A communication radiation source identification method, device, equipment and storage medium

Abstract

The application provides a communication radiation source identification method, device and equipment and a storage medium. The method comprises the following steps: obtaining a signal feature of a radio signal; inputting the signal feature into a pre-trained communication radiation source identification model to obtain a communication radiation source identification result corresponding to the radio signal; wherein the communication radiation source identification model comprises a feature reconstruction subnetwork and a feature classification subnetwork, and the feature reconstruction subnetwork is obtained through feature reconstruction training by taking the signal feature of a radio signal sample as a training sample and a training label respectively. The communication radiation source identification method greatly reduces the quantity requirement of labeled samples, can achieve the purpose of communication radiation source identification under the condition of a small number of labeled samples, and thus makes the method competent for more communication radiation source identification tasks.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a communication emitter identification method and device, equipment and storage medium. BACKGROUND

[0002] Communication emitter individual identification is a technology for identifying a specific communication emitter individual by receiving and measuring radio signals emitted by the communication emitter individual, extracting specific fingerprint information of the emitter individual from the radio signals, and identifying the specific communication emitter individual that emits the given radio signals.

[0003] In recent years, deep learning methods have been increasingly widely applied in the field of communication emitter identification. The communication emitter identification based on deep learning methods mainly extracts features by designing a neural network, and classifies and identifies communication emitter individuals. This communication emitter identification method needs to constantly train the model using training samples with emitter individual class labels, so that the neural network model can deeply mine the features of the emitter individual contained in the radio signals, thereby achieving accurate communication emitter individual identification.

[0004] The above-mentioned communication emitter identification method based on deep learning needs a large number of training samples to train the neural network model, so as to achieve an observable identification effect. However, when the number of training samples is small, the model training effect will be poor, and the communication emitter identification accuracy will be reduced. SUMMARY

[0005] Based on the above technical status, the present application proposes a communication emitter identification method, device, equipment and storage medium, which can achieve high-accuracy communication emitter identification under the condition of reducing training samples.

[0006] The first aspect of the present application proposes a communication emitter identification method, comprising: obtaining signal features of a radio signal;

[0007] inputting the signal features into a pre-trained communication emitter identification model to obtain a communication emitter identification result corresponding to the radio signal;

[0008] The communication emitter identification model comprises a feature reconstruction subnetwork and a feature classification subnetwork, and the feature reconstruction subnetwork is trained by taking the signal features of the radio signal samples as training samples and training labels respectively.

[0009] The second aspect of the present application proposes a communication emitter identification device, comprising:

[0010] a signal processing unit configured to obtain signal features of a radio signal;

[0011] The signal recognition unit is used to input the signal features into a pre-trained communication radiation source recognition model to obtain the communication radiation source recognition result corresponding to the radio signal;

[0012] The communication radiation source identification model includes a feature reconstruction subnetwork and a feature classification subnetwork. The feature reconstruction subnetwork is obtained by using the signal features of radio signal samples as training samples and training labels respectively for feature reconstruction training.

[0013] A third aspect of this application discloses a communication radiation source identification device, comprising:

[0014] A signal acquisition unit, a signal encoder connected to the signal acquisition unit, and a processor connected to the signal encoder;

[0015] The signal acquisition device is used to acquire radio signals;

[0016] The signal encoder is used to extract signal features from the radio signals collected by the signal collector to obtain the signal features of the radio signals.

[0017] The processor is configured to input the signal features output by the signal encoder into a pre-trained communication radiation source identification model to obtain a communication radiation source identification result corresponding to the radio signal; wherein, the communication radiation source identification model includes a feature reconstruction subnetwork and a feature classification subnetwork, and the feature reconstruction subnetwork is obtained by performing feature reconstruction training by using the signal features of the radio signal sample as training samples and training labels respectively.

[0018] The fourth aspect of this application discloses a communication radiation source identification device, comprising:

[0019] Memory and processor;

[0020] The memory is connected to the processor and is used to store programs;

[0021] The processor is used to implement the above-described communication radiation source identification method by running the program in the memory.

[0022] The fifth aspect of this application proposes a storage medium storing a computer program, which, when executed by a processor, implements the aforementioned communication radiation source identification method.

[0023] The proposed communication radiation source identification method employs unsupervised feature reconstruction training on the feature reconstruction subnetwork of the communication radiation source identification model. This enables the model to possess good feature prediction and reconstruction capabilities; in other words, unsupervised feature reconstruction training allows the model to accurately extract signal features beneficial for communication radiation source classification. This method significantly reduces the requirement for a large number of labeled samples, achieving communication radiation source identification with a limited sample size, thus enabling it to handle a wider range of communication radiation source identification tasks. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0025] FIG. 1 A flowchart illustrating a communication radiation source identification method provided in an embodiment of this application;

[0026] FIG. 2 This is a schematic diagram of the structure of the communication radiation source identification model provided in the embodiments of this application;

[0027] FIG. 3 A schematic diagram illustrating the training process of the communication radiation source identification model provided in this application embodiment;

[0028] FIG. 4 This is a schematic diagram of the structure of a communication radiation source identification device provided in an embodiment of this application;

[0029] FIG. 5 A schematic diagram of the structure of a communication radiation source identification device provided in an embodiment of this application;

[0030] FIG. 6 This is a schematic diagram of another communication radiation source identification device provided in an embodiment of this application. Detailed Implementation

[0031] The technical solution of this application is applicable to the application scenario of identifying communication radiation sources that transmit radio signals. Using the technical solution of this application, a high-performance communication radiation source identification model can be trained with a small number of labeled samples, thereby achieving accurate communication radiation source identification.

[0032] In traditional communication radiation source identification techniques, the most crucial task is extracting effective feature parameters that reflect the individual fingerprint of the radiation source. However, the generation mechanism of individual radiation source fingerprints is complex and lacks a precise physical definition, making it impossible to accurately model and express using mathematical tools. Furthermore, radiation source fingerprints are unintentional modulations attached to radio signals, with weaker energy compared to carrier signals, making them particularly susceptible to interference from complex channel conditions, multipath effects, and environmental noise. These problems typically prevent the extraction of true radiation source fingerprints. Therefore, many researchers have proposed extracting effective feature parameters from multiple perspectives and then performing weighted fusion at the feature level to obtain a radiation source fingerprint that approximates the true fingerprint and possesses uniqueness, independence, and stability. However, this feature fusion method increases the complexity of communication radiation source identification techniques, requiring high hardware and software barriers for widespread application.

