Radiation source individual identification method and system facing unknown receiver

Abstract

The invention discloses a radiation source individual identification method and system facing unknown receivers, and relates to the technical field of radiation source individual identification, a plurality of receivers are utilized to acquire signals of a plurality of radiation source individuals respectively, a sample set is constructed and divided into a training set and a test set, and the receivers of samples in the training set and the test set are different; the individual identification model comprises a feature extraction module, a binary mask unwrapping device, a transmitter mapping layer, a receiver mapping layer, a transmitter classifier, a receiver classifier and a reconstruction module; based on a feature decoupling thought, a training set and a test machine are utilized to train and test an individual recognition model, individual signals are decomposed into transmitter related features and receiver related features, and the model only uses the transmitter related features to predict, so that the influence of model recognition performance degradation caused by receiver changes is reduced; and finally, the robust recognition performance is realized. Effective extraction of transmitter related features in signals is realized through a binary mask unwrapping device and decoupling feature cross combination, and a reconstruction module is combined to ensure that key information is not lost in a model training process, so that the classification accuracy of the model is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of radiation source individual recognition, and in particular to a method and system for radiation source individual recognition facing an unknown receiver. Background Art

[0002] The radiation source individual recognition technology is a method of extracting unique radio frequency fingerprints in the radiation source signal and using a specific classification algorithm to identify radiation source individuals. This technology can comprehensively analyze and effectively mine the inherent differences between different radiation source individuals for the same model, batch, and working mode of radiation sources. With the rise of technologies such as the Internet of Things (IoT) and the Internet of Vehicles (IoV), ensuring the security and privacy of wireless networks has become particularly important, and the SEI technology provides strong support for effectively managing the security of the spectrum environment.

[0003] In recent years, with the rapid development of deep learning technology, these technologies have begun to show great potential in the field of radiation source individual recognition. Compared with traditional radiation source individual recognition methods, deep learning methods adopt an end-to-end design and no longer rely heavily on expert experience.

[0004] However, although the radiation source individual recognition methods based on deep learning have made remarkable progress in extracting complex signal features, in practical applications, especially in a multi-receiver environment, these methods face severe challenges. When the radio frequency fingerprint recognition model is deployed on a new receiver, due to the differences in the hardware characteristics of different receivers, the recognition accuracy of the model trained with the data of the old receiver often drops significantly. This performance loss is mainly due to the influence of the internal components of the receiver on the received signal, resulting in a change in the data distribution captured on different receivers. In addition, the hardware characteristics of the receiver may drift over time or change due to the instability of low-cost receivers, further exacerbating the performance instability. In related technologies, for example, Chinese Patent CN117113061A provides a cross-receiver radiation source fingerprint recognition method, which uses KL divergence to reduce the impact of the decline in the model recognition accuracy caused by unknown receivers, but this solution requires data collected by unknown receivers in the training stage and cannot immediately obtain a highly generalized model.

[0005] As can be seen from the above, the related technologies do not give any technical inspiration on how to obtain a highly generalized model without accessing the data captured by unknown receivers. Summary of the Invention

[0006] To overcome the above-mentioned defects in the prior art, the present invention provides a method for identifying individual radiation sources facing unknown receivers. Using the idea of feature decoupling, the individual signal is decomposed into transmitter-related features and receiver-related features, reducing the impact of receiver changes on the degradation of the model's recognition performance and achieving robust recognition performance.

[0007] To achieve the above object, the present invention adopts the following technical solutions, including:

[0008] A method for identifying individual radiation sources facing unknown receivers, including the following steps:

[0009] S1. Use multiple receivers to separately collect signals of several transmitters, i.e., individual radiation sources, and construct a sample set; divide the sample set into a training set and a test set, where the receivers in the training set and the test set are different;

[0010] S2. Construct an individual recognition model, which is used to identify individual radiation sources for signals collected by unknown receivers; the individual recognition model includes: a feature extraction module, a binary mask unwrapper, a transmitter mapping layer, a receiver mapping layer, a transmitter classifier, and a receiver classifier;

[0011] S3. Use the training set to train the individual recognition model:

[0012] The feature extraction module extracts the intermediate feature Z of the sample signal x in the training set; the binary mask unwrapper decomposes the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer maps the feature Z t to the transmitter-related feature f t ; the receiver mapping layer maps the feature Z r to the receiver-related feature f r ; the transmitter classifier classifies according to the transmitter-related feature f t and outputs the predicted label of the transmitter; the receiver classifier classifies according to the receiver-related feature f r and outputs the predicted label of the receiver;

[0013] Calculate the corresponding loss, update the model parameters, and obtain the trained individual recognition model;

[0014] S4. Use the prediction set to test the trained individual recognition model: the feature extraction module extracts the intermediate feature Z of the sample signal x in the prediction set; the binary mask unwrapper decomposes the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer maps the feature Z t to the transmitter-related feature f t; The transmitter classifier classifies according to the transmitter-related feature f t and outputs the predicted label of the transmitter;

[0015] According to the predicted label and the true label of the transmitter, calculate the prediction accuracy of the model, and update the model parameters to obtain the pre-trained individual recognition model;

[0016] S5. Use the pre-trained individual recognition model to perform radiation source individual recognition on the signals collected by the unknown receiver.

[0017] Preferably, the calculation of the corresponding loss includes:

[0018] Send the cross-combination of the transmitter-related feature f t and the receiver-related feature f r into the transmitter classifier, and send the transmitter-related feature f t alone into the transmitter classifier. According to the output results of the transmitter classifier in the two cases, calculate the decoupled feature cross-combination loss L CC as:

[0019]

[0020] where C Tx represents the transmitter classifier; f ti and f ri are the transmitter-related feature and the receiver-related feature of sample i respectively, and f rj is the receiver-related feature of sample j; MSE(·) represents the mean square error function.

[0021] Preferably, the calculation of the corresponding loss further includes:

[0022] Calculate the transmitter classification loss L Tx as:

[0023]

[0024] where is the transmitter prediction probability distribution vector of the model for sample i, y Txi is the transmitter true label one-hot vector of sample i, and M represents the number of samples;

[0025] Calculate the receiver classification loss L Rx as:

[0026]

[0027] where is the receiver prediction probability distribution vector of the model for sample i, and y Rxi is the receiver true label one-hot vector of sample i.

[0028] Preferably, the individual recognition model further includes a reconstruction module; in step S3, during model training, the reconstruction module performs signal reconstruction based on the transmitter-related feature f t to obtain the reconstructed signal x';

[0029] Calculating the corresponding loss further includes calculating the reconstruction loss L rec :

[0030]

[0031] where x i ' is the reconstructed signal obtained after the transmitter-related feature f of sample i ti passes through the reconstruction module, and x i is the original signal of sample i.

[0032] Preferably, in step S1, multiple transmitters and multiple receivers are used. The transmitters are regarded as radiation source individuals, and each receiver collects the signal data transmitted by all transmitters. The signal data collected by each receiver is used as the data of one domain; a part of the domain data is used as the training set and another part of the domain data is used as the test set;

[0033] In step S1, the data is also preprocessed, including discarding empty data packets and channel equalization processing.

[0034] Preferably, the feature extraction module includes at least one layer of network, and each layer of network is composed of a one-dimensional convolutional Conv1D layer, a BN layer, a ReLU layer, a MaxPool layer, and an AvgPool layer.

[0035] Preferably, the binary mask unwrapper consists of learnable vectors, and the decomposition formula is as follows:

[0036]

[0037] where M = [m1,..., m i ,..., m k ; denotes element-wise multiplication; σ(·) denotes the sigmoid function; is the network learnable parameter; γ is the threshold; the sigmoid function maps the input to the range (0, 1), and then performs threshold processing to determine that the value of the mask parameter m i is 0 or 1.

