Communication radiation source identification method and related device

Abstract

The present application provides a communication radiation source identification method and related devices, the communication radiation source identification method comprising: receiving a target signal, and obtaining a signal to be identified from the target signal; extracting features of the signal to be identified to obtain a CPC feature vector to be identified; decomposing the CPC feature vector to be identified using a preset noise vector and a preset content vector to obtain an eigenchannel vector; obtaining the communication link where the target signal is located, and the communication radiation sources located at both ends of the communication link according to the eigenchannel vector. Through the above method, the amount of training samples required by the present application is small, and the accuracy of communication radiation source identification can be improved.

Description

Technical Field

[0001] The present application belongs to the technical field of communication radiation source identification, and specifically relates to a communication radiation source identification method and related devices. Background Art

[0002] Communication radiation source individual identification technology is one of the core technologies in electronic reconnaissance systems and is also a cutting-edge direction in the field of electronic warfare at home and abroad. It is generally achieved by deploying a reconnaissance device to capture radio signals propagating in the air, and then identifying the characteristics of the captured radio signals to determine the corresponding communication radiation source. The accuracy of the communication radiation source identification results plays a key role in analyzing the communication network structure, determining the threat level and making tactical decisions.

[0003] Traditional communication radiation source identification methods require the collection of a large amount of data to analyze the subtle characteristics of the transistors of the communication radiation source. In the face of complex and changing electromagnetic environments, this method has the problem of low accuracy in communication radiation source identification. With the development of artificial intelligence, the use of deep learning for communication radiation source identification has made new breakthroughs and progress, but the existing technology still has shortcomings. Summary of the invention

[0004] The present application provides a communication radiation source identification method and related devices to improve the accuracy of communication radiation source identification.

[0005] In order to solve the above technical problems, a technical solution adopted in the present application is: to provide a communication radiation source identification method, including: receiving a target signal, and obtaining a signal to be identified from the target signal; performing feature extraction on the signal to be identified to obtain a CPC feature vector to be identified; decomposing the CPC feature vector to be identified using a preset noise vector and a preset content vector to obtain an eigenchannel vector; and obtaining the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link according to the eigenchannel vector.

[0006] To solve the above technical problems, another technical solution adopted in the present application is: to provide a communication radiation source identification device, including: a receiving module, used to receive a target signal and obtain a signal to be identified from the target signal; a feature extraction module, connected to the receiving module, used to extract features from the signal to be identified to obtain a CPC feature vector to be identified; a decomposition module, connected to the feature extraction module, used to decompose the CPC feature vector to be identified using a preset noise vector and a preset content vector to obtain an intrinsic channel vector; an acquisition module, connected to the decomposition module, used to obtain the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link according to the intrinsic channel vector.

[0007] To solve the above technical problems, another technical solution adopted in the present application is: to provide an electronic device, comprising a memory and a processor coupled to each other, wherein the memory stores program instructions, and the processor is used to execute the program instructions to implement the communication radiation source identification method described in any of the above embodiments.

[0008] In order to solve the above technical problems, another technical solution adopted by the present application is: providing a storage device storing program instructions that can be executed by a processor, wherein the program instructions are used to implement the communication radiation source identification method described in any of the above embodiments.

[0009] Different from the prior art, the beneficial effects of the present application are as follows: the communication radiation source identification method provided by the present application will extract features from the signal to be identified to obtain the CPC feature vector to be identified; then the CPC feature vector to be identified is decomposed using a preset noise vector and a preset content vector to obtain an intrinsic channel vector; finally, the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link are obtained according to the intrinsic channel vector. In the above-mentioned design method, a communication link identification technology based on joint factor analysis is adopted, and the individual data at both ends of the communication link are merged to obtain the CPC feature vector to be identified, and then the CPC feature vector to be identified is hierarchically decoupled to obtain the intrinsic channel characteristics, and the communication radiation source is determined by the intrinsic channel characteristics; compared with the method of parsing the individual fingerprint characteristics of the communication radiation source, this method requires a smaller amount of training samples and can improve the accuracy of communication radiation source identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work, among which:

[0011] Figure 1 This is a flow chart of an implementation method of a communication radiation source identification method of the present application;

[0012] Figure 2 It is a structural diagram of an implementation method of CPC network framework;

[0013] Figure 3 It is a schematic diagram of signal decomposition;

[0014] Figure 4 for Figure 1 A schematic diagram of a flow chart of an implementation method corresponding to step S104;

[0015] Figure 5 for Figure 4A schematic diagram of a flow chart of an implementation method corresponding to step S203;

[0016] Figure 6 Schematic diagram of a structure of an implementation method of a residual block in a residual network;

[0017] Figure 7 It is a structural schematic diagram of an implementation method of a communication link set;

