Multifunctional radar working mode identification method based on LSTCN network

Abstract

The invention discloses a multifunctional radar working mode recognition method based on an LSTCN network, and the method comprises the steps: firstly analyzing the working mode characteristics of an airborne phased array radar, generating multifunctional radar pulse group data, inputting a pulse data generation program, carrying out the data preprocessing, dividing a data set into a training set and a test set, and carrying out the recognition of the working mode of the airborne phased array radar; and then constructing a multifunctional radar LSTCN working mode identification network model, inputting the preprocessed training set data into the model for training, and finally sending the test set data into the trained model to complete the multifunctional radar working mode identification test. According to the method, the LSTCN recurrent neural network and the TCN network are combined, an LSTCN network model is provided, the problems that an existing working mode recognition method is incomprehensive in working mode feature analysis and low in recognition rate are solved, simulation results show that in a noise environment, compared with the LSTM network and the like, the LSTCN neural network has better performance, and the recognition efficiency is improved. And compared with an existing network model, the corresponding identification accuracy is higher.

Description

Technical Field

[0001] The invention belongs to the technical field of electronic countermeasures, and in particular relates to a multifunctional radar working mode recognition method based on an LSTCN network. Background Art

[0002] Radar signal classification and intention determination are important foundations of cognitive electronic countermeasures. With the advancement of electronic scanning and phased array systems, radars can adjust beam pointing in real time according to tasks to achieve multiple goals. The flexible combination of tasks forms a variety of working modes. The identification of radar working modes provides an important basis for the reconnaissance party, enabling it to dynamically adjust the electronic countermeasure strategy according to the behavior of the radar party, and thus take the initiative in the confrontation. Therefore, the LSTCN network method is applied to the identification of multi-function radar working modes.

[0003] Long short-term memory (LSTM) is a time-recurrent neural network designed to solve the long-term dependency problem of ordinary recurrent neural networks (RNNs). Its structure consists of input gates, forget gates, output gates, etc., and is designed to process time series data. In contrast, the time domain convolutional network (TCN) realizes causal time data processing through dilated convolution, and can effectively process sequence data with consistent input and output lengths.

[0004] In recent years, progress has been made in the recognition of multifunctional radar working modes. Existing methods based on template matching or statistical histogram sorting perform well in modulated radars with fixed signal parameters, but for complex multifunctional radars, due to the variable signal parameters and unclear relationship between tasks and working modes, existing methods are difficult to extract dynamic features. The literature "Ou, Jian." Research on Radar Word Extraction Technology in Multifunctional Radar Modeling." Journal of National University of Defense Technology, vol. 32, no. 2, 2010, pp. 91-96." proposed a radar signal representation method based on a syntactic model, which describes the signal characteristics of multifunctional radars through "radar words", "radar phrases" and "radar sentences", so as to better capture the hierarchical relationship of the signal.

[0005] In addition, research based on seq2one and seq2seq models has also been gradually proposed. Among them, the seq2one method is difficult to cope with multi-class mixed inputs, and although the seq2seq method can process multi-function radar pulse sequences in layers, it has deficiencies in computational complexity and feature analysis. The literature "Chen, H., Feng, K., Kong, Y., Zhang, L., Yu, X., & Yi, W. (2022). Multi-Function Radar Work Mode Recognition Based on Encoder-DecoderModel. IEEE International Geoscience and Remote Sensing Symposium (IGARSS), 1189-1192." proposed a codec model to identify radar working modes. Although good performance was achieved, the simplification of the data set and the computational overhead are still challenges. Current research mainly focuses on the optimization of network structure and lacks in-depth analysis of the data set. Summary of the invention

[0006] In order to solve the above technical problems, the present invention proposes a multi-function radar working mode recognition method based on the LSTCN network. Through simulation comparison, it is proved that the LSTCN network can have better performance than the existing LSTM network, GRU network and TCN network in a noisy environment.

[0007] The technical solution adopted by the present invention is: a multifunctional radar working mode recognition method based on LSTCN network, and the specific steps are as follows:

[0008] S1. Analyze the working mode characteristics of the multi-function radar, generate multi-function radar pulse group data, input the pulse data generation program GUI, generate a sample data set based on the working mode transfer law, perform data preprocessing, and then divide the sample data set into a training set and a test set;

[0009] The multifunctional radar pulse group data includes: PRI and PA.

[0010] S2. Combine the LSTM neural network and the TCN network to build a multi-function radar LSTCN working mode recognition network model;

[0011] The network model uses the LSTM neural network to quantize and reduce the dimension of the input pulse sequence data to extract the corresponding time series features, and uses the TCN neural network to convert the extracted features into the required output data.

[0012] S3, inputting the training set data preprocessed in step S1 into the network model of step S2 for training;

[0013] S4, sending the test set data preprocessed in step S1 to the network model trained in step S3 for testing, and completing the multi-function radar working mode recognition test.

[0014] Furthermore, the step S1 is specifically as follows:

[0015] S11, generating PRI according to the characteristics of the multi-function radar working mode, inputting into GUI and displaying the PRI type;

[0016] First, according to the characteristic parameters of different working modes, the pulse parameters are set respectively, including: pulse carrier frequency RF, distance R between the reconnaissance aircraft and the radar carrier, speed V of the reconnaissance aircraft a , Countermeasure radar carrier speed V t , sampling rate, modulation method, speed of light, working mode, number of group changes.

[0017] Among them, the modulation modes include: fixed, group-variable / staggered, and fixed / staggered; in each working mode, the number of group-variable groups of RF and PW is the same as that of PRI, and the staggered change model is similar to PRI, with the same staggered change.

[0018] Then, the pulse data generation program GUI is used to generate the pulse group characteristic working mode data under each mode, namely, PRI, and the PRI type is displayed.