[0033] In recent years, deep learning methods have achieved remarkable success in many research fields, including the identification of individual radiation sources in communication systems. Deep learning-based methods for identifying individual radiation sources do not require extensive expertise in related fields, nor do they need to define the actual physical meaning of the extracted feature parameters. They primarily extract features by designing neural network structures and then predict the category of the individual radiation source.

[0034] This method for identifying individual radiation sources is an end-to-end approach. Specifically, it involves establishing a mathematical model and algorithm for a convolutional neural network structure, training its connection weight parameters, and enabling the network to perform data-based pattern recognition and function mapping. By continuously training the network using training samples labeled with individual radiation source category information, it can deeply mine the essential fingerprint information of weak radiation sources contained in radio signals, thus accurately classifying and identifying individual radiation sources. Using deep learning methods for individual radiation source identification eliminates the need for pre-designed, finely detailed feature parameter extraction methods, reducing reliance on prior knowledge in related fields and lowering the research threshold. Therefore, it has become the mainstream technical solution in the field of individual radiation source identification in communications.

[0035] The aforementioned deep learning-based scheme for identifying individual sources of communication radiation uses a data-driven approach to train a convolutional neural network model. The size of the training sample directly determines the training effect of the convolutional neural network, thus affecting its performance in identifying individual sources of communication radiation. Without a sufficient number of training samples, a high-performing network model cannot be trained, and the accuracy of individual radiation source identification will be low.

[0036] Therefore, conventional deep learning-based individual identification schemes for communication radiation sources cannot be applied when there are few labeled samples. In particular, when there are few signal samples that can be collected for a particular communication radiation source, the network model cannot accurately identify the communication radiation source.

[0037] Based on the aforementioned technical problems, this application proposes a communication radiation source identification scheme. This scheme focuses on improving the training method of the communication radiation source identification model, reducing the required number of labeled samples for communication radiation sources. This allows the communication radiation source identification model to achieve good training results with only a small number of labeled samples, thereby accurately identifying communication radiation sources. The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0038] Exemplary method

[0039] This application first proposes a method for identifying communication radiation sources, see [link to previous document]. FIG. 1 As shown, the method includes:

[0040] S101. Obtain the signal characteristics of the radio signal.

[0041] The aforementioned radio signals refer to radio signals emitted by an individual communication radiation source into the signal propagation space. Radio signals emitted by an individual communication radiation source into the signal propagation space can be captured from this space using signal acquisition equipment such as radar and antennas.

[0042] Different individual communication radiation sources emit radio signals with different characteristics such as signal frequency, band, modulation method, carrier frequency, and amplitude-frequency characteristics. Therefore, based on the signal characteristics of the acquired radio signal, it is possible to infer which individual communication radiation source emitted the radio signal, thereby achieving the purpose of identifying the communication radiation source.

[0043] Based on the above ideas, in this embodiment of the application, after obtaining the radio signal, the signal characteristics of the radio signal are first obtained.

[0044] The signal characteristics of the radio signal can be obtained by extracting signal characteristics from the radio signal. Alternatively, if the signal characteristics of the radio signal have been pre-stored in a preset storage location, the signal characteristics of the radio signal can be directly read from that preset location.

[0045] When extracting signal features from the radio signal, any signal feature extraction method specific to radio signals can be used. In this embodiment, the short-time fourier transform (STFT) of the acquired radio signal is performed to obtain the STFT characteristic parameters of the radio signal, which are then used as the signal features of the radio signal.

[0046] S102. Input the signal features into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to the radio signal.

[0047] The communication radiation source identification model includes a feature reconstruction subnetwork and a feature classification subnetwork. The feature reconstruction subnetwork is obtained by using the signal features of radio signal samples as training samples and training labels respectively for feature reconstruction training.

[0048] For details, see FIG. 2 As shown in the embodiments of this application, the communication radiation source identification model mainly consists of a feature reconstruction subnetwork and a feature classification subnetwork. The signal features of the radio signal are first input into the feature reconstruction subnetwork for feature reconstruction, and the signal features output by the feature reconstruction subnetwork are input into the feature classification subnetwork for feature classification, thereby obtaining the communication radiation source classification and identification result.

[0049] The feature reconstruction subnetwork performs feature reconstruction on the signal features of radio signals. It mainly utilizes the effective components in the signal features to predict and reconstruct the features of the blank or noise parts in the input signal features, thereby making the signal features contain more useful components and avoiding the influence of blank noise components in the signal features on the classification and identification of communication radiation sources.

[0050] When training the feature reconstruction subnetwork described above, features are first extracted from radio signal samples to obtain their signal features, which are then input into the feature reconstruction subnetwork. During the signal feature reconstruction process, the input signal features are masked, and the subnetwork then predicts and reconstructs the masked portion. The input signal features are also used as training labels to verify the accuracy of the subnetwork's prediction and reconstruction of the masked portion. This process is repeated until the feature reconstruction subnetwork can accurately predict and reconstruct the signal features of the radio signal. At this point, the feature reconstruction subnetwork possesses the ability to accurately predict and reconstruct the signal features in the blank areas.

[0051] After training the feature reconstruction subnetwork described above, it is connected in series with the feature classification subnetwork to form a composite network that integrates feature reconstruction and feature classification. This network can then be used as a communication radiation source identification model for communication radiation source identification.

[0052] It is understandable that the training process of the feature reconstruction subnetwork of the above-mentioned communication radiation source identification model does not require communication radiation source labels corresponding to the radio signal training samples. It only needs to perform unsupervised feature reconstruction training on the feature reconstruction subnetwork based on the signal features of the radio signals to enable it to have good feature prediction and reconstruction capabilities. Thus, the above-mentioned communication radiation source identification model can achieve good feature prediction and reconstruction capabilities based on a small number of labeled samples. This allows it to comprehensively and accurately extract signal feature components that are beneficial to communication radiation source identification from the signal features of radio signals, thereby ensuring the effectiveness of communication radiation source identification.