[0018] Figure 8 This is a schematic structural diagram of an embodiment of a communication radiation source identification device of the present application;

[0019] Fig. 9 This is a schematic diagram of the structure of an embodiment of an electronic device of the present application;

[0020] Fig.10 This is a schematic structural diagram of an implementation of a storage device of the present application. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0022] In the prior art, individual identification is mainly achieved by capturing and analyzing the transient signal characteristics when the communication radiation source is turned on. In the initial establishment stage of the communication radiation source signal transmission, the internal circuit switch conversion or mode conversion moment contains rich nonlinear and non-stationary characteristics, but the transient signal noise ratio is low, and the existence time is short and difficult to capture. Compared with the transient signal characteristics, it is more difficult to extract the subtle features of the steady-state signal; because in the stable working state, the individual differences of the communication radiation source are synthesized on the signal, and it is difficult to mathematically model the impact of all internal differences on the signal. The frequency source noise of the communication radiation source will cause the frequency or phase of the spectrum signal to fluctuate randomly. The carrier frequency deviation caused by the fluctuation of this frequency source can be used as the individual fingerprint feature of the communication radiation source, and the individual fingerprint feature can be extracted by time domain, spectrum and time-frequency analysis methods. However, the above-mentioned prior art has the following disadvantages: the processor can only process the data before demodulation such as radio frequency and intermediate frequency signals, and cannot process the demodulated audio signals and text signals, so it is often impossible to really apply it due to the limitations of actual combat equipment. In addition, in an actual combat environment, it is difficult to collect a large amount of sample data, so there is a problem of low recognition rate.

[0023] To solve the above technical problems, please refer to Figure 1 , Figure 1This is a flow chart of an embodiment of a communication radiation source identification method of the present application, and the identification method specifically includes:

[0024] S101: Receive a target signal, and obtain a signal to be identified from the target signal.

[0025] Specifically, in this embodiment, the target signal emitted by the communication radiation source can be received by the broadband receiver; generally speaking, the communication radiation source can be a radio station, etc.; the broadband receiver can receive all target signals within its operating frequency range, and at this time, the broadband receiver can send each target signal to the processor to execute Figure 1 Further, the implementation process of the above step S101 includes:

[0026] A. In response to the target signal being a non-encrypted signal, an audio signal and / or a text signal is obtained from the target signal and used as the signal to be identified. Specifically, the target signal received by the broadband receiver is a radio frequency signal. After the radio frequency signal enters the receiving antenna, in order to enable the amplifier to work stably and reduce interference, the radio frequency signal is converted into an intermediate frequency signal, and then the intermediate frequency signal is converted into a baseband (I / Q signal); wherein the baseband signal can be understood as a complex variable signal of I+Qj, where I represents in-phase, Q represents orthogonal, Q and I have a phase difference of 90 degrees, and I is called the real part and J is called the imaginary part in the complex variable function. Finally, the real part information is extracted from the baseband signal and sampled into an audio signal and / or a text signal, thereby obtaining the signal to be identified.

[0027] B. In response to the target signal being an encrypted signal, an intermediate frequency signal is obtained from the target signal and used as the signal to be identified. Specifically, since the target signal is an encrypted signal, it is impossible to obtain an audio signal and / or a text signal from the target signal; in this case, the intermediate frequency signal can be directly used as the signal to be identified.

[0028] That is, in this embodiment, the signal to be identified can be an intermediate frequency signal before demodulation, or an audio signal and / or a text signal after demodulation. The communication radiation source identification method provided by the present application has a wider application range.

[0029] S102: Extract features of the signal to be identified to obtain a CPC feature vector to be identified.

[0030] Specifically, in order to make the CPC feature vector extracted in step S102 more accurate, the signal to be identified may be preprocessed before step S102. The specific preprocessing process may include denoising and / or obtaining a signal whose frequency exceeds a threshold from the signal to be identified.

[0031] For example, in response to the signal to be recognized including an audio signal, the preprocessing process may include: denoising the signal to be recognized, and obtaining a signal whose frequency exceeds a threshold from the denoised signal to be recognized; in the above manner, a real speech segment that is non-noise and non-silent can be obtained from the signal to be recognized, and the result of subsequent feature extraction using the real speech segment is more accurate. Of course, in other embodiments, in response to the signal to be recognized including an audio signal, a signal whose frequency exceeds a threshold can also be directly obtained from the signal to be recognized. For another example, in response to the signal to be recognized including an intermediate frequency signal, the signal to be recognized is denoised.

[0032] Optionally, in this embodiment, the denoising process may be implemented by spectrum shifting and low-pass filtering; the process of obtaining a signal whose frequency exceeds a threshold from the signal to be identified may be implemented by voice activity detection (VAD) and other technologies.

