Radar waveform generation method and device based on generative adversarial network

Abstract

The application discloses a radar waveform generation method and device based on a generative adversarial network, and relates to the technical field of signal processing, and comprises the following steps: obtaining text content; inputting the text content into a trained text classification model for processing to obtain labels of interference signals of a preset category described in the text content; and inputting the labels into a trained generative adversarial network for processing to obtain interference signals corresponding to the labels; wherein the trained text classification model is obtained by taking data of a first preset category as a training data set and training an initial text classification model; and the trained generative adversarial network is obtained by taking data of a second preset category as a training data set and training an initial generative adversarial network. The application can provide rich data set resources for tasks such as sorting and detection of interference signals.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of signal processing, and particularly relates to a radar waveform generation method and device based on a generative adversarial network. BACKGROUND

[0002] With the rapid development of information technology and artificial intelligence, generative adversarial networks have important application value in image generation, data enhancement, signal processing, etc. In the field of synthetic aperture radar (SAR) jamming signals, generative adversarial network technology can be used to generate and enhance signal samples, thereby providing a rich set of interference data for signal detection, sorting, identification, etc. At the same time, traditional signal generation methods require certain professional knowledge and complex parameter settings.

[0003] In the field of SAR jamming waveform generation, existing methods are overly dependent on high-quality target signals actually collected to train the network, which is difficult to obtain in actual scenarios. On the other hand, they are overly dependent on accurate labels and encoding of signal features, which is costly in terms of human resources, and feature selection often requires expert knowledge, increasing the complexity of model development.

[0004] Therefore, there is an urgent need to provide a jamming waveform generation method to solve the above-mentioned defects. SUMMARY

[0005] To solve the above-mentioned problems in the prior art, the present application provides a radar waveform generation method and device based on a generative adversarial network. The technical problems to be solved by the present application are solved by the following technical solutions:

[0006] In a first aspect, the present application provides a radar waveform generation method based on a generative adversarial network, comprising:

[0007] obtaining text content, the text content describing the demand for jamming signals of a preset category;

[0008] inputting the text content into a trained text classification model for processing to obtain labels of jamming signals of the preset category described in the text content;

[0009] inputting the labels into a trained generative adversarial network for processing to obtain jamming signals corresponding to the generated labels;

[0010] The trained text classification model is trained by taking data of a first preset category as a training data set, and the initial text classification model is obtained by training, the data of the first preset category being text data and a label corresponding to the text data, and the text data describing features of an interference signal of the preset category.

[0011] In a second aspect, the application further provides a radar waveform generation device based on a generative adversarial network, comprising:

[0012] A text acquisition module is configured to acquire text content, the text content describing requirements for an interference signal of a preset category.

[0013] A text processing module is configured to input the text content into the trained text classification model for processing to obtain a label of the interference signal of the preset category described in the text content.

[0014] A label processing module is configured to input the label into the trained generative adversarial network for processing to obtain an interference signal corresponding to the label.

[0015] The trained text classification model is trained by taking data of a first preset category as a training data set, and the initial text classification model is obtained by training, the data of the first preset category being text data and a label corresponding to the text data, and the text data describing features of an interference signal of the preset category.

[0016] The application has the following beneficial effects:

[0017] The radar waveform generation method and device based on the generative adversarial network provided by the application consider that most signal generation methods highly depend on high-quality training data sets, and it is difficult to obtain such data in actual scenarios. In addition, for some application scenarios requiring instant response, the generation speed of the prior art may be insufficient, and it is difficult to meet actual requirements. In the application, only simulated interference signals are used for network training, and complex signal feature coding is not required, which reduces certain labor costs and reduces the difficulty of signal generation, and provides rich data set resources for tasks such as sorting and detection of interference signals.

[0018] The application will be further described in detail below with reference to the accompanying drawings and embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flow chart of a radar waveform generation method based on a generative adversarial network provided by an embodiment of the application;

[0020] Figure 2 is a schematic diagram of collecting a training data set provided by an embodiment of the application;

[0021] Figure 3 is a schematic diagram of generating an interference signal provided by an embodiment of the application. DETAILED DESCRIPTION

[0022] The application will be further described in detail below in combination with specific embodiments, but the embodiments of the application are not limited thereto.

[0023] In view of the problems in the prior art, such as poor real-time performance of interference signal generation technology, difficulty in collecting real data, and high cost of manually marking data, a generative adversarial network SAR interference waveform generation method and system based on natural language processing are proposed. The system can generate interference signals with specified features according to simple natural language input by a user. The method can generate specific interference signal waveforms and has good generation capability, as indicated by a time-frequency graph. Compared with other generation technologies, the advantage of the application lies in that only simulation interference data is needed for training, without the need for complex marked data, and the operation is simple, which greatly improves the precision, efficiency and user experience of signal generation, thereby providing rich interference data sets for signal detection, sorting, identification and other tasks.

[0024] Please refer to Figure 1 , Figure 1 is a flow chart of a radar waveform generation method based on a generative adversarial network provided by an embodiment of the application, and the radar waveform generation method based on a generative adversarial network provided by the application comprises:

[0025] S101, obtaining text content, the text content describing the demand for an interference signal of a preset category.

