Deep learning-based cooperative jamming method, electronic device, and storage medium

Abstract

The application discloses a deep learning-based cooperative jamming method, an electronic device and a storage medium, relates to the technical field of radar cooperative jamming, and comprises the following steps: collecting and analyzing target radar echoes, and constructing an observation signal feature tensor and a jamming platform reliability coefficient; constructing an ideal signal feature tensor based on an ideal standard, and performing cross-modal implicit alignment on the ideal signal feature tensor and the observation tensor, and extracting a double inconsistency factor; calling a cooperative jamming strategy generator to analyze the double inconsistency factor, and outputting an initial cooperative jamming strategy; and combining the platform reliability coefficient to cooperatively schedule a power spectrum, and generating a target cooperative jamming strategy. The application solves the technical problems that the existing radar jamming method has poor adaptability and low jamming efficiency when dealing with modern radar systems, achieves the technical effects of extracting a double inconsistency factor through cross-modal implicit alignment and generating a cooperative jamming strategy, and improves the jamming adaptability and jamming efficiency for modern radar systems.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of radar cooperative jamming, in particular to a cooperative jamming method based on deep learning, an electronic device and a storage medium. BACKGROUND

[0002] In the field of radar countermeasures, traditional jamming methods mostly rely on single jamming platforms and fixed jamming modes, usually affecting target radars through noise suppression or simple deceptive signals. However, with the diversification of radar systems and the continuous enhancement of anti-jamming capabilities, single feature dimension analysis and static parameter configuration have been difficult to cope with dynamic changes in complex electromagnetic environments. Existing technologies often use explicit alignment and empirical modeling when extracting differences between radar echoes and ideal signals, which is difficult to fully capture the implicit differences in frequency domain power distribution and time sequence phase trajectory, resulting in insufficient pertinence of jamming strategies. In addition, traditional jamming methods are often inefficient and difficult to adapt to the complexity and intelligence of modern radar systems. SUMMARY

[0003] The present application provides a cooperative jamming method based on deep learning, an electronic device and a storage medium, which solves the technical problems of poor adaptability and low jamming efficiency of existing radar jamming methods in dealing with modern radar systems.

[0004] In a first aspect, the present application provides a cooperative jamming method based on deep learning, which comprises: collecting the echo of the transmitted signal of the target jamming radar through a set of jamming platforms, obtaining a set of radar echo sequences, and performing time-frequency spectrum energy distribution and range-Doppler-azimuth feature integration analysis on the set of radar echo sequences to determine an observed signal feature tensor and a set of jamming platform reliability coefficients; determining an ideal signal feature tensor based on the set of radar echo sequences and the ideal standard of pulse compression, performing cross-modal implicit alignment between the ideal signal feature tensor and the observed signal feature tensor, and extracting a set of double inconsistent factors, wherein the set of double inconsistent factors includes a first inconsistent factor and a second inconsistent factor; calling a cooperative jamming strategy generator to analyze the first inconsistent factor and the second inconsistent factor, and outputting an initial cooperative jamming strategy, wherein the cooperative jamming strategy generator is constructed based on deep learning, and the cooperative jamming strategy includes jamming type, jamming frequency range, sub-pulse selection, phase modulation mode and time sequence offset; combining the set of jamming platform reliability coefficients and the initial cooperative jamming strategy to perform power spectrum cooperative scheduling of the set of jamming platforms, obtaining a target cooperative jamming strategy, and performing cooperative jamming on the target jamming radar based on the target cooperative jamming strategy.

[0005] In a second aspect of the present application, an electronic device is provided, comprising: a processor coupled with a memory, the memory being configured to store a program, when the program is executed by the processor, to perform the method of any one of the first aspect.

[0006] In a third aspect of the present application, a computer readable storage medium is provided, the storage medium having stored thereon a computer program, the computer program being executed by a processor to implement the method of the first aspect.

[0007] The one or more technical solutions provided in the present application have at least the following technical effects or advantages:

[0008] The deep learning-based cooperative jamming method, electronic device and storage medium provided by the present application are related to the technical field of radar cooperative jamming, and through constructing an ideal signal and an observation signal feature tensor and performing cross-modal implicit alignment, a dual inconsistency factor of frequency domain and time sequence difference is extracted, a multi-dimensional jamming strategy is generated by inputting a cooperative jamming strategy generator, and the power spectrum is cooperatively scheduled in combination with a jamming platform reliability coefficient, thereby solving the technical problems of poor adaptability and low jamming efficiency of existing radar jamming methods in dealing with modern radar systems, and achieving the technical effects of extracting a dual inconsistency factor through cross-modal implicit alignment and generating a cooperative jamming strategy, improving the jamming adaptability and jamming efficiency for modern radar systems. BRIEF DESCRIPTION OF DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0010] Figure 1 A deep learning-based cooperative jamming method flowchart is provided for the embodiments of the present application.

[0011] Figure 2 A structural schematic diagram of an electronic device is provided for the present application.

