of the transmitter are combined with the recognition network through the frequency offset layer, and better recognition accuracy is achieved in both the closed set time test and the open set time test.

Description

Technical Field

[0001] This invention belongs to the field of radio frequency fingerprint recognition technology, specifically relating to an open set recognition method for ADS-B signal radio frequency fingerprints. Background Technology

[0002] The Automatic Dependent Surveillance-Broadcast (ADS-B) system uses the Global Positioning System (GPS), air-to-ground, and air-to-air data links for traffic monitoring and information dissemination. Each aircraft uses a unique 24-bit address code as its identity information in the ADS-B signal. Aircraft equipped with ADS-B periodically broadcast their status information, enabling airspace management authorities to comprehensively monitor the status of all aircraft within the area, thereby ensuring the normal operation of air traffic. However, due to the unencrypted nature of the ADS-B system and the widespread use of software-defined radio equipment, any user can send ADS-B signals. Malicious users can use known 24-bit address codes to impersonate legitimate users, leading to airspace security failures. Although traditional key encryption methods can solve the problem of malicious intrusion, they violate the openness principle of the ADS-B system and are not suitable for practical applications.

[0003] Radio frequency (RF) fingerprinting technology utilizes subtle differences in signal performance caused by hardware deviations during equipment manufacturing to identify radiation sources. These deviations arise from various factors, such as materials, processes, equipment, and environmental conditions, and can even exist within the same batch of equipment. Signal distortion is primarily caused by components in analog circuits, such as oscillators, amplifiers, and mixers, resulting in DC bias, I / Q quadrature bias errors, phase noise, and frequency bias in the signal. These deviations exhibit typical universality and uniqueness in their signal representation. Specifically, while deviations are prevalent across devices, they manifest uniquely within each device. This forms the physical basis for identifying radiation sources through radio frequency signals.

[0004] In fingerprint recognition research, ADS-B has been extensively explored. Experiments have confirmed the feasibility of the ADS-B radio frequency fingerprinting method described by US universities and research institutions. It is worth noting that when deploying this system outdoors for real-time data reception and recognition using existing algorithms, the system's recognition accuracy drops below 50%. This type of problem corresponds to the issues of open-set time testing and closed-set time testing. Closed-set time testing refers to the training and validation models using data from the same time period; this method is commonly used in scientific research experiments. A portion of the collected data is randomly selected as the training set, and the other portion as the validation set. Open-set time testing refers to the validation test data not belonging to the same time period as the training test data. Because data collected within a fixed time period cannot contain future states and their impact on the circuit due to changes in time and environment, a model trained using data within a specific time range is effective for data collected within that time period, but ineffective for new data outside the training data time distribution, even if these signals come from the same transmitter. Since circuit performance is affected by power fluctuations, ambient temperature, and interference from other radiation sources, some characteristics of the circuit will also change with time and environment. For newly acquired data in real-world scenarios, changes in circuit performance due to various factors may render the trained model ineffective. Therefore, existing technologies may experience a significant drop in recognition accuracy in practical applications. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide an ADS-B signal radio frequency fingerprint open set recognition method to meet the need for improving the recognition rate.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] According to a first aspect, embodiments of the present invention provide an open-set identification method for ADS-B signal radio frequency fingerprints, comprising: receiving and preprocessing an electromagnetic wave signal emitted by an ADS-B transmitter antenna; constructing a Gaussian mixture model based on the preprocessed electromagnetic wave signal, updating the parameters of the Gaussian mixture model using an expectation-maximization algorithm, and reconstructing an ideal signal; performing time-shifting and frequency-shifting on the ideal signal, transforming the preprocessed electromagnetic wave signal with the time-shifted and frequency-shifted ideal signal to obtain a time-frequency transformation feature matrix; establishing a time-frequency feature map based on the time-frequency transformation feature matrix; inputting the time-frequency feature map into an ADS-B radio frequency fingerprint identification model to obtain an open-set identification result for the ADS-B signal radio frequency fingerprint, wherein the ADS-B radio frequency fingerprint identification model is constructed based on a modified Swing Transformer model, with an added frequency offset layer processing structure, the frequency offset layer being constructed using the inherent frequency offset characteristics of the ADS-B transmitter as prior conditions.

[0008] Optionally, based on the preprocessed electromagnetic wave signal, a Gaussian mixture model is constructed, the parameters of the Gaussian mixture model are updated using the expectation-maximization algorithm, and the ideal signal is reconstructed. This includes: calculating the posterior probability of the pulse amplitude of the preprocessed electromagnetic wave signal data points belonging to each Gaussian distribution; updating the parameters of the Gaussian mixture model according to the posterior probability until convergence, obtaining multiple high pulse amplitude parameters and multiple low pulse amplitude parameters; calculating the average value of the multiple high pulse amplitude parameters; calculating the average value of the multiple low pulse amplitude parameters; and obtaining the ideal signal based on the average value of the high and low pulse amplitude parameters and the average value of the low pulse amplitude parameters.

