Noise waveform generation method based on radar signal and related equipment

Abstract

The invention provides a noise waveform generation method based on a radar signal and related equipment. The method comprises the following steps: acquiring an original radar signal; performing signal spectrum analysis on the original radar signal, and extracting a main frequency component; performing signal folding on the original radar signal according to the period corresponding to the main frequency component, and obtaining a two-dimensional feature image corresponding to each main frequency component; constructing a noise generation network model, and training the noise generation network model by using a pre-constructed complex radar signal data set until the loss value of the noise generation network model is smaller than a preset loss threshold value, thereby obtaining the trained noise generation network model; inputting the two-dimensional feature image corresponding to each main frequency component into the trained noise generation network model to obtain a noise waveform based on the original radar signal, and performing convolution processing on the noise waveform and the original radar signal to obtain a final noise waveform; and sending the noise waveform signal to an emission source.

Description

Technical Field

[0001] The present invention belongs to the field of radar technology, and in particular relates to a noise waveform generation method based on radar signals and related equipment. Background Art

[0002] Traditional noise waveforms, such as RF noise or noise-modulated signals, are widely used due to their simplicity of modulation and low reliance on radar signal reconnaissance information. However, modern radars widely employ waveform agility and coherent processing techniques to suppress noise waveforms. To counteract radar anti-jamming technology, active radar jamming techniques based on digital radio frequency memory (DRFM) have rapidly developed. Smart noise jamming (based on convolution modulation) was originally proposed by Schleich et al. This method, based on DRFM technology, convolves the received and stored radar signal with a noise signal, followed by amplification and forwarding. The resulting smart noise signal, convolved with the received signal, achieves the same processing gain as the radar signal and can automatically align with the radar signal's frequency band, effectively affecting detection radars. Compared to traditional noise signals, smart noise signals generated by modulating intercepted radar signals exhibit significant advantages in achieving radar processing gain, offering significant effectiveness and high engineering feasibility.

[0003] Modern intelligent radar systems are capable of flexibly adjusting operating parameters and modes based on the operating environment and mission requirements. They are also equipped with anti-interference technologies such as frequency agility and waveform agility. These technologies significantly increase the difficulty of smart noise devices impacting radars. Furthermore, for convolutional smart noise, different noise levels can significantly impact the effect. Furthermore, existing smart noise generation technologies are unable to synthesize optimal noise waveforms tailored to the external adversarial environment, making it difficult to effectively impact modern intelligent radars. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a noise waveform generation method based on radar signals and related equipment.

[0005] In a first aspect, the present invention provides a method for generating a noise waveform based on a radar signal, the method comprising the following steps: Obtaining the original radar signal emitted by the transmitting source; Perform signal spectrum analysis on the original radar signal to extract the main frequency components in the original radar signal; According to the period corresponding to the main frequency component, the original radar signal is folded to obtain the two-dimensional feature image corresponding to each main frequency component; the two-dimensional feature image is used to describe the timing characteristics of the original radar signal within each period and between adjacent periods; A noise generation network model is constructed and trained using a pre-constructed complex radar signal data set until the loss value of the noise generation network model is less than a preset loss threshold, thereby obtaining a trained noise generation network model. The noise generation network model includes an encoder, a decoder, and a noise generation module. The encoder uses depthwise separable convolution to extract the spatiotemporal joint features of the input two-dimensional feature image and uses the spatiotemporal joint features to encode the input two-dimensional feature image to obtain a coded feature map. The depthwise separable convolution is composed of depthwise convolution and pointwise convolution. The decoder is used to decode the coded feature map. The noise generation module is used to generate a noise waveform based on the decoding result and the size of the main frequency components. The two-dimensional feature image corresponding to each main frequency component is input into the trained noise generation network model to obtain a noise waveform based on the original radar signal. The noise waveform and the original radar signal are then convolved to obtain the final noise signal. The resulting noise signal is sent to the transmitting source.

[0006] Optionally, perform signal spectrum analysis on the original radar signal to extract the main frequency components in the original radar signal, including: Perform Fourier transform on the original radar signal to obtain the spectrum of the original radar signal; Extract multiple frequency components from the spectrum as main frequency components in descending order of amplitude.

[0007] Optionally, signal folding is performed on the original radar signal according to the period corresponding to the main frequency component to obtain a two-dimensional feature image corresponding to each main frequency component, including: By calculating the formula

[0008] Get the first The two-dimensional feature image corresponding to the main frequency components ;in, represents the original radar signal, Indicates zero-fill operation, Represents a folding operation, Indicates the The period corresponding to the main frequency components is Indicates the The frequency magnitude of the main frequency components, , Indicates the total number of dominant frequency components.

