Radar composite active jamming recognition method based on series BP neural network

Abstract

The application discloses a radar composite active jamming recognition method based on a series BP neural network, and comprises the following steps: extracting time domain features and frequency domain features of a signal to be recognized; performing segmented Fourier transform on the signal to be recognized, and extracting frequency domain features of the signal to be recognized after the transform; inputting the time domain features, the frequency domain features and the transformed frequency domain features into a trained series BP neural network model to obtain a jamming type of the signal to be recognized; wherein, the series BP neural network model is trained according to composite jamming signals of multiple jamming types, the series BP neural network model comprises multiple recognition modules connected in series, and the recognition type of a previous recognition module is composed of the recognition types of next recognition modules. The recognition method provided by the application can recognize the jamming type of the composite jamming signal.

Description

Technical Field

[0001] This invention belongs to the field of signal recognition technology, specifically relating to a radar composite active interference identification method based on a cascaded BP neural network. Background Technology

[0002] Radar, as a crucial sensor, is primarily responsible for detection, communication, early warning, and navigation, playing a vital role in situational awareness and precise target detection. However, with the continuous development of modern electronic technology, a plethora of active radar jamming signals, employing various interference methods such as suppression and deception in the time and frequency domains, have significantly impacted the normal operation of radar. Generally, a comprehensive radar anti-jamming system should first identify the type of radar interference, secondly, implement anti-jamming methods for different types of interference, and finally, evaluate the effectiveness of these methods to ensure the radar can accurately and efficiently combat interference, thereby guaranteeing its normal functionality. Therefore, research on radar interference type identification methods, as the first step in an anti-jamming system, has significant scientific and strategic importance.

[0003] The basic process of current mainstream radar active jamming identification methods is as follows: Figure 1 As shown, firstly, the interference signal undergoes certain signal processing and data preprocessing. Signal processing includes Fourier transform and short-time Fourier transform, while data preprocessing generally includes standardization and normalization. Secondly, feature extraction is performed on the preprocessed data. Then, the features are input into the corresponding classifier to obtain the recognition result of the interference signal.

[0004] However, traditional machine learning-based radar active jamming signal identification often deals with single jamming signals, while jamming is rarely singular and may overlap in the time and frequency domains. Therefore, traditional identification methods cannot function properly when dealing with complex jamming and cannot accurately identify the type of complex jamming signal. Summary of the Invention

[0005] This invention provides a radar composite active interference identification method based on a tandem BP neural network, which can solve the problem that current interference identification methods cannot identify interference signals.

[0006] In a first aspect, embodiments of the present invention provide a radar composite active interference identification method based on a cascaded BP neural network, the method comprising:

[0007] Extract the time-domain and frequency-domain features of the signal to be identified;

[0008] Perform a piecewise Fourier transform on the signal to be identified and extract the frequency domain features of the transformed signal;

[0009] The time-domain features, frequency-domain features, and transformed frequency-domain features are input into a trained cascaded BP neural network model to obtain the interference type of the signal to be identified.

[0010] The cascaded BP neural network model is trained based on composite interference signals that include multiple types of interference. The cascaded BP neural network model includes multi-level cascaded recognition modules, and the recognition type of the previous level recognition module is a composite of the recognition types of the next level recognition module.

[0011] Secondly, embodiments of the present invention provide a radar composite active interference identification device based on a cascaded BP neural network, the device comprising:

[0012] The basic feature extraction module is used to extract the time-domain and frequency-domain features of the signal to be identified.

[0013] The Fourier transform feature extraction module is used to perform piecewise Fourier transform on the signal to be identified and extract the frequency domain features of the transformed signal.

[0014] The interference identification module includes a tandem BP neural network model. The interference identification module is used to input time-domain features, frequency-domain features, and transformed frequency-domain features into the trained tandem BP neural network model to obtain the interference type of the signal to be identified.

[0015] The cascaded BP neural network model is trained based on composite interference signals that include multiple types of interference. The cascaded BP neural network model includes multi-level cascaded recognition modules, and the recognition type of the previous level recognition module is a composite of the recognition types of the next level recognition module.

[0016] Thirdly, embodiments of the present invention provide an electronic device, including a processor and a memory, wherein the memory is used to store a computer program; the processor can be used to execute a calculator program (instructions) stored in the memory to implement the method of the first aspect described above.