[0033] Furthermore, in this embodiment, unsupervised representation learning technology can be used to extract unsupervised prediction feature vectors from high-dimensional communication radiation source signals (intermediate frequency signals, audio signals (text signals)), that is, comparative prediction coding CPC feature vectors. CPC feature learning is a global "slow-varying" feature, which is more robust than traditional features and does not require artificial prior information. The CPC feature vector can be output by directly inputting the signal to be identified. Please refer to Figure 2 , Figure 2 The specific implementation process of the above step S102 can be: the signal to be identified is divided into multiple sub-signals according to the time sequence, and the multiple sub-signals are respectively passed through the nonlinear encoder genc and the autoregressive model gar on the corresponding branch chain to generate the context representation c related to the sub-signal t , the context representation c corresponding to the multi-segment sub-signal t The CPC feature vector to be identified is formed by concatenation.

[0034] For example, Figure 2 The CPC network framework contains four branches, each of which contains a nonlinear encoder genc and an autoregressive model gar. At this time, the signal to be recognized can be divided into four sub-signals according to the time sequence, and each sub-signal is processed by the corresponding branch to generate a context representation c related to the sub-signal. t , the four context representations c corresponding to the four sub-signals t The CPC feature vector to be identified is formed by concatenation.

[0035] It should be noted that the various parameters in the CPC network framework will not change during the application process. Before application, the training data set can be used to train the various parameters in the CPC network framework. For example, the training data set includes an intermediate frequency signal, an audio signal, and a text signal; each training data in the training data set is divided into multiple sub-signals according to the time sequence, and the sub-signals with the earlier time sequence are used as the input of the CPC network framework, and the sub-signals with the later time sequence are used as the corresponding labels. During the training process, the sub-signals with the earlier time sequence are input into the CPC network framework to generate the corresponding context representation c t Then, we use the context to represent c t Obtain the predicted value at the later time series, and update the various parameters in the CPC network framework according to the predicted value and the corresponding label according to the noise-contrastive estimation loss function.

[0036] S103: Decompose the CPC feature vector to be identified by using a preset noise vector and a preset content vector to obtain an eigenchannel vector.

[0037] Specifically, in the communication radiation source individual identification system, the basic transmission process is that the transmitter transmits the radio frequency signal, and the radio frequency signal propagates through the atmospheric medium, is reflected by the ionosphere, and is sent to the receiver. Among them, the channel information transmitted through the atmosphere can be used as the core parameter to distinguish different communication links, that is, in this embodiment, the intrinsic channel characteristics can be used to identify the communication link. Please refer to Figure 3 , Figure 3 Schematic diagram of signal decomposition. From the perspective of signal analysis, the signal can be artificially decomposed into content part, channel part and noise part. That is, the CPC feature vector μ to be identified corresponding to the signal to be identified is artificially defined to be composed of content vector Vy(s), channel vector Ux(s) and noise vector dz(s). The preset content vector Vy(s) and the preset noise vector dz(s) in the above step S103 can be obtained in advance by Gaussian distribution fitting.

[0038] In one embodiment, before the above step S103, a process of determining a preset noise vector dz(s) is included, which specifically includes:

[0039] A. Receive multiple noise data collected from at least one communication link in a database.

[0040] Specifically, it is assumed that the database used in the process of obtaining the preset noise vector dz(s) contains multiple communication radiation sources, and it is known in advance that the multiple communication radiation sources in the database can form multiple communication links. At this time, the original noise data emitted by the transmitter on each communication link can be a pre-recorded shortwave noise, and the listening device (for example, a broadband receiver) can collect noise data for more than 1 hour for multiple communication links. Generally speaking, the more noise data is collected and the wider the types covered, the better the effect.

[0041] For example, suppose that the database used in the process of obtaining the preset noise vector dz(s) contains six communication radiation sources A, B, C, D, E, and F, and Figure 7 As shown in , it is known in advance that A and D form link 1, B and E form link 2, and C and F form link 3. One of A and D can be a transmitter and the other can be a receiver; one of B and E can be a transmitter and the other can be a receiver; one of C and F can be a transmitter and the other can be a receiver. The original noise data emitted by the communication radiation source at the transmitter can be a pre-recorded shortwave noise. At this time, the listening device (for example, a broadband receiver) can collect noise data for more than 1 hour for link 1, link 2, and link 3 respectively.

[0042] B. Obtain the noise CPC feature vector corresponding to each noise data.

[0043] Specifically, in this embodiment, the process of obtaining the noise CPC feature vector corresponding to each noise data may include: for each noise data, obtaining a noise signal to be identified from each noise data (the specific process is similar to that in step S101 and will not be repeated here); preprocessing the noise signal to be identified (the specific preprocessing process is similar to that in the above embodiment and will not be repeated here); performing feature extraction on the preprocessed noise signal to be identified to obtain a noise CPC feature vector (the specific process is similar to that in the above step S102 and will not be repeated here).