[0053] Therefore, the communication radiation source identification method proposed in this application, by performing unsupervised feature reconstruction training on the feature reconstruction subnetwork of the communication radiation source identification model, enables the communication radiation source identification model to possess good feature prediction and reconstruction capabilities. In other words, through unsupervised feature reconstruction training, the communication radiation source identification model can accurately extract signal features beneficial for communication radiation source classification. This communication radiation source identification method significantly reduces the requirement for a large number of labeled samples, achieving the purpose of communication radiation source identification with a small number of labeled samples, thus enabling the method to handle a wider range of communication radiation source identification tasks.

[0054] In a preferred embodiment, when training the aforementioned communication radiation source identification model, this application first performs unsupervised feature reconstruction training on the feature reconstruction subnetwork. After the feature reconstruction subnetwork is trained, it is then concatenated with the feature classification subnetwork to form the communication radiation source identification model. Finally, the model is trained for communication radiation source classification based on labeled samples. These labeled samples are radio signal samples carrying communication radiation source labels.

[0055] Below, in conjunction with FIG. 3 The training process diagram shown illustrates the training process of the aforementioned communication radiation source identification model, including the training process of the feature reconstruction subnetwork and the training process of the feature classification subnetwork. Therefore, the relevant content regarding the training of the feature reconstruction subnetwork and the training of the feature classification subnetwork in other embodiments of this application can be found in the description of this embodiment.

[0056] See FIG. 3As shown, the training process of the aforementioned communication radiation source identification model mainly includes two stages: a model pre-training stage and a model fine-tuning stage. In the model pre-training stage, the feature reconstruction subnetwork of the communication radiation source identification model undergoes unsupervised feature reconstruction training to enable it to predict and reconstruct features. In the model fine-tuning stage, the feature classification subnetwork of the communication radiation source identification model undergoes communication radiation source classification training based on labeled samples to enable it to accurately classify communication radiation sources.

[0057] See FIG. 3 As shown, the feature reconstruction subnetworks mentioned above mainly include multi-layer convolutional neural networks (CNN), long short-term memory recurrent neural networks (LSTM), and feature prediction networks.

[0058] This application embodiment achieves unsupervised feature reconstruction training of the feature reconstruction subnetwork by randomly masking the signal features of radio signal samples and reconstructing the signal features of the radio signal samples based on the randomly masked signal features. The specific training process includes the following steps a) to i):

[0059] a) Perform a short-time Fourier transform (STFT) on the radio signals emitted by an individual communication radiation source to obtain STFT features of dimension d.

[0060] Since radio signals generally have a large bandwidth, this embodiment of the application randomly divides the STFT features of the radio signal into feature sub-segments f of dimension d0. n (n=1,2,...,N), where N is the number of feature sub-segments, N=d / d0.

[0061] b) The STFT feature sub-fragment f n The signal is fed into a multi-layer convolutional neural network (CNN) to extract the first signal feature z. t (t=1,2,...,T), where T is the number of feature frames.

[0062] c) Regarding the first signal feature z t The second signal feature is obtained by performing random masking on (t=1,2,...,T).

[0063] Specifically, the first signal feature z t Features z corresponding to certain consecutive frames in (t=1,2,...,T) t Set to 0 to obtain the second signal feature z after masking. t (t = 1, 2, ..., T). Generally, the maximum consecutive frame length for random masking is 5.

[0064] d) Extract the second signal feature z't Inputting the signal into a Long Short-Term Memory (LSTM) recurrent neural network yields the third signal feature c. t (t = 1, 2, ..., T);

[0065] e) The third signal feature c t (t=1,2,...,T) Input feature prediction network, using third signal feature c t Predict the first signal feature z t The reconstructed signal feature p corresponding to the feature sub-segment is obtained. t+k .

[0066] Specifically, the prediction result for the signal features corresponding to the k-th frame after the current t-th frame is p. t+k =W k c t W k This represents the predictive coding transform matrix for the k-th frame after the current t-th frame, which is used to reconstruct the first signal features.

[0067] f) The first signal feature z obtained according to step b) t The reconstructed signal features p obtained in step e) t+k Calculate the loss function.

[0068] First, calculate the predicted value p of the signal features in the k-th frame after the current t-th frame. t+k and the true value z t+k similarity between Where T represents the vector transpose.

[0069] Then, calculate the loss function. The numerator of this formula represents the signal characteristics c based on the current t-th frame. t The predicted signal features of the k-th frame after the current t-th frame and the corresponding signal features z of the positive sample are compared. t+k The similarity is expressed in the denominator as the similarity based on the signal features c of the current t-th frame. t The predicted signal features of the k-th frame after the current t-th frame and the corresponding signal features z of the positive sample are compared. t+k The similarity and signal features z corresponding to several other negative example samples. t+j The sum of similarities of (j=-j0,...0,...,j0).

[0070] g) Use the loss function L obtained in step f) to correct the parameters of the feature reconstruction subnetwork.

[0071] Specifically, using the loss function L calculated in step f), the error backpropagation (BP) algorithm is employed to update the model parameters.

[0072] h) Repeat steps b) to g) above for other STFT feature sub-segments to complete the prediction and reconstruction of STFT features in other frequency bands, and update the parameters of the feature reconstruction sub-network.

[0073] i) Repeat steps a) to h) to complete the feature extraction and prediction reconstruction of other radio signals. Through multiple iterations, the feature reconstruction subnetwork is trained.

[0074] After the above training, a feature reconstruction subnetwork can be obtained, which can output more accurate radio signal features through feature prediction and reconstruction.

[0075] Then, the trained feature reconstruction subnetwork is concatenated with the feature classification subnetwork, and supervised training for communication radiation source classification is performed on the feature classification subnetwork. This supervised training for communication radiation source classification involves using radio signal samples as training samples and the corresponding communication radiation source labels as training labels to perform communication radiation source classification training.