[0009] Optionally, the ideal signal is time-shifted and frequency-shifted, and the time-shifted and frequency-shifted ideal signal is transformed using the preprocessed electromagnetic wave signal to obtain the time-frequency transform characteristic matrix, including:

[0010]

[0011] Where, |χ(τ,f d ) | represents the result of transforming an ideal signal using an electromagnetic wave signal demodulated to baseband, De represents demodulating the signal to baseband, λ represents the spatial propagation attenuation coefficient, G is the nonlinear effect of the power amplifier on the signal after passing through the power amplifier, and k I and k Q It is the I / Q gain coefficient, g I (t), g Q (t) represent the outputs of the I / Q signals through the DAC, D I and D Q This indicates the I / Q DC bias, and Δf represents the frequency offset error generated by the local oscillator. d This indicates the Doppler frequency shift caused by the channel. f0 is the signal propagation frequency in space, c represents the speed of light, v is the spacecraft velocity, θ is the angle between the velocity direction and the line connecting the spacecraft and the ground receiving station, and φ is the phase error. It is phase noise generated by the local oscillator. It is an ideal signal for adding time shift and frequency shift.

[0012] Optionally, the time-frequency feature map is input into the ADS-B radio frequency fingerprint recognition model to obtain the open-set recognition result of the ADS-B signal radio frequency fingerprint. This includes: performing patch partitioning on the time-frequency feature map to divide it into multiple image blocks; the linear embedding layer maps the divided image blocks and then inputs them into the Swin Transformer to extract feature information; the feature information is input into the normalization layer and the average pooling layer for normalization and average pooling, and then input into the post-processing layer. The post-processing layer includes a fully connected layer, a frequency offset layer, and a Softmax layer. The output of the fully connected layer is used as the input of the frequency offset layer. The frequency offset layer contains the frequency offset probability matrix of various ADS-B transmitters. The Softmax layer uses the frequency offset probability matrix of various ADS-B transmitters as a priori conditions to obtain the open-set recognition result of the ADS-B signal radio frequency fingerprint.

[0013] Optionally, the received electromagnetic wave signal transmitted by the ADS-B transmitter antenna is:

[0014]

[0015] Among them, X Spa (t) is the received electromagnetic wave signal emitted by the ADS-B transmitter antenna, λ represents the spatial propagation attenuation coefficient, G is the nonlinear effect of the power amplifier on the signal after passing through the power amplifier, and k I and k Q It is the I / Q gain coefficient, g I (t), g Q (t) represent the outputs of the I / Q signals through the DAC, D I and D Q This indicates the I / Q DC bias, and Δf represents the frequency offset error generated by the local oscillator. d This indicates the Doppler frequency shift caused by the channel. f0 is the signal propagation frequency in space, c represents the speed of light, v is the spacecraft velocity, θ is the angle between the velocity direction and the line connecting the spacecraft and the ground receiving station, and φ is the phase error. It is phase noise generated by the local oscillator.

[0016] Optionally, the parameters of the Gaussian mixture model are updated based on the posterior probability, including:

[0017]

[0018] Where, N k Let λ represent the number of data points for the k-th Gaussian function, I represent the number of parameters, γ(ik) represent the posterior probability of the impulse amplitude of the i-th data point belonging to the k-th Gaussian distribution, and λ represent the number of data points for the k-th Gaussian function. k μ represents the weight of the k-th Gaussian model. kσ represents the mean of the Gaussian components. k A represents the variance of the Gaussian components. i This represents the pulse amplitude of the i-th data point.

[0019] Optionally, the average value of multiple high pulse amplitude parameters is calculated, and the ideal signal is obtained based on the average value of the high pulse amplitude parameters; or, the average value of multiple low pulse amplitude parameters is calculated, and the ideal signal is obtained based on the average value of the low pulse amplitude parameters, including:

[0020]

[0021] in, The ideal signal representing reconstruction, A HL Indicates pulse amplitude as A H Or A L A H A represents the average value of the high pulse amplitude parameter. L denoted by , where represents the average value of the low pulse amplitude parameter, N represents the total number of data points, and p(t-iT) represents the pulse of the i-th point at time iT.

[0022] Optionally, the modified SwinTransformer includes: the SW-MSA module and the W-MSA module.

[0023] According to a second aspect, an embodiment of the present invention provides an electronic device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor performs the steps of the ADS-B signal radio frequency fingerprint open set identification method described in the first aspect or any embodiment of the first aspect.