[0009] Optionally, the spatiotemporal joint features include spatial features and channel features; Spatial features are extracted by depth-wise convolution; depth-wise convolution performs independent convolution on each input channel of the input two-dimensional feature image; Channel features are extracted by point-by-point convolution; point-by-point convolution aggregates the convolution results of all input channels.

[0010] Optionally, a noise waveform is generated based on the decoding result and the magnitude of the main frequency components, including: By calculating the formula

[0011]

[0012]

[0013] Get the noise waveform ;in, Indicates the The weights of the main frequency components, Indicates the The decoding result of the two-dimensional feature image corresponding to the main frequency components, Denotes a decoding operation, Indicates the The encoding feature map corresponding to the main frequency components, Represents the encoding operation, represents batch normalization, Represents the activation function.

[0014] Optionally, the complex radar signal data set includes multiple complex radar signals; the complex radar signals are generated by jointly modulating simulated radar pulse waveforms using multiple modulation methods.

[0015] Optionally, the loss value is calculated as:

[0016] in, represents the loss value, express The constant false alarm threshold height at all times, represents variance calculation, Indicates mean calculation.

[0017] In a second aspect, the present invention provides a noise waveform generator based on a radar signal, comprising: A receiving module is used to obtain the original radar signal sent by the transmitting source; An extraction module is used to perform signal spectrum analysis on the original radar signal and extract the main frequency components in the original radar signal; The signal folding module is used to fold the original radar signal according to the period corresponding to the main frequency component to obtain a two-dimensional feature image corresponding to each main frequency component; the two-dimensional feature image is used to describe the timing characteristics of the original radar signal within each period and between adjacent periods; A model processing module is used to construct a noise generation network model and train the noise generation network model using a pre-constructed complex radar signal data set until the loss value of the noise generation network model is less than a preset loss threshold, thereby obtaining a trained noise generation network model; wherein, the noise generation network model includes an encoder, a decoder, and a noise generation module, the encoder uses depthwise separable convolution to extract the spatiotemporal joint features of the input two-dimensional feature image, and uses the spatiotemporal joint features to encode the input two-dimensional feature image to obtain a coded feature map, the depthwise separable convolution is composed of depthwise convolution and pointwise convolution, the decoder is used to decode the coded feature map, and the noise generation module is used to generate a noise waveform based on the decoding result and the size of the main frequency component; The noise generation module is used to input the two-dimensional feature image corresponding to each main frequency component into the trained noise generation network model to obtain a noise waveform based on the original radar signal, and convolve the noise waveform with the original radar signal to obtain the final noise signal; The transmitting module is used to send the final noise signal to the transmitting source.

[0018] In a third aspect, the present invention provides a terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method when executing the computer program.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program, which implements the above method when executed by a processor.

[0020] The beneficial effects of the present invention are: The radar signal-based noise waveform generation method provided by the present invention performs signal spectrum analysis on the original radar signal to extract the main frequency components in the original radar signal, which can only use the frequencies containing information and avoid meaningless noise. The original radar signal is folded according to the period corresponding to the main frequency components, which can simultaneously capture the changes of the original radar signal within each period and between adjacent periods, which is conducive to improving the efficiency of feature extraction. The two-dimensional structure of the two-dimensional feature image can effectively describe local correlation and periodic consistency, thereby enhancing the interpretability of time series features. The constructed noise generation network model has an encoder that separates the calculation of depth-by-depth convolution and point-by-point convolution, which greatly reduces redundant operations and improves the efficiency of noise signal generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flowchart of a method for generating a noise waveform based on a radar signal in one embodiment of the present application; Figure 2 This is a structural diagram of a noise generation network model in one embodiment of the present application; Figure 3 This is a processing flow chart of an encoder in one embodiment of the present application; Figure 4 This is a processing flow chart of a decoder in one embodiment of the present application; Figure 5 This is a schematic diagram of the CFAR detection threshold in one embodiment of the present application; Figure 6 This is a comparison diagram of the effects of the radar signal-based noise waveform generation method proposed by the present invention and the genetic algorithm in one embodiment of the present application; Figure 7 This is a structural diagram of a noise waveform generator in one embodiment of the present application; Figure 8 This is a schematic diagram of the structure of a terminal device in one embodiment of the present application. DETAILED DESCRIPTION