[0076] See FIG. 3 As shown, the training of the feature classification subnetwork described above can be achieved through steps j) to n):

[0077] j) Generate a radio signal based on the acquired radio signal sample to obtain an amplified radio signal corresponding to the acquired radio signal.

[0078] Due to the limited number of training sample data for radio signals emitted by labeled individual communication radiation sources, it is necessary to augment the existing small number of labeled radio signal samples. Specifically, the acquired radio signal samples are processed by increasing / decreasing the transmission rate, adding random noise, and simulating multipath fading to generate new radio signals, which are the augmented radio signals.

[0079] The above data augmentation process can increase the diversity of training data coverage scenarios, which is beneficial for improving network training performance and increasing the diversity of applicable scenarios for the network.

[0080] Since the amplified radio signals mentioned above are all generated based on the acquired radio signals, when these amplified radio signals are used as training samples, their corresponding sample labels are all the sample labels corresponding to the acquired radio signals.

[0081] Therefore, after performing the above data amplification process, the acquired radio signals and the amplified radio signals corresponding to the acquired radio signals can be used as training samples, and the communication radiation source labels corresponding to the acquired radio signals can be used as training labels to train the feature classification subnetwork for communication radiation source classification.

[0082] k) Based on the feature reconstruction subnetwork, extract the signal features of the acquired radio signal and the amplified radio signal corresponding to the acquired radio signal.

[0083] Specifically, referring to step a) above, the acquired radio signal and its corresponding amplified radio signal are subjected to Short Time Fourier Transform (STFT) to obtain STFT features of dimension d. Furthermore, the signal features are divided into feature sub-segments to obtain feature sub-segments f of dimension d0. n (n = 1, 2, ..., N).

[0084] Then, the feature sub-segments are input into the feature reconstruction sub-network trained above to obtain the predicted and reconstructed signal features c output by the network. t (t = 1, 2, ..., T).

[0085] l) Input the signal features of the acquired radio signal into the feature classification subnetwork, and input the signal features of the amplified radio signal corresponding to the acquired radio signal into the auxiliary classification subnetwork to obtain the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork.

[0086] See FIG. 3 As shown, the feature classification subnetwork mainly consists of a residual neural network (ResNet), fully connected layers, and a softmax layer. When the signal features of the acquired radio signal are input into the ResNet, they are processed through multiple layers of residual convolutional connections to obtain the representation vector w of the corresponding individual communication radiation source. i This vector can be considered as reflecting the unique identity information of an individual communication radiation source.

[0087] The above representation vector w i The representation vector w is obtained by using a fully connected layer and performing softmax to calculate the posterior probability q. i The corresponding individual classification label of the radiation source; based on the posterior probability q and the individual label of the communication radiation source corresponding to the input radio signal, calculate the loss function L0 = -log(q).

[0088] The auxiliary classification subnetwork is also mainly composed of residual neural networks (ResNet), fully connected layers, and softmax. It amplifies the radio signal corresponding to the acquired radio signal. The signal feature c is predicted and reconstructed by the feature reconstruction subnetwork. t Inputting (t = 1, 2, ..., T) into the auxiliary classification subnetwork yields its corresponding representation vector. in, These represent the amplified radio signals obtained by randomly altering (increasing / decreasing) the transmission rate of the acquired radio signal, adding random noise, and adding multipath fading processing, respectively.

[0089] Amplified radio signals Their respective representation vectors The representation vector is obtained by using a fully connected layer and performing softmax to calculate the posterior probabilities q1, q2, and q3. The corresponding individual classification label for radiation sources.

[0090] The dimension of the aforementioned radiation source individual classification labels is the same as the total number of radiation source individual categories. Each bit of the radiation source individual classification label output by the aforementioned feature classification subnetwork and auxiliary classification subnetwork represents the probability that the communication radiation source individual corresponding to the current radio signal belongs to the corresponding radiation source individual category.

[0091] For example, assuming there are 10 categories of all communication radiation sources, the radiation source individual classification labels output by the feature classification subnetwork and the auxiliary classification subnetwork above will have 10 bits. Each bit corresponds to a communication radiation source individual category, and the value of each bit represents the probability that the currently identified communication radiation source belongs to the communication radiation source individual category corresponding to that bit.

[0092] Based on the aforementioned individual classification labels for radiation sources, the individual category of the communication radiation source corresponding to the label with the highest probability value can be used as the communication radiation source identification result. Theoretically, in the individual classification labels of radiation sources output by a communication radiation source identification model with sufficiently high accuracy, the label with the highest value should be the label corresponding to the correct individual category of the communication radiation source.

[0093] m), calculate the loss function based at least on the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork, and the communication radiation source labels corresponding to the acquired radio signals.

[0094] Specifically, based on the posterior probability q output by the feature classification subnetwork and the individual label of the communication radiation source corresponding to the input radio signal, the first loss function L0 = -log(q) is calculated.

[0095] Simultaneously, based on the posterior probabilities q1, q2, and q3 output by the auxiliary classification subnetwork and the individual labels of the communication radiation sources corresponding to the input radio signals, the second loss functions L1 = -log(q1), L2 = -log(q2), and L3 = -log(q3) are calculated.

[0096] Furthermore, amplification of radio signals should not alter the representation vectors corresponding to the training samples. Therefore, to ensure the consistency of the representation vectors before and after data amplification, this embodiment utilizes the minimum mean square error constraint to determine the difference between the representation vectors output by the feature classification subnetwork and the auxiliary classification subnetwork, thus obtaining a third loss function, namely...

[0097] Finally, the first, second, and third loss functions are weighted to obtain the overall training loss function L = L0 + L1 + L2 + L3 + αL 01 +βL 02 +γL 03 , where α, β and γ represent preset weighting coefficients.

[0098] n) Based on the calculated loss function, the parameters of the feature classification subnetwork and the auxiliary classification subnetwork are adjusted.

[0099] Specifically, based on the overall training loss function L calculated in step m), the error backpropagation (BP) algorithm is used to update the parameters of the feature classification subnetwork and the auxiliary classification subnetwork.

