Training method of complex domain layered radar anti-jamming pattern joint parameter decision network

Abstract

The application provides a training method of a complex domain layered radar anti-interference pattern combined parameter decision network, and the method comprises the following steps: constructing a complex domain interference feature extraction network; constructing a layered anti-interference decision model HCD2QNet comprising an outer anti-interference pattern decision network and an inner anti-interference parameter decision network; cascading the complex domain interference feature extraction network and the layered anti-interference decision model HCD2QNet to obtain the complex domain layered radar anti-interference pattern combined parameter decision network; receiving active interference signals transmitted by an interference machine, obtaining radar echoes with complex interference signals according to the received signals, and training the complex domain layered radar anti-interference pattern combined parameter decision network according to the radar echoes, so that the obtained complex domain layered radar anti-interference pattern combined parameter decision network realizes the integration of perception and decision from original radar echo signals to anti-interference measures, and enhances the flexibility and efficiency of radar anti-interference decision confrontation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of communication, and particularly relates to a training method of a complex domain layered radar anti-jamming pattern joint parameter decision network. BACKGROUND

[0002] In modern electronic warfare, the interference environment faced by the radar system is complex and changeable, and it is difficult for manual extraction of interference features to meet the demand of radar for accurate and real-time perception of the battlefield environment, which affects the subsequent radar anti-jamming decision countermeasure effect; the traditional radar anti-jamming decision method relies on expert experience to manually design countermeasures, and has insufficient autonomous decision-making ability, poor countermeasure performance and low real-time performance, and cannot cope with the current flexible and changeable intelligent interference scene; the existing radar anti-jamming decision focuses on the anti-jamming pattern level (or only decides a single parameter but ignores the anti-jamming pattern), ignores the influence of the anti-jamming parameter on the anti-jamming effect, destroys the relevance of the anti-jamming pattern and the anti-jamming parameter, and leads to poor application effect of the anti-jamming strategy in the countermeasure environment.

[0003] However, the prior art has the following disadvantages:

[0004] 1. The current radar anti-jamming decision method mostly manually extracts feature information such as interference bandwidth, power and angle to construct the environment state, and then uses methods such as Q learning or deep Q network to obtain the Q value of different anti-jamming patterns, and decides the final anti-jamming pattern according to the maximum Q value. The disadvantage of this kind of method is that the type of manually extracted features depends on expert experience, and the quality of the features is difficult to guarantee, which reduces the accuracy of the anti-jamming decision result; the type of manually extracted features is fixed, which cannot cope with the current flexible and changeable intelligent interference scene, and reduces the radar autonomous decision-making ability; the type of manually selected features is limited, which cannot adapt to complex interference signals, has low interference environment perception ability, and is difficult to fully represent the complex interference countermeasure environment; the above disadvantages of input features directly limit the accuracy of the anti-jamming decision algorithm.

[0005] 2、The existing radar anti-jamming decision method mainly focuses on using intelligent models such as deep Q network or decision tree to select anti-jamming patterns, such as document [1] Zhao Jiachen, Zhang Jingdong, Li Ziyu. Radar intelligent decision generation algorithm based on deep reinforcement learning [J]. Modern Radar, 2022, 44 (12): 25-33. Blind source separation anti-jamming pattern is manually extracted for feature information such as number of interference sources, power and interference type, or single anti-jamming parameter is optimized in the case of fixed anti-jamming pattern, such as document [2] Y. Fang, S. Wei, L. Zhang, Z. Wu and J. Wu, "Online Emission Policy Selection for Radar Antijamming Using Bandit-Optimized Policy Search," in IEEE Transactions on Aerospace and Electronic Systems, vol. 60, no. 3, pp. 3132-3147, June 2024, doi: 10.1109 / TAES.2024.3358793. Radar carrier frequency in Ss. The disadvantage of this method is that only the anti-jamming pattern or the anti-jamming parameter is independently decided, which destroys the coupling between the two, weakens the effectiveness and flexibility of the radar autonomous anti-jamming decision, and thus reduces the radar jamming countermeasure performance and limits the application of the anti-jamming decision method to the interference scene.

