A radar emitter identification system based on time domain graph tensor attention network

Abstract

The application discloses a radar radiation source identification system based on a time domain graph tensor attention network, realizes the connection between different sampling points and different radar features, and thus accurately identifies the radar radiation source. The identification system is composed of a radar acquisition module, a database and an upper computer. The radar acquisition module acquires the radar radiation source signal and stores the radar radiation source signal into the database. The upper computer collects the data in the database, performs signal transformation on the collected data, models a robust attention identification model based on the time domain graph tensor data after signal transformation, and detects new radar radiation source signals by using the robust attention identification model. The application realizes online identification of the radar radiation source identification with strong intelligence, high precision and high accuracy, and solves the problems of low identification precision of the radar radiation source and the fact that the time domain correlation and feature correlation are not considered.

Description

Technical Field

[0001] This invention relates to the field of radar radiation source identification, and in particular, to a radar radiation source identification system based on a time-domain graph tensor attention network. Background Technology

[0002] How to quickly and accurately identify radar radiation source signals has always been a hot research topic for scholars both at home and abroad. Radar technology is widely used in communication, cognitive radio, self-organizing networks, and other scenarios. Utilizing the electromagnetic pulse signals transmitted by radar transmitters, and determining the individual transmitter that generates the signal based on prior information, high-precision identification of radar radiation source signals is a problem and challenge that urgently needs to be solved.

[0003] To improve the quality and speed of radar source identification, a high-efficiency radar source identification system is a prerequisite. Therefore, fast and accurate radar source identification technology has become a core research area in the radar field. In recent years, with the rapid development of deep learning, deep learning models such as convolutional neural networks and recurrent neural networks have been widely used in radar source identification. However, these methods do not consider the temporal correlation of each feature and the relationships between features. In the temporal indicators of radar source signals, amplitude, instantaneous frequency, and instantaneous phase are interrelated. Transforming the signal into a time-domain graph tensor and employing a robust attention model can learn the temporal correlation and feature correlation of each feature. Therefore, inventing a radar source identification system with high accuracy, high intelligence, and the ability to handle relationships between features is of great significance. Summary of the Invention

[0004] To overcome the shortcomings of existing radar source identification methods that fail to consider the temporal correlation of each feature and the relationships between features, and considering the interrelationships among the temporal indices of radar source signals (amplitude, instantaneous frequency, and instantaneous phase), this invention provides a high-accuracy, highly intelligent radar source identification system capable of handling inter-feature relationships. Based on a time-domain graph tensor attention network, this radar source identification system acquires radar source signals through a radar acquisition module and stores them in a database. A host computer collects data from the database, performs signal transformation on the acquired data, and builds a robust attention recognition model based on the transformed time-domain graph tensor data. This robust attention recognition model is then used to detect new radar source signals. This invention achieves highly intelligent, high-precision, and high-accuracy online radar source identification, solving the problems of low accuracy and failure to consider temporal and feature correlations in radar source identification.

[0005] The technical solution adopted by this invention to solve its technical problem is:

[0006] A radar radiation source identification system based on a time-domain graph tensor attention network comprises a radar acquisition module, a database, and a host computer. These three modules and the system are interconnected to form a complete radar radiation source identification system. The host computer includes: a signal transformation module, a robust attention recognition model building module, a robust attention recognition module, and a recognition result output module.

[0007] Furthermore, the host computer collects data from the database, performs signal transformation on the collected data, and builds a robust attention recognition model based on the time-domain graph tensor data after signal transformation. It can also update the database in real time, thereby updating the robust attention recognition model in real time. At the same time, it uses the robust attention recognition model to detect new radar radiation source signals.

[0008] Furthermore, the signal transformation module transforms the data in the radar radiation source database into time-domain graph tensors, which is accomplished through the following process:

[0009] The original radar radiation source signal is a one-dimensional signal. The i-th original radar signal can be represented as:

[0010]

[0011] Where L represents the length of the signal.

[0012] Through empirical mode decomposition, it is decomposed into three intrinsic mode functions (IMF1). i IMF2 i IMF3 i :

[0013]

[0014]

[0015]

[0016] in Indicates IMF1 i The decomposed value of the l-th pulse of the i-th original radar signal in the set. Indicates IMF2 i The decomposed value of the l-th pulse of the i-th original radar signal in the set. Indicates IMF3 i The decomposition value of the l-th pulse of the i-th original radar signal in the set.