[0022] In response to the problem that traditional radar signal noise waveform generation methods have poor effects, the present invention discloses a radar signal-based noise waveform generation method and related equipment. The method extracts the main frequency components in the original radar signal by performing signal spectrum analysis on the original radar signal, and can only use the frequencies containing information to avoid meaningless noise; the original radar signal is folded according to the period corresponding to the main frequency component, and the changes of the original radar signal within each period and between adjacent periods can be captured at the same time, which is conducive to improving the efficiency of feature extraction. The two-dimensional structure of the two-dimensional feature image can effectively describe local correlation and periodic consistency, and enhance the interpretability of time series features; the constructed noise generation network model, whose encoder separates the calculation of depth-by-depth convolution and point-by-point convolution, greatly reduces redundant operations and improves the efficiency of signal generation.

[0023] The noise waveform generation method based on radar signals provided by the present invention is described below.

[0024] like Figure 1 As shown, the noise waveform generation method based on radar signal includes the following steps: Step 11: Acquire the original radar signal emitted by the transmitting source.

[0025] In the embodiment of the present invention, the original radar signal emitted by the emission source can be intercepted by electronic reconnaissance equipment.

[0026] In one feasible implementation, an antenna system (such as a broadband directional antenna) can be used to intercept the original radar signal. It should be noted that the antenna must have high sensitivity and dynamic range to accommodate the needs of receiving both weak signals at long distances and strong signals at close range.

[0027] In another feasible implementation, the original radar signal may be intercepted by a receiver (superheterodyne receiver, direct sampling receiver, digital channelized receiver).

[0028] Step 12: Perform signal spectrum analysis on the original radar signal to extract the main frequency components in the original radar signal.

[0029] In an embodiment of the present invention, the above step 12 includes steps 12.1 and 12.2.

[0030] Step 12.1: Perform Fourier transform on the original radar signal to obtain the spectrum of the original radar signal.

[0031] Step 12.2: Extract multiple frequency components from the spectrum in descending order of amplitude as main frequency components.

[0032] In order to utilize only frequencies containing information and avoid the generation of meaningless noise, in an embodiment of the present invention, only the first k amplitudes in the Fourier transform result are taken, representing the k frequency components with the largest proportion in the signal, as the main frequency components.

[0033] In a feasible implementation, the process of extracting the main frequency components includes steps 12.2.1 to 12.2.2.

[0034] Step 12.2.1: Obtain the intensity of each frequency component in the spectrum.

[0035] Specifically, by calculating the formula

[0036] Get the intensity of each frequency component ;in, Indicates the amplitude value calculation, represents the fast Fourier transform, Represents the original radar signal, which is a one-dimensional time series. Represents the number of sampling points in a one-dimensional time series, Indicates mean calculation.

[0037] Step 12.2.2, select the one with the highest intensity frequency components as the main frequency components.

[0038] Specifically, by calculating the formula

[0039] Get the main frequency components .in, Indicates taking the front It should be noted that since the frequency calculated by Fourier transform is symmetrical, only the frequency from the previous Select the frequency with the highest intensity frequency components.

[0040] Step 13: fold the original radar signal according to the period corresponding to the main frequency component to obtain a two-dimensional feature image corresponding to each main frequency component.

[0041] Since the signal waveform emitted by modern radars is usually modulated based on modulation methods such as linear frequency modulation, it has periodicity. For any periodic time series, each point involves two types of time changes: one is the change between adjacent moments in the same period, and the other is the change between the same moments in adjacent periods, that is, intra-period changes and inter-period changes. However, the original one-dimensional time series structure can only present the changes between adjacent time points. If one-dimensional convolution is used to extract the features of a periodic time series, a deeper network is required to make the receptive field of the convolution kernel cover adjacent periods, which will make the computational complexity of the neural network difficult to control and affect the feasibility and engineering value of the neural network. Therefore, in order to enable the neural network to better capture time patterns originating from different periods, the present invention proposes a signal folding method similar to the fast time and slow time processing of radar.

[0042] Specifically, by calculating the formula

[0043] Get the first The two-dimensional feature image corresponding to the main frequency components ;in, represents the original radar signal, Indicates zero-fill operation, Represents a folding operation, Indicates the The period corresponding to the main frequency components is Indicates the The frequency magnitude of the main frequency components, , Indicates the total number of dominant frequency components.

[0044] In the embodiment of the present invention, the two-dimensional feature image is used to describe the time series characteristics of the original radar signal within each cycle and between adjacent cycles.