[0019] The corresponding PRI value generation in each working mode is as follows:

[0020] (1) The corresponding PRI value in the working mode of the medium-frequency group transformer;

[0021] First, the corresponding signal round-trip time T in the working mode of the medium-frequency group transformer is obtained. A , the expression is as follows:

[0022]

[0023] Among them, R d represents the current distance between the radar and the target, c represents the propagation speed of electromagnetic waves, that is, the speed of light, τ s Indicates the duration of the signal pulse, that is, the pulse width.

[0024] After the signal round trip time is known, the pulse repetition interval PRI of the intermediate frequency group and the repetition frequency group is defined as follows:

[0025]

[0026] Among them, PRI m represents the pulse repetition interval of the mth group of the intermediate frequency group, T BIndicates the round trip time of the signal in the corresponding working mode, D1 indicates the constant parameter of the intermediate frequency group, and m indicates the group number. n represents the pulse repetition interval of the nth group of the repetition frequency group, D2 represents the constant parameter of the repetition frequency group, n represents the group number, Δ n Represents a small perturbation used to adjust the repetition rate group to ensure the target visibility reaches the desired level.

[0027] (2) PRI value in high repetition rate group transformer working mode;

[0028]

[0029] Among them, R c Indicates the target distance. η channel represents the introduced channel loss parameter, that is, the loss of the signal during propagation, and is less than 1. After considering the multipath effect and attenuation, the corresponding high repetition frequency group edge working mode PRI value expression is as follows:

[0030]

[0031] Among them, λ represents the wavelength of the signal, a t and a r represent the acceleration of target and radar respectively, and t represents time.

[0032] Then the pulse drift effect is introduced to obtain the pulse repetition interval after the drift effect. The expression is as follows:

[0033]

[0034] Where PRI1 represents the pulse repetition interval before being affected by the drift effect, PRI2 represents the pulse repetition interval after being affected by the drift effect, and δ drift Indicates the drift amplitude, that is, the fluctuation amplitude of the pulse interval, f drift Represents the drift frequency, which is used to simulate the fluctuation in pulse time.

[0035] Introducing phase noise and Doppler effect, the expression is as follows:

[0036]

[0037] in, represents the pulse repetition interval after the introduction of phase noise and Doppler effect, v r Represents the radial velocity of the target, calculated using the Doppler shift formula.

[0038] For the pulse sequence model, the contamination mechanism is introduced, and pulse loss, false pulses and measurement noise are considered. The pulse sequence model expression is as follows:

[0039]

[0040] Where S represents the pulse sequence of the working mode under non-ideal conditions with measurement noise, pulse loss, and false pulses, It means that a pulse with noise or deviation is measured at time t. represents the pulse deviation introduced by the measurement noise, Pulses indicating the presence of false pulses, Indicates false pulses introduced due to channel contamination or equipment failure. Indicates the presence of both false pulses and pulse loss.

[0041] Then introduce the signal-to-noise ratio SNR and noise power P at the same time noise , the expression is as follows:

[0042]

[0043] Among them, P signal represents the signal power, η noise represents the noise factor, that is, the noise level, k represents the Boltzmann constant, T represents the system noise temperature, and B represents the signal bandwidth.

[0044] Finally, the target detection probability P is introduced det , the expression is as follows:

[0045]

[0046] Where Q represents the Q function of the standard normal distribution, which is used to calculate the detection probability, threshold represents the detection threshold, and σ noise represents the standard deviation of the measurement noise.

[0047] (3) Staggered PRI and jittered PRI generation;

[0048] Only the pulse group staggered PRI mode is considered, a reference pulse repetition interval PRI is defined for each pulse group, and then a staggered effect is introduced into the pulse group, so that the interval of each pulse in the pulse group deviates from the reference PRI, but the structural integrity of the overall pulse group is maintained.

[0049] First, we introduce the reference pulse group pulse repetition interval PRI group , the expression is defined as follows:

[0050]

[0051] Among them, T group Represents the duration of the entire pulse group, N pluses Indicates the number of pulses in a pulse group.

[0052] The pulse group deviation is applied to each pulse, rather than the interval between each pulse. That is, the PRI of each pulse in the pulse group is the base PRI plus a deviation term, expressed as follows:

[0053] PRI group,i =PRI group (1+δ group,i )

[0054] Among them, PRI group,i represents the PRI of the ith pulse in the pulse group, δ group,i Represents the variation of the ith pulse.

[0055] The variance is then defined as a random variable that obeys a specific distribution, and the expression is as follows:

[0056]

[0057] or,

[0058]

[0059] Among them, σ group The standard deviation of the pulse group difference is used to describe the degree of random deviation of each pulse repetition interval (PRI) in the pulse group. max represents the maximum value of the deviation, Indicates that the variance has a mean of 0 and a standard deviation of σ group Gaussian distribution. Indicates that the difference is -δ max to δ max Evenly distributed between.

[0060] In addition to the pulse group stagger, each pulse in the pulse group is also affected by random jitter. Then the jitter effect is introduced on the basis of the pulse group stagger. The expression is as follows:

[0061]

[0062] in, Represents pulse jitter, with a mean of 0 and a standard deviation of σ jitter Gaussian noise.

[0063] S12, generating a pulse amplitude PA according to the characteristics of the multi-function radar working mode;

[0064] The PA data corresponding to each working mode is determined by the following parameters, including: the distance between the reconnaissance aircraft and the adversary radar carrier in the current mode, the gain of the adversary radar transmitting antenna, and the gain parameter of the reconnaissance aircraft receiving antenna. The specific expression is as follows:

[0065]

[0066] Among them, P r (t) represents the signal power received by the reconnaissance aircraft, P t Represents the corresponding radar transmission signal power, G r ,G t , λ respectively represent the receiving antenna gain of the reconnaissance aircraft, the transmitting antenna gain of the adversary radar and the corresponding signal wavelength, L represents the system loss, F t represents the corresponding radar emission pattern, F t (θ(t), φ(t)) represents the directional gain function of the transmitting antenna pattern of the adversary radar, and θ(t) and φ(t) represent the angles in the elevation and azimuth directions, respectively.