[0017] The Hilbert transform is used to transform the original signal and its three intrinsic mode functions. The original signal after the Hilbert transform is... and three inherent pattern functions It can be described as:

[0018]

[0019]

[0020]

[0021]

[0022] in express The l-th value of the i-th radar signal in the set. express The l-th value of the i-th radar signal in the set. express The i-th radar signal in the set is the l-th value, and the analysis signal at each time point consists of a real part and an imaginary part.

[0023] Calculate the envelope (instantaneous amplitude), instantaneous phase, and instantaneous frequency. Using the original signal as an example, its instantaneous amplitude... Instantaneous phase and instantaneous frequency It can be represented as:

[0024]

[0025]

[0026]

[0027] in, The instantaneous amplitude of the i-th original radar signal at the l-th sampling time can be calculated using the following formula:

[0028]

[0029] The instantaneous phase of the i-th original radar signal at the l-th sampling time can be calculated using the following formula:

[0030]

[0031] The instantaneous frequency of the i-th original radar signal at the l-th sampling time can be calculated using the following formula:

[0032]

[0033] f s This indicates the sampling frequency. Similarly, the instantaneous amplitude of the three IMFs can be obtained. Instantaneous phase and instantaneous frequency

[0034] To ensure that the length of each feature is equal, instantaneous amplitudes are discarded. Instantaneous phase and the original signal The first sampling point is set so that the length of each feature is L-1. These three features are then concatenated with the original signal to obtain the node features of the original image, IMF1 image, IMF2 image, and IMF3 image.

[0035]

[0036]

[0037]

[0038]

[0039] Wherein, the l-th node of the original graph is The l-th node of the IMF1 diagram is The l-th node of the IMF2 diagram is The l-th node of the IMF3 diagram is The edges between nodes consist of learnable attention coefficients, which together form a time-domain graph tensor.

[0040] Furthermore, the robust attention recognition model modeling module utilizes the time-domain graph tensor obtained after signal transformation to establish a high-precision robust attention recognition model, and automatically learns how to effectively perform recognition using training set data in the database. The novel robust attention recognition model proposed in this invention includes one intra-graph propagation and one inter-graph propagation. Intra-graph propagation can efficiently fuse temporal correlations, while inter-graph propagation can effectively fuse feature correlations, thereby effectively improving the model's recognition accuracy. The robust attention recognition model modeling module is completed through the following process:

[0041] During intra-graph propagation, each node aggregates all its neighbors and itself. This aggregation is precisely to preserve the original features. In the case of the original graph, the intra-graph propagation process is as follows:

[0042]

[0043]

[0044] in, It represents the l-th node of the i-th radar radiation source signal after propagation within the original graph. Indicates from z1 to z K Vector concatenation, It is the l-th node and the m-th node The normalized dot product attention coefficients are calculated for the k-th attention interval. These are learnable transformation weights.

[0045] Using the same intra-frame propagation method, the nodes of the IMF1, IMF2, and IMF3 graphs can be updated to... and

[0046]

[0047] Inter-graph propagation spreads information between different graphs in a time-domain graph tensor, gradually blending heterogeneous information from different graphs into consistent information. The original signal and features from different IMFs are used for message propagation at the same sampling point. A self-loop is added during the aggregation process to preserve the original features. Similarly, taking the original graph as an example, inter-graph propagation is as follows:

[0048]

[0049]

[0050] in, It is represented by the l-th node of the i-th radar signal after intra-graph and inter-graph propagation in the original graph, and is derived from the l-th node. and the nth node The normalized dot product attention coefficients are calculated for the q-th attention interval. Represents nodes In the set of nodes with the same sampling points These are learnable transformation weights. The inter-graph propagation process takes into account the relationships between features, effectively improving the recognition accuracy of radar radiation source identification systems.

[0051] The updated node representation includes not only the relationships between different sampling points, but also the relationships between different features at the same sampling point. All updated nodes are flattened into a one-dimensional matrix. After a fully connected module and softmax operation, the radar signal type is obtained:

[0052]

[0053] Among them, the learnable transformation weights W 3 ∈R 16(L-1)×C , This represents the predicted probability for each type of radar source signal. C is the number of radar source signal types. The category with the highest predicted probability is selected as the final identification category.