[0012] Explanation of reference signs: electronic device 300, memory 301, processor 302, communication interface 303, bus architecture 304. DETAILED DESCRIPTION

[0013] The deep learning-based cooperative jamming method, electronic device and storage medium provided by the present application are used to solve the technical problems of poor adaptability and low jamming efficiency of existing radar jamming methods in dealing with modern radar systems.

[0014] With reference to the drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of the present application.

[0015] It should be noted that the terms "first", "second", etc. in the specification of the present application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or server including a series of steps or units does not have to be limited to only those steps or units clearly listed, but can include other steps or modules not clearly listed or inherent to these processes, methods, products or devices.

[0016] Embodiment one, as shown in the present application provides a deep learning based cooperative jamming method, which comprises: Figure 1

[0017] P10: Collecting the echo of the transmission signal of the target jamming radar by the jamming platform set, obtaining a radar echo sequence set, and performing time-frequency spectrum energy distribution and range-Doppler-azimuth feature integration analysis on the radar echo sequence set to determine an observation signal feature tensor and a jamming platform reliability coefficient set.

[0018] Further, the step P10 of the embodiments of the present application further comprises:

[0019] P11: performing fast Fourier transform on the radar echo signal sequence set to obtain a time-frequency spectrum energy distribution set; P12: performing pulse Doppler processing and azimuth estimation on the radar echo signal sequence set to generate a range-Doppler-azimuth feature set; P13: mapping and encoding the time-frequency spectrum energy distribution set and the range-Doppler-azimuth feature set to determine the observation signal feature tensor, and iteratively identifying the signal reception quality of the radar echo signal sequence set to obtain the jamming platform reliability coefficient set.

[0020] It should be understood that the echo of the transmission signal of the target jamming radar is collected by the jamming platform set, the radar echo sequence set is obtained, and multi-dimensional analysis is performed on the radar echo sequence set to generate an observation signal feature tensor and a jamming platform reliability coefficient set for subsequent processing. ​

[0021] First, the radar echo signal sequence set is processed, and the corresponding time-frequency spectrum energy distribution set is obtained by using Fast Fourier Transform (FFT). Fast Fourier Transform is an efficient algorithm for converting time-domain signals to frequency domain, which can reveal the energy distribution of signals at different frequency components. Through this step, the energy change rule of the target radar echo signal in time and frequency dimensions can be effectively captured, thereby providing a mathematical basis for identifying abnormal frequency components and potential interference traces. The time-frequency spectrum energy distribution here refers to the joint energy representation of the signal on the time axis and the frequency axis, which essentially describes the distribution matrix of signal energy at different time-frequency positions.

[0022] Second, the radar echo signal sequence set is further processed by pulse Doppler and azimuth estimation to generate a range-Doppler-azimuth feature set. Pulse Doppler processing is a typical radar signal processing technique used to obtain the radial velocity information of the target through pulse compression and Doppler filtering; azimuth estimation is based on array antenna or beamforming technology to infer the azimuth angle information of the target. By iterating the radar echo signal sequence set, pulse Doppler processing and azimuth estimation are performed to generate a range-Doppler-azimuth feature set. This set contains the key features of the target's range, Doppler frequency, and azimuth angle, and is an important basis for constructing the multi-dimensional feature space of the observation signal.

[0023] Subsequently, the time-frequency spectrum energy distribution set obtained above is mapped and encoded with the range-Doppler-azimuth feature set to determine the observation signal feature tensor. The feature tensor is a multi-dimensional data structure that simultaneously carries the multi-dimensional feature information of the signal in the time domain, frequency domain, and spatial domain. Compared with the traditional two-dimensional matrix, it can more comprehensively represent the multi-modal features of the target radar signal. At the same time, the radar echo signal sequence set is iterated for signal reception quality identification to obtain a set of interference platform reliability coefficients. The interference platform reliability coefficient is a quantitative indicator of the stability of the interference platform's transmission and reception link, including power coverage, interference signal consistency, and time delay controllability. The power coverage reflects the power range and coverage ability of the interference platform; the interference signal consistency measures the stability and consistency of the interference signal; the time delay controllability represents the adjustability of the interference signal time delay. Through comprehensive analysis of the above coefficients, the reliability of different interference platforms can be weighted in the subsequent interference strategy generation and scheduling process, thereby improving the efficiency and stability of the overall coordinated interference.

[0024] P20: determining an ideal signal feature tensor based on the radar echo sequence set and the pulse compression ideal standard, performing cross-modal implicit alignment between the ideal signal feature tensor and an observed signal feature tensor, and extracting a set of double inconsistency factors, wherein the set of double inconsistency factors includes a first-order inconsistency factor and a second-order inconsistency factor.