[0067] Then the amplitude of the received pulse is obtained by taking the square root of the received power, that is,

[0068] S13, based on steps S11-S12, obtain multifunctional radar pulse group data, combine the working mode transfer rule, integrate the data set, and generate a sample data set;

[0069] Based on steps S11-S12, pulse group data within each working mode is generated. On the basis of the pulse group data, the working mode transfer is performed in combination with the working mode transfer rules between the corresponding working modes to generate corresponding working mode data with working mode transfer relationship, that is, the corresponding sample data set.

[0070] S14, performing data preprocessing based on the sample data set obtained in step S13, and dividing the sample data set into a training set and a test set after preprocessing;

[0071] The pulse data sets generated in different working modes are distinguished by working mode labels, and then the pulse data sets are filled with data. Meanwhile, the pollution of the data sets by the corresponding non-ideal conditions in the real electromagnetic environment is considered, and a reconnaissance aircraft intercepted pulse sequence model is proposed to deal with data pollution of the filled data sets.

[0072] The pulse data set is the sample data set obtained in step S13; the pulse sequence model intercepted by the reconnaissance aircraft is the pulse sequence model in step S12.

[0073] Finally, the preprocessed dataset is divided into training set and test set in a ratio of 7:3.

[0074] Furthermore, the step S3 is specifically as follows:

[0075] The training set data generated in step S1 is input into the network model described in step S2 for training, and the hyperparameters of the neural network are adjusted according to the accuracy of the training set, including: learning rate, batch size, until the neural network model converges, the loss function size remains unchanged, and the model training is completed.

[0076] Among them, the basic cross entropy loss function loss fc The expression is as follows:

[0077]

[0078] Among them, N train represents the number of samples in the training set, p t represents the output of the network at time t, that is, the possible probability of each working mode judged by the network at time t, p t Indicates the actual working mode at time t; L i represents the length of the pulse sequence under the i-th training sample, s t represents the true label at time t; λ1 represents the corresponding regularization factor, and ω2 represents the L2 regularization weight vector.

[0079] Based on the basic cross entropy loss function, an adaptive weight adjustment mechanism and a multi-task learning strategy are introduced, and the expression is as follows:

[0080] loss total =α·loss fc +β·loss noise +γ·loss multi-task

[0081] Among them, loss total Represents the total loss, loss fc Represents the loss for pulse repetition interval variation. noise represents the loss term for noise and channel effects. multi-task Represents the multi-task learning loss, which is used for comprehensive optimization of different radar modes. α, β, and γ represent adaptive weight parameters, which are dynamically adjusted according to the importance of different tasks during training.

[0082] Furthermore, the step S4 is specifically as follows:

[0083] The test set data generated in step S1 is sent to the network model trained in step S2. The result of the model output is the possible label of the radar working mode. The corresponding working mode recognition accuracy is calculated according to the label to obtain the corresponding pulse-level multi-function radar working mode recognition result, and the multi-function radar working mode recognition test is completed.

[0084] Among them, the weighted accuracy formula is used to calculate the corresponding working mode recognition accuracy, and the expression is as follows:

[0085]

[0086] Among them, acc p represents the accuracy, w t Represents the weight of each sample, which is adjusted according to the category importance or sample difficulty, N test represents the total number of corresponding test samples, Indicates the result label of the working mode after recognition, y t Indicates the actual working mode label.

[0087] Beneficial effects of the present invention: The method of the present invention first analyzes the working mode characteristics of the airborne phased array radar, generates multifunctional radar pulse group data, inputs the pulse data generation program and performs data preprocessing, and divides the data set into a training set and a test set, and then constructs a multifunctional radar LSTCN working mode recognition network model, inputs the preprocessed training set data into the model for training, and finally sends the test set data to the trained model to complete the multifunctional radar working mode recognition test. The method of the present invention combines the LSTM recurrent neural network and the TCN network, and proposes a network model of LSTCN, which solves the problem that the existing working mode recognition method does not fully analyze the working mode characteristics and has a low recognition rate. The simulation results show that in a noisy environment, compared with the LSTM network, the LSTCN neural network has better performance, and compared with the existing network model, the corresponding recognition accuracy is higher. BRIEF DESCRIPTION OF THE DRAWINGS

[0088] Figure 1 The present invention is a flowchart of a method for identifying a multi-function radar working mode based on an LSTCN network.

[0089] Figure 2 This is a structural diagram of the network model for identifying the working mode of the multifunctional radar LSTCN according to an embodiment of the present invention.

[0090] Figure 3 Schematic diagram of the working mode transfer rule of the multi-function radar in an embodiment of the present invention.

[0091] Figure 4 This is a specific pattern diagram of signal generation in an embodiment of the present invention.

[0092] Figure 5 It is a time domain diagram and a distribution diagram of a data set in VS mode in an embodiment of the present invention.

[0093] Figure 6 This is a diagram showing the effect of multi-function radar working mode recognition in an embodiment of the present invention. DETAILED DESCRIPTION

[0094] The method of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0095] like Figure 1 As shown, a flow chart of a multifunctional radar working mode recognition method based on an LSTCN network of the present invention, the specific steps are as follows:

[0096] S1. Analyze the working mode characteristics of the multi-function radar, generate multi-function radar pulse group data, input the pulse data generation program GUI, generate a sample data set based on the working mode transfer law, perform data preprocessing, and then divide the sample data set into a training set and a test set;

[0097] The multifunctional radar pulse group data includes: PRI and PA.