[0006] 3、The current radar anti-jamming decision technology decomposes the interference environment perception (i.e. interference environment feature extraction) and the anti-jamming decision into two independent parts, without considering the correlation between the interference feature extraction and the anti-jamming decision, resulting in low efficiency of the radar anti-jamming decision and difficulty in effectively counteracting the time-varying interference environment. SUMMARY

[0007] In order to solve the above problems existing in the prior art, the application provides a training method of a complex domain layered radar anti-jamming pattern joint parameter decision network, which specifically comprises:

[0008] In the first aspect, the application provides a training method of a complex domain layered radar anti-jamming pattern joint parameter decision network, which comprises:

[0009] S1, a complex domain interference feature extraction network is constructed, the complex domain interference feature extraction network comprises a complex interference echo real part network branch and a complex interference echo imaginary part network branch, the complex interference echo real part network branch is used for extracting echo real part features, and the complex interference echo imaginary part network branch is used for extracting echo imaginary part features;

[0010] S2, a hierarchical anti-interference decision model HCD2QNet is constructed, the hierarchical anti-interference decision model HCD2QNet includes an outer anti-interference pattern decision network and an inner anti-interference parameter decision network, the outer anti-interference pattern decision network is used for determining an optimal anti-interference pattern according to an interference environment state vector output by the complex domain interference feature extraction network, and the inner anti-interference parameter decision network is used for determining an optimal anti-interference parameter according to the interference environment state vector output by the complex domain interference feature extraction network and the optimal anti-interference pattern;

[0011] S3, the complex domain interference feature extraction network and the hierarchical anti-interference decision model HCD2QNet are cascaded, and a complex domain hierarchical radar anti-interference pattern joint parameter decision network is obtained, wherein the output of the complex domain interference feature extraction network is used as the input of the outer anti-interference pattern decision network of the hierarchical anti-interference decision model HCD2QNet, and the output of the complex domain interference feature extraction network and the output of the outer anti-interference pattern decision network of the hierarchical anti-interference decision model HCD2QNet are used as the input of the inner anti-interference parameter decision network;

[0012] S4, an active jamming signal transmitted by a jammer is received, a radar echo with a complex interference signal is obtained according to the received signal, and the complex domain hierarchical radar anti-interference pattern joint parameter decision network is trained according to the radar echo with the complex interference signal.

[0013] In a second aspect, the present application further provides a complex domain hierarchical radar anti-interference pattern joint parameter decision method, comprising:

[0014] An active jamming signal transmitted by a jammer is received, and a radar echo with a complex interference signal S t ′ is obtained;

[0015] The radar echo with the complex interference signal S t ′ is input into the complex domain hierarchical radar anti-interference pattern joint parameter decision network, and an optimal anti-interference pattern and an optimal anti-interference parameter are obtained, the complex domain hierarchical radar anti-interference pattern joint parameter decision network is obtained through the training method of any complex domain hierarchical radar anti-interference pattern joint parameter decision network provided in the first aspect;

[0016] The optimal anti-interference pattern and the optimal anti-interference parameter are implemented, and the countermeasures are completed.

[0017] In a third aspect, the present application further provides an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus;

[0018] The memory is used for storing a computer program;

[0019] A processor is configured to implement any method provided by the first aspect or the second aspect when executing a program stored in a memory.

[0020] Advantages of the present application:

[0021] The training method of the complex domain layered radar anti-jamming pattern and parameter decision network provided by the present application comprises the following steps: constructing a complex domain jamming feature extraction network, the complex domain jamming feature extraction network comprising a complex jamming echo real part network branch and a complex jamming echo imaginary part network branch, the complex jamming echo real part network branch being configured to extract echo real part features, and the complex jamming echo imaginary part network branch being configured to extract echo imaginary part features; constructing a layered anti-jamming decision model HCD2QNet, the layered anti-jamming decision model HCD2QNet comprising an outer-layer anti-jamming pattern decision network and an inner-layer anti-jamming parameter decision network, the outer-layer anti-jamming pattern decision network being configured to determine an optimal anti-jamming pattern according to an output jamming environment state vector of the complex domain jamming feature extraction network, and the inner-layer anti-jamming parameter decision network being configured to determine optimal anti-jamming parameters according to the output jamming environment state vector of the complex domain jamming feature extraction network and the optimal anti-jamming pattern; cascading the complex domain jamming feature extraction network and the layered anti-jamming decision model HCD2QNet to obtain a complex domain layered radar anti-jamming pattern and parameter decision network, wherein the output of the complex domain jamming feature extraction network is taken as the input of the outer-layer anti-jamming pattern decision network of the layered anti-jamming decision model HCD2QNet, and the output of the complex domain jamming feature extraction network and the output of the outer-layer anti-jamming pattern decision network of the layered anti-jamming decision model HCD2QNet are taken as the input of the inner-layer anti-jamming parameter decision network; receiving active jamming signals transmitted by a jammer, obtaining radar echoes with complex jamming signals according to the received signals, and training the complex domain layered radar anti-jamming pattern and parameter decision network according to the radar echoes with complex jamming signals, so that the obtained complex domain layered radar anti-jamming pattern and parameter decision network avoids separation of jamming feature extraction and anti-jamming decision steps, retains the relevance between the two, realizes perception and decision integration from original radar echo signals to anti-jamming measures, enhances the flexibility and efficiency of radar anti-jamming decision confrontation, and has the advantage of being widely applied to complex time-varying intelligent jamming scenarios.