[0026] S102, inputting the text content into a trained text classification model for processing to obtain a label of the interference signal of the preset category described in the text content.

[0027] S103, inputting the label into a trained generative adversarial network for processing to obtain an interference signal corresponding to the label.

[0028] The trained text classification model is trained by taking data of a first preset category as a training data set, the data of the first preset category being text data and labels corresponding to the text data, the text data describing features of an interference signal of the preset category; and the trained generative adversarial network is trained by taking data of a second preset category as a training data set, the data of the second preset category being an interference signal of the preset category and a label corresponding to the interference signal of the preset category.

[0029] Specifically, before training the initial text classification model, the method further includes:

[0030] The text classification model is constructed based on a convolutional neural network, different sizes of convolution kernels are used to perform convolution operations on input text, a maximum pooling operation is performed on the results, and finally a category probability is output through a fully connected layer and a Softmax layer.

[0031] The initial text classification model is trained, including:

[0032] The text data set and labels corresponding to the text data set are obtained, the text data set is cleaned and standardized, and the cleaned text data set is divided into 80% training set and 20% test set.

[0033] Optionally, the data set cleaning and standardization includes character segmentation, removal of blank space characters, text label segmentation, text length limitation, and word embedding.

[0034] The initial text classification model is trained using the training text data set, and the text classification model is adjusted using the test text data set to make the model achieve optimal performance.

[0035] Optionally, there are 1000 texts of narrowband interference signals and linear frequency modulation interference signals, the model training stage inputs texts with a length of pad_side of 32 and labels corresponding thereto, the learning rate is 0.001, and the training is performed for 20 epochs.

[0036] Before training the initial generative adversarial network, the method further includes:

[0037] The generative adversarial network is constructed.

[0038] The generative adversarial network includes a generator and a discriminator, and the output of the generator is connected to the input of the discriminator.

[0039] In the embodiment, the initial generative adversarial network is trained, including:

[0040] Obtaining second preset category data, constructing a training data set, the training data set comprising a plurality of training samples, each training sample comprising a simulation generated preset category interference signal and a preset category interference signal corresponding label; wherein, the preset category interference signal comprises a narrowband interference signal and a linear frequency modulation interference signal; as shown in Figure 2 Figure 2 is a schematic diagram provided by an embodiment of the present application for collecting a training data set, and matlab is used to simulate a narrowband interference signal and a linear frequency modulation interference signal.

[0041] Optionally, 5000 narrowband interference signals and linear frequency modulation interference signals of different carrier frequencies, the length of the interference signal is 1201*1, the label of the narrowband interference signal is 1, and the label of the linear frequency modulation interference is 2.

[0042] The training data set is input into an initial generative adversarial network, so that the generator learns the corresponding relationship between the simulation generated preset category interference signal in the training sample and the preset category interference signal corresponding label, and outputs the learned interference signal; the interference signal learned by the generator is input into the discriminator, so that the discriminator connects the input interference signal with the label, and outputs a preset probability, so that the iteration cycle is repeated within a preset number of iterations, until the loss function of the generative adversarial network reaches a minimum, and a trained generative adversarial network is obtained.

[0043] It can be understood that the simulation interference signal sample of 1201*1 and its label are input into the initial generation network, so that the generator learns the corresponding relationship between the signal and the label, and the generator trains a process of converting noise 1*100 noise and label into false signal. Comparison and analysis are performed by using the discriminator and the label, the network is iteratively trained according to the loss function, and the target of the discriminator is to distinguish between real data and generated data.

[0044] In the use of the trained generation network model, the generator input is the label and random noise, the label corresponding interference signal is generated, and the discriminator is used to distinguish whether the input signal and the input label correspond, as shown in Figure 3 Figure 3 is a schematic diagram provided by an embodiment of the present application for generating an interference signal, and by using the method provided by the embodiment, a narrowband interference signal and a linear frequency modulation interference signal can be generated, Figure 3 The narrowband interference signal, the narrowband interference signal spectrum and the narrowband interference signal time-frequency graph, the linear frequency modulation interference signal, the linear frequency modulation signal spectrum and the linear frequency modulation signal time-frequency graph are displayed.

[0045] In addition, the trained generation network is also tested by a test data set, the trained generation network outputs a time domain interference waveform, and then the time domain interference waveform is converted into a time-frequency graph by Fourier transform, so as to evaluate the performance of the trained generation network.

[0046] ​​In this embodiment, the expression of the loss function of the generative adversarial network is:

[0047]

[0048] lossDiscriminator=lossReal+lossGenerated;

[0049]

[0050] wherein lossGenerated represents the loss of the discriminator on the generated data, N represents the number of training samples, i represents the number of training batches, probGenerated i represents the output prediction probability of the discriminator on the generated data, lossReal represents the loss of the discriminator on the real data, probReal i represents the output prediction probability of the discriminator on the real data, lossDiscriminator represents the total loss of the discriminator, and lossGenerator represents the loss of the generator.