[0017] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program that, when executed, can implement the method described in the first aspect above.

[0018] The beneficial effects of this invention compared to existing technologies are as follows: Since the cascaded BP neural network model in this invention is trained based on composite signals and consists of multiple cascaded recognition modules, each module can perform a more detailed identification of the interference type of the signal. Therefore, through multi-level recognition, the interference category of the composite signal can be identified. Furthermore, by extracting the piecewise Fourier features of the signal to identify it, the model's recognition performance can be enhanced, and the signal recognition effect can be improved. Attached Figure Description

[0019] Figure 1 This is a schematic diagram illustrating the implementation process of an interference signal identification method.

[0020] Figure 2 This is a schematic diagram of the structure of a cascaded BP neural network model provided in an embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram illustrating the implementation process of a training method for a cascaded BP neural network model provided in an embodiment of the present invention.

[0022] Figure 4 This is a schematic diagram of the time-domain moment skewness of a noise amplitude modulation interference + deceptive interference signal provided in an embodiment of the present invention;

[0023] Figure 5 This is a schematic diagram of the frequency domain moment skewness of a noise sweep frequency interference + deceptive interference signal provided in an embodiment of the present invention;

[0024] Figure 6a , Figure 6b These are schematic diagrams of the time-domain and frequency-domain kurtosis of a noise amplitude modulation interference + deceptive interference signal provided in an embodiment of the present invention.

[0025] Figure 7 A schematic diagram of the time-domain envelope variability of a deceptive interference and its composite interference signals provided in an embodiment of the present invention;

[0026] Figure 8 A schematic diagram of the frequency domain envelope undulation of a suppression interference and its composite interference signals provided in an embodiment of the present invention;

[0027] Figure 9 This is a schematic diagram of the normalized instantaneous amplitude spectrum maximum value of a deceptive interference and its composite interference signals, provided in an embodiment of the present invention.

[0028] Figure 10 This is a schematic diagram of the normalized instantaneous absolute value of a noise sweep frequency interference + deceptive interference signal provided in an embodiment of the present invention;

[0029] Figure 11A schematic diagram of the instantaneous absolute phase of a suppression interference and its composite interference signals provided in an embodiment of the present invention;

[0030] Figure 12 A schematic diagram of the standard deviation of the instantaneous phase of a suppression interference and its composite interference signal provided in an embodiment of the present invention;

[0031] Figure 13 This is a schematic diagram of the parameters for identifying fast intra-pulse modulation of noise sweeping interference and deceptive interference signals, provided in an embodiment of the present invention.

[0032] Figure 14 A schematic diagram of the first additive white Gaussian noise factor of a deceptive interference and its composite interference signal provided in an embodiment of the present invention;

[0033] Figure 15 A schematic diagram of the second additive white Gaussian noise factor for a noise-frequency modulation interference + deceptive interference type signal provided in an embodiment of the present invention;

[0034] Figure 16 This is a schematic diagram of the carrier frequency factor of a noise amplitude modulation interference + deceptive interference signal provided in an embodiment of the present invention;

[0035] Figure 17 This invention provides a schematic diagram of the instantaneous phase threshold probability of suppressed interference and its composite interference signals.

[0036] Figure 18 A schematic diagram of the normalized 3dB bandwidth of a deceptive interference and its composite interference signals provided in an embodiment of the present invention;

[0037] Figure 19 This is a schematic diagram illustrating a scenario for extracting piecewise Fourier transform features according to an embodiment of the present invention.

[0038] Figure 20a , Figure 20b , Figure 20c , Figure 20d , Figure 20e , Figure 20f , Figure 20g A schematic diagram of a transformed frequency domain feature provided in an embodiment of the present invention;

[0039] Figure 21 A flowchart illustrating a radar composite active interference identification method based on a tandem BP neural network, provided in an embodiment of the present invention;

[0040] Figure 22 A schematic diagram of the structure of a radar composite active interference identification device based on a tandem BP neural network provided in an embodiment of the present invention;

[0041] Figure 23a , Figure 23b This is a schematic diagram illustrating a recognition probability provided in an embodiment of the present invention;

[0042] Figure 24 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0043] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0044] Figure 2 The diagram shown is a schematic representation of a cascaded BP neural network model provided in an embodiment of the present invention.

[0045] For example, model 200 can be used to identify the type of interference in an active interference signal whose intra-pulse modulation method is linear frequency modulation.