[0045] For the above two-dimensional vector Each column and row corresponds to adjacent time points and cycle-adjacent time points in the time domain, similar to fast time and slow time in radar signal processing. Adjacent time points and cycles often contain similar temporal variations. This structure clearly displays variations within and between cycles, providing a significant advantage in representational capabilities. Furthermore, because this two-dimensional tensor exhibits two-dimensional locality, 2D convolutions can capture temporal information and learn deeper features of radar signals.

[0046] Step 14: construct a noise generation network model and train the noise generation network model using a pre-constructed complex radar signal data set until the loss value of the noise generation network model is less than a preset loss threshold, thereby obtaining a trained noise generation network model.

[0047] In the embodiment of the present invention, Figure 2 As shown, the noise generation network model includes an encoder 201 , a decoder 202 and a noise generation module 203 .

[0048] The encoder 201 uses depthwise separable convolution to extract the spatiotemporal joint features of the input two-dimensional feature image, and uses the spatiotemporal joint features to encode the input two-dimensional feature image to obtain an encoded feature map.

[0049] The above-mentioned depth-wise separable convolution is composed of depth-wise convolution and point-wise convolution. The spatiotemporal joint features include spatial features and channel features.

[0050] Specifically, spatial features are extracted by depth-wise convolution, which performs independent convolution on each input channel of the input two-dimensional feature image. Channel features are extracted by point-wise convolution; point-wise convolution aggregates the convolution results of all input channels. The encoder processing flow is as follows: Figure 3 shown.

[0051] In an embodiment of the present invention, depth-wise convolution includes C convolution kernels of size K*K*1, and point-wise convolution includes O convolution kernels of size 1*1*C.

[0052] Accordingly, the computational complexity ratio V of the depthwise separable convolution kernel designed in the present invention and the standard convolution can be calculated by the following formula:

[0053] K is usually taken as 3, which means that the depthwise separable convolution designed by the present invention can save more than 80% of the multiplication and addition calculation amount compared with the standard convolution.

[0054] The decoder 202 is used to decode the encoded feature map.

[0055] In order to ensure the decoding capability of the neural network, in the embodiment of the present invention, the decoder 202 does not use depthwise separable convolution. The specific process is as follows Figure 4 It should be noted that, in the embodiment of the present invention, batch normalization and activation function are used to connect the encoder 201 and the decoder 202.

[0056] The noise generating module 203 is used to generate a noise waveform according to the decoding result and the magnitude of the main frequency component.

[0057] Since the energy proportions of different frequency components in the original radar signal are different, the feature maps (i.e., encoding and decoding results) obtained by segmentation according to different frequencies need to be weighted to obtain an accurate noise waveform.

[0058] In a feasible implementation, the relative energy values A of different frequencies may be subjected to a Softmax function to obtain a weighted value, which is multiplied by the encoding and decoding result and then summed to obtain an accurate noise waveform.

[0059] Specifically, by calculating the formula

[0060]

[0061]

[0062] Get the noise waveform ;in, Indicates the The weights of the main frequency components, Indicates the The decoding result of the two-dimensional feature image corresponding to the main frequency components, Denotes a decoding operation, Indicates the The encoding feature map corresponding to the main frequency components, Represents the encoding operation, represents batch normalization, Represents the activation function.

[0063] The following describes the training process of the noise generation network model in an embodiment of the present invention.

[0064] Specifically, the noise generation network model is trained using a pre-constructed complex radar signal data set until the loss value of the noise generation network model is less than a preset loss threshold, thereby obtaining a trained noise generation network model.

[0065] In a complex electromagnetic environment, there may be radar signals with a variety of different modulation modes and parameter settings. Since the parameters of radar signals in confrontation are not public and difficult to obtain, the present invention simulates multiple possibilities by randomly selecting radar signal modulation modes and modulation parameters. Specifically, radar pulse waveforms are generated through simulation, including linear frequency modulation signals, frequency coded signals, nonlinear frequency modulation signals, and joint modulation of random combinations of different modulation modes. The specific radar signal modulation modes are shown in Table 1, and the radar signal modulation parameters are shown in Table 2. The data set contains some mixed modulated radar signals, including overlapping signals with the same modulation mode but different frequencies and randomly overlapping signals with different modulation modes. When generating radar signals, the program will randomly select the modulation mode and modulation parameters.

[0066] Table 1

[0067] Table 2

[0068] The noise generation network model constructed was trained using this complex radar signal dataset. The parameters k=8, d=16, and the network input sequence length T=512 were set. The radar signal sequence in the dataset is 1000 points long, so the first 512 points are used as input. The loss value is calculated as:

[0069] in, represents the loss value, express The constant false alarm threshold height at all times, represents variance calculation, Indicates mean calculation.