[0044] C. Decompose multiple noise CPC feature vectors to obtain corresponding noise vectors, perform Gaussian fitting on all noise vectors, and use the noise vector with the highest probability value as the preset noise vector.

[0045] Specifically, in the process of decomposing the noise CPC feature vector, the content vector Vy(s) can be regarded as nothing; and since the communication link is known in advance, the channel vector Ux(s) corresponding to each communication link can be regarded as known, so the noise vector corresponding to each noise CPC feature vector can be obtained through the decomposition process.

[0046] In addition, the probability density function formula of the Gaussian mixture model used in the above step C is as follows:

[0047]

[0048] Among them, μ i is the mean vector of the input samples, ∑ i is the covariance matrix of the input sample. x is the input sample vector, and the dimension is n. The dimension n is generally set artificially according to the size of the training data, and the dimension n is the same as the dimension of the CPC feature vector to be identified. For example, n is greater than 128, and n is 256. In this embodiment, multiple noise vectors can be used as input sample vectors for Gaussian fitting according to the above-mentioned probability density function formula, and the noise vector with the highest probability value is used as the preset noise vector. At this time, the preset noise vector is also an n-dimensional vector, and the preset noise vector can be treated as a fixed constant vector.

[0049] In yet another embodiment, before the above step S103, a process of determining a preset content vector Vy(s) is included, which specifically includes:

[0050] A. Receive multiple content data collected by at least one communication radiation source.

[0051] Specifically, it is assumed that the database used in the process of obtaining the preset content vector Vy(s) contains multiple communication radiation sources, and it is known in advance that the multiple communication radiation sources in the database can form multiple communication links with each other. At this time, the original content data sent by the communication radiation source at the transmitting end can be pre-recorded data with clear content, and the listening device (for example, a broadband receiver) can collect content data for more than 1 hour for multiple links respectively. Generally speaking, the more content data is collected and the wider the types covered, the better the effect.

[0052] For example, suppose that the database used in the process of obtaining the preset content vector Vy(s) contains six communication radiation sources A, B, C, D, E, and F, and Figure 7 As shown in , it is known in advance that A and D form link 1, B and E form link 2, and C and F form link 3. One of A and D can be a transmitter and the other can be a receiver; one of B and E can be a transmitter and the other can be a receiver; one of C and F can be a transmitter and the other can be a receiver. The original content data emitted by the communication radiation source at the transmitter can be pre-recorded data with clear content. At this time, the listening device (for example, a broadband receiver) can collect content data of more than 1 hour for link 1, link 2, and link 3 respectively.

[0053] B. Obtain the content CPC feature vector corresponding to each content data.

[0054] Specifically, in this embodiment, the process of obtaining the content CPC feature vector corresponding to each content data may include: for each content data, obtaining a content signal to be identified from each content data (the specific process is similar to that in step S101 and will not be repeated here); preprocessing the content signal to be identified (the specific preprocessing process is similar to that in the above embodiment and will not be repeated here); performing feature extraction on the preprocessed content signal to be identified to obtain a content CPC feature vector (the specific process is similar to that in the above step S102 and will not be repeated here).

[0055] C. Perform vector decomposition on multiple content CPC feature vectors to obtain corresponding content vectors, perform Gaussian fitting on all content vectors, and use the content vector with the highest probability value as the preset content vector.

[0056] Specifically, in the process of decomposing the content CPC feature vector, the noise vector dz(s) can be regarded as non-existent; and since the communication link is known in advance, the channel vector Ux(s) corresponding to each communication link can be regarded as known, so the content vector corresponding to each content CPC feature vector can be obtained through the decomposition process.

[0057] In addition, the probability density function formula of the Gaussian mixture model used in the above step C is as follows:

[0058]

[0059] Among them, μ i is the mean vector of the input samples, ∑ i is the covariance matrix of the input sample. x is the input sample vector, and the dimension is n. The dimension n is generally set manually according to the size of the training data, and the dimension n is the same as the dimension of the CPC feature vector to be identified. For example, n is greater than 128, and n is 256. In this embodiment, multiple content vectors can be used as input sample vectors for Gaussian fitting according to the above-mentioned probability density function formula, and the content vector with the highest probability value is used as the preset content vector. At this time, the preset content vector is also an n-dimensional vector, and the preset content vector can be treated as a fixed constant vector.