[0022] The present application will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 A flowchart of the training method of the complex domain layered radar anti-jamming pattern and parameter decision network provided by the present application is shown in the figure.

[0024] Figure 2 An architecture diagram of the complex domain layered radar anti-jamming pattern and parameter decision network provided by the present application is shown in the figure.

[0025] Figure 3 An emulation result schematic diagram provided by the present application. DETAILED DESCRIPTION

[0026] The present application will be further described in conjunction with specific embodiments, but the embodiments of the present application are not limited thereto.

[0027] The present application aims at the deficiencies of the prior art, and proposes a complex domain layered radar anti-jamming pattern joint parameter decision method: a layered model HCD2QNet is constructed, the anti-jamming pattern is decided through the outer layer model, the anti-jamming parameters are selected through the inner layer model, the anti-jamming best pattern joint parameter common decision is realized; a complex domain feature extraction network is designed, and the complex interference environment features are fully perceived; the perception decision integration of the original radar echo signal to the anti-jamming measure is realized through cascading the complex domain feature extraction network and the layered model HCD2QNet, so as to avoid the problems of manual extraction of interference signal features, low decision efficiency, low accuracy, poor autonomy and insufficient flexibility in the anti-jamming decision process, and the like, and the method can effectively resist various complex radar interference scenes.

[0028] Figure 1 A flowchart of a training method of a complex domain layered radar anti-jamming pattern joint parameter decision network provided by the present application is shown as shown in Figure 1 The method comprises the following steps:

[0029] S1, a complex domain interference feature extraction network is constructed.

[0030] The complex domain interference feature extraction network comprises a complex interference echo real part network branch and a complex interference echo imaginary part network branch.

[0031] The complex interference echo real part network branch is used for extracting echo real part features.

[0032] The complex interference echo imaginary part network branch is used for extracting echo imaginary part features.

[0033] As shown in Figure 2 Optionally, the complex interference echo real part comprises a first convolutional pooling layer, a second convolutional pooling layer and a first global maximum pooling layer connected in sequence.

[0034] Optionally, the complex interference echo imaginary part comprises a third convolutional pooling layer, a fourth convolutional pooling layer and a second global maximum pooling layer connected in sequence, and the first global maximum pooling layer in the complex interference echo real part and the second global maximum pooling layer in the complex interference echo imaginary part are feature spliced.

[0035] Optionally, the detailed parameters of each layer of the complex domain interference feature extraction network are shown in Table 1.

[0036] Table 1

[0037]

[0038] The complex domain interference feature extraction network designed in the application perceives the interference environment, overcomes the limitations of manual feature extraction of existing anti-interference decision algorithms and the problem that the existing anti-interference decision algorithms cannot adapt to complex interference signals, fully characterizes the interference environment, and improves the accuracy of subsequent radar anti-interference decision

[0039] S2, a hierarchical anti-interference decision model HCD2QNet is constructed.

[0040] The hierarchical anti-interference decision model HCD2QNet includes an outer anti-interference pattern decision network and an inner anti-interference parameter decision network.

[0041] The outer anti-interference pattern decision network is used to determine the best anti-interference pattern according to the interference environment state vector output by the complex domain interference feature extraction network.

[0042] The inner anti-interference parameter decision network is used to determine the best anti-interference parameter according to the interference environment state vector output by the complex domain interference feature extraction network and the best anti-interference pattern.

[0043] As shown in Figure 2 Optionally, the outer anti-interference pattern decision network includes a first full connection layer, a second full connection layer, a third full connection layer, a fourth full connection layer, and a first element addition module, wherein the output of the first full connection layer is the input of the second full connection layer; the output of the second full connection layer is respectively the input of the third full connection layer and the fourth full connection layer; the output of the third full connection layer and the output of the fourth full connection layer are subjected to element addition operation in the first element addition module.

[0044] As shown in Figure 2 Optionally, the inner anti-interference parameter decision network includes a fifth full connection layer, a sixth full connection layer, a seventh full connection layer, an eighth full connection layer, and a second element addition module; wherein the output of the first element addition module and the output of the fifth full connection layer are the input of the sixth full connection layer; the output of the sixth full connection layer is respectively the input of the seventh full connection layer and the eighth full connection layer; the output of the seventh full connection layer and the output of the eighth full connection layer are subjected to element addition operation in the second element addition module.