[0024] According to a third aspect, embodiments of the present invention provide a computer storage medium storing computer instructions thereon, which, when executed by a processor, implement the steps of the ADS-B signal radio frequency fingerprint open set identification method described in the first aspect or any embodiment of the first aspect.

[0025] This method provides an open-set RF fingerprinting method for ADS-B signals. In this method, multiple parameters such as phase noise, frequency shift, and time delay in the signal are represented through the signal's time-frequency feature map. Compared to traditional methods that only use signal I / Q data or simple spectrum data as input to the neural network, the scheme of this invention considers the study of the signal's own properties, more intuitively representing the RF fingerprint features that may exist in the signal. Furthermore, a post-processing layer is introduced into the ADS-B RF fingerprinting model to replace the original SoftMax layer output. The post-processing layer includes a fully connected layer, a frequency offset layer, and a softmax layer. By combining the transmitter's characteristics with the recognition network through the frequency offset layer, better recognition accuracy is achieved in both closed-set time and open-set time tests compared to traditional methods.

[0026] Other advantages, objectives, and features of the invention will be set forth in the following description and will be apparent to those skilled in the art in some respects, or may be learned by practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0027] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration:

[0028] Figure 1 This is a schematic diagram of a typical radio frequency fingerprint system design process;

[0029] Figure 2 This is a flowchart illustrating a specific example of an ADS-B signal radio frequency fingerprint open set recognition method according to the present invention;

[0030] Figure 3 This is a schematic diagram of the basic circuit of the ADS-B transmitter of the present invention;

[0031] Figure 4 This is a schematic diagram of the overall process of the time-frequency feature map construction method of the present invention;

[0032] Figure 5a This is a diagram of the electromagnetic wave signals received by the present invention;

[0033] Figure 5b This is the ideal signal diagram reconstructed by the present invention;

[0034] Figure 6 This invention presents time-frequency characteristic maps of different types of transmitters established using time-frequency transformation feature matrices.

[0035] Figure 7 This is a schematic diagram of the ADS-B radio frequency fingerprint recognition model structure of the present invention;

[0036] Figure 8 This is a schematic diagram of the model structure of two consecutive Swing Transformers connected in series according to the present invention;

[0037] Figure 9 This is a flowchart illustrating the overall process framework of ADS-B signal transmission and processing in this invention.

[0038] Figure 10 This is a schematic block diagram of a specific example of an electronic device in an embodiment of the present invention. Detailed Implementation

[0039] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can also refer to the internal connection of two components; and they can refer to a wireless connection or a wired connection. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0041] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0042] Further research and experiments revealed that previous studies were conducted within a closed time set, meaning the model training and testing data occurred within the same period. However, circuit performance is affected by various factors such as power supply fluctuations, ambient temperature, and interference from other radiation sources. As time and the environment change, certain circuit characteristics also change accordingly. Previous studies on specified datasets essentially involved randomly selecting a portion of the data for training, with the remainder used for testing. The training and testing sets had the same time distribution and were within the same period. However, newly acquired data from real-world scenarios can introduce various factors that cause changes in circuit performance, ultimately rendering the trained model ineffective. Therefore, research on RF fingerprinting technology based on closed-time sets is no longer valid.

[0043] In related research, there is little focus on open set time, with most studies concentrating on the open set of data categories. How to enable models to recognize newly arriving categories typically involves research related to transfer learning, lifelong learning, and zero-shot learning, categorized as studies addressing the openness and non-stationarity of real-world scenarios. This also applies to open set time research in ADS-B RF fingerprinting. As time and environment change, data collected within a fixed period cannot encompass the future state of the circuit and its impact. Therefore, a model trained on data from a specific time range is effective for data collected within that time period, but even if the signal originates from the same transmitter, if the temporal distribution exceeds the training data, such new data is invalid. In this case, simply using signal I / Q data or simple spectral data as input to the neural network is insufficient to solve the practical problem. The study of the signal's inherent properties should also be considered in model development.

[0044] Typical RF fingerprint system design such as Figure 1 As shown, the signal research mainly involves the study of transmitter circuit characteristics. By analyzing the main radio frequency (RF) characteristics affected by hardware features and combining them with the signal transmission method, the signal is modeled. The system design must meet the application requirements of the RF fingerprint recognition system, incorporating signal reception, processing, identification, storage, and algorithm optimization functions into the entire system. Feature extraction and clustering algorithm design utilizes collected high signal-to-noise ratio (SNR) data. The clustering algorithm groups signals with similar RF characteristics to identify different transmission sources. Then, data with the lowest possible SNR is used to verify the algorithm's performance, maintaining high recognition accuracy under open-set time conditions. Algorithm verification requires processing real data under the above conditions and in conjunction with the actual application environment to verify the algorithm's effectiveness.