[0025] Further, the step P20 of the embodiments of the present application further includes:

[0026] P21: performing radar system parameter identification on the radar echo sequence set to obtain target interference radar system parameters, and determining an ideal signal feature tensor in combination with the pulse compression ideal standard; P22: performing spatial embedding mapping of the ideal signal feature tensor and the observed signal feature tensor through a shared encoder to obtain an ideal signal feature vector and an observed signal feature vector; P23: performing cross-modal implicit alignment and double difference analysis on the ideal signal feature vector and the observed signal feature vector to obtain the set of double inconsistency factors.

[0027] Optionally, a multi-dimensional correspondence is established between the ideal signal and the observed signal, and a set of key interference factors is extracted through double inconsistency analysis, so as to provide input basis for subsequent interference strategy generation and scheduling.

[0028] Firstly, radar system parameter identification is performed on the radar echo sequence set, which is a prerequisite for determining the ideal signal feature tensor. The radar system parameters include operating frequency, pulse width, pulse repetition frequency, etc., which directly affect the characteristics of the radar signal. After obtaining the above parameters, the ideal signal feature tensor can be constructed in combination with the pulse compression ideal standard, that is, a multi-dimensional signal representation structure generated under the ideal conditions of no interference and no noise according to the radar signal model (such as SAR imaging process or pulse compression output). The feature tensor is used as a reference standard to compare the deviation of the actual observed signal.

[0029] Next, the ideal signal feature tensor and the observed signal feature tensor obtained in the previous step are input into the shared encoder for spatial embedding mapping. The shared encoder is a multi-layer nonlinear mapping network constructed based on deep learning, which can extract and compress features of different signal tensors through unified weight parameters, and embed the original high-dimensional signal data into a latent feature space. By using the shared encoder, the ideal signal feature vector and the observed signal feature vector can be obtained, and they are projected into the same embedding space, so that the deviation caused by inconsistent feature distribution is reduced on the basis of maintaining the integrity of the original semantics, thereby ensuring the comparability between signals of different sources and providing a unified data representation form for subsequent difference analysis.

[0030] Subsequently, cross-modal implicit alignment and double difference analysis are performed on the obtained ideal signal feature vector and the observed signal feature vector. Cross-modal implicit alignment is a way of matching in the latent feature space. By measuring the similarity of the distribution in the latent space, the corresponding relationship between signals in different feature spaces is found to realize the implicit mapping relationship between different signal modalities, thereby revealing the inconsistency between signals.

[0031] After completing the implicit alignment, further double difference analysis is carried out, that is, the above inconsistency is further quantified and decomposed into two main inconsistent factor sets. The double inconsistent factor set includes two levels of inconsistent features. The first inconsistent factor mainly reflects the energy focusing deviation of the signal in the imaging domain, such as resolution reduction, false enhancement or disappearance of echo scattering points, etc., representing the difference between the spatial focusing effect of the observed signal and the ideal imaging result. The second inconsistent factor reflects the deviation of the signal in the frequency spectrum and phase characteristics, such as delay drift, phase jitter, and false positioning of scatterers, etc., representing the abnormal difference of the observed signal in the frequency domain consistency and phase trajectory. Through the extraction and induction of the above two types of inconsistent factors, the deviation degree of the target jamming radar echo signal compared with the ideal standard can be fully described, so as to obtain a high-discriminative interference factor set, providing accurate and quantifiable input conditions for subsequent deep learning-based interference strategy generation.

[0032] Further, the step P23 of the embodiment of the present application further includes:

[0033] P23-1: constructing a cross-modal relationship matrix for the ideal signal feature vector and the observed signal feature vector, and performing pooling dimension reduction weight analysis on the obtained cross-modal relationship matrix to obtain a cross-modal attention weight; P23-2: multiplying the cross-modal attention weight with the observed signal feature vector to obtain an implicit alignment feature vector; P23-3: performing frequency domain power distribution difference identification based on the implicit alignment feature vector and the ideal signal feature vector to obtain a first inconsistent factor; P23-4: performing time sequence phase trajectory difference identification based on the implicit alignment feature vector and the ideal signal feature vector to obtain a second inconsistent factor; P23-5: summarizing the first inconsistent factor and the second inconsistent factor to obtain a double inconsistent factor set.

[0034] Specifically, the extraction process of the inconsistency between the ideal signal feature vector and the observed signal feature vector can be further refined.

[0035] Firstly, the cross-modal relationship matrix is constructed between the ideal signal feature vector and the observed signal feature vector. The cross-modal relationship matrix is used to measure the mutual dependence and matching degree between different modal features, and the matrix elements are calculated by the similarity function (such as cosine similarity, inner product or weighted similarity function) between vectors. In order to reduce the redundancy and computational complexity of the relationship matrix in high-dimensional space, the obtained cross-modal relationship matrix is further reduced by dimensionality reduction and weight analysis is carried out in the process to extract the importance weight of different relationship elements, thereby obtaining the cross-modal attention weight. The attention weight can highlight the feature area with the most discriminative degree in the comparison between the ideal and observed signals, and ensure that the key features are given enough attention.