[0070] The design idea of the loss function is as follows: Assuming that the original radar signal is a linear frequency modulation signal, its expression is:

[0071] in, is the carrier frequency, is the pulse width, is the signal bandwidth.

[0072] In order to simplify the problem, the target covered by the equipment is regarded as an ideal point target, and the distance from the radar is , the signal propagation delay is , the effective scattering cross-sectional area of the target is , then the target response function is ; The target reflected echo is ; The smart noise signal is obtained by convolving the radar signal sampled by the device with the noise.

[0073] Assume that the modulation forwarding delay of the device is , the noise signal is , then the expression of the smart noise signal is: .

[0074] Assuming that the radar uses matched filtering to process the echo signal, the pulse compression coefficient of the radar signal processor is , after passing through the matched filter, the pulse pressure expression is:

[0075] According to the derivation, Approximate a fixed carrier frequency signal with a sinc function as the envelope. The specific expression is as follows:

[0076] when When , the matched filter output of the echo can be written as:

[0077] According to the above formula, the echo signal output after matched filter processing is The relationship is closely related, especially when the radar signal adopts a more complex modulation method. This relationship is difficult to describe clearly through precise mathematical expressions, and the effects caused by different noise h(t) are significantly different.

[0078] In order to quantify the effect, the present invention uses constant false alarm rate (CFAR) detection to evaluate the effect. CFAR detection is a commonly used radar target detection technology that can adaptively adjust the target detection threshold. Its expression is:

[0079] in, They represent the number of unilateral protection units and averaging units respectively. The unilateral protection unit is used to eliminate the influence of the target signal itself on the noise estimation, and the averaging unit is used to calculate the average energy of the background noise. Indicates the false alarm rate.

[0080] According to the principle of matched filtering, to achieve optimal local suppression of the signal after pulse compression, the energy distribution of the signal after pulse compression should be as close to uniform as possible. At the same time, care should be taken to avoid excessive power loss after pulse compression, which could reduce effectiveness. Specifically, to effectively disrupt enemy radar detection and mask the true target signal, the amplitude of the output signal from all reference units in the detection process needs to be increased. In other words, the higher the average constant false alarm threshold and the smaller the variance, the better the effect. This is how the above loss value calculation formula is derived.

[0081] In step 15, the two-dimensional feature image corresponding to each main frequency component is input into the trained noise generation network model to obtain a noise waveform based on the original radar signal, and the noise waveform and the original radar signal are convolved to obtain the final noise signal.

[0082] The functional processing steps of each module have been described in detail in step 14 and will not be repeated here. Please refer to the relevant description of step 14 for details.

[0083] Step 16: Send the final noise signal to the transmitting source.

[0084] In a feasible embodiment, since the radar signal is modulated, the transmitting source is no longer able to detect the target through the received radar echo, thereby ensuring that the target is not exposed.

[0085] In one embodiment of the present invention, to demonstrate the optimized value of the loss function constructed by the present invention, a Monte Carlo simulation was performed. A linear frequency modulation (LFM) signal with a duration of 10 μs, a bandwidth of 40 MHz, and 1000 sampling points was created. 100,000 random Gaussian white noise signals were generated, each with 1000 sampling points. Each noise signal was convolved with the LFM signal, and the loss function value was calculated while maintaining the same signal power. The resulting loss function value approximated a normal distribution, with a mean of 4.94 and a variance of 5.76.

[0086] In order to intuitively demonstrate the impact of different noises on the effect, two different noises are used while ensuring the same signal power. Convolve the linear frequency modulation signal with noise to generate interference waveform , after passing through the matched filter and CFAR detector, the detection thresholds 1 and 2 are obtained, as shown in Figure 5 shown. Figure 5 In the example, the maximum value of the matched filter of the real target echo is Near the target position, threshold 1 is obviously higher than threshold 1 and is flatter, indicating that the noise The noise convolution generated can effectively increase the detection threshold, reduce the probability of radar detecting the shielded target, and make it more difficult for the radar to distinguish. and The calculated loss function values are 2.52 and 13.21 respectively, proving that the function can effectively reflect the effect.

[0087] Change the radar signal modulation mode to frequency coded signal and recalculate the noise and The loss values are 3.52 and 16.8 respectively, which proves that for different radar signals, specific noise, that is, excellent noise, is needed to modulate and generate a noise waveform with the best effect.