[0060] In this embodiment, the above-mentioned preset content vector Vy(s) is essentially irrelevant to the actual content meaning contained in the content data, and can be understood as being related to the baseband signal of the transmitting end. And the content vector Vy(s) corresponding to different types of communication radiation sources may be different, and the above-mentioned preset content vector Vy(s) is equivalent to the comprehensive content vector Vy(s) obtained by each communication radiation source in the comprehensive database. In addition, in the process of obtaining the preset content vector Vy(s) and the preset noise vector dz(s), the number of listening devices can be one or more, and the types of listening devices can be the same or different. And the type of listening device used in the process of obtaining the preset content vector Vy(s) and the preset noise vector dz(s) can be the same or different from the type of listening device used in the actual use process, that is, the listening device has little influence on the results of obtaining the preset content vector Vy(s) and the preset noise vector dz(s).

[0061] Furthermore, the specific implementation process of the above step S103 includes:

[0062] A. Standardize the CPC feature vector to be identified.

[0063] Specifically, the above-mentioned standardization process can eliminate the differences in magnitude and dimension between variables. The specific process of the above-mentioned standardization process includes: obtaining the standard deviation of each factor in the CPC feature vector to be identified; obtaining the ratio of each factor in the CPC feature vector to be identified to the standard deviation, so as to obtain the CPC feature vector to be identified after standardization.

[0064] For example, the CPC feature vector to be identified is an n-dimensional vector, and the CPC feature vector to be identified x = [x1, x2, ..., x n ] T ; The standard deviation of each factor in the CPC feature vector to be identified is σ=sqrt(((x1-x) 2 +(x2-x) 2 +......(x n -x) 2 ) / n) 2 ; Divide each factor in the feature vector to be identified by the standard deviation to obtain the standardized CPC feature vector to be identified

[0065] B. Subtract the preset noise vector and the preset content vector from the CPC feature vector to be identified after the standardization processing to obtain the intrinsic channel vector.

[0066] Specifically, it is expressed by the formula: Ux(s)=X-Vy(s)-dz(s); wherein X is the standardized CPC feature vector to be identified, Vy(s) is the preset content vector, and dz(s) is the preset noise vector; and the above-mentioned CPC feature vector to be identified, preset content vector and preset noise vector are all n-dimensional vectors, and the corresponding eigenchannel vector is also an n-dimensional vector.

[0067] S104: Obtain the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link according to the eigenchannel vector.

[0068] Specifically, in this embodiment, please refer to Figure 4 , Figure 4 for Figure 1 The flowchart of an implementation method corresponding to step S104 in FIG. 1 is a flowchart of an implementation method corresponding to step S104 in FIG. 1 , wherein step S104 specifically includes:

[0069] S201: Obtain the score of each factor in the eigenchannel vector using a regression algorithm.

[0070] For example, the Bartlett estimation method may be used to obtain the score of each factor in the eigenchannel vector, and the importance of each factor in the eigenchannel vector may be evaluated according to the score.

[0071] S202: Select some factors with higher scores to form an input vector.

[0072] Specifically, the factors in the eigenchannel vector may be sorted in order of scores from high to low or from low to high, and then the first M factors with higher scores are selected to form an input vector; wherein the eigenchannel vector is an n-dimensional vector, and the value of M is less than n. For example, n is greater than 128, n may be 256, etc., and the value of M may be 128, etc.

[0073] In addition, the present application does not limit the order of the factors in the input vector. The order of the factors in the input vector can be the same as or different from the order of the factors in the eigenchannel. For example, suppose four factors with higher scores are selected from the CPC feature vector to be identified, namely: x1, x3, x5, x6. The subscripts of the four factors represent their order positions in the CPC feature vector to be identified. At this time, the order of the four factors in the input vector can be arbitrary, for example, the input vector P = [x1, x3, x5, x6] T , or input vector P = [x6, x1, x3, x5] T , or input vector P = [x6, x3, x1, x5] T wait.

[0074] S203: Obtain the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link according to the input vector.

[0075] Specifically, see Figure 5 , Figure 5 for Figure 4 The flowchart of an implementation method corresponding to step S203 in the embodiment of the present invention may include:

[0076] S301: Input the input vector into the residual network to obtain the probability that the input vector belongs to each communication link in the database, and obtain the maximum probability from all the probabilities.

[0077] Specifically, the classification of intrinsic channel features is mainly a process of pattern classification, so the design of the classifier also plays a vital role in the effect of communication radiation source identification. At present, there are typical classifiers such as the square decision analysis gate (QDA) and linear decision analysis (LDA). Commonly used classifiers such as Parzen window (PW) and K nearest neighbor (KNN) are non-parametric classifiers for probability density function estimation. However, these classifiers generally have poor robustness and poor recognition effect. The residual network used in this application is a convolutional neural network, which is easy to optimize and can improve the accuracy by increasing the depth of the network layer. Figure 6 As shown, Figure 6 Schematic diagram of a structure of a residual block in a residual network. Figure 6 As can be seen from the figure, the residual blocks in the residual network use skip connections, which alleviates the gradient vanishing problem caused by increasing the depth in the deep neural network.