[0038] The present invention also provides a radiation source individual recognition system for an unknown receiver, which is applicable to the above-mentioned radiation source individual recognition method for an unknown receiver; the system uses an individual recognition model to perform radiation source individual recognition;

[0039] The individual recognition model includes: a feature extraction module, a binary mask unwrapper, a transmitter mapping layer, a receiver mapping layer, a transmitter classifier, a receiver classifier, and a reconstruction module;

[0040] The feature extraction module is used to extract the intermediate feature Z of the signal x; the binary mask unwrapper is used to decompose the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer is used to map the feature Z t to the transmitter-related feature f t ; the receiver mapping layer is used to map the feature Z r to the receiver-related feature f r ; the transmitter classifier is used to classify according to the transmitter-related feature f t and output the predicted label of the transmitter; the receiver classifier is used to classify according to the receiver-related feature f r and output the predicted label of the receiver; the reconstruction module is used to perform signal reconstruction on the transmitter-related feature f t to obtain the reconstructed signal x'.

[0041] The present invention also provides an electronic device, which includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the above-mentioned radiation source individual recognition method for an unknown receiver.

[0042] The present invention also provides a computer program product, which includes a computer program / instructions. When the computer program / instructions are executed by a processor, they implement the above-mentioned radiation source individual recognition method for an unknown receiver.

[0043] The advantages of the present invention are as follows:

[0044] (1) By using the idea of feature decoupling, the present invention decomposes the individual signal into transmitter-related features and receiver-related features, enabling the model to perform predictions only using the transmitter-related features, thereby reducing the impact of the degradation of the model recognition performance caused by the change of the receiver, and finally achieving a robust recognition performance.

[0045] (2) The present invention effectively extracts the transmitter-related features from the original signal through a binary mask untangler and decoupled feature cross-combinations, and combines a reconstruction module to ensure that key information is not lost during the model training process, optimizing the extraction of transmitter-related features by the model, thereby further improving the generalization performance of the model.

[0046] (3) The present invention uses a binary mask untangler, feature cross-combinations, and a reconstruction process to decompose the input individual signal of the radiation source into transmitter-related features and receiver-related features, and minimizes the influence of receiver-related features on the transmitter classifier, thereby improving the classification accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic flowchart of a method for identifying individual radiation sources facing unknown receivers according to the present invention.

[0048] Figure 2 It is a schematic diagram of the principle of a method for identifying individual radiation sources facing unknown receivers according to the present invention.

[0049] Figure 3 It is a schematic diagram comparing the recognition accuracy of the traditional ResNet model and the model of the present invention for data facing unknown receivers.

[0050] Figure 4 It is a schematic diagram of the transmitter confusion matrix of the traditional ResNet model.

[0051] Figure 5 It is a schematic diagram of the transmitter confusion matrix of the model of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0053] Embodiment 1

[0054] As Figure 1 shown, a method for identifying individual radiation sources facing unknown receivers is as follows:

[0055] S1, constructing and preprocessing a sample set.

[0056] Use multiple receivers to separately collect signals from several transmitters, i.e., radiation source individuals, and perform preprocessing to construct a sample set. The samples include transmitters, receivers, and corresponding signals. Divide the sample set into a training set and a test set, where the receivers in the training set and the test set are different.

[0057] The specific process is as follows:

[0058] S101, collect signal data.

[0059] Prepare 6 transmitters and 8 receivers. Let the transmitters follow the IEEE 802.11a / g standard and work on WiFi channel 11 with a center frequency of 2462 MHz and a bandwidth of 20 MHz together with a WiFi access point. The sampling rate of the receivers is 25 Msps, and each receiver simultaneously collects the signals transmitted by all transmitters.

[0060] S102, data preprocessing, including discarding empty data packets and channel equalization processing.

[0061] To eliminate the bias that may be introduced by the messages carried by the signals, extract the exactly same parts, the preambles Legacy Long Training Field (L-LTF) and Legacy Short Training Field (L-STF), in each WiFi data packet as signal samples, and perform channel equalization on each signal sample according to the preamble.

[0062] S103, sample set production:

[0063] Set the number of points of each signal sample to 256×2. Each signal sample consists of an in-phase component and a quadrature component. Each receiver collects 1000 data samples of each transmitter, and each receiver collects a total of 6000 data samples. There are 48000 data samples in total for 8 receivers.