[0088] In another embodiment of the present invention, in order to verify the effectiveness of the noise generation network model constructed by the present invention, the Monte Carlo method is used to evaluate the optimization efficiency of the noise generation network model. radar signals, and the noise generated by the noise generation network model is combined with the random noise (noise waveform) The loss value of the convolution smart noise modulated by the optimal noise and random noise is calculated respectively. , For the entire validation set, The average value is about 4.94, The average value is 2.41, which means that the excellent noise has achieved a 51.2% improvement in effect.

[0089] statistics The median is less than , calculate its proportion Since the smaller the loss function value, the better the effect, when The smaller it is, the better the N generated by the noise generation network model is than random noise. Traverse all radar signals in the validation set and calculate A distribution diagram was obtained. The results show that for 93.7% of radar signals in the validation set, the probability that randomly generated noise outperformed the optimal noise generated using the noise generation network model was less than 20%. Furthermore, for 88.3% of radar signals in the validation set, the probability that random noise outperformed the optimal noise generated using the noise generation network model was less than 1%. This demonstrates that the optimal noise generated using the noise generation network model is significantly superior to random noise in terms of performance and stability in complex electromagnetic environments.

[0090] In another embodiment of the present invention, the computational complexity and real-time feasibility of the noise generation network model provided by the present invention were verified. Specifically, the noise generation network model contains 585 pre-trained parameters in 32-bit floating-point format, and the model size is approximately 15,000 bytes. Running the network once requires a total of approximately 1.4M multiplication-addition computations (the sum of the number of multiplications and additions). The noise generation network model takes an average of 1.73ms on a CPU (i7) and 0.54ms on a GPU (3060). The noise generation network model can be divided into two parts based on the computational complexity: the FFT and sorting part, and the k-group neural network part consisting of an encoder and decoder. Because the neural network involves convolution operations on large matrices, the computational complexity is concentrated in the second part. Parallelizing the second part can save a significant amount of computational time. In this case, the longest computational path within a folded computation module only includes five convolutional layers. According to calculations, if the noise generation network model is implemented on an FPGA with a frequency of 200MHz, the computation can be completed within microseconds. It is proved that when hardware programming resources are sufficient, the noise generation network model can be made fully feasible by trading space for time.

[0091] In traditional noise signal generation, genetic algorithms are a commonly used solution. Genetic algorithms originate from computer simulations of biological system development. Drawing on Darwin's theory of evolution and Mendel's genetics, they are search and optimization methods that mimic natural development. Based on the problem's objective function, they construct a fitness function, evaluate, genetically manipulate, and select a population of multiple solutions. After multiple generations of reproduction, the individual with the best fitness value is obtained as the optimal solution to the problem. To compare the superiority of the method proposed in this invention, a comparison was conducted with the genetic algorithm under the same conditions. Figure 6 The comparison results of the calculation speed and optimization effect of the two are shown. Figure 6 The folding method in the present invention refers to the noise waveform generation method provided by the present invention.

[0092] like Figure 6 As shown, after approximately 20 generations of population iteration, the genetic algorithm achieved an objective function value comparable to that of the radar signal-based noise waveform generation method proposed in this paper, and converged around 50 generations. By applying the genetic algorithm to all radar signal samples in the validation set, statistical results showed that the average number of iterations required to achieve model-equivalent output was 18.1 generations, with an average of 905.34 loss function calculations per run and an average optimization time of 1.58 seconds.

[0093] The loss function requires two convolution operations and CFAR checks, and each call to the loss function requires 6.71M multiplications and additions. The genetic algorithm, on average, requires approximately 6000M multiplications and additions. In comparison, the proposed method achieves the same effect, reducing computational effort by approximately 4000 times and time consumption by 1216 times.

[0094] In summary, the noise waveform generation method based on radar signals provided by the present invention performs signal spectrum analysis on the original radar signal to extract the main frequency components in the original radar signal, and can only use the frequencies containing information to avoid meaningless noise; according to the period corresponding to the main frequency components, the original radar signal is folded, and the changes of the original radar signal within each period and between adjacent periods can be captured at the same time, which is conducive to improving the efficiency of feature extraction, and the two-dimensional structure of the two-dimensional feature image can effectively describe local correlation and periodic consistency, thereby enhancing the interpretability of the time series features; the constructed noise generation network model, whose encoder separates the calculation of depth-by-depth convolution and point-by-point convolution, greatly reduces redundant operations and improves the efficiency of noise signal generation.

[0095] The noise waveform generator provided by the present invention is described below.