[0046] For example, interference types can be divided into two main categories: suppression interference and deception interference. Suppression interference can include three types: Noise Amplitude Modulation Jamming (NAMJ), Noise Frequency Modulation Jamming (NFMJ), and Noise Sweep Jamming (NSJ). Deception interference can also include three types: Distance Deception Jamming (DDJ), Interrupted Sampling Repeater Jamming (ISRJ), and Smeared Spectrum Jamming (SMSP).

[0047] In one possible implementation, model 200 may include multi-level serial recognition modules (only two levels are shown here), where the recognition type of the upper-level recognition module may be composed of the recognition types of the lower-level recognition modules.

[0048] In one example, when model 200 includes two cascaded identification modules, the first-level identification module 210 can identify major interference types, such as: suppression interference and its composite interference, spoofing interference and its composite interference, noise amplitude modulation interference and spoofing interference, noise frequency modulation interference and spoofing interference, and noise sweep frequency interference and spoofing interference. The second-level identification module 220 includes multiple sub-identification modules, such as: a suppression-suppression composite sub-identification module, a spoofing-spoofing composite sub-identification module, a noise amplitude modulation-spoofing sub-identification module, a noise frequency modulation-spoofing sub-identification module, and a noise sweep frequency-spoofing sub-identification module. Each sub-identification module can identify a smaller level of interference type.

[0049] For example, the interference signal is of the type of noise FM interference + noise sweep frequency interference. When the interference signal is input into model 200, the first-level identification module 210 identifies the interference signal as a type of suppression interference and its composite interference. Therefore, the suppression-suppression composite sub-identification module in the second-level identification module is called to further identify the interference signal, and the specific interference type of the interference signal is identified as: noise FM interference + noise sweep frequency interference.

[0050] It should be understood that the present invention does not limit the specific number of layers in the recognition module of the cascaded BP neural network model.

[0051] Figure 3 The diagram illustrates the implementation flow of a training method for a cascaded BP neural network model provided by an embodiment of the present invention. Method 300 can be used to train the aforementioned model 200. As an example and not a limitation, method 300 may include steps S301-S303, which are described below.

[0052] S301 extracts the time-domain and frequency-domain features of the sample signal.

[0053] For example, the sample signal may include a composite interference sample signal and a single interference sample signal.

[0054] For example, a composite interference sample signal can be composed of two or three interference signals.

[0055] For example, the interference types of composite interference sample signals can include: noise amplitude modulation interference + noise frequency modulation interference, noise amplitude modulation interference + noise frequency modulation interference + noise sweep frequency interference.

[0056] In one possible implementation, the time-domain features may include one or more of the following: time-domain moment skewness, time-domain moment kurtosis, time-domain envelope undulation, standard deviation of normalized instantaneous amplitude absolute value, standard deviation of instantaneous phase absolute value, standard deviation of instantaneous phase, fast intra-pulse modulation mode identification parameters, and instantaneous phase threshold probability.

[0057] In one possible implementation, the frequency domain features may include one or more of the following: frequency domain moment skewness, frequency domain moment kurtosis, frequency domain envelope ripple, normalized instantaneous amplitude spectrum maximum value, additive white Gaussian noise factor, carrier frequency factor, and normalized 3dB bandwidth.

[0058] In one example, the moment skewness coefficient reflects the degree of symmetry in the statistical distribution of a random variable, and it can satisfy the following formula:

[0059]

[0060] Where μ represents the mean of random variable X, and σ represents the standard deviation of random variable X. Random variable X can be time-domain or frequency-domain data of sample signals, and a3 can be the time-domain moment skewness or the frequency-domain moment skewness.

[0061] Specifically, see Figure 4 The time-domain moment skewness variation curve of medium-noise amplitude-modulated interference + deceptive interference signals, and Figure 5 The curves showing the frequency domain moment skewness variation of noise-sweeping interference and deceptive interference signals are presented. AM represents noise amplitude modulation interference, and JNR is the ratio of remaining interference power to noise power when the radar is in an interference environment and anti-interference measures are implemented. It can be seen that the trends and amplitudes of the time-domain and frequency-domain moment skewness variations differ for different types of signals. By extracting the variation characteristics of the time-domain and frequency-domain moment skewness, the type of interference can be effectively distinguished.