[0078] In this embodiment, a residual network training process may be included before the above step S203. The specific training process may include:

[0079] A. Construct a training sample set. Figure 7 As shown, Figure 7 The structure diagram of an implementation method of a communication link set is shown in FIG. 1 . Assume that the current database contains six communication radiation sources, namely A, B, C, D, E and F; and it is known in advance that A and D can communicate with each other to form link 1, B and E can communicate with each other to form link 2, and C and F can communicate with each other to form link 3. Further, the channel characteristics corresponding to each link can be obtained through steps S101 to S103. The channel characteristics can be an eigenchannel vector or an input vector obtained from the eigenchannel vector. Among them, the training sample set can include multiple groups of channel characteristics, communication link labels corresponding to the channel characteristics, and communication radiation sources at both ends of the corresponding communication links.

[0080] B. Input the channel feature into the residual network to obtain the probability that the channel feature belongs to each communication link in the database.

[0081] C. Use this probability and the corresponding communication link label to adjust the parameters in the residual network.

[0082] D. In response to reaching a condition for stopping training, outputting the trained residual network; optionally, reaching the condition for stopping training may include reaching the number of training times, or the loss function convergence, etc.

[0083] In the subsequent application process, the input vector obtained in step S202 is directly input into the trained residual network to obtain the probability that the input vector belongs to each communication link in the database, and further the maximum probability can be obtained from all the probabilities.

[0084] S302: In response to the maximum probability being greater than or equal to a threshold, outputting a communication link corresponding to the maximum probability and communication radiation sources located at both ends of the communication link.

[0085] Specifically, in this embodiment, the threshold value can be set manually, for example, the threshold value can be 0.1, etc. Figure 7 For example, assuming that the current maximum probability is greater than 0.1, and the communication link corresponding to the maximum probability is link 1, then it can be considered that the target signal is propagating between communication radiation sources A and D, and link 1 can be precisely attacked and interfered with later.

[0086] S303: In response to the maximum probability being less than a threshold, determining that the communication link where the target signal is located is outside the database.

[0087] For example, Figure 7 For example, assuming that the current maximum probability is less than 0.1, and the communication link corresponding to the maximum probability is link 1; then it can be considered that the target signal is propagating on a link outside the database used to train the residual network, and the target signal can be classified into other categories.

[0088] In the above embodiment, in order to improve the accuracy of residual network discrimination, some factors with higher scores are selected to form an input vector, and then the input vector is input into the residual network for discrimination. Of course, in other embodiments, the eigenchannel vector can also be directly input into the residual network for discrimination, and this application does not limit this.

[0089] In the prior art, it is necessary to collect a large amount of training data for each communication radiation source, and use machine learning methods to extract the subtle features of the communication radiation source; however, in actual combat, it is already very difficult to find many target signals, and it is impossible to collect a large amount of data. This application adopts a communication link identification technology based on joint factor analysis, combines the individual data at both ends of the communication link to obtain the CPC feature vector, and then hierarchically decouples the CPC feature vector to obtain the intrinsic channel characteristics, and determines the communication radiation source through the intrinsic channel characteristics; compared with the prior art, only very few samples are needed to determine which communication link the target signal belongs to, and its accuracy is relatively high.

[0090] See also Figure 8 , Figure 8 This is a schematic structural diagram of an embodiment of a communication radiation source identification device of the present application. The communication radiation source identification device includes a receiving module 10 , a feature extraction module 12 , a decomposition module 14 and an acquisition module 16 .

[0091] The receiving module 10 is used to receive the target signal and obtain the signal to be identified from the target signal. Optionally, the receiving module 10 is specifically used to obtain the intermediate frequency signal from the target signal in response to the target signal being an encrypted signal and use it as the signal to be identified; in response to the target signal being an unencrypted signal, obtain the audio signal and / or text signal from the target signal and use it as the signal to be identified.

[0092] The feature extraction module 12 is connected to the receiving module 10 and is used to extract features from the signal to be identified to obtain a CPC feature vector to be identified.

[0093] The decomposition module 14 is connected to the feature extraction module 12, and is used to decompose the CPC feature vector to be identified using a preset noise vector and a preset content vector to obtain an eigenchannel vector. The acquisition module 16 is connected to the decomposition module 14, and is used to obtain the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link according to the eigenchannel vector.