[0045] Optionally, the detailed parameters of each layer of the hierarchical anti-interference decision model HCD2QNet are shown in Table 2:

[0046] Table 2

[0047]

[0048] The layered anti-interference decision model HCD2QNet constructed by the application, wherein the outer layer model decides the anti-interference pattern, and the inner layer model synchronously selects the corresponding anti-interference parameters, solves the problem that the existing anti-interference decision method cannot simultaneously decide the anti-interference pattern and the anti-interference parameters, realizes the joint matching decision of the optimal anti-interference pattern-anti-interference parameters, and improves the precision and the countermeasure performance of the radar countermeasure decision.

[0049] S3, cascade the complex domain interference feature extraction network and the layered anti-interference decision model HCD2QNet to obtain the complex domain layered radar anti-interference pattern joint parameter decision network.

[0050] The output of the complex domain interference feature extraction network is used as the input of the outer layer anti-interference pattern decision network of the layered anti-interference decision model HCD2QNet, and the output of the complex domain interference feature extraction network and the output of the outer layer anti-interference pattern decision network of the layered anti-interference decision model HCD2QNet are used as the input of the inner layer anti-interference parameter decision network.

[0051] S4, receiving the active jamming signal transmitted by the jammer, obtaining the radar echo with the complex interference signal according to the received signal, and training the complex domain layered radar anti-interference pattern joint parameter decision network according to the radar echo with the complex interference signal.

[0052] In a possible implementation manner, the step S4 includes the following steps S41-S43:

[0053] S41, setting the anti-interference reward function of the outer layer anti-interference pattern decision network and the anti-interference reward function of the inner layer anti-interference parameter decision network, and the corresponding expressions are as follows:

[0054]

[0055] Wherein, R o represents the anti-interference reward function of the outer layer anti-interference pattern decision network, J t+1 and J t represent the interference signals suffered at t+1 and the current t respectively, JSR t+1 and JSR t represent the jamming-to-signal ratios of the interference signals suffered at t+1 and the current t respectively,

[0056] R I = 10R ISR (J t ,a t ),

[0057] represents the anti-interference reward function of the inner layer anti-interference parameter decision network, R ISR (J t ,at represents the interference signal J suffered by the radar t Anti-jamming pattern is executed the change rate of the interference energy suppression ratio R ISR (J t , a t The calculation formula is:

[0058]

[0059] represents the anti-jamming action, the anti-jamming pattern executed at the current time t and the anti-jamming parameter together constitute, and respectively represent the maximum value of the residual interference signal amplitude in the radar echo and the maximum value of the target signal amplitude after executing the anti-jamming action.

[0060] S42, construct a training loss function.

[0061] Optionally, the training loss function is constructed, comprising:

[0062] The training loss function of the outer anti-jamming pattern decision network is constructed, and the corresponding expression is:

[0063]

[0064] Wherein, L outer represents the training loss function of the outer anti-jamming pattern decision network, represents the anti-jamming reward of the outer anti-jamming pattern decision network, and respectively represent the output of the outer anti-jamming pattern decision network at time t and t+1, θ represents the parameter of the outer anti-jamming pattern decision network, and γ represents the first discount factor, the value range is [0, 1];

[0065] The training loss function of the inner anti-jamming parameter decision network is constructed:

[0066]

[0067] Wherein, L outer represents the training loss function of the inner anti-jamming parameter decision network, represents the anti-jamming reward of the inner anti-jamming parameter decision network, and respectively represent the output of the inner anti-jamming parameter decision network at time t and t+1, σ represents the parameter of the inner anti-jamming parameter decision network, and λ represents the second discount factor, the value range is [0, 1].

[0068] S43, set the training parameters of the complex domain layered radar anti-jamming pattern joint parameter decision network, the training parameters including the iteration number M, the network learning rate setting, the batch size, and the update frequency interval H of the outer anti-jamming pattern decision network in the layered anti-jamming decision model HCD2QNet.

[0069] For example, the iteration number M is set to 10000, the network learning rate is set to 0.0002, the batch size is 64, and the update frequency interval H of the outer anti-jamming pattern decision network in the layered anti-jamming decision model HCD2QNet is set to 10.

[0070] S44, based on the anti-jamming reward function of the outer anti-jamming pattern decision network, the anti-jamming reward function of the inner anti-jamming parameter decision network, the training loss function, and the training parameters of the complex domain layered radar anti-jamming pattern joint parameter decision network, receive the active jamming signal transmitted by the jammer, obtain the radar echo with complex jamming signal according to the received signal, and train the complex domain layered radar anti-jamming pattern joint parameter decision network according to the radar echo with complex jamming signal.