[0045] Based on the above analysis and description, embodiments of the present invention provide an ADS-B signal radio frequency fingerprint open set identification method, such as... Figure 2 As shown, it includes:

[0046] S101 receives and preprocesses the electromagnetic wave signal emitted by the ADS-B transmitter antenna;

[0047] For example, the electromagnetic wave signal transmitted by the ADS-B transmitter antenna may be the result of DC offset, I / Q imbalance, frequency offset error, and phase error in the basic circuitry of the ADS-B transmitter. After receiving the electromagnetic wave signal, preprocessing can be performed, such as down-conversion, to obtain the down-converted signal. Demodulation can then be performed to extract the original baseband signal from the down-converted signal and recover the original information from the transmitting end.

[0048] Basic circuit of ADS-B transmitter as follows Figure 3As shown, the I / Q signals are output through the DAC, and are represented as g respectively. I (t), g Q (t). Considering the potential DC offset and I / Q imbalance in the circuit, at the DAC output G DAC-OUT This effect is introduced in (t):

[0049] G DAC-OUT (t)=k I (g I (t)+D I )+k Q (g Q (t)+D Q )

[0050] Where k I and k Q It is the I / Q gain coefficient, D I and D Q This indicates I / Q DC bias.

[0051] For the output signal, the carrier generated by the local oscillator is up-converted in the mixer. In reality, the carrier frequency is not always the same as the preset value. This is due to differences in materials and processes during semiconductor device manufacturing, which cause slight variations in the local oscillator frequency of each device. The deviation between the actual carrier frequency and the expected frequency caused by these variations is called carrier frequency offset (CFO). After passing through the mixer, we introduce frequency offset error and phase error.

[0052]

[0053] Among them G mix-OUT (t) represents the output of the mixer, Δf represents the frequency offset error, and φ is the phase error, which is a random process that changes with time. The nonlinear effect of the power amplifier on the signal after passing through the power amplifier is summarized as G.

[0054]

[0055] Where X Tr (t) is the electromagnetic wave signal emitted from the ADS-B transmitter antenna.

[0056] The above is the circuit model of the entire transmitter. For radio frequency fingerprinting research, the main factors to consider should include the above effects, while other effects are not fully discussed here.

[0057] For an aircraft traveling at a speed of v and an altitude of H, assume that θ is the angle between the direction of the velocity and the line connecting the aircraft and the ground receiving station.

[0058]

[0059] Among them, f d Let f0 represent the Doppler frequency shift caused by the channel, f0 be the signal propagation frequency in space, and c be the speed of light. Since the receiver is typically located in an open airport or on a mountaintop, multipath effects are negligible. Therefore, considering the channel's influence on the transmitted signal, the received electromagnetic wave signal emitted by the ADS-B transmitter antenna is:

[0060]

[0061] Among them, X Spa (t) is the received electromagnetic wave signal emitted by the ADS-B transmitter antenna, λ represents the spatial propagation attenuation coefficient, G is the nonlinear effect of the power amplifier on the signal after passing through the power amplifier, and k I and k Q It is the I / Q gain coefficient, g I (t), g Q (t) represent the outputs of the I / Q signals through the DAC, D I and D Q This indicates the I / Q DC bias, and Δf represents the frequency offset error generated by the local oscillator. d This indicates the Doppler frequency shift caused by the channel. f0 is the signal propagation frequency in space, c represents the speed of light, v is the spacecraft velocity, θ is the angle between the velocity direction and the line connecting the spacecraft and the ground receiving station, and φ is the phase error. It is phase noise generated by the local oscillator.

[0062] S102. Based on the preprocessed electromagnetic wave signal, a Gaussian mixture model is constructed, the parameters of the Gaussian mixture model are updated using the expectation-maximization algorithm, and the ideal signal is reconstructed.

[0063] For example, the Gaussian mixture model is:

[0064]

[0065] Where GMM is a Gaussian mixture model, λ k N(μ) represents the weights of the k-th Gaussian model. k ,σ k ) represents the distribution of the k-th Gaussian model, where μ k Represents the mean and σ k Represents variance.

[0066] The process of reconstructing the ideal signal is as follows Figure 4As shown, first, the number of clusters K is determined, corresponding to the number of categories. Since the ideal ADS-B signal only has high and low amplitude levels, K = 2 is sufficient. Then, a Gaussian distribution is initialized by selecting an initial parameter set, such as the mean and variance of the Gaussian distribution. Next, the Expectation-Maximization (EM) iterative algorithm is used to determine which Gaussian distribution the data point belongs to. The data point can be the pulse amplitude of the pre-processed electromagnetic wave signal at a specific instant in the time series. Expectation-Maximization (EM) consists of two steps: the E-step (expectation step), which calculates the posterior probability of each data point belonging to each Gaussian distribution; and the M-step (maximization step), which updates the mean and variance parameters of the Gaussian mixture model based on the posterior probability of each data point belonging to each Gaussian distribution, maximizing the log-likelihood function. These steps are repeated iteratively until the convergence condition is met.