[0094] In one embodiment, the recognition device provided by the present application also includes a fitting module 18, and both ends of the fitting module 18 are connected to the feature extraction module 12 and the decomposition module 14; wherein, before the decomposition module 14 uses a preset noise vector and a preset content vector to decompose the CPC feature vector to be identified, the receiving module 10 is also used to receive multiple noise data collected from at least one communication link in the database; the feature extraction module 12 is also used to obtain the noise CPC feature vector corresponding to each noise data; the decomposition module 14 is also used to decompose the multiple noise CPC feature vectors to obtain the corresponding noise vector; the fitting module 18 is used to perform Gaussian fitting on multiple noise vectors, take the noise vector with the highest probability value as the preset noise vector, and pass the preset noise vector to the decomposition module 14.

[0095] In addition, before the decomposition module 14 uses the preset noise vector and the preset content vector to decompose the CPC feature vector to be identified, the receiving module 10 is also used to receive multiple content data collected from at least one communication link in the database; the feature extraction module 12 is also used to obtain the content CPC feature vector corresponding to each content data; the decomposition module 14 is also used to decompose the multiple content CPC feature vectors to obtain the corresponding content vector; the fitting module 18 is also used to perform Gaussian fitting on the multiple content vectors, and use the content vector with the highest probability value as the preset content vector, and pass the preset content vector to the decomposition module 14.

[0096] Optionally, in this embodiment, the above-mentioned decomposition module 14 specifically includes a standardization sub-module and a decomposition sub-module that are interconnected; wherein the standardization sub-module is used to perform standardization processing on the CPC feature vector to be identified; the decomposition sub-module is used to obtain the intrinsic channel vector after subtracting the preset noise vector and the preset content vector from the CPC feature vector to be identified after the standardization processing.

[0097] Furthermore, the above-mentioned standardization submodule is specifically used to: obtain the standard deviation of each factor in the CPC feature vector to be identified; obtain the ratio of each factor in the CPC feature vector to be identified to the standard deviation, so as to obtain the CPC feature vector to be identified after standardization.

[0098] In another embodiment, the above-mentioned acquisition module 16 includes a first acquisition submodule, a selection submodule and a second acquisition submodule which are sequentially connected to each other; wherein the first acquisition submodule is used to obtain the scores of each factor in the intrinsic channel vector by using a regression algorithm; the selection submodule is used to select some factors with higher scores to form an input vector; the second acquisition submodule is used to obtain the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link according to the input vector.

[0099] Furthermore, the above-mentioned second acquisition submodule is specifically used to input the input vector into the residual classification network residual network to obtain the probability that the input vector belongs to each communication link in the database, and obtain the maximum probability from all probabilities; in response to the maximum probability being greater than or equal to the threshold, the communication link corresponding to the maximum probability and the communication radiation sources located at both ends of the communication link are output; in response to the maximum probability being less than the threshold, it is determined that the communication link where the target signal is located is outside the database.

[0100] In another embodiment, the identification device provided by the present application may also include a preprocessing module 11, located between the receiving module 10 and the feature extraction module 12, and the preprocessing module 11 is used to preprocess the signal to be identified and input the preprocessed signal to be identified into the feature extraction module 12; wherein the preprocessing includes denoising and / or obtaining a signal whose frequency exceeds a threshold from the signal to be identified.

[0101] See also Fig. 9 , Fig. 9 This is a structural diagram of an embodiment of an electronic device of the present application, and the electronic device specifically includes: a memory 30 and a processor 32 coupled to each other, the memory 30 stores program instructions, and the processor 32 is used to execute the program instructions to implement the steps in any of the above-mentioned communication radiation source identification methods. Specifically, the electronic device includes but is not limited to: a desktop computer, a laptop computer, a tablet computer, a server, etc., which are not limited here. In addition, the processor 32 can also be called a CPU (Central Processing Unit). The processor 32 may be an integrated circuit chip with signal processing capabilities. The processor 32 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc. In addition, the processor 32 can be implemented by an integrated circuit chip.

[0102] See also Fig.10 , Fig.10 This is a schematic diagram of the structure of an embodiment of a storage device of the present application. The storage device 50 stores program instructions 52 that can be executed by a processor. The program instructions 52 are used to implement the steps in any of the above-mentioned communication radiation source identification methods.