[0098] S2. Combine the LSTM neural network and the TCN network to build a multi-function radar LSTCN working mode recognition network model;

[0099] Among them, the multifunctional radar LSTCN working mode recognition network model structure is as follows Figure 2 As shown, the data mining and prediction capabilities of the recurrent convolutional neural network and the corresponding data decoding capabilities unique to the TCN time-domain convolutional neural network are used to combine the LSTM neural network with the TCN neural network. The network model uses the LSTM neural network to quantize and reduce the dimension of the input pulse sequence data to extract the corresponding time series features, and uses the TCN neural network to convert the extracted features into the required output data.

[0100] S3, inputting the training set data preprocessed in step S1 into the network model of step S2 for training;

[0101] S4, sending the test set data preprocessed in step S1 to the network model trained in step S3 for testing, and completing the multi-function radar working mode recognition test.

[0102] In this embodiment, the step S1 is specifically as follows:

[0103] S11, generating PRI according to the characteristics of the multi-function radar working mode, inputting into GUI and displaying the PRI type;

[0104] In this embodiment, in the Matlab2022b environment, firstly, the pulse parameters are set according to the characteristic parameters of different working modes, including: pulse carrier frequency RF, distance R between the reconnaissance aircraft and the radar carrier, speed V of the reconnaissance aircraft a , Countermeasure radar carrier speed V t , sampling rate, modulation method, speed of light, working mode, number of group changes.

[0105] The working modes of this embodiment include four working modes: VS, RWS, TAS, and STT. The VS mode refers to the speed search mode, and the RWS mode (search and range measurement mode) can be divided into two sub-modes: HRWS (High-PRF Range While Search) mode and MRWS (Medium-PRF Range While Search) sub-mode. The TAS (search and tracking mode) mode can also be divided into two sub-states: search sub-state (TAS (S)) and tracking sub-state (TAS (T)), and STT refers to single target tracking mode.

[0106] The modulation modes include: fixed, group-variable / staggered, and fixed / staggered; the number of group-variable groups of RF and PW in each working mode is the same as that of PRI, and the staggered change model is similar to PRI, with the same staggered change. The working mode definition and the corresponding working mode modulation mode are shown in Table 1.

[0107] Table 1

[0108]

[0109] The specific values ​​of the working modes are shown in Table 2, that is, the value range of the corresponding data under each working mode.

[0110] Table 2

[0111]

[0112] Among them, U represents uniform distribution.

[0113] Then, the pulse data generation program GUI is used to generate the pulse group characteristic working mode data under each mode (ie, the corresponding PRI data is calculated according to the above-mentioned information set), and the PRI type is displayed.

[0114] The corresponding PRI value generation in each working mode is as follows:

[0115] (1) The corresponding PRI value in the working mode of the medium-frequency group transformer;

[0116] First, the corresponding signal round trip time under the working mode of the medium-frequency group transformer is obtained, and the expression is as follows:

[0117]

[0118] Among them, R d represents the current distance between the radar and the target, c represents the propagation speed of electromagnetic waves, that is, the speed of light, τ s Indicates the duration of the signal pulse, that is, the pulse width.

[0119] Signal round trip time T A:The improved signal round trip time takes into account the current distance R between the radar and the target d , rather than the designed maximum clear distance. This is more suitable for real-time calculations in dynamic environments.

[0120] After the signal round trip time is known, different constants D1 and D2 are introduced to represent the characteristics of the intermediate frequency group and the repetition frequency group. For the pulse repetition interval PRI of the intermediate frequency group and the repetition frequency group, the expression is defined as follows:

[0121]

[0122] Among them, PRI m represents the pulse repetition interval of the mth group of the intermediate frequency group, T B Indicates the signal round trip time of the corresponding working mode, D1 indicates the constant parameter of the intermediate frequency group, and m indicates the group number. n represents the pulse repetition interval of the nth group of the repetition frequency group, D2 represents the constant parameter of the repetition frequency group, n represents the group number, Δ n Represents a small disturbance used to adjust the repetition frequency group, that is, the PRI adjustment in the repetition frequency group. When the group change is switched, the disturbance is used to ensure the target detection stability and visibility of the repetition frequency group, ensuring that the target visibility reaches an ideal level. Unlike the conventional formula, the group numbers m and n are added with multiple factors, so that different frequency distributions can be maintained between the intermediate frequency and the repetition frequency to avoid frequency overlap. Through the above settings, the values ​​of m and n are no longer limited to odd and even numbers, but a wider range of group change sequences are generated by a factor of 3, further expanding the flexibility of the group change.

[0123] Based on the above design, PRI m , PRI n It provides more possibilities for dynamic adjustment of the PRI value generation of the medium-frequency group variable, can better adapt to different environments and target distances, and can provide more complex and realistic training data for the multi-function radar working mode recognition method model based on the LSTCN network.

[0124] (2) PRI value in high repetition rate group transformer working mode;

[0125]

[0126] In actual situations, radar signals are affected by multipath effects, noise interference, attenuation and other channel effects during transmission, so it is necessary to add channel influence parameters in the high repetition rate group variable working mode. c Indicates the target distance. η channel represents the introduced channel loss parameter, that is, the loss of the signal during propagation, and is less than 1. After considering the multipath effect and attenuation, the corresponding high repetition frequency group edge working mode PRI value expression is as follows:

[0127]

[0128] Among them, λ represents the wavelength of the signal, a t and a r They represent the acceleration of the target and the radar respectively, considering the velocity change in long-term tracking, and t represents time, considering the relative motion change between the radar and the target over a long period of time.