[0067] The specific process of calculating the posterior probability of the impulse amplitude belonging to each Gaussian distribution for each data point in the E-step (expectation step) is as follows:

[0068] Using Bayes' theorem, we can calculate:

[0069]

[0070] Where γ(ik) indicates that the i-th data point belongs to the k-th Gaussian distribution.

[0071] The specific process for updating the mean and variance parameters of the Gaussian mixture model is as follows:

[0072]

[0073] Where, N k Let λ represent the number of data points for the k-th Gaussian function, I represent the number of parameters, γ(ik) represent the posterior probability of the impulse amplitude of the i-th data point belonging to the k-th Gaussian distribution, and λ represent the number of data points for the k-th Gaussian distribution. k μ represents the weight of the k-th Gaussian model. k σ represents the mean of the Gaussian components. k A represents the variance of the Gaussian components. i This represents the pulse amplitude of the i-th data point.

[0074] If convergence is not achieved, the parameters of the Gaussian distribution are updated and the EM algorithm is repeated iteratively. After convergence, two clusters are obtained: one with multiple high pulse amplitude parameters and the other with multiple low pulse amplitude parameters. The average value is then taken to obtain A. H and A L Calculate the average of multiple high-pulse amplitude parameters, calculate the average of multiple low-pulse amplitude parameters, and obtain the ideal signal based on the average of the high-pulse amplitude parameters and the average of the low-pulse amplitude parameters.

[0075]

[0076] in, A represents the ideal signal for reconstruction. HL Indicates pulse amplitude as A H Or A L A H A represents the average value of the high pulse amplitude parameter. L denoted by , where represents the average value of the low pulse amplitude parameter, N represents the total number of data points, and p(t-iT) represents the pulse of the i-th point at time iT.

[0077] Received electromagnetic wave signals such as Figure 5a As shown, the reconstructed ideal signal is as follows: Figure 5b As shown.

[0078] This embodiment proposes a method for constructing an ideal signal based on the received signal. Compared with traditional methods, since the Gaussian mixture model can fit the probability distribution of the data, it can be used for anomaly detection, eliminating erroneous sampling points that may have large deviations in the signal, and making the reconstructed signal more accurate.

[0079] S103, Time-shift and frequency-shift the ideal signal, and transform the preprocessed electromagnetic wave signal with the time-shifted and frequency-shifted ideal signal to obtain the time-frequency transformation feature matrix;

[0080] For example, after reconstructing the ideal signal, such as Figure 4 As shown, by performing time and frequency shifts on an ideal signal, the ideal signal transformation is: The preprocessed electromagnetic wave signal is the demodulated electromagnetic wave signal to baseband, which can be represented as: De[X] Spa [(t)], De represents demodulating the transmitted signal to baseband, and then using the demodulated electromagnetic wave signal to perform a convolution operation on the ideal signal that has undergone time and frequency shifting. The specific formula for the transformation is as follows:

[0081]

[0082] Where, |χ(τ,f d ) | represents the result of transforming an ideal signal using an electromagnetic wave signal demodulated to baseband, De represents demodulating the signal to baseband, λ represents the spatial propagation attenuation coefficient, G is the nonlinear effect of the power amplifier on the signal after passing through the power amplifier, and k I and k Q It is the I / Q gain coefficient, g I (t), g Q (t) represent the outputs of the I / Q signals through the DAC, D I and D QThis indicates the I / Q DC bias, and Δf represents the frequency offset error generated by the local oscillator. d This indicates the Doppler frequency shift caused by the channel. f0 is the signal propagation frequency in space, c represents the speed of light, v is the spacecraft velocity, θ is the angle between the velocity direction and the line connecting the spacecraft and the ground receiving station, and φ is the phase error. It is phase noise generated by the local oscillator. It is an ideal signal for adding time shift and frequency shift.

[0083] The time offset and rate offset are defined as having ranges of (τ1, τ2, ..., τ). K )∈[-T,T] and (f1,f2...f K Given that f ∈ [-f, f], the time-frequency transform characteristic matrix can be obtained:

[0084]

[0085] Where W represents the time-frequency transformation feature matrix constructed from the signal, and K is a positive integer.