[0062] In one example, the kurtosis coefficient reflects the sharpness of the statistical distribution of a random variable X, and it can satisfy the following formula:

[0063]

[0064] Where a4 can represent the time-domain moment kurtosis (see...) Figure 6a ) and frequency domain moment kurtosis (see Figure 6b ).

[0065] Specifically, see Figure 6a The time-domain moment kurtosis variation curve of medium-noise amplitude modulation interference + deceptive interference signal and Figure 6b The curves showing the frequency domain kurtosis variation of a signal with medium-noise FM interference and deceptive interference are presented, where FM represents medium-noise FM interference. It can be seen that the trends and amplitudes of the time-domain and frequency-domain kurtosis variations differ among different signal types. By extracting the variation characteristics of the time-domain and frequency-domain kurtosis, the type of interference can be effectively distinguished.

[0066] In one example, envelope undulation can reflect the degree of change in the signal envelope. If the sample signal x(t) satisfies Where A(t) is the amplitude, and ω0 is the angular frequency. Let x(n) represent the discrete sequence of the sampled signal. Performing a Fast Fourier Transform on the discrete sampled sequence yields the Discrete Fourier Transform sequence X(m). Then the square of the instantaneous envelope of the signal |x(n)| 2 It can be represented as: |x(n)| 2 =Re 2 [x(n)]+Im 2 [x(n)], where Re represents the real part and Im denotes the imaginary part. The square of the signal's frequency domain envelope |X(m)| 2 It can be represented as |X(m)| 2 =Re 2 [X(m)]+Im 2 [X(m)] can then be used to calculate the envelope undulation:

[0067]

[0068] Where μ is the mean of the instantaneous square envelope of the signal or the square envelope of the signal in the frequency domain, σ is the standard deviation of the instantaneous square envelope of the signal or the square envelope of the signal in the frequency domain, and R is the time-domain envelope variability or the frequency-domain envelope variability.

[0069] Specifically, see Figure 7 The time-domain envelope undulation curves of deceptive interference and its composite interference signals and Figure 8 The curves showing the variation of the frequency domain envelope undulation of medium-suppression interference and its composite interference signals are presented. It can be seen that the trends and amplitudes of the variation of the time-domain and frequency-domain envelope undulations differ among different types of signals. By extracting the variation characteristics of the time-domain and frequency-domain envelope undulations, the type of interference can be effectively distinguished.

[0070] In one example, if the sample signal is x(t), and its discrete sampled sequence is represented as x(n), performing a Hilbert transform on the discrete sampled sequence yields its analytical expression as follows: in Let z(n) be the Hilbert transform of x(n), where z(n) is the analytical expression of x(n) after the Hilbert transform. Let be the imaginary part of this analytical expression. Then the instantaneous amplitude of the sample signal can be expressed as: The maximum value of the normalized instantaneous amplitude spectrum can satisfy the following formula:

[0071]

[0072] Where, γ max The maximum value of the normalized instantaneous amplitude spectrum (see Figure 9 ), Ns a is the number of sampling points. cn (n) = a(n) / E[a(n)]⁻¹, where DFT represents the Hilbert transform. a(n) is the instantaneous amplitude of the sample signal. Here, the instantaneous amplitude is normalized using the average value to eliminate the influence of signal gain. The maximum value of the normalized instantaneous amplitude spectrum can be used to distinguish between signals containing amplitude fluctuation information and signals that do not contain amplitude fluctuation information.

[0073] Specifically, see Figure 9 The variation curves of the normalized instantaneous amplitude spectrum maximum value of deceptive interference and its composite interference signals show that the variation trends and amplitudes of the normalized instantaneous amplitude spectrum maximum value of different types of signals are different. By extracting the variation characteristics of the normalized instantaneous amplitude spectrum maximum value, the interference type of the signal can be effectively distinguished, especially for ISRJ type interference signals.

[0074] In one example, the standard deviation of the normalized instantaneous absolute value of amplitude can satisfy the following formula:

[0075]

[0076] Where, σ aa The standard deviation of the normalized instantaneous amplitude absolute value (see Figure 10 ).

[0077] Specifically, see Figure 10 The variation curves of the standard deviation of the normalized instantaneous amplitude absolute value of signals with medium-noise sweep frequency interference and deceptive interference show that the variation trends and amplitudes of the standard deviation of the normalized instantaneous amplitude absolute value are different for different types of signals. By extracting the variation characteristics of the standard deviation of the normalized instantaneous amplitude absolute value, the interference type of the signal can be effectively distinguished.