[0129] In order to simulate the pulse behavior more complexly, the pulse drift effect is introduced (this effect will cause a slight fluctuation in the pulse repetition interval), and the pulse repetition interval after the drift effect is obtained. The expression is as follows:

[0130]

[0131] Where PRI1 represents the pulse repetition interval before being affected by the drift effect, PRI2 represents the pulse repetition interval after being affected by the drift effect, and δ drift Indicates the drift amplitude, that is, the fluctuation amplitude of the pulse interval, f drift Represents the drift frequency, which is used to simulate the fluctuation in pulse time.

[0132] The behavior of the pulse is more complex by introducing phase noise and Doppler effect. Phase noise is one of the sources of uncertainty in the signal, while the Doppler effect is related to the target speed. The expression is as follows:

[0133]

[0134] in, represents the pulse repetition interval after the introduction of phase noise and Doppler effect, v r Represents the radial velocity of the target, calculated using the Doppler shift formula. The Doppler effect causes a change in the frequency of the radar received signal, so the pulse repetition interval will also be affected.

[0135] For the pulse sequence model, the contamination mechanism is introduced, and pulse loss, false pulses and measurement noise are considered. The pulse sequence model expression is as follows:

[0136]

[0137] Where S represents the pulse sequence of the working mode under non-ideal conditions with measurement noise, pulse loss, and false pulses, It means that a pulse with noise or deviation is measured at time t. represents the pulse deviation introduced by the measurement noise, Pulses indicating the presence of false pulses, Indicates false pulses introduced due to channel contamination or equipment failure. Indicates the presence of both false pulses and pulse loss.

[0138] Then introduce the signal-to-noise ratio SNR and noise power P at the same time noise , further simulate the data set. The expression is as follows:

[0139]

[0140] Among them, P signal represents the signal power, η noise represents the noise factor, i.e. the noise level, k represents the Boltzmann constant, T represents the system noise temperature, and B represents the signal bandwidth. SNR determines the relative strength of the signal and noise, affecting the quality and complexity of the data set.

[0141] In order to better simulate the detection task in the training set, the target detection probability P is finally introduced det , the expression is as follows:

[0142]

[0143] Where Q represents the Q function of the standard normal distribution, which is used to calculate the detection probability, threshold represents the detection threshold, and σ noise represents the standard deviation of the measurement noise.

[0144] This embodiment can generate a more complex and realistic data set by introducing more parameters and influencing factors (channel loss, acceleration, pulse drift, Doppler effect, measurement noise, target detection probability, etc.). This data set can better simulate radar detection tasks in real environments, is suitable for training multi-function radar working mode recognition methods based on LSTCN networks, can improve the model's ability to recognize complex working modes, and improve the generalization performance of the model.

[0145] (3) Staggered PRI and jittered PRI generation;

[0146] Pulse group stagger is different from pulse inter-pulse stagger. According to the common mode of current airborne radar, the data set of this embodiment only considers the pulse group stagger PRI mode, defines a reference pulse repetition interval PRI for each pulse group, and then introduces the stagger effect in the pulse group, so that the interval of each pulse of the pulse group deviates from the reference PRI, but maintains the structural integrity of the overall pulse group.

[0147] First, we introduce the reference pulse group pulse repetition interval PRI group , the expression is defined as follows:

[0148]

[0149] Among them, T groupRepresents the duration of the entire pulse group, N pluses Indicates the number of pulses in a pulse group. The base PRI of a pulse group is equal to the total time of the pulse group divided by the number of pulses.

[0150] The pulse group deviation is applied to each pulse, rather than the interval between each pulse. That is, the PRI of each pulse in the pulse group is the base PRI plus a deviation term, expressed as follows:

[0151] PRI group,i =PRI group (1+δ group,i )

[0152] Among them, PRI group,i represents the PRI of the ith pulse in the pulse group, δ group,i Indicates the deviation of the ith pulse, which is set according to the actual situation.

[0153] In order to make the pulse group variation have a realistic effect, the variation is defined as a random variable that obeys a specific distribution, including Gaussian distribution or uniform distribution. The expression is as follows:

[0154]

[0155] or,

[0156]

[0157] Among them, σ group The standard deviation of the pulse group difference is used to describe the degree of random deviation of each pulse repetition interval (PRI) in the pulse group. max represents the maximum value of the deviation, Indicates that the variance has a mean of 0 and a standard deviation of σ group Gaussian distribution. Indicates that the difference is -δ max to δ max This allows each pulse in the pulse group to have a slight deviation, but does not deviate from the structure of the entire pulse group, thereby generating more realistic radar signal data.

[0158] In addition to the pulse group stagger, each pulse in the pulse group is also affected by random jitter. Then the jitter effect is introduced on the basis of the pulse group stagger. The expression is as follows:

[0159]

[0160] in, Represents pulse jitter, with a mean of 0 and a standard deviation of σ jitter Gaussian noise.

[0161] This jitter effect can reflect the random changes between each pulse in the pulse group due to the instability of the electronic system or external interference. The training data set containing pulse group jitter can make staggered adjustments for the pulses in each pulse group, rather than adjusting the interval between each pulse individually, which can better reflect the complex changes between radar signal groups. The training data set containing pulse group jitter adds random jitter to each pulse on the basis of pulse group jitter, making the signal more random and complex. These improvements will provide more complex and rich data for the LSTCN network model of multi-function radar working mode recognition, and enhance the model's ability to adapt to complex real-world scenarios.