[0100] Following the above training process, other feature sub-segments and other radio signal samples are input for multiple rounds of iterative training to obtain a communication radiation source identification model that can be used to identify individual communication radiation sources.

[0101] After the above training, a communication radiation source identification model is finally obtained, consisting of CNN, LSTM, feature prediction network, ResNet, fully connected layer, and softmax layer. Among them, CNN, LSTM, and feature prediction network are combined to form a feature reconstruction subnetwork, which is used to reconstruct the features of radio signals to obtain the reconstructed signal features; ResNet, fully connected layer, and softmax layer are combined to form a feature classification subnetwork, which is used to classify the signal features output by the feature reconstruction subnetwork to determine the individual identification result of the communication radiation source.

[0102] In reality, the signal bandwidth of radio signals emitted by communication radiation sources is generally large, and direct processing of them requires a large amount of computing resources and is slow. In order to improve the processing speed and the accuracy of communication radiation source identification, corresponding to the above training process, in the embodiments of this application, when using the communication radiation source identification model to identify communication radiation sources, after obtaining the signal features of the radio signal, the signal features of the radio signal are first divided into multiple signal sub-features.

[0103] For example, the STFT feature (d-dimensional) of a radio signal is segmented (each sub-segment has a dimension of d0) to obtain N feature sub-segments f. n (n = 1, 2, ..., N).

[0104] Then, each signal sub-feature is input into the communication radiation source identification model obtained through the above training to obtain the communication radiation source identification result corresponding to each signal sub-feature.

[0105] For example, each feature sub-fragment f n Input the communication radiation source identification model trained above to obtain the posterior probability score vector s corresponding to each feature sub-segment. n .

[0106] Finally, based on the communication radiation source identification results corresponding to each signal sub-feature, the communication radiation source identification results corresponding to the radio signal are determined.

[0107] As mentioned above, the communication radiation source identification model outputs a radiation source individual classification label containing multiple tag bits. The dimension of this radiation source individual classification label is the same as the preset number of communication radiation source individual categories. When each signal sub-feature is input into the above-mentioned communication radiation source identification model, a radiation source individual classification label is obtained from the model output. The radiation source individual category labels corresponding to each signal sub-feature are accumulated to obtain the total score vector.

[0108] Based on this, the communication radiation source individual category corresponding to the highest-scoring label bit is selected from the total score vector as the communication radiation source individual category to which the acquired radio signal belongs, thus obtaining the communication radiation source identification result corresponding to the acquired radio signal.

[0109] Exemplary apparatus

[0110] Accordingly, this application also provides a communication radiation source identification device, see [link to relevant documentation]. FIG. 4 As shown, the device includes:

[0111] Signal processing unit 001 is used to acquire the signal characteristics of radio signals;

[0112] The signal recognition unit 002 is used to input the signal features into a pre-trained communication radiation source recognition model to obtain the communication radiation source recognition result corresponding to the radio signal;

[0113] The communication radiation source identification model includes a feature reconstruction subnetwork and a feature classification subnetwork. The feature reconstruction subnetwork is obtained by using the signal features of radio signal samples as training samples and training labels respectively for feature reconstruction training.

[0114] As an optional implementation, the feature reconstruction subnetwork is trained before the feature classification subnetwork is trained;

[0115] The feature classification subnetwork is obtained by training the communication radiation source identification model, which includes the trained feature reconstruction subnetwork, to classify communication radiation sources.

[0116] As an optional implementation, the training process of the feature reconstruction sub-network includes:

[0117] By randomly masking the signal features of radio signal samples and enabling the feature reconstruction subnetwork to reconstruct the signal features of the radio signal samples based on the randomly masked signal features, unsupervised feature reconstruction training of the feature reconstruction subnetwork is achieved.

[0118] As an optional implementation, the feature reconstruction sub-network includes a multi-layer convolutional neural network, a long short-term memory recurrent neural network, and a feature prediction network;

[0119] The unsupervised feature reconstruction training of the feature reconstruction subnetwork, achieved by randomly masking the signal features of the radio signal samples and reconstructing the signal features of the radio signal samples based on the randomly masked signal features, includes:

[0120] The signal features of the radio signal sample are input into the multi-layer convolutional neural network of the feature reconstruction sub-network to obtain the first signal feature output by the multi-layer convolutional neural network;

[0121] The first signal feature is subjected to random masking to obtain the second signal feature;

[0122] The second signal feature is input into the long short-term memory recurrent neural network of the feature reconstruction subnetwork to obtain the third signal feature;

[0123] The third signal feature is input into the feature prediction network to obtain the reconstructed signal features corresponding to the radio signal sample;

[0124] Calculate the loss function based on the first signal features and the reconstructed signal features;

[0125] The parameters of the feature reconstruction subnetwork are corrected using the calculated loss function.

[0126] As an optional implementation, the feature classification subnetwork is obtained by using radio signal samples as training samples and the corresponding communication radiation source labels as training labels to perform communication radiation source classification training.

[0127] As an optional implementation, the training process of the feature classification sub-network includes:

[0128] A radio signal is generated based on the acquired radio signal sample, and an amplified radio signal corresponding to the acquired radio signal is obtained.

[0129] The acquired radio signals and their corresponding amplified radio signals are used as training samples, and the communication radiation source labels corresponding to the acquired radio signals are used as training labels to train the feature classification subnetwork for communication radiation source classification.

[0130] As an optional implementation, the step of using the acquired radio signal and the amplified radio signal corresponding to the acquired radio signal as training samples, and using the communication radiation source label corresponding to the acquired radio signal as the training label, to train the feature classification sub-network for communication radiation source classification includes:

[0131] Based on the aforementioned features, a sub-network is reconstructed to extract the signal features of the acquired radio signals and the amplified radio signals corresponding to the acquired radio signals.

[0132] The signal features of the acquired radio signal are input into the feature classification subnetwork, and the signal features of the amplified radio signal corresponding to the acquired radio signal are input into the auxiliary classification subnetwork to obtain the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork; wherein, the auxiliary classification subnetwork has the same structure as the feature classification subnetwork;

[0133] The loss function is calculated based at least on the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork, as well as the communication radiation source labels corresponding to the acquired radio signals;

[0134] Based on the calculated loss function, the parameters of the feature classification subnetwork and the auxiliary classification subnetwork are adjusted.