[0036] Next, the cross-modal attention weight is multiplied element by element with the observed signal feature vector to adjust the weight of each feature in the observed signal feature vector, so that it is closer to the ideal signal feature vector, thereby obtaining the implicit alignment feature vector. In this way, the attention weight can be used to strengthen the part highly related to the ideal signal in the feature space, while suppressing the feature area without significance, so that the observed signal forms a closer implicit alignment relationship with the ideal signal in the latent space.

[0037] Next, based on the comparison between the implicit alignment feature vector and the ideal signal feature vector, the frequency domain power distribution difference is identified, and the first inconsistent factor is obtained by comparing the power distribution in the frequency domain. The frequency domain power distribution difference mainly manifests as the deviation of the two in the spectral energy concentration, bandwidth consistency and peak power distribution, such as resolution reduction, spectral peak position drift or energy diffusion. These features can reflect the gap between the energy focusing effect of the target radar echo signal and the ideal standard.

[0038] Further, based on the implicit alignment feature vector and the ideal signal feature vector, the timing phase trajectory difference is identified, and the second inconsistent factor is obtained by comparing the phase trajectory in time sequence. The timing phase trajectory difference mainly manifests as delay drift, phase jitter, sudden phase inversion and false scattering point positioning. This kind of factor can reveal the deviation of the target radar signal in the time and phase evolution process, thereby depicting the abnormal performance of the signal in the dynamic characteristics.

[0039] Finally, the first inconsistent factor and the second inconsistent factor are summarized and integrated to form a double inconsistent factor set. The set simultaneously contains the frequency domain power distribution difference and the timing phase trajectory difference in a structured manner, which can provide high discriminant input features for the subsequent deep learning interference strategy generator, thereby realizing accurate interference strategy design based on double inconsistent modeling.

[0040] Further, the step P23-3 of the embodiment of the present application further comprises:

[0041] P23-31: constructing an energy spectrum Gaussian distribution model of the implicit alignment feature vector and an energy spectrum Gaussian distribution model of the ideal signal feature vector based on the frequency domain and power distribution features of the implicit alignment feature vector and the ideal signal feature vector;

[0042] Optionally, the process of identifying the frequency domain power distribution difference based on the implicit alignment feature vector and the ideal signal feature vector can be further refined.

[0043] First, based on the frequency domain features and power distribution features of the implicit alignment feature vector and the ideal signal feature vector, corresponding energy spectrum Gaussian distribution models are constructed. These two models respectively represent the energy distribution of the implicit alignment feature vector and the ideal signal feature vector in the frequency domain. Among them, the energy spectrum Gaussian distribution model of the implicit alignment feature vector reflects the energy distribution features of the observed signal in the frequency domain after cross-modal implicit alignment processing. The energy spectrum Gaussian distribution model of the ideal signal feature vector represents the frequency energy distribution of the radar signal under ideal conditions, i.e., without interference and other noise. By constructing the energy spectrum Gaussian distribution model of the implicit alignment feature vector, the actual frequency spectrum distribution of the observed signal under the interference environment can be obtained; and by constructing the energy spectrum Gaussian distribution model of the ideal signal feature vector, the standard frequency spectrum representation of the signal under the condition of no interference can be obtained. The combination of the two forms a pair of reference models for comparison.

[0044] Then, Kullback-Leibler divergence (abbreviated as KL divergence) is used to identify the distribution difference of the above two energy spectrum Gaussian distribution models. KL divergence is a commonly used probability distribution difference measurement method, which is used to measure the deviation of one distribution compared with another distribution. When the KL divergence value is larger, it means that the energy spectrum distribution of the implicit alignment feature vector and the energy spectrum distribution of the ideal signal feature vector are more different. By calculating the KL divergence between the energy spectrum Gaussian distribution model of the implicit alignment feature vector and the energy spectrum Gaussian distribution model of the ideal signal feature vector, the difference in frequency domain power distribution between the two can be identified.

[0045] These differences are quantified as a first-order inconsistency factor, whose output indicators include frequency drift amount, sub-pulse power deviation, and bandwidth coverage difference. Among them, the frequency drift amount reflects the degree of signal frequency deviation, the sub-pulse power deviation describes the deviation of sub-pulse power from the ideal state, and the bandwidth coverage difference measures the difference between the signal bandwidth and the ideal bandwidth. These indicators can reveal the energy focusing deviation in the radar imaging domain, such as resolution degradation, false enhancement or disappearance of scattering points, etc.

[0046] Through the above steps, not only can the frequency domain power distribution difference be qualitatively identified, but also quantitative analysis can be achieved through probability statistical modeling and information theory measurement, thereby ensuring that the first-order inconsistency factor has clear, verifiable and executable reference value in the subsequent interference strategy generation link.

[0047] Further, the step P23-4 of the embodiment of the present application further comprises:

[0048] P23-41: respectively projecting the implicit alignment feature vector and the ideal signal feature vector in the time dimension and the vector dimension to obtain a symbol timing trajectory sequence and a phase trajectory sequence; P23-42: using a shared weight twin network to perform high-dimensional feature extraction on the symbol timing trajectory sequence and the phase trajectory sequence to obtain an implicit alignment feature vector embedding representation and an ideal signal feature vector embedding representation; P23-43: performing difference identification on the obtained implicit alignment feature vector embedding representation and the ideal signal feature vector embedding representation to obtain a second-order inconsistency factor.