[0064] Among them, the receiver numbers are 1, 2, 3, 4, 5, 6, 7, 8. Combine receiver 1 with any other receiver to obtain 7 combinations as the training set, and the test set is the data of the other receivers not participating in the training.

[0065] S2, construct an individual recognition model, which is used to perform radiation source individual recognition on the signals collected by unknown receivers. As Figure 2 shown, the individual recognition model includes: a feature extraction module, a binary mask unwrapper, a transmitter mapping layer, a receiver mapping layer, a transmitter classifier, a receiver classifier, and a reconstruction module.

[0066] Among them, the feature extraction module is used to extract the intermediate feature Z of the signal x; the binary mask unwrapper is used to decompose the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer is used to map the feature Z t to the transmitter-related feature f t ; the receiver mapping layer is used to map the feature Z r to the receiver-related feature f r ; the transmitter classifier is used to classify according to the transmitter-related feature f t and output the predicted label of the transmitter; the receiver classifier is used to classify according to the receiver-related feature f r and output the predicted label of the receiver; the reconstruction module is used to reconstruct the signal from the transmitter-related feature f t to obtain the reconstructed signal x'.

[0067] An individual recognition model based on feature decoupling is constructed by building each module. Specifically, the feature extraction module includes at least one layer of network, and each layer of network is composed of a one-dimensional convolutional Conv1D layer, a BN layer, a ReLU layer, a MaxPool layer, and an AvgPool layer. In this embodiment, the feature extraction module includes 4 layers of network. The binary mask unwrapper consists of a learnable vector. The mapping layer and the classifier are both composed of Linear. The specific structure of the individual recognition model is shown in Table 1 below.

[0068] Table 1 Schematic table of the specific structure of the individual recognition model

[0069]

[0070] The baseline model uses the classical artificial intelligence model ResNet, and optimizes and adjusts the size and number of convolutional kernels to adapt to the length, width, and height of the input signal.

[0071] S3. Use the training set to train the individual recognition model and optimize the model parameters to obtain a model with high generalization performance for unknown receiver data.

[0072] Training the individual recognition model is to select an appropriate loss function and an optimizer, and initialize the training configuration. Based on the idea of feature decoupling, the samples are separated into transmitter-related features and receiver-related features, and the corresponding losses are calculated to update the model parameters to complete the training of the model, so that when facing unknown receiver data, the model has good recognition accuracy. Among them, the training configuration includes a learning rate lr of 0.0001, a training number of epochs of 100, and a batch size of 512 for each batch of samples, etc.

[0073] First, input the sample signal x in the training set into the feature extraction module to obtain the intermediate feature Z.

[0074] Secondly, decompose the intermediate feature Z into two mutually orthogonal and complementary features through a binary mask unwrapper. The decomposition formula of the unwrapper is as follows:

[0075]

[0076] where M = [m1,..., m i ,..., m k ; denotes element-wise multiplication; σ(·) denotes the sigmoid function; are network-learnable parameters; the sigmoid function maps the input to the range (0, 1), thus providing a smooth probability interpretation of the value. Then, set the threshold to 0.5 and perform threshold processing on this value to determine that the value of the mask parameter m i is 0 or 1, and decompose the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ;

[0077] Then, send Z t and Z r into the transmitter classification branch and the receiver classification branch respectively, reduce the feature dimensions through the transmitter mapping layer and the receiver mapping layer respectively, and obtain the transmitter-related feature f t and the receiver-related feature f r .

[0078] Finally, the transmitter-related feature f t and the receiver-related feature f r pass through the transmitter classifier and the receiver classifier respectively to obtain the predicted label of the transmitter and the predicted label of the receiver.

[0079] In addition, the reconstruction module also performs signal reconstruction based on the transmitter-related feature f t to obtain the reconstructed signal x'.

[0080] Calculate the corresponding losses, including the transmitter classification loss, the receiver classification loss, the decoupled feature cross-combination loss, and the reconstruction loss, for updating the model parameters.

[0081] The transmitter classification loss L Tx is calculated from the predicted label and the true label output by the transmitter classifier, and is used to ensure the correct classification of the transmitter. The formula is as follows:

[0082]

[0083] where is the transmitter prediction probability distribution vector of the network model for sample i, y Txi is the transmitter true label one-hot vector of sample i, and M represents the number of samples.