[0086] S104, Establish a time-frequency feature map based on the time-frequency transformation feature matrix;

[0087] For example, constructing a time-frequency feature map based on the time-frequency transform feature matrix can be achieved using appropriate visualization tools, such as Matplotlib or Seaborn, to plot the feature matrix as an image. The horizontal axis of the feature map is typically time, the vertical axis is frequency, and the color represents energy or amplitude. Time-frequency feature maps of different types of transmitters constructed using the time-frequency transform feature matrix are shown below. Figure 6 As shown, there are six categories, each with different image features. This embodiment does not limit the method of establishing the time-frequency feature map based on the time-frequency transform feature matrix; those skilled in the art can determine the method as needed.

[0088] For different time shifts and frequency shifts, there must exist τ and f. τ Make the coherence between the two signals |χ(τ,f d The time and frequency shifts added to the ideal signal simulate the signal frequency changes caused by hardware factors in the RF circuit, which are reflected in the two-dimensional time-frequency characteristic map.

[0089] S105, input the time-frequency feature map into the ADS-B radio frequency fingerprint recognition model to obtain the ADS-B signal radio frequency fingerprint open set recognition result. The ADS-B radio frequency fingerprint recognition model is constructed based on the modified Swing Transformer model, with the addition of a frequency offset layer processing structure. This frequency offset layer is constructed using the inherent frequency offset characteristics of the ADS-B transmitter as a priori conditions.

[0090] For example, the ADS-B radio frequency fingerprint recognition model can be adopted as follows: Figure 7 The network shown uses the Swin Transformer as its main architecture. It divides the time-frequency feature map into multiple image blocks by performing patch partitioning. The linear embedding layer maps the divided image blocks and feeds them into the Swin Transformer to extract feature information. The feature information is then input to the normalization layer and the average pooling layer for normalization and average pooling, and then input to the post-processing layer. The post-processing layer includes a fully connected layer, a frequency offset layer, and a softmax layer. The output of the fully connected layer is used as the input of the frequency offset layer. The frequency offset layer contains the frequency offset probability matrix of various ADS-B transmitters. The softmax layer uses the frequency offset probability matrix of various ADS-B transmitters as a priori conditions to obtain the open set recognition result of the ADS-B signal radio frequency fingerprint.

[0091] Specifically, the linear embedding layer maps the pre-defined image blocks and feeds them into a Swing Transformer block to extract feature information. By repeatedly stacking Swing Transformer blocks, further feature information within the image is extracted. For classification problems, the output layer includes Layer Normalization (LN), Average Pooling (AP), Post-processing (BP), and the output layer itself. The BP layer contains a fully connected (FC) layer, a frequency offset layer, and a Softmax layer. Based on the aforementioned characteristic of frequency offset in signals, a frequency offset probability matrix is ​​constructed as a priori condition for recognition. When recognizing a Class A transmitter, it is assumed that there are L different frequency offsets.

[0092]

[0093] Among them W f P represents the frequency offset probability matrix. AL This represents the probability that the frequency offset of a Class A transmitter is L.

[0094] Output K of the FC layer:

[0095] K = [K1 K2 ··· K] A ]

[0096] Where K represents the output of the FC layer, K A This represents the A-th element output by the FC layer, corresponding to the A-th category.

[0097] Taking frequency offset into account, the corresponding SoftMax output is:

[0098]

[0099] in This indicates that the parameters (z1, z2, ..., z) are included.A The vector input of the SoftMax function, W f-i (n) represents the frequency deviation as W. f The element in the i-th row of the n-th column, K i It is the i-th element corresponding to the output of the FC layer.

[0100] By inputting the constructed time-frequency feature map into an optimized Swing-Transformer recognition network architecture, which fully utilizes the characteristics of the signal and introduces a frequency offset processing layer, recognition accuracy is improved. High-accuracy video fingerprint recognition is achieved in both open-set and closed-set time tests.

[0101] The ADS-B signal RF fingerprint open-set recognition method provided in this invention reflects various parameters in the signal, such as phase noise, frequency shift, and time delay, through the signal's time-frequency feature map. Compared with traditional methods that only use signal I / Q data or simple spectrum data as input to the neural network, the scheme of this invention considers the study of the signal's own properties, more intuitively representing the RF fingerprint features that may exist in the signal. Furthermore, a post-processing layer is introduced into the ADS-B RF fingerprint recognition model to replace the original SoftMax layer output. The post-processing layer includes a fully connected layer, a frequency offset layer, and a softmax layer. By combining the transmitter's characteristics with the recognition network through the frequency offset layer, better recognition accuracy is achieved in both closed-set time and open-set time tests compared to traditional methods.

[0102] As an optional implementation, an ADS-B signal radio frequency fingerprint open set identification method, the modified SwinTransformer includes: SW-MSA module and W-MSA module.