[0071] Optionally, based on the anti-jamming reward function of the outer anti-jamming pattern decision network, the anti-jamming reward function of the inner anti-jamming parameter decision network, the training loss function, and the training parameters of the complex domain layered radar anti-jamming pattern joint parameter decision network, receive the active jamming signal transmitted by the jammer, obtain the radar echo with complex jamming signal according to the received signal, and train the complex domain layered radar anti-jamming pattern joint parameter decision network according to the radar echo with complex jamming signal, including:

[0072] Iteratively execute the following steps S81-S84, and execute step S85 once every H times after executing S83, until the set iteration number M is reached, H being the update frequency interval of the outer anti-jamming pattern decision network in the layered anti-jamming decision model HCD2QNet;

[0073] S81, receive the active jamming signal transmitted by the jammer at the current time t, and obtain the radar echo with complex jamming signal S t , and input the radar echo with complex jamming signal S t into the complex domain interference feature extraction network to extract the interference environment state vector V t , and input the interference environment state vector V t into the outer anti-jamming pattern decision network of the layered anti-jamming decision model HCD2QNet, and determine the anti-jamming pattern in the maximum value output by the outer anti-jamming pattern decision network as the optimal anti-jamming pattern this time.

[0074] S82, input the interference environment state vector V tThe inner-layer anti-jamming parameter decision network in the hierarchical anti-jamming decision model HCD2QNet combined with the current best anti-jamming pattern is used to determine the anti-jamming parameter in the maximum value output by the inner-layer anti-jamming parameter decision network as the current anti-jamming parameter index, and determine the current best anti-jamming parameter according to the current best anti-jamming pattern and the current anti-jamming parameter index, and the correspondence between the anti-jamming pattern, the anti-jamming parameter index and the anti-jamming parameter.

[0075] Optionally, the correspondence between the anti-jamming pattern, the anti-jamming parameter index and the anti-jamming parameter is shown in Table 3:

[0076] Table 3

[0077]

[0078]

[0079] S83, re-receive the active jamming signal transmitted by the jammer, and obtain a new radar echo S with complex interference signals according to the received new signal t+1 .

[0080] S84, according to the current best anti-jamming pattern, the current best anti-jamming parameter, the radar echo S with complex interference signals t and the radar echo S with complex interference signals t+1 , determine the network loss value of the training loss function of the inner-layer anti-jamming parameter decision network, and obtain the first gradient according to the network loss value of the training loss function of the inner-layer anti-jamming parameter decision network and the gradient descent algorithm, and update the network parameters by propagating the first gradient to the inner-layer anti-jamming parameter decision network in the hierarchical anti-jamming decision model HCD2QNet through the Adam optimization algorithm.

[0081] S85, according to the current best anti-jamming pattern, the radar echo S with complex interference signals t and the radar echo S with complex interference signals t+1 , determine the network loss value of the training loss function of the outer-layer anti-jamming pattern decision network, and obtain the second gradient according to the network loss value of the training loss function of the outer-layer anti-jamming pattern decision network and the gradient descent algorithm, and update the network parameters by propagating the second gradient to the outer-layer anti-jamming pattern decision network and the complex domain interference feature extraction network of the hierarchical anti-jamming decision model HCD2QNet through the Adam optimization algorithm.

[0082] The application provides a training method of a complex domain layered radar anti-interference pattern combined parameter decision network, which comprises the following steps: constructing a complex domain interference feature extraction network, wherein the complex domain interference feature extraction network comprises a complex interference echo real part network branch and a complex interference echo imaginary part network branch, the complex interference echo real part network branch is used for extracting echo real part features, and the complex interference echo imaginary part network branch is used for extracting echo imaginary part features; constructing a layered anti-interference decision model HCD2QNet, wherein the layered anti-interference decision model HCD2QNet comprises an outer-layer anti-interference pattern decision network and an inner-layer anti-interference parameter decision network, the outer-layer anti-interference pattern decision network is used for determining an optimal anti-interference pattern according to an interference environment state vector output by the complex domain interference feature extraction network, and the inner-layer anti-interference parameter decision network is used for determining optimal anti-interference parameters according to the interference environment state vector output by the complex domain interference feature extraction network and the optimal anti-interference pattern; and cascading the complex domain interference feature extraction network and the layered anti-interference decision model HCD2QNet to obtain the complex domain layered radar anti-interference pattern combined parameter decision network, wherein the output of the complex domain interference feature extraction network is used as the input of the outer-layer anti-interference pattern decision network of the layered anti-interference decision model HCD2QNet, and the output of the complex domain interference feature extraction network and the output of the outer-layer anti-interference pattern decision network of the layered anti-interference decision model HCD2QNet are used as the input of the inner-layer anti-interference parameter decision network; receiving active interference signals transmitted by an interference machine, obtaining radar echoes with complex interference signals according to the received signals, and training the complex domain layered radar anti-interference pattern combined parameter decision network according to the radar echoes with complex interference signals, so that the obtained complex domain layered radar anti-interference pattern combined parameter decision network avoids separation of interference feature extraction and anti-interference decision steps, retains the correlation between the two, realizes perception decision integration from original radar echo signals to anti-interference measures, enhances the flexibility and efficiency of radar anti-interference decision confrontation, and has the advantage of being widely applied to complex time-varying intelligent interference scenes.