[0084] The receiver classification loss L Rx is calculated from the predicted label and the true label output by the receiver classifier, and is used to ensure the correct classification of the receiver. The formula is as follows:

[0085]

[0086] where is the receiver prediction probability distribution vector of the network model for sample i, y Rxi is the receiver true label one-hot vector of sample i.

[0087] The transmitter-related feature f t and the receiver-related feature f r are cross-combined and fed into the transmitter classifier, and the transmitter-related feature f t is fed into the transmitter classifier alone. According to the output results of the transmitter classifier in the two cases, the decoupled feature cross-combination loss L CC is calculated to ensure that the transmitter classifier is not affected by the receiver-related features. The formula is as follows:

[0088]

[0089] where C Tx represents the transmitter classifier; f ti and f ri are the transmitter-related feature and the receiver-related feature of sample i respectively, and f rj is the receiver-related feature of sample j, and MSE(·) represents the mean square error function.

[0090] The reconstruction loss L rec is obtained by comparing the transmitter-related feature f r after passing through the reconstruction module with the original sample, and is used to avoid the loss of main features. The formula is as follows:

[0091]

[0092] where x i ' is the reconstructed signal obtained after the transmitter-related feature f ti of sample i passes through the reconstruction module, and x i is the original signal of sample i, and M represents the number of samples.

[0093] Calculate all losses in each iteration to update the optimizer until the loss value reaches the set threshold or the number of iterations reaches the set number, and save the optimal model parameters.

[0094] S4. Use the prediction set to test the trained individual recognition model.

[0095] Collect data using a receiver that has not been used in the training stage to test the individual recognition model, so as to reflect that the model has good model recognition accuracy when facing data from unknown receivers.

[0096] Specifically, after training is completed, the receiver classification branch and the reconstruction module will no longer be used. That is, the data passes through the feature extractor, the binary mask unwrapper, the transmitter mapping layer, and the transmitter classifier in sequence to obtain the final predicted label of the transmitter.

[0097] According to the predicted label and the true label of the transmitter, calculate the prediction accuracy of the model, and return to step S3 to continue updating the model parameters until the preset prediction accuracy or the maximum number of iterations is reached, and obtain the pre-trained individual recognition model.

[0098] S5. Use the pre-trained individual recognition model to perform radiation source individual recognition on the signals collected by the unknown receiver: the feature extraction module extracts the intermediate feature Z of the signal x; the binary mask unwrapper decomposes the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer maps the feature Z t to the transmitter-related feature f t ; the transmitter classifier classifies according to the transmitter-related feature f t and outputs the predicted label of the transmitter.

[0099] In this embodiment, under 7 groups of training sets, the individual recognition model of the present invention and the traditional ResNet model are obtained respectively. As Figure 3 shown, it is a comparison schematic diagram of the recognition accuracies of the individual recognition model (our scheme) of the present invention and the traditional ResNet model on their respective test sets. The two columns respectively represent the classification accuracies of the two models, so as to reflect that the model of the present invention has better performance on unknown receivers.

[0100] Figure 4 It is a schematic diagram of the transmitter confusion matrix of the traditional ResNet model. Figure 5 It is a schematic diagram of the transmitter confusion matrix of the individual recognition model of the present invention. As Figure 4 and Figure 5 shown, the ResNet model cannot correctly classify the transmitter Tx2 and the transmitter Tx6, while the method we proposed effectively alleviates the misclassification of these two categories.

[0101] In summary, the method for identifying individual radiation sources facing unknown receivers according to the present invention uses a binary mask unwrapper, feature cross - combination, and reconstruction process to decompose the input individual radiation source signal into transmitter - related features and receiver - related features, and minimizes the influence of receiver - related features on the transmitter classifier, thereby improving the classification accuracy of the model.