[0078] In one example, the instantaneous phase Φ(n) of the sample signal can satisfy... However, since the phase calculated by the arctangent function is defined in the interval... Therefore, phase aliasing will occur, and Φ(n) needs to be de-aliased to transform the signal into the [0, 2π] interval. The de-aliasing operation process is as follows:

[0079]

[0080] After the operation, the dealiased instantaneous phase sequence Φ can be obtained. NL (n), from which we can derive that the standard deviation of the absolute value of the central nonlinear component of the instantaneous phase satisfies the following formula:

[0081]

[0082] Where, σ dp The standard deviation of the absolute value of the instantaneous phase (see Figure 11 ), a t This is an amplitude decision threshold for judging weak signals. If it is below this threshold, the estimation of the instantaneous phase is very sensitive to noise. c is Φ NL(n) The condition a is satisfied n (n)>a t The number of signal points.

[0083] Specifically, see Figure 11 The curves showing the variation of the standard deviation of the instantaneous phase absolute value of the medium-suppression interference and its composite interference signals reveal that the variation trends and amplitudes of the standard deviation of the instantaneous phase absolute value differ among different types of signals. By extracting the variation characteristics of the standard deviation of the instantaneous phase absolute value, the interference type of the signal can be effectively distinguished.

[0084] In one example, the standard deviation of the instantaneous phase can satisfy the following formula:

[0085]

[0086] Where, σ dp The standard deviation of the instantaneous phase (see Figure 12 (Standard deviation of the instantaneous phase of medium-suppression interference and its composite interference signals).

[0087] In one example, the fast intra-pulse modulation scheme identification parameters can satisfy the following formula:

[0088]

[0089] Where C is the fast intra-pulse modulation mode identification parameter (see...) Figure 13 The parameters for identifying fast intra-pulse modulation of signals with medium noise sweep frequency interference and deceptive interference are as follows: R(n) is the autocorrelation function of the sample signal, and n1, n2, and n3 are the locations of the sampling points of the autocorrelation function, and the distance between them is equal.

[0090] In one example, for typical additive white Gaussian noise, the envelope of the signal spectrum changes relatively smoothly. However, for other forms of signal envelopes, the changes are generally more pronounced, and the degree of envelope change is usually related to the form of the signal. Therefore, the additive white Gaussian noise factor can satisfy the following formula:

[0091]

[0092] Where A1 and A2 are the first additive Gaussian white noise factor and the second additive Gaussian white noise factor, respectively (see...). Figure 14The first additive white Gaussian noise factor of deceptive interference and its composite interference signals and Figure 15 The second additive white Gaussian noise factor of the medium noise FM interference + deceptive interference signal, E1 is the mean of the power spectrum of the input sample signal, E2 is the mean of the power spectrum greater than E1, and E2 is the mean of the power spectrum greater than E2.

[0093] In one example, performing a Fast Fourier Transform (FFT) on the sample signal x(n) yields the Discrete Fourier Transform (DFT) X(n). The carrier frequency factor is obtained by counting the maximum and second-largest values ​​of X(n). The carrier frequency factor satisfies the following formula:

[0094]

[0095] Where R2 is the carrier frequency factor (see...) Figure 16 Carrier frequency factor of medium noise amplitude modulation interference + deceptive interference signal), X(n) max ) is the maximum value of X(n), X(n) min ) is the second largest value of X(n).

[0096] In one example, the probability within the instantaneous phase threshold (see...) Figure 17 The instantaneous phase threshold probability of a signal in the mid-suppression interference and its composite interference types is the probability that the instantaneous phase of the signal will appear near the upper threshold of the phase mean. The instantaneous phase threshold probability can satisfy the following formula:

[0097]

[0098] Where P is the probability within the instantaneous phase threshold, θ0(n) is the instantaneous phase of the sample signal, and θ mean Let θ be the mean of the instantaneous phase. g Let num{·} be the instantaneous phase threshold, and let num{·} represent the number of points where the instantaneous phase of the signal appears near the upper or lower threshold of the phase mean.

[0099] In one example, different types of interference occupy different bandwidths, and these differences in bandwidth can be used to construct characteristic parameters.