[0096] like Figure 7 As shown, the noise waveform generator 700 includes: The receiving module 701 is used to obtain the original radar signal sent by the transmitting source; Extraction module 702, used to perform signal spectrum analysis on the original radar signal and extract the main frequency components in the original radar signal; The signal folding module 703 is used to perform signal folding on the original radar signal according to the period corresponding to the main frequency component, and obtain a two-dimensional feature image corresponding to each main frequency component; the two-dimensional feature image is used to describe the timing characteristics of the original radar signal within each period and between adjacent periods; The model processing module 704 is used to construct a noise generation network model and train the noise generation network model using a pre-constructed complex radar signal data set until the loss value of the noise generation network model is less than a preset loss threshold, thereby obtaining a trained noise generation network model. The noise generation network model includes an encoder, a decoder, and a noise generation module. The encoder uses depthwise separable convolution to extract the spatiotemporal joint features of the input two-dimensional feature image and encodes the input two-dimensional feature image using the spatiotemporal joint features to obtain a coded feature map. The depthwise separable convolution is composed of depthwise convolution and pointwise convolution. The decoder is used to decode the coded feature map. The noise generation module is used to generate a noise waveform based on the decoding result and the size of the main frequency components. The noise generation module 705 is configured to input the two-dimensional feature image corresponding to each main frequency component into the trained noise generation network model to obtain a noise waveform based on the original radar signal, and perform convolution processing on the noise waveform and the original radar signal to obtain a final noise waveform; The transmitting module 706 is configured to send the final noise waveform to a transmitting source.

[0097] In one possible embodiment, a noise waveform generator is attached to a target for emitting smart noise.

[0098] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiments of the present application. Their specific functions and technical effects can be found in the method embodiment section and will not be described in detail here. Those skilled in the art will clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the above-mentioned method embodiment and will not be described in detail here.

[0099] like Figure 8 As shown, an embodiment of the present invention provides a terminal device, such as Figure 8 As shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 8 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above-mentioned method embodiments when executing the computer program D102.

[0100] Specifically, when the processor D100 executes the computer program D102, it obtains the original radar signal emitted by the transmitting source; performs signal spectrum analysis on the original radar signal to extract the main frequency components in the original radar signal; performs signal folding on the original radar signal according to the period corresponding to the main frequency components to obtain a two-dimensional feature image corresponding to each main frequency component; constructs a noise generation network model, and uses a pre-constructed complex radar signal data set to train the noise generation network model until the loss value of the noise generation network model is less than a preset loss threshold, thereby obtaining a trained noise generation network model; inputs the two-dimensional feature image corresponding to each main frequency component into the trained noise generation network model to obtain a noise waveform based on the original radar signal, and performs convolution processing on the noise waveform and the original radar signal to obtain a final noise signal; and sends the final noise signal to the transmitting source. Among them, by performing signal spectrum analysis on the original radar signal and extracting the main frequency components in the original radar signal, only the frequencies containing information can be used to avoid meaningless noise; according to the period corresponding to the main frequency components, the original radar signal is folded, which can simultaneously capture the changes of the original radar signal within each period and between adjacent periods, which is conducive to improving the efficiency of feature extraction, and the two-dimensional structure of the two-dimensional feature image can effectively describe local correlation and periodic consistency, enhancing the interpretability of the temporal features; the constructed noise generation network model, whose encoder separates the calculations through depth-by-depth convolution and point-by-point convolution, greatly reduces redundant operations and improves the efficiency of noise signal generation.

[0101] The processor D100 may be a central processing unit (CPU), or may be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0102] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device D10. Furthermore, the memory D101 may include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, application programs, a boot loader, data, and other programs, such as the program code of the computer program. The memory D101 may also be used to temporarily store data that has been output or is about to be output.

[0103] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned various method embodiments can be implemented.

[0104] An embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0105] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of protection of the present application is limited to these examples. In line with the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of different aspects of one or more embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0106] The one or more embodiments of this application are intended to encompass all such substitutions, modifications, and variations that fall within the broad scope of this application. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of one or more embodiments of this application should be included in the scope of protection of this application.