[0054] Furthermore, the robust attention recognition module is used to directly input the newly acquired radar radiation source signal after the signal transformation module into the robust attention recognition model to obtain the category of the radar radiation source signal.

[0055] Furthermore, the identification result output module outputs the newly acquired radar radiation source signal category results.

[0056] The technical concept of this invention is as follows: Addressing the shortcomings of existing radar source identification methods that fail to consider the temporal correlation of each feature and the relationships between features, this invention provides a radar source identification system with high accuracy, high intelligence, and the ability to handle relationships between features. Based on a time-domain graph tensor attention network, the radar source identification system acquires radar source signals through a radar acquisition module and stores them in a database. A host computer collects data from the database, performs signal transformation on the acquired data, and builds a robust attention recognition model based on the transformed time-domain graph tensor data. This robust attention recognition model is then used to detect new radar source signals. This invention achieves highly intelligent, high-precision, and high-accuracy online radar source identification, solving the problems of low accuracy and failure to consider temporal and feature correlation in radar source identification.

[0057] The beneficial effects of this invention are mainly reflected in the following aspects: 1. Intra-graph propagation can efficiently fuse temporal correlations, thereby effectively improving the model's recognition accuracy; 2. Inter-graph propagation can effectively fuse feature correlations, thereby effectively improving the model's recognition accuracy; 3. The radar radiation source recognition system based on the temporal domain graph tensor attention network can automatically learn according to the training data, has strong intelligence, and is less affected by human factors. Attached Figure Description

[0058] Figure 1 Hardware connection diagram of a radar radiation source identification system based on a time-domain graph tensor attention network;

[0059] Figure 2 A host computer diagram of a radar radiation source identification system based on a time-domain graph tensor attention network. Detailed Implementation

[0060] The present invention will now be further described with reference to the accompanying drawings.

[0061] refer to Figure 1 A radar radiation source identification system based on a time-domain graph tensor attention network is described, wherein a radar acquisition module 1, a database 2, and a host computer 3 are sequentially connected. (Reference) Figure 2 The host computer 2 includes a signal transformation module 4, a robust attention recognition model modeling module 5, a robust attention recognition module 6, and a recognition result output module 7.

[0062] The robust attention recognition model modeling module 5 is based on data in the database and can update the database in real time, thereby updating the robust attention recognition model in real time.

[0063] Furthermore, the signal transformation module 4 collects data from the database, performs signal transformation on the collected data, and builds a robust attention recognition model based on the time domain tensor data after signal transformation. It can also update the database in real time, thereby updating the robust attention recognition model in real time. At the same time, it uses the robust attention recognition model to detect new radar radiation source signals.

[0064] Furthermore, the signal transformation module 4 is used to transform the data in the radar radiation source database into a time-domain graph tensor, which is accomplished through the following process:

[0065] The original radar radiation source signal is a one-dimensional signal. The i-th original radar signal can be represented as:

[0066]

[0067] Where L represents the length of the signal.

[0068] Through empirical mode decomposition, it is decomposed into three intrinsic mode functions (IMF1). i IMF2 i IMF3 i :

[0069]

[0070]

[0071]

[0072] in Indicates IMF1 i The decomposed value of the l-th pulse of the i-th original radar signal in the set. Indicates IMF2 i The decomposed value of the l-th pulse of the i-th original radar signal in the set. Indicates IMF3 i The decomposition value of the l-th pulse of the i-th original radar signal in the set.

[0073] The Hilbert transform is used to transform the original signal and its three intrinsic mode functions. The original signal after the Hilbert transform is... and three inherent pattern functions It can be described as:

[0074]

[0075]

[0076]

[0077]

[0078] in express The l-th value of the i-th radar signal in the set. express The l-th value of the i-th radar signal in the set. express The i-th radar signal in the set is the l-th value, and the analysis signal at each time point consists of a real part and an imaginary part.

[0079] Calculate the envelope (instantaneous amplitude), instantaneous phase, and instantaneous frequency. Using the original signal as an example, its instantaneous amplitude... Instantaneous phase and instantaneous frequency It can be represented as:

[0080]

[0081]

[0082]

[0083] in, The instantaneous amplitude of the i-th original radar signal at the l-th sampling time can be calculated using the following formula:

[0084]

[0085] The instantaneous phase of the i-th original radar signal at the l-th sampling time can be calculated using the following formula:

[0086]

[0087] The instantaneous frequency of the i-th original radar signal at the l-th sampling time can be calculated using the following formula:

[0088]

[0089] f s This indicates the sampling frequency. Similarly, the instantaneous amplitude of the three IMFs can be obtained. Instantaneous phase and instantaneous frequency

[0090] To ensure that the length of each feature is equal, instantaneous amplitudes are discarded. Instantaneous phase and the original signal The first sampling point is set so that the length of each feature is L-1. These three features are then concatenated with the original signal to obtain the node features of the original image, IMF1 image, IMF2 image, and IMF3 image.