[0103] For example, such as Figure 8 As shown, two consecutive Swing Transformers are connected in series, and each Swing Transformer is connected using SW-MSA and W-MSA modules.

[0104] Two consecutive Swin Transformer modules are typically represented as:

[0105]

[0106] in and z l These are the output characteristics of SW-MSA / W-MSA and MLP in the l-th block.

[0107] In this embodiment, the SW-MSA and W-MSA modules are connected to process information on the local window, which efficiently reduces the number of parameters. Furthermore, its hierarchical construction method effectively extracts global features while also allowing for efficient parallel computation.

[0108] Based on the above embodiments, the overall flowchart of ADS-B signal transmission and processing is as follows: Figure 9 Electromagnetic wave signals are transmitted through a transmitter circuit and transmitted to a receiver via a channel. The receiver can be an aircraft or other device. After receiving the electromagnetic wave signals, the receiver preprocesses them, including down-conversion and demodulation, to extract the original baseband signal. The extracted baseband signal is then analyzed and stored for subsequent training and validation of the ADS-B RF fingerprint recognition model. Simultaneously, the demodulated electromagnetic wave signals are used to reconstruct an ideal signal, which undergoes time-shifting and frequency-shifting. The preprocessed electromagnetic wave signals are then used to transform the time-shifted and frequency-shifted ideal signal to obtain a time-frequency transformation feature matrix, thereby constructing a time-frequency feature map. This time-frequency feature map is then input into the ADS-B RF fingerprint recognition model for identification, yielding the recognition result.

[0109] This application also provides an electronic device, such as... Figure 10 As shown, processor 501 and memory 502 are connected via a bus or other means.

[0110] Processor 501 can be a central processing unit (CPU). Processor 501 can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or combinations of the above types of chips.

[0111] The memory 502, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and modules, such as the program instructions / modules corresponding to the ADS-B signal radio frequency fingerprint open set identification method in this embodiment of the invention. The processor executes various functional applications and data processing by running the non-transitory software programs, instructions, and modules stored in the memory.

[0112] Memory 502 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory 502 may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0113] The one or more modules are stored in the memory 502, and when executed by the processor 501, they perform actions such as... Figure 1 The ADS-B signal radio frequency fingerprint open set recognition method in the illustrated embodiment.

[0114] For specific details regarding the aforementioned electronic devices, please refer to the relevant references. Figure 2 The relevant descriptions and effects in the illustrated embodiments are for understanding purposes only and will not be repeated here.

[0115] This embodiment also provides a computer storage medium storing computer-executable instructions that can execute the ADS-B signal radio frequency fingerprint open set identification method in any of the above method embodiments. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk drive (HDD), or solid-state drive (SSD), etc.; the storage medium may also include combinations of the above types of memory.

[0116] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A method for open-set identification of ADS-B signal radio frequency fingerprints, characterized in that, include: Receive and preprocess electromagnetic wave signals emitted by the ADS-B transmitter antenna; Based on the preprocessed electromagnetic wave signal, a Gaussian mixture model is constructed, the parameters of the Gaussian mixture model are updated using the expectation-maximization algorithm, and the ideal signal is reconstructed. The ideal signal is time-shifted and frequency-shifted, and the preprocessed electromagnetic wave signal is transformed with the time-shifted and frequency-shifted ideal signal to obtain the time-frequency transformation characteristic matrix. Establish a time-frequency feature map based on the time-frequency transform feature matrix; The time-frequency feature map is input into the ADS-B radio frequency fingerprint recognition model to obtain the open set recognition result of the ADS-B signal radio frequency fingerprint. The ADS-B radio frequency fingerprint recognition model is built based on the modified Swing Transformer model, with the addition of a frequency offset layer processing structure. This frequency offset layer is constructed using the inherent frequency offset characteristics of the ADS-B transmitter as a priori conditions.

2. The ADS-B signal radio frequency fingerprint open set identification method according to claim 1, characterized in that, Based on the preprocessed electromagnetic wave signal, a Gaussian mixture model is constructed. The parameters of the Gaussian mixture model are updated using the expectation-maximization algorithm, and the ideal signal is reconstructed, including: Calculate the posterior probability of pulse amplitude belonging to each Gaussian distribution for the preprocessed electromagnetic wave signal data points. Based on the posterior probability, update the parameters of the Gaussian mixture model until convergence, and obtain multiple high pulse amplitude parameters and multiple low pulse amplitude parameters. Calculate the average value of multiple high pulse amplitude parameters, calculate the average value of multiple low pulse amplitude parameters, and obtain the ideal signal based on the average value of the high pulse amplitude parameters and the average value of the low pulse amplitude parameters.