[0083] The application further provides a flowchart of a complex domain layered radar anti-interference pattern combined parameter decision method, which comprises the following steps 1-3.

[0084] 1. Receiving active interference signals transmitted by an interference machine to obtain radar echoes with complex interference signals S t ′.

[0085] 2. Inputting the radar echoes with complex interference signals S t ′ into a complex domain layered radar anti-interference pattern combined parameter decision network to obtain optimal anti-interference patterns and optimal anti-interference parameters, wherein the complex domain layered radar anti-interference pattern combined parameter decision network is obtained by training any complex domain layered radar anti-interference pattern combined parameter decision network provided in the above embodiment.

[0086] 3. Implementing the optimal anti-interference pattern and the optimal anti-interference parameter to complete the countermeasure.

[0087] The effects of the present application can be further illustrated by the following simulation experiment.

[0088] 1. Simulation conditions

[0089] The hardware platform of the simulation experiment of the present application is: Intel(R) Core(TM) i7-10700 CPU, 2.90GHz, memory is 64G, and GPU is NVIDIA GeForce RTX 3090.

[0090] The software platform of the simulation experiment of the present application is: Pycharm2021.

[0091] 2. Simulation results and analysis

[0092] Under the above simulation conditions, the radar echo subjected to linear frequency shift interference is counteracted 5000 times using the present application, and the results of the HCD2QNet decision of the optimal anti-interference pattern and the anti-interference parameter are shown in Figure 3 and Table 4. Among them Figure 3 Figures (a) and (b) in Table 4 are the yield curves of the outer anti-interference decision network and the inner anti-interference parameter, respectively, and Table 4 is the anti-interference pattern and specific parameter decided by 5000 times of countermeasure.

[0093] Table 4

[0094]

[0095] From Figure 3 it can be seen that after 1000 times of countermeasure, HCD2QNet can quickly realize high-yield decision, that is, select the anti-interference pattern and parameter with high reward value. The results in Table 4 show that HCD2QNet can accurately decide the optimal anti-interference pattern and anti-interference parameter.

[0096] In summary, the complex domain layered radar anti-interference pattern joint parameter decision method provided by the present application realizes the integration of perception and decision of the original radar echo signal to the anti-interference measure. Through the outer interference pattern decision network in the layered anti-interference decision model HCD2QNet, the interference pattern is selected, and through the inner anti-interference parameter decision network, the anti-interference parameter is selected, realizing the joint decision of the optimal anti-interference pattern and parameter, and improving the accuracy, efficiency and flexibility of the radar anti-interference decision.

[0097] The application can be applied to synthetic aperture, frequency agile and other radar platforms, and can perform efficient and accurate anti-jamming pattern combined anti-jamming parameter decision on various types of jamming signals, and improve the flexibility and intelligent degree of radar jamming countermeasures.

[0098] The application also provides a structure of an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus,

[0099] The memory is used for storing a computer program.

[0100] The processor is used for executing the program stored on the memory, and realizes the steps provided in the method embodiments.

[0101] The communication interface is used for communication between the electronic device and other devices.

[0102] For the electronic device embodiment, because it is basically similar to the method embodiment, the description is relatively simple, and the specific content, beneficial effects and related parts are referred to the part of the method embodiment.

[0103] The terms "first", "second" are only used for description purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second" can explicitly or implicitly include one or more features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0104] The above is a further detailed description of the application in combination with specific preferred embodiments, and the specific implementation of the application cannot be limited to these descriptions. For ordinary skilled persons in the art to which the application belongs, without departing from the concept of the application, a number of simple deductions or replacements can be made, and all should be regarded as falling within the protection scope of the application.