[0135] As an optional implementation, at least based on the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork, and the communication radiation source labels corresponding to the acquired radio signals, a loss function is calculated, including:

[0136] The first loss function is calculated based on the communication radiation source classification results output by the feature classification subnetwork and the communication radiation source labels corresponding to the acquired radio signals.

[0137] The second loss function is calculated based on the communication radiation source classification results output by the auxiliary classification sub-network and the communication radiation source labels corresponding to the acquired radio signals.

[0138] The third loss function is calculated based on the mean square error between the signal features output by the fully connected layer of the feature classification subnetwork and the signal features output by the fully connected layer of the auxiliary classification subnetwork.

[0139] As an optional implementation, the signal features are input into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to the radio signal, including:

[0140] The signal features are divided into multiple signal sub-features;

[0141] Each signal sub-feature is input into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to each signal sub-feature;

[0142] Based on the communication radiation source identification results corresponding to each signal sub-feature, the communication radiation source identification results corresponding to the radio signal are determined.

[0143] The communication radiation source identification device provided in this embodiment belongs to the same concept as the communication radiation source identification method provided in the above embodiments of this application. It can execute the communication radiation source identification method provided in any of the above embodiments of this application and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in this embodiment can be found in the specific processing content of the communication radiation source identification method provided in the above embodiments of this application, and will not be repeated here.

[0144] Exemplary electronic device

[0145] This application also proposes a communication radiation source identification device, see [link to relevant documentation]. FIG. 5 As shown, the device includes:

[0146] The signal acquisition unit 011, the signal encoder 012 connected to the signal acquisition unit, and the processor 013 connected to the signal encoder;

[0147] The signal acquisition device 011 is used to acquire radio signals;

[0148] The signal encoder 012 is used to extract signal features from the radio signals collected by the signal collector to obtain the signal features of the radio signals.

[0149] The processor 013 is used to input the signal features output by the signal encoder into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to the radio signal; wherein, the communication radiation source identification model includes a feature reconstruction subnetwork and a feature classification subnetwork, and the feature reconstruction subnetwork is obtained by performing feature reconstruction training by using the signal features of the radio signal sample as training samples and training labels respectively.

[0150] As an optional implementation, the feature reconstruction subnetwork is trained before the feature classification subnetwork is trained;

[0151] The feature classification subnetwork is obtained by training the communication radiation source identification model, which includes the trained feature reconstruction subnetwork, to classify communication radiation sources.

[0152] As an optional implementation, the training process of the feature reconstruction sub-network includes:

[0153] By randomly masking the signal features of radio signal samples and enabling the feature reconstruction subnetwork to reconstruct the signal features of the radio signal samples based on the randomly masked signal features, unsupervised feature reconstruction training of the feature reconstruction subnetwork is achieved.

[0154] As an optional implementation, the feature reconstruction sub-network includes a multi-layer convolutional neural network, a long short-term memory recurrent neural network, and a feature prediction network;

[0155] The unsupervised feature reconstruction training of the feature reconstruction subnetwork, achieved by randomly masking the signal features of the radio signal samples and reconstructing the signal features of the radio signal samples based on the randomly masked signal features, includes:

[0156] The signal features of the radio signal sample are input into the multi-layer convolutional neural network of the feature reconstruction sub-network to obtain the first signal feature output by the multi-layer convolutional neural network;

[0157] The first signal feature is subjected to random masking to obtain the second signal feature;

[0158] The second signal feature is input into the long short-term memory recurrent neural network of the feature reconstruction subnetwork to obtain the third signal feature;

[0159] The third signal feature is input into the feature prediction network to obtain the reconstructed signal features corresponding to the radio signal sample;

[0160] Calculate the loss function based on the first signal features and the reconstructed signal features;

[0161] The parameters of the feature reconstruction subnetwork are corrected using the calculated loss function.

[0162] As an optional implementation, the feature classification subnetwork is obtained by using radio signal samples as training samples and the corresponding communication radiation source labels as training labels to perform communication radiation source classification training.

[0163] As an optional implementation, the training process of the feature classification sub-network includes:

[0164] A radio signal is generated based on the acquired radio signal sample, and an amplified radio signal corresponding to the acquired radio signal is obtained.

[0165] The acquired radio signals and their corresponding amplified radio signals are used as training samples, and the communication radiation source labels corresponding to the acquired radio signals are used as training labels to train the feature classification subnetwork for communication radiation source classification.

[0166] As an optional implementation, the step of using the acquired radio signal and the amplified radio signal corresponding to the acquired radio signal as training samples, and using the communication radiation source label corresponding to the acquired radio signal as the training label, to train the feature classification sub-network for communication radiation source classification includes:

[0167] Based on the aforementioned features, a sub-network is reconstructed to extract the signal features of the acquired radio signals and the amplified radio signals corresponding to the acquired radio signals.

[0168] The signal features of the acquired radio signal are input into the feature classification subnetwork, and the signal features of the amplified radio signal corresponding to the acquired radio signal are input into the auxiliary classification subnetwork to obtain the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork; wherein, the auxiliary classification subnetwork has the same structure as the feature classification subnetwork;

[0169] The loss function is calculated based at least on the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork, as well as the communication radiation source labels corresponding to the acquired radio signals;

[0170] Based on the calculated loss function, the parameters of the feature classification subnetwork and the auxiliary classification subnetwork are adjusted.

[0171] As an optional implementation, at least based on the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork, and the communication radiation source labels corresponding to the acquired radio signals, a loss function is calculated, including:

[0172] The first loss function is calculated based on the communication radiation source classification results output by the feature classification subnetwork and the communication radiation source labels corresponding to the acquired radio signals.

[0173] The second loss function is calculated based on the communication radiation source classification results output by the auxiliary classification sub-network and the communication radiation source labels corresponding to the acquired radio signals.

[0174] The third loss function is calculated based on the mean square error between the signal features output by the fully connected layer of the feature classification subnetwork and the signal features output by the fully connected layer of the auxiliary classification subnetwork.