3. The ADS-B signal radio frequency fingerprint open set identification method according to claim 1, characterized in that, The ideal signal is time-shifted and frequency-shifted. A transformation is then performed between the preprocessed electromagnetic wave signal and the time-shifted and frequency-shifted ideal signal to obtain the time-frequency transform characteristic matrix, which includes: Where, |χ(τ,f d ) | represents the result of transforming an ideal signal using an electromagnetic wave signal demodulated to baseband, De represents demodulating the signal to baseband, λ represents the spatial propagation attenuation coefficient, G is the nonlinear effect of the power amplifier on the signal after passing through the power amplifier, and k I and k Q It is the I / Q gain coefficient, g I (t), g Q (t) represent the outputs of the I / Q signals through the DAC, D I and D Q This indicates the I / Q DC bias, and Δf represents the frequency offset error generated by the local oscillator. d This indicates the Doppler frequency shift caused by the channel. f0 is the signal propagation frequency in space, c represents the speed of light, v is the spacecraft velocity, θ is the angle between the velocity direction and the line connecting the spacecraft and the ground receiving station, and φ is the phase error. It is phase noise generated by the local oscillator. It is an ideal signal for adding time shift and frequency shift.

4. The ADS-B signal radio frequency fingerprint open set identification method according to claim 1, characterized in that, The time-frequency feature map is input into the ADS-B radio frequency fingerprint recognition model to obtain the open-set recognition result of the ADS-B signal radio frequency fingerprint, including: The time-frequency feature map is processed by patchpartition to divide it into multiple image blocks; The linear embedding layer maps the divided image blocks and feeds them into the Swing Transformer to extract feature information; Feature information is input to a normalization layer and an average pooling layer for normalization and average pooling, and then input to a post-processing layer. The post-processing layer includes a fully connected layer, a frequency offset layer, and a Softmax layer. The output of the fully connected layer is used as the input to the frequency offset layer, which contains the frequency offset probability matrix of various ADS-B transmitters. The Softmax layer uses the frequency offset probability matrix of various ADS-B transmitters as a priori conditions to obtain the open set recognition result of ADS-B signal radio frequency fingerprint.

5. The ADS-B signal radio frequency fingerprint open set identification method according to claim 1, characterized in that, The received electromagnetic wave signal emitted by the ADS-B transmitter antenna is: Among them, X Spa (t) is the received electromagnetic wave signal emitted by the ADS-B transmitter antenna, λ represents the spatial propagation attenuation coefficient, G is the nonlinear effect of the power amplifier on the signal after passing through the power amplifier, and k I and k Q It is the I / Q gain coefficient, g I (t), g Q (t) represent the outputs of the I / Q signals through the DAC, D I and D Q Indicates I / Q DC bias, Δf represents the frequency offset error generated by the local oscillator, f d This indicates the Doppler frequency shift caused by the channel. f0 is the signal propagation frequency in space, c represents the speed of light, v is the spacecraft velocity, θ is the angle between the velocity direction and the line connecting the spacecraft and the ground receiving station, and φ is the phase error. It is phase noise generated by the local oscillator.

6. The ADS-B signal radio frequency fingerprint open set identification method according to claim 2, characterized in that, Based on the posterior probability, update the parameters of the Gaussian mixture model, including: Where, N k Let λ represent the number of data points for the k-th Gaussian function, I represent the number of parameters, γ(ik) represent the posterior probability of the impulse amplitude of the i-th data point belonging to the k-th Gaussian distribution, and λ represent the number of data points for the k-th Gaussian distribution. k μ represents the weight of the k-th Gaussian model. k σ represents the mean of the Gaussian components. k A represents the variance of the Gaussian components. i This represents the pulse amplitude of the i-th data point.

7. The ADS-B signal radio frequency fingerprint open set identification method according to claim 2, characterized in that, Calculate the average value of multiple high pulse amplitude parameters, calculate the average value of multiple low pulse amplitude parameters, and obtain the ideal signal based on the average values ​​of the high and low pulse amplitude parameters, including: in, The ideal signal representing reconstruction, A HL Indicates pulse amplitude as A H Or A L A H A represents the average value of the high pulse amplitude parameter. L denoted by , where represents the average value of the low pulse amplitude parameter, N represents the total number of data points, and p(t-iT) represents the pulse of the i-th point at time iT.

8. The ADS-B signal radio frequency fingerprint open set identification method according to claim 4, characterized in that, The modified SwinTransformer includes: SW-MSA module and W-MSA module.

9. An electronic device, the device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor performs the steps of the ADS-B signal radio frequency fingerprint open set identification method according to any one of claims 1-8.

10. A computer storage medium storing computer instructions thereon, characterized in that, When executed by the processor, this instruction implements the steps of the ADS-B signal radio frequency fingerprint open set identification method according to any one of claims 1-8.

Images