[0100] For example, the normalized spectrum of a sample signal can satisfy the following formula:

[0101]

[0102] Among them, S u S(n) is the normalized spectrum, S(n) = |FFT(x(n))| is the magnitude of the Fast Fourier Transform of the received signal, and N is the sampling length.

[0103] Then the normalized spectrum has a 3dB bandwidth (see...) Figure 18The normalized spectrum (3dB bandwidth) of deceptive interference and its composite interference signals can satisfy the following formula:

[0104]

[0105] Among them, B ω For the normalized spectrum 3dB bandwidth, The normalized spectrum threshold is set, typically 0.707.

[0106] Specifically, see Figures 11-18 The variation curves of the standard deviation of the instantaneous phase, the fast intra-pulse modulation identification parameters, the first additive white Gaussian noise factor, the second additive white Gaussian noise factor, the carrier frequency factor, the instantaneous phase threshold probability, and the normalized spectrum 3dB bandwidth show that the variation trends and amplitudes of different parameters of different types of signals are different. By extracting the variation characteristics of each parameter, the interference type of the signal can be effectively distinguished.

[0107] S302, extract the transformed frequency domain features of the sample signal.

[0108] In one possible implementation, see Figure 19 The sample signal within a pulse can be sliced ​​to obtain multiple sub-pulse signals. A Fast Fourier Transform (FFT) is then performed on each of these sub-pulse signals to obtain the transformed sub-pulse signals. The frequency domain features of the transformed sub-pulse signals are extracted to obtain segmented frequency domain features (i.e., segmented Fourier Transform features). These segmented frequency domain features are then concatenated to obtain the transformed frequency domain features of the sample signal.

[0109] In one example, the mean of the segmented frequency domain features of all sub-pulse signals can be determined as the frequency domain features after the sample signal transformation, so as to complete the splicing process of the segmented frequency domain features.

[0110] Traditional methods using the Fast Fourier Transform (FFT) to extract frequency domain signal features represent the overall characteristics of a signal over a pulse duration, but cannot capture the signal's characteristics over a short time. Furthermore, time-frequency analysis of the signal is relatively complex. In this invention, by segmenting the complete pulse signal and performing Fourier Transforms on each segment, the short-time characteristics of the signal can be obtained more easily, improving the distinguishability between different types of signals.

[0111] Specifically, see Figure 20a , Figure 20b , Figure 20c , Figure 20d , Figure 20e , Figure 20f , Figure 20gThe curves showing the changes in the frequency domain moment skewness, frequency domain moment kurtosis, carrier factor, first additive white Gaussian noise factor, second additive white Gaussian noise factor, normalized 3dB bandwidth, and frequency domain envelope ripple after the transformation are presented. Figure 12 , Figure 17 In comparison, it can be seen that the transformed frequency domain features have higher distinguishability.

[0112] S303, based on the time-domain characteristics, frequency-domain characteristics, and transformed frequency-domain characteristics of the sample signal, a tandem BP neural network model is trained to obtain a trained tandem BP neural network model.

[0113] According to the model training method provided by the present invention, the model is trained by using composite interference sample signals and single interference sample signals, so that the trained model can identify composite interference signals.

[0114] The wireless blockchain network sharding method provided in this embodiment of the invention can be applied to electronic devices such as mobile terminals, personal laptops, and supercomputers. This embodiment of the invention does not impose any restrictions on the specific type of electronic device.

[0115] Figure 21 The diagram shown is a schematic flowchart of a radar composite active interference identification method based on a cascaded BP neural network according to an embodiment of the present invention. As an example and not a limitation, method 21 may include steps S211-S213, and method 21 can be applied to the aforementioned electronic device. The steps are described below.

[0116] S211, extract the time-domain and frequency-domain features of the signal to be identified.

[0117] S212, perform piecewise Fourier transform on the signal to be identified, and extract the frequency domain features of the transformed signal.

[0118] The steps S211 and S212 in method 21 are the same as steps S301 and S302 in method 300, except that the execution object is different. In method 300, the execution object is the sample signal, while in method 21 it is the signal to be identified. Please refer to the relevant description in method 300 for details, which will not be repeated here.

[0119] S213, input the time domain features, frequency domain features and transformed frequency domain features into the trained cascaded BP neural network model to obtain the interference type of the signal to be identified.