[0102] Embodiment 2

[0103] An individual radiation source identification system facing unknown receivers is applicable to the method for identifying individual radiation sources facing unknown receivers in the above - mentioned Embodiment 1. The system uses an individual identification model to identify individual radiation sources. Figure 2 As shown, the individual identification model includes: a feature extraction module, a binary mask unwrapper, a transmitter mapping layer, a receiver mapping layer, a transmitter classifier, a receiver classifier, and a reconstruction module.

[0104] The feature extraction module is used to extract the intermediate feature Z of the signal x; the binary mask unwrapper is used to decompose the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer is used to map the feature Z t to the transmitter - related feature f t ; the receiver mapping layer is used to map the feature Z r to the receiver - related feature f r ; the transmitter classifier is used to classify according to the transmitter - related feature f t and output the predicted label of the transmitter; the receiver classifier is used to classify according to the receiver - related feature f r and output the predicted label of the receiver; the reconstruction module is used to perform signal reconstruction on the transmitter - related feature f t to obtain the reconstructed signal x'.

[0105] Embodiment 3

[0106] An electronic device includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the method for identifying individual radiation sources facing unknown receivers in the above - mentioned Embodiment 1.

[0107] The electronic device in the embodiment of the present application can be the movable device itself or a stand - alone device independent of it. The stand - alone device can communicate with the movable device to receive the input signals collected from them and send the selected target decision behaviors to them.

[0108] The electronic device includes one or more processors and a memory. The processor can be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and can control other components in the electronic device to perform desired functions. The memory can include one or more computer program products, and the computer program products can include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory can include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory can include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions can be stored on the computer-readable storage media, and the processor can run the program instructions to implement the decision-making behavior decision method of each embodiment of the present application described above and / or other desired functions.

[0109] The electronic device may further include an input device and an output device.

[0110] Embodiment 4

[0111] In addition to the above methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions that, when run by a processor, cause the processor to execute the steps in the decision-making behavior decision method according to various embodiments of the present application described in Embodiment 1 above of this specification.

[0112] The computer program product can be written in any combination of one or more programming languages to write program code for performing the operations of the embodiments of the present application. The programming languages include object-oriented programming languages, such as Java, C++, etc., and also include conventional procedural programming languages, such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computing device, partially on the user's device, executed as an independent software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0113] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for identifying individual radiation sources facing unknown receivers, characterized in that, Including the following steps: S1. Use multiple receivers to separately collect signals of several transmitters, i.e., radiation source individuals, and construct a sample set; divide the sample set into a training set and a test set, where the receivers in the training set and the test set are different; S2. Construct an individual recognition model, which is used to recognize radiation source individuals for signals collected by an unknown receiver; The individual recognition model includes: a feature extraction module, a binary mask unwrapper, a transmitter mapping layer, a receiver mapping layer, a transmitter classifier, and a receiver classifier; S3. Use the training set to train the individual recognition model: The feature extraction module extracts the intermediate feature Z of the sample signal x in the training set; the binary mask unwrapper decomposes the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer maps the feature Z t to the transmitter-related feature f t ; the receiver mapping layer maps the feature Z r to the receiver-related feature f r ; the transmitter classifier classifies according to the transmitter-related feature f t and outputs the predicted label of the transmitter; the receiver classifier classifies according to the receiver-related feature f r and outputs the predicted label of the receiver; Calculate the corresponding loss, update the model parameters, and obtain the trained individual recognition model; S4. Use the prediction set to test the trained individual recognition model: The feature extraction module extracts the intermediate feature Z of the sample signal x in the prediction set; the binary mask unwrapper decomposes the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer maps the feature Z t to the transmitter-related feature f t ; the transmitter classifier classifies according to the transmitter-related feature f t and outputs the predicted label of the transmitter; According to the predicted label and the true label of the transmitter, calculate the prediction accuracy of the model, and update the model parameters to obtain the pre-trained individual recognition model; S5. Use the pre-trained individual recognition model to recognize radiation source individuals for signals collected by an unknown receiver.