[0049] In a possible embodiment of the present application, the process of timing phase trajectory difference identification based on the implicit alignment feature vector and the ideal signal feature vector can be further refined.

[0050] First, respectively project the implicit alignment feature vector and the ideal signal feature vector in the time dimension and the vector dimension. Through this projection operation, the original high-dimensional vector can be decomposed in the time axis and the vector space, and then a symbol timing trajectory sequence and a phase trajectory sequence are obtained. The symbol timing trajectory sequence is used to represent the timing characteristics of the signal in the transmission process, such as symbol interval, symbol drift and symbol synchronization deviation; the phase trajectory sequence is used to describe the phase evolution law of the signal in the modulation and propagation process, such as phase accumulation, jitter and sudden phase inversion. These two sequences respectively represent the timing information and phase change information of the signal in time.

[0051] Then, the symbol timing trajectory sequence and the phase trajectory sequence are respectively subjected to high-dimensional feature extraction by a shared-weight twin network. The twin network is a special type of neural network that can learn the similarity or difference between two input sequences by sharing weights. In this embodiment, the twin network is used to extract high-dimensional features of the implicit alignment feature vector and the ideal signal feature vector, thereby obtaining their embedding representations, and outputting the embedding representation of the implicit alignment feature vector and the embedding representation of the ideal signal feature vector, respectively. The former contains the actual feature performance of the observation signal in the time sequence and phase dimensions, and the latter represents the standardized representation of the signal in the same dimensions under ideal conditions.

[0052] Finally, the difference between the obtained embedding representation of the implicit alignment feature vector and the embedding representation of the ideal signal feature vector is identified, that is, the difference between the two embedding representations is compared, thereby obtaining a second inconsistency factor. The difference identification process can be realized in various ways, such as calculating the Euclidean distance between vectors, the cosine similarity, or outputting the deviation degree through a classification discrimination network. In this process, if the difference between the implicit alignment feature and the ideal feature exceeds a certain threshold value, it indicates that there is a significant deviation in the timing accuracy or phase trajectory consistency of the signal, and then a second inconsistency factor can be generated. The second inconsistency factor mainly involves the mismatch of the spectrum and the phase trajectory, such as time delay drift, phase disturbance, or false positioning of scatterers. These problems may be caused by multipath effects in the signal propagation process, the influence of moving targets, or other interference sources, which pose challenges to the accurate analysis of radar signals.

[0053] The finally obtained second inconsistency factor provides a structured and quantifiable discrimination basis for feature extraction of the interference signal, and in combination with the first inconsistency factor, it can form a complete dual inconsistency factor set to provide more comprehensive input conditions for subsequent interference strategy generation.

[0054] P30: calling a cooperative jamming strategy generator to analyze the first inconsistency factor and the second inconsistency factor, and outputting an initial cooperative jamming strategy, wherein the cooperative jamming strategy generator is constructed based on deep learning, and the cooperative jamming strategy includes a jamming type, a jamming frequency range, a sub-pulse selection, a phase modulation mode, and a timing offset.

[0055] Further, the step P30 of the embodiment of the present application further includes:

[0056] P31: constructing a strategy analysis input feature vector based on the first inconsistency factor and the second inconsistency factor; P32: calling an encoder and a decoder of the cooperative jamming strategy generator to analyze the strategy analysis input feature vector, and obtaining the initial cooperative jamming strategy, wherein the decoder has multiple output heads.

[0057] It should be understood that the first and second heavy inconsistency factors obtained in the preceding steps are analyzed into an executable jamming strategy, so as to realize directional suppression and jamming of the target jamming radar.

[0058] In the specific implementation process, first, a strategy analysis input feature vector is constructed based on the aforementioned first and second heavy inconsistency factors. The feature vector integrates the frequency domain power distribution difference (including frequency drift, sub-pulse power deviation, and bandwidth coverage difference, etc.) and the time sequence phase trajectory difference (including delay drift, phase jitter, and false scatterer positioning deviation, etc.) through splicing, forming a multi-dimensional input that can comprehensively represent the difference between the observed signal and the ideal signal. The feature vector not only covers the deviation of the signal in the frequency domain and the time domain, but also has a measurable value in the numerical value, thereby ensuring that the subsequent strategy generation process has a stable input basis.

[0059] Then, in the strategy generation link, the strategy analysis input feature vector is input into the encoder and decoder structure of the cooperative jamming strategy generator. The encoder first performs nonlinear mapping and dimensionality reduction compression on the input feature, that is, converts the input feature vector into a representation form inside the model, extracts potential high-order feature representations while retaining key information. Subsequently, the decoder uses a multi-output head structure to complete the prediction of the jamming parameters on different channels. Each output head corresponds to a dimension of the jamming strategy, so that multiple parameters of the jamming scheme can be generated in parallel. Through this structure design, the independence and complementarity between different jamming parameters can be guaranteed, avoiding overfitting and coupling bias that may be caused by a single output path.