[0120] For example, the trained cascaded BP neural network model can be model 200 trained by method 300.

[0121] Since the cascaded BP neural network model in this invention is trained based on composite signals and consists of multiple cascaded recognition modules, each module can perform a more detailed identification of the interference type of the signal. Therefore, through multi-level recognition, the interference category of the composite signal can be identified. Furthermore, by extracting the piecewise Fourier features of the signal to identify the signal, the model's recognition performance can be enhanced, and the signal recognition effect can be improved.

[0122] Figure 22 The diagram shown is a schematic representation of a radar composite active interference identification device based on a tandem BP neural network, provided in an embodiment of the present invention.

[0123] As an example and not a limitation, device 22 may include:

[0124] The basic feature extraction module 221 is used to extract the time-domain and frequency-domain features of the signal to be identified.

[0125] Fourier transform feature extraction module 222 is used to perform piecewise Fourier transform on the signal to be identified and extract the frequency domain features of the transformed signal.

[0126] Interference identification module 223 includes a tandem BP neural network model. The interference identification module is used to input time-domain features, frequency-domain features and transformed frequency-domain features into the trained tandem BP neural network model to obtain the interference type of the signal to be identified.

[0127] The cascaded BP neural network model is trained based on composite interference signals that include multiple types of interference. The cascaded BP neural network model includes multi-level cascaded recognition modules, and the recognition type of the previous level recognition module is a composite of the recognition types of the next level recognition module.

[0128] To better illustrate the beneficial effects of the present invention, the following simulation experiments were conducted:

[0129] For example, the radar parameters used in the simulation experiment are shown in Table 1 below, and its signal waveform uses a linear frequency modulated signal. The parameters of the interference signal are shown in Table 2 below.

[0130] Table 1 Simulation Radar Parameters

[0131] Parameter name Parameter value radar bandwidth 10MHz Sampling frequency 200MHz Pulse width 10us

[0132] Table 2 Simulation parameters of six types of interference and their combined interference signals

[0133]

[0134]

[0135] Figure 23 shows a schematic diagram of an identification probability provided by an embodiment of the present invention.

[0136] For example, in the simulation experiment, the cascaded BP neural network model proposed in this invention can be used to randomly generate 1000 samples for each type of interference within the 0-20dB interference-to-noise ratio range, and extract their time-domain, frequency-domain, and piecewise Fourier transform domain feature parameters as the training dataset. Furthermore, for each interference signal within the -10-20dB range, 500 samples are generated for each interference signal under each integer interference-to-noise ratio condition as the validation set. The simulation uses the neural network toolkit included in Matlab. All networks are configured as four-layer networks, each containing 16 neurons, and the activation function is the tanh function. Additionally, the momentum factor is set to 0.9, and the learning rate is set to 0.25*e. -i +1 -6 , where i is the number of training iterations. 80% of the samples in the training set are randomly selected for training, and the remaining 20% ​​are used for testing. Training is considered complete when the change in the recognition rate on the test set does not exceed 0.5% for ten consecutive times, and the trained model is saved. Finally, the validation samples are input into the trained cascaded backpropagation neural network model.

[0137] See Figure 23a and Figure 23b It can be seen that the identification results of the three major types of interference are relatively ideal: suppression interference and its composite interference, deceptive interference and its composite interference, and noise amplitude modulation interference + deceptive interference. The identification rate can reach more than 90% under the condition of greater than 3dB. In the identification of noise frequency modulation interference + deceptive interference, except for noise amplitude modulation interference + distance deceptive interference + spectrum dispersion interference which can achieve a recognition rate of more than 90% after greater than 7dB, other types of interference in noise frequency modulation interference + deceptive interference and noise sweep frequency interference + deceptive interference can all achieve a recognition rate of more than 90% after greater than 5dB.

[0138] Table 3 Comparison of recognition rates for 16 features and 23 features

[0139]

[0140]

[0141] For example, Table 3 above shows the recognition rates of various types of interference signals when 16 time-domain and frequency-domain features are input into Model 200 within a JNR range of 0–20 dB; and the comparison of the recognition rates of various types of interference signals when 23 features (16 time-domain features, frequency-domain features, and 7 transformed frequency-domain features) are input into Model 200. Referring to Table 3, it can be concluded that the recognition rate using the 23 transformed frequency-domain features is greater than the recognition rate using the 16 features, with a total recognition rate improvement of 4%. This demonstrates that extracting piecewise Fourier features of the signal for signal recognition can enhance the model's recognition performance and improve the signal recognition effect.