2. The method for identifying individual radiation sources facing an unknown receiver according to claim 1, wherein The calculation of the corresponding loss includes: The transmitter-related feature f t and the receiver-related feature f r are cross-combined and fed into the transmitter classifier, and the transmitter-related feature f t is separately fed into the transmitter classifier. According to the output results of the transmitter classifier in the two cases, the decoupled feature cross-combination loss L CC is as follows: Among them, C Tx represents the transmitter classifier; f ti and f ri are the transmitter-related feature and receiver-related feature of sample i respectively, and f rj is the receiver-related feature of sample j; MSE(·) represents the mean square error function.

3. The method for identifying individual radiation sources facing an unknown receiver according to claim 2, characterized in that, The calculation of the corresponding loss further includes: Calculate the transmitter classification loss L Tx as follows: Among them, is the transmitter prediction probability distribution vector of the model for sample i, and y Txi is the transmitter true label one-hot vector of sample i, and M represents the number of samples; Calculate the classification loss L of the receiver Rx as follows: Among them, is the receiver prediction probability distribution vector of the model for sample i, and y Rxi is the one-hot vector of the true label of the receiver for sample i.

4. A method for identifying individual radiation sources facing an unknown receiver according to claim 3, characterized in that The individual recognition model further includes a reconstruction module; in step S3, during model training, the reconstruction module performs signal reconstruction based on the transmitter-related feature f t to obtain the reconstructed signal x'. Said calculating the corresponding loss further includes calculating a reconstruction loss L rec : where x i ' is the transmitter-related feature f of sample i ti The reconstructed signal obtained after passing through the reconstruction module, x i is the original signal of sample i.

5. A method for identifying individual radiation sources facing an unknown receiver according to claim 1, characterized in that In step S1, multiple transmitters and multiple receivers are used. The transmitters are used as radiation source individuals. Each receiver collects signal data transmitted by all transmitters, and the signal data collected by each receiver is used as data of one domain; part of the data of the domain is used as the training set and another part of the data of the domain is used as the test set; In step S1, the data is also preprocessed, including discarding empty data packets and channel equalization processing.

6. A method for identifying individual radiation sources facing unknown receivers according to claim 1, characterized in that, The feature extraction module includes at least one layer of network, and each layer of network is composed of a one-dimensional convolutional Conv1D layer, a BN layer, a ReLU layer, a MaxPool layer, and an AvgPool layer.

7. A method for identifying individual radiation sources facing an unknown receiver according to claim 1, characterized in that, The binary mask unwrapper consists of learnable vectors, and the decomposition formula is as follows: where \(M = [m_1,\cdots,m i ,\cdots,m k \); denotes element-wise multiplication; \(\sigma(\cdot)\) denotes the sigmoid function; are learnable parameters of the network; \(\gamma\) is the threshold; the sigmoid function maps the input to the range \((0, 1)\), and then thresholding is performed to determine that the value of the mask parameter \(m i is either 0 or 1.

8. A radiation source individual recognition system for an unknown receiver, characterized in that, Applicable to a method for identifying radiation source individuals facing an unknown receiver according to any one of claims 1-7 above; the system uses an individual recognition model to identify radiation source individuals; The individual recognition model includes: a feature extraction module, a binary mask unwrapper, a transmitter mapping layer, a receiver mapping layer, a transmitter classifier, a receiver classifier, and a reconstruction module; The feature extraction module is used to extract the intermediate feature Z of the signal x; the binary mask unwrapper is used to decompose the intermediate feature Z into two mutually orthogonal and complementary features Z t and Z r ; the transmitter mapping layer is used to map the feature Z t to the transmitter-related feature f t ; the receiver mapping layer is used to map the feature Z r to the receiver-related feature f r ; the transmitter classifier is used to classify according to the transmitter-related feature f t and output the predicted label of the transmitter; the receiver classifier is used to classify according to the receiver-related feature f r and output the predicted label of the receiver; the reconstruction module is used to perform signal reconstruction on the transmitter-related feature f t to obtain the reconstructed signal x'.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements a method for identifying radiation source individuals facing an unknown receiver according to any one of claims 1-7.

10. A computer program product, characterized in that, It includes a computer program / instructions, and when the computer program / instructions are executed by a processor, it implements a method for identifying radiation source individuals facing an unknown receiver according to any one of claims 1-7.

Images