[0162] S12, generating a pulse amplitude PA according to the characteristics of the multi-function radar working mode;

[0163] Since the pulse amplitude is affected by many factors, such as the corresponding radar platform related parameters, the relative distance factor between the carrier and the target, the radar cross-sectional area of ​​the target, etc. In this embodiment, the PA data corresponding to each working mode is determined by the following parameters, including: the distance between the reconnaissance aircraft and the counter-radar carrier in the current mode, the gain of the counter-radar transmitting antenna, and the gain parameter of the reconnaissance aircraft receiving antenna. The specific expression is as follows:

[0164]

[0165] Among them, P r (t) represents the signal power received by the reconnaissance aircraft, P t Represents the corresponding radar transmission signal power, G r ,G t , λ respectively represent the receiving antenna gain of the reconnaissance aircraft, the transmitting antenna gain of the adversary radar and the corresponding signal wavelength, L represents the system loss, F t represents the corresponding radar emission pattern, F t (θ(t), φ(t)) represents the directional gain function of the adversary radar transmitting antenna pattern, which is used to quantify the power loss caused by path loss or other factors during signal transmission. θ(t) and φ(t) represent the angles in elevation and azimuth, respectively.

[0166] Then the amplitude of the received pulse is obtained by taking the square root of the received power, that is,

[0167] S13, based on steps S11-S12, obtain multifunctional radar pulse group data, combine the working mode transfer rule, integrate the data set, and generate a sample data set;

[0168] Based on steps S11-S12, pulse group data within each working mode is generated. On the basis of the pulse group data, the working mode transfer is performed in combination with the working mode transfer rules between the corresponding working modes to generate corresponding working mode data with working mode transfer relationship, that is, the corresponding sample data set.

[0169] This embodiment generates data in each of the above working modes based on the working mode transfer according to a certain working mode transfer relationship, and generates corresponding working mode data with working mode transfer relationship. The specific working mode transfer relationship is as follows: Figure 3 As shown, the transition relationship corresponding to the four working modes in this embodiment is that VS and RWS alternately search to realize the search of the airspace and target detection, and when the target is found, it switches to the corresponding TAS mode or the corresponding STT working mode, which is as follows:

[0170] Since the VS mode has serious distance ambiguity in the distance dimension, it needs to be alternated with the RWS mode to achieve target deambiguation. At the same time, the alternation of the VS mode and the RWS mode can also obtain the full attitude detection capability of the target. When the target is found in the VS mode, the HRWS waveform in the RWS mode and the corresponding MRWS waveform need to be emitted to establish tracking that meets the track update accuracy.

[0171] Therefore, when a target is found, the VS mode always needs to be switched to the RWS mode to achieve distance measurement and corresponding deambiguation processing. If the target threat level is low, it will be switched to the TAS tracking state to achieve corresponding target tracking. If the target threat level is high, it will be switched to the corresponding STT mode to achieve high-precision tracking. When the target is lost in the tracking state, it is necessary to switch to the corresponding search state to achieve re-search of the target. Similarly, when the target threat level decreases, the corresponding tracking is switched from the STT mode to the corresponding TAS mode.

[0172] S14, performing data preprocessing based on the sample data set obtained in step S13, and dividing the sample data set into a training set and a test set after preprocessing;

[0173] The pulse data sets generated in different working modes are distinguished by working mode labels, and then the pulse data sets are filled with data. Meanwhile, the pollution of the data sets by the corresponding non-ideal conditions in the real electromagnetic environment is considered, and a reconnaissance aircraft intercepted pulse sequence model is proposed to deal with data pollution of the filled data sets.

[0174] The pulse data set is the sample data set obtained in step S13; the pulse sequence model intercepted by the reconnaissance aircraft is the pulse sequence model in step S12.

[0175] Finally, the preprocessed dataset is divided into training set and test set in a ratio of 7:3.

[0176] In this embodiment, step S3 is specifically as follows:

[0177] The training set data generated in step S1 is input into the network model described in step S2 for training, and the hyperparameters of the neural network are adjusted according to the accuracy of the training set, including: learning rate, batch size, until the neural network model converges, the loss function size remains unchanged, and the model training is completed.

[0178] Among them, the basic cross entropy loss function loss fc The expression is as follows:

[0179]

[0180] Among them, N train represents the number of samples in the training set, p t represents the output of the network at time t, that is, the possible probability of each working mode judged by the network at time t, p t Indicates the actual working mode at time t; L i represents the length of the pulse sequence under the i-th training sample, s t represents the true label at time t; λ1 represents the corresponding regularization factor, and ω2 represents the L2 regularization weight vector.

[0181] However, since the generated data set contains jagged and group variations, in order to further extract the characteristic form of the network, the loss function is adjusted accordingly based on the basic cross entropy loss function. An adaptive weight adjustment mechanism and multi-task learning strategy are introduced to improve the adaptability of the model to different modes. The expression is as follows:

[0182] loss total =α·loss fc +β·loss noise +γ·loss multi-task

[0183] Among them, loss total Represents the total loss, loss fc Represents the loss for pulse repetition interval variation. noise represents the loss term for noise and channel effects. multi-task Represents the multi-task learning loss, which is used for comprehensive optimization of different radar modes. α, β, and γ represent adaptive weight parameters, which are dynamically adjusted according to the importance of different tasks during training.

[0184] In this embodiment, step S4 is specifically as follows:

[0185] The test set data generated in step S1 is sent to the network model trained in step S2. The result of the model output is the possible label of the radar working mode. The corresponding working mode recognition accuracy is calculated according to the label to obtain the corresponding pulse-level multi-function radar working mode recognition result, and the multi-function radar working mode recognition test is completed.

[0186] Among them, the weighted accuracy formula is used to calculate the corresponding working mode recognition accuracy, and the expression is as follows:

[0187]

[0188] Among them, acc p represents the accuracy, w t Represents the weight of each sample, which is adjusted according to the category importance or sample difficulty, N test represents the total number of corresponding test samples, Indicates the result label of the working mode after recognition, y t Indicates the actual working mode label. The specific style of signal generation is as follows Figure 4 As shown, the corresponding PRI distribution characteristics in VS mode are as follows Figure 5 As shown in the figure, the time domain diagram and distribution diagram of the data set in VS mode, the recognition effect of the multi-function radar working mode is as follows Figure 6 shown.