[0175] As an optional implementation, the signal features are input into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to the radio signal, including:

[0176] The signal features are divided into multiple signal sub-features;

[0177] Each signal sub-feature is input into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to each signal sub-feature;

[0178] Based on the communication radiation source identification results corresponding to each signal sub-feature, the communication radiation source identification results corresponding to the radio signal are determined.

[0179] The communication radiation source identification device provided in this embodiment belongs to the same concept as the communication radiation source identification method provided in the above embodiments of this application. It can execute the communication radiation source identification method provided in any of the above embodiments of this application and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in this embodiment can be found in the specific processing content of the communication radiation source identification method provided in the above embodiments of this application, and will not be repeated here.

[0180] Another embodiment of this application also proposes another communication radiation source identification device, see [link to relevant documentation] FIG. 6 As shown, the device includes:

[0181] Memory 200 and processor 210;

[0182] The memory 200 is connected to the processor 210 and is used to store programs;

[0183] The processor 210 is used to implement the communication radiation source identification method disclosed in any of the above embodiments by running the program stored in the memory 200.

[0184] Specifically, the aforementioned communication radiation source identification device may also include: a bus, a communication interface 220, an input device 230, and an output device 240.

[0185] The processor 210, memory 200, communication interface 220, input device 230, and output device 240 are interconnected via a bus. Among them:

[0186] A bus can include a pathway for transmitting information between various components of a computer system.

[0187] The processor 210 can be a general-purpose processor, such as a general-purpose central processing unit (CPU), a microprocessor, etc., or an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control the execution of the program of the present invention. It can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0188] Processor 210 may include a main processor, as well as a baseband chip, modem, etc.

[0189] The memory 200 stores a program that executes the technical solution of this invention, and may also store an operating system and other key business functions. Specifically, the program may include program code, which includes computer operation instructions. More specifically, the memory 200 may include read-only memory (ROM), other types of static storage devices capable of storing static information and instructions, random access memory (RAM), other types of dynamic storage devices capable of storing information and instructions, disk storage, flash memory, etc.

[0190] Input device 230 may include a device for receiving user input data and information, such as a keyboard, mouse, camera, scanner, light pen, voice input device, touch screen, pedometer, or gravity sensor.

[0191] Output device 240 may include devices that allow information to be output to a user, such as a display screen, printer, speaker, etc.

[0192] The communication interface 220 may include a device that uses any transceiver to communicate with other devices or communication networks, such as Ethernet, Radio Access Network (RAN), Wireless Local Area Network (WLAN), etc.

[0193] The processor 210 executes the program stored in the memory 200 and calls other devices, which can be used to implement the various steps of any of the communication radiation source identification methods provided in the above embodiments of this application.

[0194] Exemplary computer program product and storage medium

[0195] In addition to the methods and devices described above, embodiments of this application may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the communication radiation source identification method described in the "Exemplary Methods" section of this specification.

[0196] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this application. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone 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.

[0197] Furthermore, embodiments of this application may also be storage media storing a computer program, which is executed by a processor in the steps of the communication radiation source identification method described in the "Exemplary Methods" section of this specification.

[0198] For the foregoing method embodiments, in order to simplify the description, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0199] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0200] The steps in the methods of the various embodiments of this application can be adjusted, merged, or deleted in order according to actual needs, and the technical features described in each embodiment can be replaced or combined.

[0201] The modules and sub-modules in the various embodiments of the present application's devices and terminals can be merged, divided, and deleted according to actual needs.

[0202] It should be understood that the disclosed terminals, devices, and methods can be implemented in other ways, given the several embodiments provided in this application. For example, the terminal embodiments described above are merely illustrative. For instance, the division of modules or sub-modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple sub-modules or modules may be combined or integrated into another module, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0203] The modules or submodules described as separate components may or may not be physically separate. The components that constitute a module or submodule may or may not be physical modules or submodules; that is, they may be located in one place or distributed across multiple network modules or submodules. Some or all of the modules or submodules can be selected to achieve the purpose of this embodiment's solution, depending on actual needs.

[0204] Furthermore, the functional modules or sub-modules in the various embodiments of this application can be integrated into one processing module, or each module or sub-module can exist physically separately, or two or more modules or sub-modules can be integrated into one module. The integrated modules or sub-modules described above can be implemented in hardware or in the form of software functional modules or sub-modules.

[0205] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0206] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software unit executed by a processor, or a combination of both. The software unit can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0207] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0208] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying communication radiation sources, characterized in that, include: Acquire the signal characteristics of radio signals; The signal features are input into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to the radio signal; The communication radiation source identification model includes a feature reconstruction subnetwork and a feature classification subnetwork. The feature reconstruction subnetwork is trained by using the signal features of radio signal samples as training samples and training labels, respectively. During the training of the feature classification subnetwork, the feature classification subnetwork classifies the acquired radio signals as communication radiation sources, and the auxiliary classification subnetwork classifies the amplified radio signals as communication radiation sources. The parameters of the feature classification subnetwork and the auxiliary classification subnetwork are updated based on a first loss function, a second loss function, and a third loss function. The first loss function is calculated based on the communication radiation source classification result output by the feature classification subnetwork and the communication radiation source label corresponding to the acquired radio signal. The second loss function is calculated based on the communication radiation source classification result output by the auxiliary classification subnetwork and the communication radiation source label. The third loss function is calculated based on the mean square error between the signal features output by the fully connected layer of the feature classification subnetwork and the signal features output by the fully connected layer of the auxiliary classification subnetwork. The amplified radio signal is generated based on the acquired radio signal, and the auxiliary classification subnetwork has the same structure as the feature classification subnetwork.

2. The method according to claim 1, characterized in that, The feature reconstruction subnetwork is trained before the feature classification subnetwork is trained. The feature classification subnetwork is obtained by training the communication radiation source identification model, which includes the trained feature reconstruction subnetwork, to classify communication radiation sources.

3. The method according to claim 1 or 2, characterized in that, The training process of the feature reconstruction subnetwork includes: By randomly masking the signal features of radio signal samples and enabling the feature reconstruction subnetwork to reconstruct the signal features of the radio signal samples based on the randomly masked signal features, unsupervised feature reconstruction training of the feature reconstruction subnetwork is achieved.