[0103] In the several embodiments provided in the present application, it should be understood that the disclosed methods and devices can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0104] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0105] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0106] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) or a processor (processor) to perform all or part of the steps of each implementation method of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0107] The above description is only an embodiment of the present application and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly used in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A communication radiation source identification method, characterized in that: include: receiving a target signal, and obtaining a signal to be identified from the target signal; Extracting features of the signal to be identified to obtain a CPC feature vector to be identified; Decomposing the CPC feature vector to be identified by using a preset noise vector and a preset content vector to obtain an intrinsic channel vector; wherein the preset noise vector is pre-obtained through Gaussian distribution fitting based on multiple noise data collected by the communication link in the database, and the preset content vector is pre-obtained through Gaussian distribution fitting based on multiple content data collected by the communication link in the database; the step of decomposing the CPC feature vector to be identified by using the preset noise vector and the preset content vector to obtain the intrinsic channel vector comprises: normalizing the CPC feature vector to be identified; obtaining the intrinsic channel vector after subtracting the preset noise vector and the preset content vector from the CPC feature vector to be identified after the normalization; The communication link where the target signal is located and the communication radiation sources at both ends of the communication link are obtained according to the intrinsic channel vector; wherein the step of obtaining the communication link where the target signal is located and the communication radiation sources at both ends of the communication link according to the intrinsic channel vector comprises: obtaining the score of each factor in the intrinsic channel vector by using a regression algorithm; selecting some factors with higher scores to form an input vector; and obtaining the communication link where the target signal is located and the communication radiation sources at both ends of the communication link according to the input vector.

2. The identification method according to claim 1, characterized in that: Before the step of decomposing the CPC feature vector to be identified by using a preset noise vector and a preset content vector, the method includes: Receiving a plurality of noise data collected from at least one communication link in a database; Obtaining a noise CPC feature vector corresponding to each of the noise data; Decompose the multiple noise CPC feature vectors to obtain corresponding noise vectors, perform Gaussian fitting on all the noise vectors, and use the noise vector with the highest probability value as the preset noise vector.

3. The identification method according to claim 1, characterized in that: Before the step of decomposing the CPC feature vector to be identified by using a preset noise vector and a preset content vector, the method includes: Receiving a plurality of content data collected from at least one communication link in a database; Obtaining a content CPC feature vector corresponding to each of the content data; Decompose the plurality of content CPC feature vectors to obtain corresponding content vectors, perform Gaussian fitting on all the content vectors, and use the content vector with the highest probability value as the preset content vector.

4. The identification method according to claim 1, characterized in that: The step of normalizing the CPC feature vector to be identified comprises: Obtaining the standard deviation of each factor in the CPC feature vector to be identified; The ratio of each factor in the CPC feature vector to be identified to the standard deviation is obtained to obtain the CPC feature vector to be identified after standardization.

5. The identification method according to claim 1, characterized in that: The step of obtaining the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link according to the input vector comprises: Inputting the input vector into a residual network to obtain the probability that the input vector belongs to each communication link in a database, and obtaining the maximum probability from all the probabilities; In response to the maximum probability being greater than or equal to a threshold, outputting a communication link corresponding to the maximum probability and communication radiation sources located at both ends of the communication link; In response to the maximum probability being smaller than a threshold, it is determined that the communication link where the target signal is located is outside the database.

6. The communication radiation source identification method according to claim 1, characterized in that: The step of receiving a target signal and obtaining a signal to be identified from the target signal comprises: in response to the target signal being an encrypted signal, obtaining an intermediate frequency signal from the target signal and using it as the signal to be identified; in response to the target signal being an unencrypted signal, obtaining an audio signal and / or a text signal from the target signal and using it as the signal to be identified; and / or, Before the step of extracting features from the signal to be identified to obtain the CPC feature vector to be identified, it includes: preprocessing the signal to be identified, wherein the preprocessing includes denoising and / or obtaining a signal whose frequency exceeds a threshold from the signal to be identified.

7. A communication radiation source identification device, characterized in that: include: A receiving module, used for receiving a target signal and obtaining a signal to be identified from the target signal; A feature extraction module, connected to the receiving module, for extracting features from the signal to be identified to obtain a CPC feature vector to be identified; A decomposition module, connected to the feature extraction module, for decomposing the CPC feature vector to be identified by using a preset noise vector and a preset content vector to obtain an eigenchannel vector; wherein the preset noise vector is pre-obtained through Gaussian distribution fitting based on multiple noise data collected by the communication link in the database, and the preset content vector is pre-obtained through Gaussian distribution fitting based on multiple content data collected by the communication link in the database; the eigenchannel vector is obtained by normalizing the CPC feature vector to be identified; and the CPC feature vector to be identified after normalization is subtracted from the preset noise vector and the preset content vector; An acquisition module is connected to the decomposition module and is used to obtain the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link according to the intrinsic channel vector; wherein the communication link where the target signal is located and the communication radiation sources located at both ends of the communication link are obtained by using a regression algorithm to obtain the scores of each factor in the intrinsic channel vector; some factors with higher scores are selected to form an input vector and obtained according to the input vector.

8. An electronic device, characterized in that: It comprises a memory and a processor coupled to each other, wherein the memory stores program instructions, and the processor is used to execute the program instructions to implement the communication radiation source identification method according to any one of claims 1 to 6.

9. A storage device, characterized in that: Program instructions that can be executed by a processor are stored, and the program instructions are used to implement the communication radiation source identification method according to any one of claims 1 to 6.

Images