Claims

1. A training method of a complex domain layered radar anti-jamming pattern joint parameter decision network, characterized in that, Comprise: S1, construct a complex domain interference feature extraction network, the complex domain interference feature extraction network comprises a complex interference echo real part network branch and a complex interference echo imaginary part network branch, the complex interference echo real part network branch is used for extracting echo real part features, and the complex interference echo imaginary part network branch is used for extracting echo imaginary part features; S2, a hierarchical anti-jamming decision model HCD2QNet is constructed, the hierarchical anti-jamming decision model HCD2QNet comprises an outer layer anti-jamming pattern decision network and an inner layer anti-jamming parameter decision network, the outer layer anti-jamming pattern decision network is used for determining the best anti-jamming pattern according to the interference environment state vector of the output of the complex domain interference feature extraction network, and the inner layer anti-jamming parameter decision network is used for determining the best anti-jamming parameter according to the interference environment state vector of the output of the complex domain interference feature extraction network and the best anti-jamming pattern; The outer layer anti-jamming pattern decision network comprises a first full connection layer, a second full connection layer, a third full connection layer, a fourth full connection layer and a first element addition module; wherein the output of the first full connection layer is the input of the second full connection layer; the outputs of the second full connection layer are respectively the inputs of the third full connection layer and the fourth full connection layer; the output of the third full connection layer and the output of the fourth full connection layer are subjected to element addition operation in the first element addition module; The inner layer anti-jamming parameter decision network comprises a fifth full connection layer, a sixth full connection layer, a seventh full connection layer, an eighth full connection layer and a second element addition module; wherein the output of the first element addition module and the output of the fifth full connection layer are the input of the sixth full connection layer; the outputs of the sixth full connection layer are respectively the inputs of the seventh full connection layer and the eighth full connection layer; the output of the seventh full connection layer and the output of the eighth full connection layer are subjected to element addition operation in the second element addition module; S3, the complex domain interference feature extraction network and the hierarchical anti-jamming decision model HCD2QNet are cascaded to obtain a complex domain hierarchical radar anti-jamming pattern joint parameter decision network, wherein the output of the complex domain interference feature extraction network is used as the input of the outer layer anti-jamming pattern decision network of the hierarchical anti-jamming decision model HCD2QNet, and the output of the complex domain interference feature extraction network and the output of the outer layer anti-jamming pattern decision network of the hierarchical anti-jamming decision model HCD2QNet are used as the input of the inner layer anti-jamming parameter decision network; S4, receiving the active jamming signal transmitted by a jammer, obtaining the radar echo with the complex interference signal according to the received signal, and training the complex domain hierarchical radar anti-jamming pattern joint parameter decision network according to the radar echo with the complex interference signal.

2. The training method of claim 1, wherein, The complex interference echo real part comprises: A first convolutional pooling layer, a second convolutional pooling layer and a first global maximum pooling layer are connected in sequence.

3. The training method of claim 2, wherein, The complex interference echo imaginary part comprises: The first global maximum pooling layer in the real part of the complex interference echo and the second global maximum pooling layer in the imaginary part of the complex interference echo are sequentially connected with the third convolutional pooling layer, the fourth convolutional pooling layer and the second global maximum pooling layer.

4. The method of claim 3, wherein, The S4 comprises: The anti-interference reward functions of the outer anti-interference mode decision network and the inner anti-interference parameter decision network are set, and the corresponding expressions are: wherein R o represents the outer layer anti-interference pattern decision network anti-interference reward function, J t+1 and J t respectively represent the interference signals suffered at t+1 time and current t time, JSR t+1 and JSR t respectively represent the interference to signal ratios of the interference signals suffered at t+1 time and current t time, R I = 10R ISR (J t , a t ), represents the inner-layer anti-jamming parameter decision network anti-jamming reward function, R ISR (J t , a t ) represents the interference signal J t executes the anti-jamming pattern after the interference energy suppression ratio change rate, R ISR (J t , a t ) calculation formula is: denotes the anti-jamming action, the anti-jamming pattern performed by the current time t and the anti-jamming parameter together constitute, and respectively denote the maximum value of the residual jamming signal amplitude and the maximum value of the target signal amplitude in the radar echo after performing the anti-jamming action The training loss function is constructed. The training parameters of the complex domain hierarchical radar anti-interference mode joint parameter decision network are set, and the training parameters comprise an iteration number M, a network learning rate setting, a batch size and an update frequency interval H of the outer anti-interference mode decision network in the hierarchical anti-interference decision model HCD2QNet. Based on the anti-interference reward functions of the outer anti-interference mode decision network and the inner anti-interference parameter decision network, the training loss function and the training parameters of the complex domain hierarchical radar anti-interference mode joint parameter decision network, an active jamming signal transmitted by a jammer is received, a radar echo with a complex interference signal is obtained according to the received signal, and the complex domain hierarchical radar anti-interference mode joint parameter decision network is trained according to the radar echo with the complex interference signal.