Claims

1. A noise waveform generation method based on radar signals, characterized in that: include: Obtaining the original radar signal emitted by the transmitting source; Performing signal spectrum analysis on the original radar signal to extract the main frequency components in the original radar signal; performing signal folding on the original radar signal according to the period corresponding to the main frequency component to obtain a two-dimensional feature image corresponding to each main frequency component; the two-dimensional feature image is used to describe the timing characteristics of the original radar signal within each period and between adjacent periods; Constructing a noise generation network model and training the noise generation network model using a pre-constructed complex radar signal data set until the loss value of the noise generation network model is less than a preset loss threshold, thereby obtaining a trained noise generation network model; wherein the noise generation network model includes an encoder, a decoder, and a noise generation module, the encoder uses depthwise separable convolution to extract the spatiotemporal joint features of the input two-dimensional feature image, and uses the spatiotemporal joint features to encode the input two-dimensional feature image to obtain a coded feature map, the depthwise separable convolution is composed of depthwise convolution and pointwise convolution, the decoder is used to decode the coded feature map, and the noise generation module is used to generate a noise waveform based on the decoding result and the size of the main frequency component; Inputting the two-dimensional feature image corresponding to each of the main frequency components into the trained noise generation network model to obtain a noise waveform based on the original radar signal, and performing convolution processing on the noise waveform and the original radar signal to obtain a final noise signal; The final noise signal is sent to the transmitting source.

2. The noise waveform generation method based on radar signal according to claim 1, characterized in that: The performing signal spectrum analysis on the original radar signal to extract the main frequency components in the original radar signal includes: Performing Fourier transform on the original radar signal to obtain a frequency spectrum of the original radar signal; A plurality of frequency components are extracted from the frequency spectrum in descending order of amplitude as the main frequency components.

3. The noise waveform generation method based on radar signal according to claim 2, characterized in that: The performing signal folding on the original radar signal according to the period corresponding to the main frequency component to obtain a two-dimensional feature image corresponding to each main frequency component includes: By calculating the formula Get the first The two-dimensional feature image corresponding to the main frequency components ;in, represents the original radar signal, Indicates zero-fill operation, Represents a folding operation, Indicates the The period corresponding to the main frequency components is Indicates the The frequency magnitude of the main frequency components, , Indicates the total number of dominant frequency components.

4. The method for generating noise waveform based on radar signal according to claim 3, characterized in that: The spatiotemporal joint features include spatial features and channel features; The spatial features are extracted by the depth-wise convolution; the depth-wise convolution performs independent convolution on each input channel of the input two-dimensional feature image; The channel features are extracted by the point-by-point convolution; the point-by-point convolution aggregates the convolution results of all input channels.

5. The method for generating noise waveform based on radar signal according to claim 4, characterized in that: The step of generating a noise waveform according to the decoding result and the magnitude of the main frequency components includes: By calculating the formula Get the noise waveform ;in, Indicates the The weights of the main frequency components, Indicates the The decoding result of the two-dimensional feature image corresponding to the main frequency components, Denotes a decoding operation, Indicates the The encoding feature map corresponding to the main frequency components, Represents the encoding operation, represents batch normalization, Represents the activation function.

6. The method for generating noise waveform based on radar signal according to claim 1, characterized in that: The complex radar signal data set includes multiple complex radar signals; the complex radar signals are generated by jointly modulating simulated radar pulse waveforms using multiple modulation methods.

7. The method for generating a noise waveform based on a radar signal according to claim 1, wherein: The calculation formula of the loss value is: in, represents the loss value, express The constant false alarm threshold height at all times, represents variance calculation, Indicates mean calculation.

8. A noise waveform generator based on radar signals, characterized in that: include: A receiving module is used to obtain the original radar signal sent by the transmitting source; an extraction module, configured to perform signal spectrum analysis on the original radar signal and extract the main frequency components in the original radar signal; a signal folding module, configured to perform signal folding on the original radar signal according to the period corresponding to the main frequency component, and obtain a two-dimensional feature image corresponding to each of the main frequency components; the two-dimensional feature image is used to describe the timing characteristics of the original radar signal within each period and between adjacent periods; A model processing module is used to construct a noise generation network model and train the noise generation network model using a pre-constructed complex radar signal data set until the loss value of the noise generation network model is less than a preset loss threshold, thereby obtaining a trained noise generation network model; wherein, the noise generation network model includes an encoder, a decoder, and a noise generation module, the encoder uses depthwise separable convolution to extract the spatiotemporal joint features of the input two-dimensional feature image, and uses the spatiotemporal joint features to encode the input two-dimensional feature image to obtain a coded feature map, the depthwise separable convolution is composed of depthwise convolution and pointwise convolution, the decoder is used to decode the coded feature map, and the noise generation module is used to generate a noise waveform based on the decoding result and the size of the main frequency component; a noise generation module, configured to input the two-dimensional feature image corresponding to each of the main frequency components into the trained noise generation network model to obtain a noise waveform based on the original radar signal, and perform convolution processing on the noise waveform and the original radar signal to obtain a final noise signal; A transmitting module is configured to send the final noise signal to the transmitting source.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images