[0091]

[0092]

[0093]

[0094]

[0095] Wherein, the l-th node of the original graph is The l-th node of the IMF1 diagram is The l-th node of the IMF2 diagram is The l-th node of the IMF3 diagram is The edges between nodes consist of learnable attention coefficients, which together form a time-domain graph tensor.

[0096] Furthermore, the robust attention recognition model modeling module 5 utilizes the time-domain graph tensor obtained after signal transformation to establish a high-precision robust attention recognition model, and automatically learns how to effectively perform recognition using the training set data in the database. The novel robust attention recognition model proposed in this invention includes one intra-graph propagation and one inter-graph propagation. Intra-graph propagation can efficiently fuse temporal correlations, while inter-graph propagation can effectively fuse feature correlations, thereby effectively improving the model's recognition accuracy. The robust attention recognition model modeling module is completed through the following process:

[0097] During intra-graph propagation, each node aggregates all its neighbors and itself. This aggregation is precisely to preserve the original features. In the case of the original graph, the intra-graph propagation process is as follows:

[0098]

[0099]

[0100] in, It represents the l-th node of the i-th radar radiation source signal after propagation within the original graph. Indicates from z1 to z K Vector concatenation, It is the l-th node and the m-th node The normalized dot product attention coefficients are calculated for the k-th attention interval. These are learnable transformation weights.

[0101] Using the same intra-frame propagation method, the nodes of the IMF1, IMF2, and IMF3 graphs can be updated to... and

[0102]

[0103] Inter-graph propagation spreads information between different graphs in a time-domain graph tensor, gradually blending heterogeneous information from different graphs into consistent information. The original signal and features from different IMFs are used for message propagation at the same sampling point. A self-loop is added during the aggregation process to preserve the original features. Similarly, taking the original graph as an example, inter-graph propagation is as follows:

[0104]

[0105]

[0106] in, It is represented by the l-th node of the i-th radar signal after intra-graph and inter-graph propagation in the original graph, and is derived from the l-th node. and the nth node The normalized dot product attention coefficients are calculated for the q-th attention interval. Represents nodes In the set of nodes with the same sampling points These are learnable transformation weights. The inter-graph propagation process takes into account the relationships between features, effectively improving the recognition accuracy of radar radiation source identification systems.

[0107] The updated node representation includes not only the relationships between different sampling points, but also the relationships between different features at the same sampling point. All updated nodes are flattened into a one-dimensional matrix. After a fully connected module and softmax operation, the radar signal type is obtained:

[0108]

[0109] Among them, the learnable transformation weights W 3 ∈R 16(L-1)×C , This represents the predicted probability for each type of radar source signal. C is the number of radar source signal types. The category with the highest predicted probability is selected as the final identification category.

[0110] Furthermore, the robust attention recognition module 6 is used to directly input the newly acquired radar radiation source signal after the signal transformation module into the trained robust attention recognition model to obtain the category of the radar radiation source signal.

[0111] Furthermore, the identification result output module 7 outputs the newly acquired radar radiation source signal category results.