[0060] In the specific jamming strategy output by the decoder, first, the type of jamming is determined. The generator can select noise suppression jamming, imaging jamming, etc. according to the input feature. Then, the jamming frequency range is output, ensuring that the generated jamming signal has coverage and pertinence within the target radar operating frequency band. In the case of limited spectrum resources, the decoder also outputs sub-pulse or sub-band selection information, so as to realize selective jamming of part of the frequency band or part of the pulse, thereby improving the power utilization efficiency. At the same time, the generator outputs the phase modulation mode, which disturbs the jamming signal in the phase domain, such as phase randomization, inter-pulse phase shift, and false target phase modulation, etc., to enhance the concealment and confusion of the jamming. Finally, the decoder outputs the time sequence offset information, which controls the delay of the jamming signal based on the pulse repetition interval parameter, forming an erroneous distance or Doppler target, so that the target jamming radar has a deviation in the time sequence and velocity measurement.

[0061] Through the process, the cooperative jamming strategy generator can output a complete initial cooperative jamming strategy containing jamming type, jamming frequency range, sub-pulse selection, phase modulation mode and timing offset. The strategy has a directly executable parameter configuration format, can be called and implemented by the jamming platform set, and can be further optimized when combined with platform reliability coefficients for power and spectrum scheduling.

[0062] P40: Perform power spectrum cooperative scheduling of the jamming platform set in combination with the initial cooperative jamming strategy and the set of jamming platform reliability coefficients, obtain a target cooperative jamming strategy, and perform cooperative jamming on the target jamming radar based on the target cooperative jamming strategy.

[0063] Further, the step P40 of the embodiments of the present application further includes:

[0064] P41: Obtain a set of signal transmission capability thresholds of the jamming platform set, combine the overall power threshold and the frequency threshold in the initial cooperative jamming strategy to construct a cooperative scheduling constraint; P42: Based on the cooperative scheduling constraint, combine the set of jamming platform reliability coefficients to perform power spectrum cooperative scheduling of the jamming platform set, and obtain a target cooperative jamming strategy.

[0065] Specifically, on the basis of the initial cooperative jamming strategy, the reliability coefficients and resource constraints of each jamming platform are combined to perform cooperative scheduling of power and spectrum, thereby forming a final executable target cooperative jamming strategy, and the jamming platform set implements jamming on the target radar.

[0066] First, a set of signal transmission capability thresholds of the jamming platform set is obtained, which reflects the maximum transmission capability of each jamming platform at different frequencies and power levels, such as the upper limit performance of each platform in terms of maximum output power, spectrum coverage range, delay control capability, etc. At the same time, in combination with the overall power threshold and the frequency threshold set in the initial cooperative jamming strategy, a cooperative scheduling constraint condition is generated to ensure the overall feasibility of the jamming scheme in the power and frequency dimensions, and to ensure that the transmission capability of a single platform is not overused, thereby avoiding platform overload or unstable jamming effect.

[0067] Then, based on the cooperative scheduling constraint, the power spectrum of the interference platform set is scheduled with the maximum interference quality as the optimization goal. Specifically, first, the reliability coefficient is used to quantitatively evaluate the contribution weight of each platform, for example, higher power and spectrum resources are allocated to platforms with high power coverage, strong signal consistency, and good controllability of time delay; then, through optimization algorithm, the global scheduling of multi-platform resources is carried out under the overall constraint condition, so that the interference signal reaches the optimal combination in terms of power distribution, spectrum coverage and timing coordination. That is, under the premise of meeting all constraint conditions, the power and frequency resources of each interference platform are allocated to achieve the best interference effect.

[0068] Finally, according to the scheduling optimization result, a target cooperative interference strategy is formed, which specifies the specific power and frequency resource allocation for each interference platform, as well as the parameters such as interference type, sub-pulse selection, phase modulation mode and timing offset. These parameters jointly determine the specific implementation of cooperative interference.

[0069] Finally, based on the target cooperative interference strategy, the target radar is subjected to cooperative interference. In this stage, each interference platform implements interference operation synchronously according to the allocated resources and parameters. Through such cooperative interference, effective power coverage can be achieved in the frequency range of the target radar, and interference can be carried out on the selected sub-pulse or frequency band, thereby significantly improving the bit error rate and reducing the communication quality of the target radar.

[0070] In summary, the embodiments of the present application have at least the following technical effects:

[0071] Through cross-modal implicit alignment of radar echo sequence and ideal signal, the present application can extract two types of inconsistency factors, frequency domain power distribution difference and timing phase trajectory difference, to form double inconsistency features; the cooperative interference strategy generator based on deep learning can take double inconsistency factors as input, automatically analyze and output a multi-dimensional initial interference strategy covering interference type, interference frequency range, sub-pulse selection, phase modulation mode and timing offset; combined with the power and spectrum cooperative scheduling of the interference platform reliability coefficient, intelligent cooperation among multiple platforms can be realized, and the pertinence and execution efficiency of the interference scheme are improved.