[0142] Therefore, the identification method provided by the present invention can identify the interference type of composite signals, and by extracting the segmented Fourier features of the signal for identification, the signal identification effect can be improved.

[0143] Figure 24 The diagram shown is a structural schematic of an electronic device provided in an embodiment of the present invention. Figure 24 The illustrated electronic device 2400 may include: at least one processor 2410 ( Figure 24 The diagram shows only one processor, a memory 2420, and a computer program 2430 stored in the memory 2420 and executable on the at least one processor 2410, which, when executing the computer program 2430, implements the steps in any of the above method embodiments.

[0144] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps described in the various method embodiments above.

[0145] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

Claims

1. A radar composite active interference identification method based on a cascaded BP neural network, characterized in that, include: Extract the time-domain and frequency-domain features of the signal to be identified; Perform a piecewise Fourier transform on the signal to be identified, and extract the frequency domain features of the transformed signal; The time-domain features, the frequency-domain features, and the transformed frequency-domain features are input into a trained cascaded BP neural network model to obtain the interference type of the signal to be identified. The cascaded BP neural network model is trained based on composite interference signals including multiple interference types. The cascaded BP neural network model includes multi-level cascaded recognition modules, and the recognition type of the previous level recognition module is a composite of the recognition types of the next level recognition module.

2. The method according to claim 1, characterized in that, The time-domain features include one or more of the following: Time-domain moment skewness, time-domain moment kurtosis, time-domain envelope undulation, standard deviation of normalized instantaneous amplitude absolute value, standard deviation of instantaneous phase absolute value, standard deviation of instantaneous phase, fast intra-pulse modulation mode identification parameters, and instantaneous phase threshold probability.

3. The method according to claim 1, characterized in that, The frequency domain features include one or more of the following: Frequency domain moment skewness, frequency domain moment kurtosis, frequency domain envelope ripple, normalized instantaneous amplitude spectrum maximum value, additive white Gaussian noise factor, carrier frequency factor, normalized spectrum 3dB bandwidth.

4. The method according to claim 1, characterized in that, The step of performing a piecewise Fourier transform on the signal to be identified and extracting the frequency domain features of the transformed signal includes: The signal to be identified within a pulse is sliced ​​to obtain multiple sub-pulse signals; Perform Fast Fourier Transform on each of the sub-pulse signals to obtain the transformed sub-pulse signals; Extract the frequency domain features of the transformed sub-pulse signal to obtain the segmented frequency domain features of the sub-pulse signal; The segmented frequency domain features are spliced ​​together to obtain the transformed frequency domain features.

5. The method according to claim 4, characterized in that, The process of concatenating the segmented frequency domain features to obtain the transformed frequency domain features includes: The mean of the segmented frequency domain features is determined as the transformed frequency domain features.

6. The method according to claim 1, characterized in that, The cascaded BP neural network model includes a two-level recognition module, and the training method of the cascaded BP neural network model includes: Extract the time-domain features, frequency-domain features, and transformed frequency-domain features of the composite interference sample signal and the single interference sample signal; Based on the time-domain features, frequency-domain features, and transformed frequency-domain features of the composite interference sample signal and the single interference sample signal, the tandem BP neural network model is trained to obtain the trained tandem BP neural network model.

7. The method according to claim 6, characterized in that, The composite interference signal is composed of two or three interference signals.

8. A radar composite active interference identification device based on a cascaded BP neural network, characterized in that, The device includes: A basic feature extraction module is used to extract the time-domain and frequency-domain features of the signal to be identified. The Fourier transform feature extraction module is used to perform a segmented Fourier transform on the signal to be identified and extract the frequency domain features of the transformed signal. An interference identification module, comprising a cascaded BP neural network model, is used to input the time-domain features, the frequency-domain features, and the transformed frequency-domain features into a trained cascaded BP neural network model to obtain the interference type of the signal to be identified. The cascaded BP neural network model is trained based on composite interference signals including multiple interference types. The cascaded BP neural network model includes multi-level cascaded recognition modules, and the recognition type of the previous level recognition module is a composite of the recognition types of the next level recognition module.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by an electronic device, it implements the method as described in any one of claims 1-7.

Images