[0112] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention, but it should not be construed as limiting the specific implementation of the invention to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A radar radiation source identification system based on a time-domain graph tensor attention network, characterized in that: The system consists of three interconnected modules: a radar acquisition module, a database, and a host computer. The host computer includes a signal transformation module, a robust attention recognition model modeling module, a robust attention recognition module, and a recognition result output module. The host computer acquires data from the database, performs signal transformation on the acquired data, models a robust attention recognition model based on the transformed time-domain tensor data, updates the robust attention recognition model in real time, and uses the robust attention recognition model to detect new radar radiation source signals. The signal transformation module is used to transform the data in the radar radiation source database into a time-domain graph tensor, which is accomplished through the following process: The original radar radiation source signal is a one-dimensional signal; the i-th original radar signal Represented as: (1) Where L represents the length of the signal; Through empirical mode decomposition, it is decomposed into three intrinsic mode functions. , , : (2) (3) (4) in express The decomposed value of the l-th pulse of the i-th original radar signal in the set. express The decomposed value of the l-th pulse of the i-th original radar signal in the set. express The decomposed value of the l-th pulse of the i-th original radar signal in the set; The Hilbert transform is used to transform the original signal and its three intrinsic mode functions; the original signal after the Hilbert transform. and three inherent pattern functions , , Described as: (5) (6) (7) (8) in express The l-th value of the i-th radar signal in the set. express The l-th value of the i-th radar signal in the set. express The i-th radar signal in the set is the l-th value, and the analysis signal at each time point consists of a real part and an imaginary part. , , , ; Calculate the instantaneous amplitude, instantaneous phase, and instantaneous frequency of the original signal; its instantaneous amplitude Instantaneous phase and instantaneous frequency Represented as: (9) (10) (11) in, The instantaneous amplitude of the i-th original radar signal at the l-th sampling time is calculated using the following formula: (12) The instantaneous phase of the i-th original radar signal at the l-th sampling time is calculated using the following formula: (13) The instantaneous frequency of the i-th original radar signal at the l-th sampling time is calculated using the following formula: (14) Indicates the sampling frequency; obtains the instantaneous amplitude of the three IMFs. Instantaneous phase and instantaneous frequency ; To ensure that the length of each feature is equal, instantaneous amplitudes are discarded. , , , Instantaneous phase ( , , , ) and the original signal ( , , , The first sampling point is used to make the length of each feature L-1; these three features are concatenated with the original signal to obtain the node features of the original image, IMF1 image, IMF2 image, and IMF3 image: (15) (16) (17) (18) Wherein, the l-th node of the original graph is The l-th node of the IMF1 diagram is The l-th node of the IMF2 diagram is The l-th node of the IMF3 diagram is The edges between nodes consist of learnable attention coefficients, which together form a time-domain graph tensor. The robust attention recognition model modeling module utilizes the time-domain graph tensor obtained after signal transformation to establish a high-precision robust attention recognition model, and automatically learns how to effectively perform recognition using training set data in the database. The robust attention recognition model modeling module is completed through the following process: (1) During the intragraph propagation process, each node will aggregate all its neighboring nodes and itself; the aggregation itself is to preserve the original features; in the case of the original graph, the intragraph propagation process is as follows: (19) (20) in, It represents the l-th node of the i-th radar radiation source signal after propagation within the original graph. Indicates from arrive Vector concatenation, It is the l-th node of the original graph and the m-th node The normalized dot product attention coefficients are calculated for the k-th attention interval. These are learnable transformation weights; Using the same intra-frame propagation method, the nodes of the IMF1, IMF2, and IMF3 graphs are updated to... , and ; (2) Inter-graph propagation propagates information between different graphs in the time-domain graph tensor, gradually mixing heterogeneous information from different graphs into consistent information; the original signal and different IMFs features are used for message propagation at the same sampling point; a self-loop is added during the aggregation process to preserve the original features; inter-graph propagation is as follows: (21) (22) in, It represents the l-th node of the i-th radar signal after intra-map and inter-map propagation in the original graph, and is derived from the l-th node of the i-th radar radiation source signal after intra-map propagation in the original graph. and the nth node The normalized dot product attention coefficients are calculated for the q-th attention interval. Represents nodes In the set of nodes with the same sampling points These are learnable transformation weights; (3) The updated node representation includes not only the relationship between different sampling points, but also the relationship between different features at the same sampling point; flatten all updated nodes into a one-dimensional matrix; after the fully connected module and softmax operation, obtain the type of radar signal: (23) Among them, learnable transformation weights , is the predicted probability for each type of radar source signal; C is the number of radar source signal types; the category with the highest predicted probability is selected as the final identification category.

2. The radar radiation source identification system based on time-domain graph tensor attention network according to claim 1, characterized in that: The robust attention recognition module is used to directly input the radar radiation source signal newly acquired by the signal transformation module into the robust attention recognition model to obtain the category of the radar radiation source signal.

3. The radar radiation source identification system based on time-domain graph tensor attention network according to claim 1, characterized in that: The recognition result output module outputs the newly acquired radar radiation source signal category results obtained by the robust attention recognition module.

Images