[0189] In summary, the method of the present invention first analyzes the working mode characteristics of the airborne phased array radar, generates multifunctional radar pulse group data, inputs a data set generation program and performs data preprocessing, and divides the data set into a training set and a test set, and then constructs a multifunctional radar LSTCN working mode recognition model, inputs the preprocessed training set data into the model for training, and finally sends the test set data to the trained model to complete the multifunctional radar working mode recognition test. The method of the present invention combines the LSTM recurrent neural network and the TCN network, proposes a network model of LSTCN, solves the problem that the existing working mode recognition method does not fully analyze the working mode characteristics and has a low recognition rate, and the simulation results show that in a noisy environment, compared with the LSTM network, the LSTCN neural network has better performance.

[0190] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A multi-function radar working mode recognition method based on LSTCN network, the specific steps are as follows: S1. Analyze the working mode characteristics of the multi-function radar, generate multi-function radar pulse group data, input the pulse data generation program GUI, generate a sample data set based on the working mode transfer law, perform data preprocessing, and then divide the sample data set into a training set and a test set; in, The multifunctional radar pulse group data includes: PRI and PA; S2. Combine the LSTM neural network and the TCN network to build a multi-function radar LSTCN working mode recognition network model; The network model uses an LSTM neural network to quantize and reduce the dimension of the input pulse sequence data to extract the corresponding time series features, and uses a TCN neural network to convert the extracted features into the required output data; S3, inputting the training set data preprocessed in step S1 into the network model of step S2 for training; S4, sending the test set data preprocessed in step S1 to the network model trained in step S3 for testing, and completing the multi-function radar working mode recognition test.

2. A multifunctional radar working mode recognition method based on LSTCN network according to claim 1, characterized in that: The step S1 is specifically as follows: S11, generating PRI according to the characteristics of the multi-function radar working mode, inputting into GUI and displaying the PRI type; First, according to the characteristic parameters of different working modes, the pulse parameters are set respectively, including: pulse carrier frequency RF, distance R between the reconnaissance aircraft and the radar carrier, speed V of the reconnaissance aircraft a , Countermeasure radar carrier speed V t , sampling rate, modulation method, light speed, working mode, number of groups; Among them, the modulation modes include: fixed, group-variable / staggered, and fixed / staggered. In each working mode, the number of group-variable groups of RF and PW is the same as that of PRI, and the staggered change model is similar to that of PRI, with the same staggered change. Then, the pulse data generation program GUI is used to generate the pulse group characteristic working mode data in each mode, namely, PRI, and the PRI type is displayed; The corresponding PRI value generation in each working mode is as follows: (1) The corresponding PRI value in the working mode of the medium-frequency group transformer; First, the corresponding signal round-trip time T in the working mode of the medium-frequency group transformer is obtained. A , the expression is as follows: Among them, R d represents the current distance between the radar and the target, c represents the propagation speed of electromagnetic waves, that is, the speed of light, τ s Indicates the duration of the signal pulse, that is, the pulse width; After the signal round trip time is known, the pulse repetition interval PRI of the intermediate frequency group and the repetition frequency group is defined as follows: Among them, PRI m represents the pulse repetition interval of the mth group of the intermediate frequency group, T B Indicates the round trip time of the signal corresponding to the working mode, D1 indicates the constant parameter of the intermediate frequency group, and m indicates the group number; PRI n represents the pulse repetition interval of the nth group of the repetition frequency group, D2 represents the constant parameter of the repetition frequency group, n represents the group number, Δ n It represents a small disturbance used to adjust the repetition rate group to ensure that the target visibility reaches the ideal level; (2) PRI value in high repetition rate group transformer working mode; Among them, R c Indicates the target distance; η channel represents the introduced channel loss parameter, that is, the loss of the signal during propagation, and is less than 1. After considering the multipath effect and attenuation, the corresponding high repetition frequency group edge working mode PRI value expression is as follows: Among them, λ represents the wavelength of the signal, a t and a r represent the acceleration of the target and radar respectively, and t represents the time; Then the pulse drift effect is introduced to obtain the pulse repetition interval after the drift effect. The expression is as follows: Among them, PRI1 represents the pulse repetition interval before being affected by the drift effect, PRI2 represents the pulse repetition interval after being affected by the drift effect, and δ drift Indicates the drift amplitude, that is, the fluctuation amplitude of the pulse interval, f drift Represents the drift frequency, which is used to simulate the fluctuation of pulse time; Introducing phase noise and Doppler effect, the expression is as follows: in, represents the pulse repetition interval after the introduction of phase noise and Doppler effect, v r Represents the radial velocity of the target, calculated using the Doppler shift formula; For the pulse sequence model, the contamination mechanism is introduced, and pulse loss, false pulses and measurement noise are considered. The pulse sequence model expression is as follows: Where S represents the pulse sequence of the working mode under non-ideal conditions with measurement noise, pulse loss, and false pulses, It means that a pulse with noise or deviation is measured at time t. represents the pulse deviation introduced by the measurement noise, Pulses indicating the presence of false pulses, Indicates false pulses introduced due to channel contamination or equipment failure. Indicates the presence of both false pulses and pulse loss; Then introduce the signal-to-noise ratio SNR and noise power P at the same time noise , the expression is as follows: P.S noise Hη noise ·k·T·B Among them, P signal represents the signal power, η noise represents the noise factor, i.e. the noise level, k represents the Boltzmann constant, T represents the system noise temperature, and B represents the signal bandwidth; Finally, the target detection probability P is introduced det , the expression is as follows: Where Q represents the Q function of the standard normal distribution, which is used to calculate the detection probability, threshold represents the detection threshold, and σ noise represents the standard deviation of the measurement noise; (3) Staggered PRI and jittered PRI generation; Only the pulse group staggered PRI mode is considered, a reference pulse repetition interval PRI is defined for each pulse group, and then a staggered effect is introduced within the pulse group, so that the interval of each pulse of the pulse group deviates from the reference PRI, but the structural integrity of the overall pulse group is maintained; First, we introduce the reference pulse group pulse repetition interval PRI group , the expression is defined as follows: Among them, T group Represents the duration of the entire pulse group, N pluses Indicates the number of pulses in a pulse group; The pulse group deviation is applied to each pulse, rather than the interval between each pulse. That is, the PRI of each pulse in the pulse group is the base PRI plus a deviation term, expressed as follows: PRI group,i =PRI group (1+δ group,i ) Among them, PRI group,i represents the PRI of the ith pulse in the pulse group, δ group,i represents the variation of the ith pulse; The variance is then defined as a random variable that obeys a specific distribution, and the expression is as follows: or, Among them, σ group The standard deviation of the pulse group difference is used to describe the degree of random deviation of each pulse repetition interval (PRI) in the pulse group. max represents the maximum value of the deviation, It means that the variance has a mean of 0 and a standard deviation of σ group Gaussian distribution of Indicates that the difference is -δ max to δ max Evenly distributed between In addition to the pulse group stagger, each pulse in the pulse group is also affected by random jitter. Then the jitter effect is introduced on the basis of the pulse group stagger. The expression is as follows: in, Represents pulse jitter, with a mean of 0 and a standard deviation of σ jitter Gaussian noise; S12, generating a pulse amplitude PA according to the characteristics of the multi-function radar working mode; The PA data corresponding to each working mode is determined by the following parameters, including: the distance between the reconnaissance aircraft and the adversary radar carrier in the current mode, the gain of the adversary radar transmitting antenna, and the gain parameter of the reconnaissance aircraft receiving antenna. The specific expression is as follows: Among them, P r (t) represents the signal power received by the reconnaissance aircraft, P t Represents the corresponding radar transmission signal power, G r ,G t , λ respectively represent the receiving antenna gain of the reconnaissance aircraft, the transmitting antenna gain of the adversary radar and the corresponding signal wavelength, L represents the system loss, F t represents the corresponding radar emission pattern, F t (θ(t), φ(t)) represents the directional gain function of the transmitting antenna pattern of the adversary radar, θ(t) and φ(t) represent the angles in the elevation and azimuth directions respectively; Then the amplitude of the received pulse is obtained by taking the square root of the received power, that is, S13, based on steps S11-S12, obtain multifunctional radar pulse group data, combine the working mode transfer rule, integrate the data set, and generate a sample data set; Based on steps S11-S12, pulse group data in each working mode is generated. On the basis of the obtained pulse group data, the working mode is transferred in combination with the working mode transfer rule between the corresponding working modes to generate corresponding working mode data with working mode transfer relationship, that is, the corresponding sample data set; S14, performing data preprocessing based on the sample data set obtained in step S13, and dividing the sample data set into a training set and a test set after the preprocessing; The pulse data sets generated in different working modes are distinguished by working mode labels, and then the pulse data sets are filled with data. Meanwhile, the pollution of the data sets caused by the corresponding non-ideal conditions in the real electromagnetic environment is considered, and a reconnaissance aircraft intercepted pulse sequence model is proposed to deal with data pollution of the filled data sets. The pulse data set is the sample data set obtained in step S13; the pulse sequence model intercepted by the reconnaissance aircraft is the pulse sequence model in step S12; Finally, the preprocessed dataset is divided into training set and test set in a ratio of 7:

3.

3. A multifunctional radar working mode recognition method based on LSTCN network according to claim 1, characterized in that: The step S3 is specifically as follows: Input the training set data generated in step S1 into the network model described in step S2 for training, and adjust the hyperparameters of the neural network according to the accuracy of the training set, including: learning rate and batch size, until the neural network model converges, the loss function size remains unchanged, and the model training is completed; Among them, the basic cross entropy loss function loss fc The expression is as follows: Among them, N train represents the number of samples in the training set, p t represents the output of the network at time t, that is, the possible probability of each working mode judged by the network at time t, p t Indicates the actual working mode at time t; L i represents the length of the pulse sequence under the i-th training sample, s t represents the true label at time t; λ1 represents the corresponding regularization factor, ω2 represents the L2 regularization weight vector; Based on the basic cross entropy loss function, an adaptive weight adjustment mechanism and a multi-task learning strategy are introduced, and the expression is as follows: loss total =a·loss fc +β·loss noise +γ·loss multi-task Among them, loss total Represents the total loss, loss fc Represents the loss for changes in pulse repetition interval; loss noise Represents the loss term for noise and channel effects; loss multi-task represents the multi-task learning loss, which is used for the comprehensive optimization of different radar modes; α, β, and γ represent adaptive weight parameters, which are dynamically adjusted according to the importance of different tasks during the training process.

4. The method for identifying a multifunctional radar working mode based on an LSTCN network according to claim 1, characterized in that: The step S4 is specifically as follows: The test set data generated in step S1 is fed into the network model trained in step S2. The result output by the model is the radar working mode possibility label. The corresponding working mode recognition accuracy is calculated according to the label to obtain the corresponding pulse-level multi-function radar working mode recognition result, and the multi-function radar working mode recognition test is completed. Among them, the weighted accuracy formula is used to calculate the corresponding working mode recognition accuracy, and the expression is as follows: Among them, acc p represents the accuracy, w t Represents the weight of each sample, which is adjusted according to the category importance or sample difficulty, N test represents the total number of corresponding test samples, Indicates the result label of the working mode after recognition, y t Indicates the actual working mode label.

Images