4. The method according to claim 3, characterized in that, The feature reconstruction subnetwork includes a multi-layer convolutional neural network, a long short-term memory recurrent neural network, and a feature prediction network; The unsupervised feature reconstruction training of the feature reconstruction subnetwork, achieved by randomly masking the signal features of the radio signal samples and reconstructing the signal features of the radio signal samples based on the randomly masked signal features, includes: The signal features of the radio signal sample are input into the multi-layer convolutional neural network of the feature reconstruction sub-network to obtain the first signal feature output by the multi-layer convolutional neural network; The first signal feature is subjected to random masking to obtain the second signal feature; The second signal feature is input into the long short-term memory recurrent neural network of the feature reconstruction subnetwork to obtain the third signal feature; The third signal feature is input into the feature prediction network to obtain the reconstructed signal features corresponding to the radio signal sample; Calculate the loss function based on the first signal features and the reconstructed signal features; The parameters of the feature reconstruction subnetwork are corrected using the calculated loss function.

5. The method according to claim 1 or 2, characterized in that, The training process of the feature classification subnetwork includes: A radio signal is generated based on the acquired radio signal sample, and an amplified radio signal corresponding to the acquired radio signal is obtained. The acquired radio signals and their corresponding amplified radio signals are used as training samples, and the communication radiation source labels corresponding to the acquired radio signals are used as training labels to train the feature classification subnetwork for communication radiation source classification.

6. The method according to claim 5, characterized in that, The step of using the acquired radio signals and the amplified radio signals corresponding to the acquired radio signals as training samples, and using the communication radiation source labels corresponding to the acquired radio signals as training labels, to train the feature classification sub-network for communication radiation source classification includes: Based on the aforementioned features, a sub-network is reconstructed to extract the signal features of the acquired radio signals and the amplified radio signals corresponding to the acquired radio signals. The signal features of the acquired radio signal are input into the feature classification subnetwork, and the signal features of the amplified radio signal corresponding to the acquired radio signal are input into the auxiliary classification subnetwork to obtain the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork. The loss function is calculated based at least on the communication radiation source classification results output by the feature classification subnetwork and the auxiliary classification subnetwork, as well as the communication radiation source labels corresponding to the acquired radio signals; Based on the calculated loss function, the parameters of the feature classification subnetwork and the auxiliary classification subnetwork are adjusted.

7. The method according to claim 1, characterized in that, The signal features are input into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to the radio signal, including: The signal features are divided into multiple signal sub-features; Each signal sub-feature is input into a pre-trained communication radiation source identification model to obtain the communication radiation source identification result corresponding to each signal sub-feature; Based on the communication radiation source identification results corresponding to each signal sub-feature, the communication radiation source identification results corresponding to the radio signal are determined.

8. A communication radiation source identification device, characterized in that, include: The signal processing unit is used to acquire the signal characteristics of radio signals; The signal recognition unit is used to input the signal features into a pre-trained communication radiation source recognition model to obtain the communication radiation source recognition result corresponding to the radio signal; The communication radiation source identification model includes a feature reconstruction subnetwork and a feature classification subnetwork. The feature reconstruction subnetwork is trained by using the signal features of radio signal samples as training samples and training labels, respectively. During the training of the feature classification subnetwork, the feature classification subnetwork classifies the acquired radio signals as communication radiation sources, and the auxiliary classification subnetwork classifies the amplified radio signals as communication radiation sources. The parameters of the feature classification subnetwork and the auxiliary classification subnetwork are updated based on a first loss function, a second loss function, and a third loss function. The first loss function is calculated based on the communication radiation source classification result output by the feature classification subnetwork and the communication radiation source label corresponding to the acquired radio signal. The second loss function is calculated based on the communication radiation source classification result output by the auxiliary classification subnetwork and the communication radiation source label. The third loss function is calculated based on the mean square error between the signal features output by the fully connected layer of the feature classification subnetwork and the signal features output by the fully connected layer of the auxiliary classification subnetwork. The amplified radio signal is generated based on the acquired radio signal, and the auxiliary classification subnetwork has the same structure as the feature classification subnetwork.

9. A communication radiation source identification device, characterized in that, include: A signal acquisition unit, a signal encoder connected to the signal acquisition unit, and a processor connected to the signal encoder; The signal acquisition device is used to acquire radio signals; The signal encoder is used to extract signal features from the radio signals collected by the signal collector to obtain the signal features of the radio signals. The processor is configured to input the signal features output by the signal encoder into a pre-trained communication radiation source identification model to obtain a communication radiation source identification result corresponding to the radio signal; wherein, the communication radiation source identification model includes a feature reconstruction subnetwork and a feature classification subnetwork, the feature reconstruction subnetwork being trained by using the signal features of the radio signal samples as training samples and training labels respectively; during the training process of the feature classification subnetwork, the feature classification subnetwork performs communication radiation source classification on the acquired radio signal, and the auxiliary classification subnetwork performs communication radiation source classification on the amplified radio signal, based on a first loss function, a second loss function, and a third loss function. The parameters of the feature classification subnetwork and the auxiliary classification subnetwork are updated. The first loss function is calculated based on the communication radiation source classification result output by the feature classification subnetwork and the communication radiation source label corresponding to the acquired radio signal. The second loss function is calculated based on the communication radiation source classification result output by the auxiliary classification subnetwork and the communication radiation source label. The third loss function is calculated based on the mean square error between the signal features output by the fully connected layer of the feature classification subnetwork and the signal features output by the fully connected layer of the auxiliary classification subnetwork. The amplified radio signal is generated based on the acquired radio signal. The auxiliary classification subnetwork has the same structure as the feature classification subnetwork.

10. A communication radiation source identification device, characterized in that, include: Memory and processor; The memory is connected to the processor and is used to store programs; The processor is configured to implement the communication radiation source identification method as described in any one of claims 1 to 7 by running a program in the memory.

11. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the communication radiation source identification method as described in any one of claims 1 to 7.

Images