[0072] The technical effects of improving the interference adaptability and interference efficiency of modern radar systems by extracting double inconsistency factors through cross-modal implicit alignment and generating a cooperative interference strategy are achieved.

[0073] Embodiment two, based on the same inventive concept as the cooperative interference method based on deep learning in the foregoing embodiments, the present application also provides a computer readable storage medium, the storage medium stores a computer program, and the computer program is executed by a processor to realize the method in embodiment one.

[0074] The person skilled in the art can clearly understand the deep learning based cooperative interference method, the electronic device and the storage medium in the embodiment according to the foregoing detailed description of the deep learning based cooperative interference method. Therefore, for the sake of brevity of the specification, no further description is given herein. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the related parts are described in the method part.

[0075] The foregoing description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0076] Embodiment three, exemplary electronic device.

[0077] The electronic device of the embodiments of the present application will be described below with reference to Figure 2

[0078] Based on the same inventive concept as the deep learning based cooperative interference method in the foregoing embodiments, the present application further provides a related electronic device, which comprises a processor coupled with a memory, the memory being configured to store a program, when the program is executed by the processor, to perform the steps of the method of embodiment one.

[0079] The electronic device 300 comprises a processor 302, a communication interface 303 and a memory 301. Optionally, the electronic device 300 can further comprise a bus architecture 304. The communication interface 303, the processor 302 and the memory 301 can be connected with each other through the bus architecture 304. The bus architecture 304 can be a peripheral component interconnect (PCI) bus or an extended industry Standard Architecture (EISA) bus, etc. The bus architecture 304 can be divided into an address bus, a data bus, a control bus, etc. For the sake of brevity, Figure 2 In the figure, only one thick line is used to represent the bus architecture 304, but it does not mean that there is only one bus or only one type of bus.

[0080] The processor 302 can be a CPU, a microprocessor, an ASIC, or one or more integrated circuits for controlling the execution of programs of the present application.

[0081] ​The communication interface 303, using any transceiver-like mechanism, is used to communicate with other devices or communication networks such as an Ethernet, a radio access network (RAN), a wireless local area network (WLAN), a wireline access network, etc.

[0082] The memory 301 can be a ROM, or other type of static storage device that can store static information and instructions; a RAM, or other type of dynamic storage device that can store information and instructions for execution by the processor; and / or an electrically erasable programmable read only memory (EEPROM), a compact disc read only memory (CD ROM) or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium capable of storing instructions or data that can be accessed by a computer, but not limited to. The memory can exist independently of the processor, and be connected to the processor via the bus architecture 304. The memory can also be integrated with the processor.

[0083] The memory 301 is configured to store computer-executable instructions for implementing the solutions of the present application, and the processor 302 is configured to control the execution of the computer-executable instructions stored in the memory 301. The processor 302 is configured to execute the computer-executable instructions stored in the memory 301, thereby implementing the deep learning-based cooperative interference method provided in the above-described embodiments of the present application.

[0084] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or can be advantageous.

[0085] The above only describes the preferred embodiments of the present application, and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

[0086] The specification and drawings are, of course, to be regarded in an illustrative rather than a restrictive sense. It is to be understood that any such modifications, variations, combinations or equivalents that fall within the scope of the application are intended to be embraced herein.