5. The method of claim 4, wherein, The training loss function is constructed. The training loss function of the outer anti-interference mode decision network is constructed, and the corresponding expression is: wherein L outer represents the training loss function of the outer anti-interference pattern decision network, represents the anti-interference reward of the outer anti-interference pattern decision network, and respectively represent the output of the outer anti-interference pattern decision network at time t and time t+1, θ represents the parameter of the outer anti-interference pattern decision network, and γ represents the first discount factor, which takes a value in the range [0, 1]. The training loss function of the inner anti-interference parameter decision network is constructed. wherein L outer represents the training loss function of the inner anti-interference parameter decision network, represents the anti-interference reward of the inner anti-interference parameter decision network, and respectively represent the output of the inner anti-interference parameter decision network at time t and time t+1, σ represents the parameter of the inner anti-interference parameter decision network, and λ represents the second discount factor, with a value range [0, 1].

6. The method of claim 5, wherein, The anti-interference reward functions of the outer anti-interference mode decision network and the inner anti-interference parameter decision network, the training loss function and the training parameters of the complex domain hierarchical radar anti-interference mode joint parameter decision network are based on, an active jamming signal transmitted by a jammer is received, a radar echo with a complex interference signal is obtained according to the received signal, and the complex domain hierarchical radar anti-interference mode joint parameter decision network is trained according to the radar echo with the complex interference signal, which comprises: The following steps S81-S84 are iteratively executed, and step S85 is executed once every H times after the execution of S83 is completed, until the set iteration number M is reached, and H is the set update frequency interval of the outer anti-interference mode decision network in the hierarchical anti-interference decision model HCD2QNet. S81, receive the active jamming signal transmitted by the jammer at the current time t to obtain a radar echo S with a complex jamming signal t and output the radar echo S with the complex jamming signal t input the complex domain jamming feature extraction network to extract an interference environment state vector V t and output the interference environment state vector V t input the outer anti-jamming pattern decision network of the hierarchical anti-jamming decision model HCD2QNet, and determine the anti-jamming pattern in the maximum value output by the outer anti-jamming pattern decision network as the optimal anti-jamming pattern this time S82, input the interference environment state vector V t input the inner-layer anti-interference parameter decision network in the hierarchical anti-interference decision model HCD2QNet in combination with the current optimal anti-interference pattern, determine the anti-interference parameter in the maximum value output by the inner-layer anti-interference parameter decision network as the current anti-interference parameter index, and determine the current optimal anti-interference parameter according to the current optimal anti-interference pattern, the current anti-interference parameter index, and the correspondence relationship among anti-interference patterns, anti-interference parameter indexes, and anti-interference parameters; S83, re-receiving the active jamming signal transmitted by the jammer, according to the received new signal, obtaining a new radar echo with complex interference signal t+1 ; S84, according to the best anti-interference pattern this time, the best anti-interference parameter this time, the radar echo S with multiple interference signals t and the radar echo S with multiple interference signals t+1 , determine the network loss value of the training loss function of the inner layer anti-interference parameter decision network, obtain the first gradient according to the network loss value and gradient descent algorithm of the training loss function of the inner layer anti-interference parameter decision network, and update the network parameters by propagating the first gradient to the inner layer anti-interference parameter decision network in the hierarchical anti-interference decision model HCD2QNet through the Adam optimization algorithm. S85, according to the best anti-interference pattern this time, the radar echo S with multiple interference signals t and the radar echo S with multiple interference signals t+1 , determine the network loss value of the training loss function of the outer anti-interference pattern decision network, obtain the second gradient according to the network loss value of the training loss function of the outer anti-interference pattern decision network and the gradient descent algorithm, and propagate the second gradient to the outer anti-interference pattern decision network of the hierarchical anti-interference decision model HCD2QNet and the complex domain interference feature extraction network through the Adam optimization algorithm to update the network parameters.

7. A method for complex domain layered radar anti-jamming pattern joint parameter decision, characterized in that, Comprise: receiving the active jamming signal transmitted by the jammer to obtain a radar echo S with complex jamming signal t ′ S t The method comprises: inputting the radar echo S with the plurality of interference signals into a complex domain layered radar anti-interference pattern joint parameter decision network to obtain an optimal anti-interference pattern and an optimal anti-interference parameter, wherein the complex domain layered radar anti-interference pattern joint parameter decision network is obtained through the training method of the complex domain layered radar anti-interference pattern joint parameter decision network according to any one of claims 1-6. The optimal anti-interference mode and the optimal anti-interference parameter are implemented to complete the confrontation.

8. An electronic device, comprising: The processor, the communication interface, the memory and the communication bus are connected through the communication bus to realize mutual communication. The memory is used to store the computer program. The processor is used to execute the program stored in the memory to realize the method in any one of claims 1-7.

Images