Claims

1. A deep learning based cooperative jamming method, characterized in that, The method comprises: Collecting the echo of the transmission signal of the target jamming radar by interfering with the platform set, obtaining a radar echo sequence set, and performing time-frequency spectrum energy distribution and distance-Doppler-azimuth feature integration analysis on the radar echo sequence set to determine an observation signal feature tensor and a jamming platform reliability coefficient set; Based on the radar echo sequence set and the ideal standard of pulse compression, an ideal signal feature tensor is determined, the ideal signal feature tensor and the observation signal feature tensor are cross-modal implicitly aligned, and a double inconsistency factor set is extracted, wherein the double inconsistency factor set includes a first inconsistency factor and a second inconsistency factor; An initial cooperative jamming strategy is output by calling a cooperative jamming strategy generator to analyze the first inconsistency factor and the second inconsistency factor, wherein the cooperative jamming strategy generator is constructed based on deep learning, and the cooperative jamming strategy includes the jamming type, the jamming frequency range, the sub-pulse selection, the phase modulation mode and the timing offset; The power spectrum of the jamming platform set is cooperatively scheduled in combination with the jamming platform reliability coefficient set and the initial cooperative jamming strategy to obtain a target cooperative jamming strategy, and the target jamming radar is cooperatively jammed based on the target cooperative jamming strategy; Wherein, based on the ideal standard of pulse compression, an ideal signal feature tensor is determined, the ideal signal feature tensor and the observation signal feature tensor are cross-modal implicitly aligned, and a double inconsistency factor set is extracted, wherein the double inconsistency factor set includes a first inconsistency factor and a second inconsistency factor, including: The radar system parameter identification is performed on the radar echo sequence set to obtain the target jamming radar system parameter, and the ideal signal feature tensor is determined in combination with the ideal standard of pulse compression; The ideal signal feature tensor and the observation signal feature tensor are spatially embedded and mapped by a shared encoder to obtain an ideal signal feature vector and an observation signal feature vector; The ideal signal feature vector and the observation signal feature vector are cross-modal implicitly aligned and double difference analyzed to obtain the double inconsistency factor set; Wherein, the ideal signal feature vector and the observation signal feature vector are cross-modal implicitly aligned and double difference analyzed to obtain the double inconsistency factor set, including: The cross-modal relationship matrix is constructed for the ideal signal feature vector and the observation signal feature vector, and the cross-modal relationship matrix is analyzed by pooling and dimension reduction weight to obtain cross-modal attention weight; The cross-modal attention weight is multiplied by the observation signal feature vector to obtain an implicitly aligned feature vector; Based on the implicitly aligned feature vector and the ideal signal feature vector, the frequency domain power distribution difference is identified to obtain the first inconsistency factor; Based on the implicitly aligned feature vector and the ideal signal feature vector, the timing phase trajectory difference is identified to obtain the second inconsistency factor; The first inconsistency factor and the second inconsistency factor are summarized to obtain the double inconsistency factor set; The frequency domain power distribution difference identification is performed based on the implicit alignment feature vector and the ideal signal feature vector, and a first heavy inconsistency factor is obtained, including: Based on the frequency domain and power distribution features of the implicit alignment feature vector and the ideal signal feature vector, an implicit alignment feature vector energy spectrum Gaussian distribution model and an ideal signal feature vector energy spectrum Gaussian distribution model are constructed; The distribution difference identification is performed on the implicit alignment feature vector energy spectrum Gaussian distribution model and the ideal signal feature vector energy spectrum Gaussian distribution model by using KL divergence, and the first heavy inconsistency factor is obtained; The time sequence phase trajectory difference identification is performed based on the implicit alignment feature vector and the ideal signal feature vector, and a second heavy inconsistency factor is obtained, including: The subspace projection is performed on the implicit alignment feature vector and the ideal signal feature vector in the time dimension and the vector dimension respectively, and a symbol timing trajectory sequence and a phase trajectory sequence are obtained; The high-dimensional feature extraction is performed on the symbol timing trajectory sequence and the phase trajectory sequence by using a shared weight twin network, and an implicit alignment feature vector embedding representation and an ideal signal feature vector embedding representation are obtained; The difference identification is performed on the obtained implicit alignment feature vector embedding representation and the ideal signal feature vector embedding representation, and the second heavy inconsistency factor is obtained.

2. The deep learning based co-interference method of claim 1, wherein, The strategy analysis is performed on the first heavy inconsistency factor and the second heavy inconsistency factor by calling a cooperative jamming strategy generator, and an initial cooperative jamming strategy is output, including: Based on the first heavy inconsistency factor and the second heavy inconsistency factor, a strategy analysis input feature vector is constructed; The strategy analysis input feature vector is analyzed by calling the encoder and the decoder of the cooperative jamming strategy generator, and the initial cooperative jamming strategy is obtained, wherein the decoder has a plurality of output heads.

3. The deep learning based co-interference method of claim 1, wherein, The power spectrum cooperative scheduling of the interference platform set is performed in combination with the initial cooperative jamming strategy and the interference platform reliability coefficient set, and a target cooperative jamming strategy is obtained, including: The signal transmission capability threshold set of the interference platform set is obtained, and the cooperative scheduling constraint is constructed in combination with the overall power threshold and the frequency threshold in the initial cooperative jamming strategy; Based on the cooperative scheduling constraint, the power spectrum cooperative scheduling is performed on the interference platform set in combination with the interference platform reliability coefficient set, and the target cooperative jamming strategy is obtained, with the goal of maximizing the interference quality.

4. The deep learning based co-interference method of claim 1, wherein, Including: The fast Fourier transform is performed on the radar echo signal sequence set, and a time-frequency spectrum energy distribution set is obtained; The pulse Doppler processing and azimuth estimation are performed on the radar echo signal sequence set, and a range-Doppler-azimuth feature set is generated; The time-frequency spectrum energy distribution set and the range-Doppler-azimuth feature set are mapped and encoded to determine the observation signal feature tensor, and the interference platform reliability coefficient set is obtained by identifying the signal reception quality of the radar echo signal sequence set.

5. An electronic device, comprising: Including: A processor coupled with a memory, the memory being used to store a program, when the program is executed by the processor, to perform the steps of the method of any one of claims 1 to 4.

6. A computer-readable storage medium, characterized in that, The storage medium has stored thereon a computer program which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 4.

Images