[0051] The loss function thereof is composed of two parts, the loss of the discriminator on the real data and the loss of the discriminator on the generated data. The discriminator wants to make lossReal and lossGenerated as small as possible, so as to effectively distinguish the real data and the generated data. The goal of the generator is to generate samples that can deceive the discriminator, that is, the discriminator judges that the generated data is real, at this time, probGenerated is close to 1, and log(probGenerated) is close to 0. In the training process, the learning rate is set to 1e-5, and the minimum batch processing is 256.

[0052] In the trained generative adversarial network in this embodiment, the expression of the output table of the ideal generator and the ideal discriminator is:

[0053]

[0054] wherein G * represents the output of the ideal generator, D * represents the output of the ideal discriminator, V(D,G) represents the value function, G represents the output of the generator, D represents the output of the discriminator, E represents the expectation, x represents a data point of the real data, p data represents the real data distribution, p z represents the prior distribution, and D(x) represents the output of the discriminator on the real data sample x, and G(z) represents the output on the input noise z.

[0055] In this embodiment, the label is input into the trained generative adversarial network for processing to obtain a generated label corresponding to the jamming signal, including:

[0056] The label is input into the trained generative adversarial network, the generator reshapes the noise array through a custom layer, simultaneously converts the input label into an embedding vector, and then trains the reshaped noise array and the embedding vector through a one-dimensional transpose convolution layer to generate a jamming signal corresponding to the label; the label and the jamming signal corresponding to the label are input into the discriminator, and a one-dimensional convolution layer and ReLU are used for processing to output a prediction probability.

[0057] It can be understood that the generator is a double-input network, which reshapes the input 1*100 noise array through a custom layer, simultaneously converts the input 1*1 classification label into an embedding vector, connects the two inputs after reshaping, and trains through multiple one-dimensional transpose convolution layers to generate a 1201*1 generated signal; the discriminator is also a double-input network, and the input is a group of signals and their corresponding labels; the pre-trained discriminator has the ability to judge true and false signals; the discriminator network connects the input signal and the label, and finally outputs a 1*1 prediction probability through a one-dimensional convolution layer and a ReLU activation function; when the prediction probability is 0.5, it indicates that the generator has the ability to generate realistic jamming signal waveforms.

[0058] In this embodiment, for narrowband jamming, its spectrum is usually concentrated in a narrow frequency range, and shows a certain peak in the frequency domain; in general, it can be regarded as the superposition of a series of single-frequency signals; the preset category of jamming signals includes narrowband jamming signals, and the expression is:

[0059]

[0060] wherein, I NB (k) represents a narrowband jamming signal, k represents the serial number of the distance unit, N represents the number of narrowband jamming, A n represents the amplitude of the nth narrowband jamming, f n represents the frequency of the nth narrowband jamming.

[0061] In this embodiment, for linear frequency modulation jamming, it occupies a certain bandwidth, and shows a certain bandwidth protrusion in the frequency domain; the preset category of jamming signals includes linear frequency modulation jamming signals, and the expression is:

[0062]

[0063] wherein, I CM (k) represents a linear frequency modulation jamming signal, k represents the serial number of the distance unit, M represents the number of frequency modulation jamming, A m represents the amplitude of the mth frequency modulation jamming, f mdenotes the frequency of the mth frequency-modulated jamming, g m denotes the frequency-modulation rate of the mth frequency-modulated jamming.

[0064] In summary, the present application provides a radar waveform generation method based on a generative adversarial network, which takes into account most signal generation methods that are highly dependent on high-quality training data sets. In actual scenarios, it is difficult to obtain such data. In addition, for some application scenarios that require immediate response, the generation speed of the prior art may not be sufficient to meet actual needs. In the present application, only simulated jamming signals are used for network training, and complex signal feature encoding is not required, reducing certain labor costs and reducing the difficulty of signal generation, providing rich data set resources for jamming signal sorting, detection, and other tasks.

[0065] Based on the same inventive concept, the present application also provides a radar waveform generation device based on a generative adversarial network, which is used to implement the method provided in the above embodiments of the present application. For embodiments of the method, please refer to the above description, which will not be repeated here. The device comprises:

[0066] A text acquisition module is configured to acquire text content describing the requirements of jamming signals of a preset category.

[0067] A text processing module is configured to input the text content into a trained text classification model for processing to obtain labels of jamming signals of the preset category described in the text content.

[0068] A label processing module is configured to input the labels into a trained generative adversarial network for processing to obtain jamming signals corresponding to the labels.

[0069] The trained text classification model is obtained by training an initial text classification model using data of a first preset category as a training data set. The data of the first preset category comprises text data and labels corresponding to the text data, and the text data describes the features of jamming signals of the preset category. The trained generative adversarial network is obtained by training an initial generative adversarial network using data of a second preset category as a training data set. The data of the second preset category comprises jamming signals of the preset category and labels corresponding to the jamming signals.

[0070] Specifically, in this embodiment, a SAR jamming signal waveform generation model based on simulated jamming signals is constructed by training the generative adversarial network. Experimental results show that the model achieves good SAR jamming signal waveform generation. Compared with other signal production methods, the present application does not require a large number of real jamming signals, does not require complex signal feature encoding, can generate specified jamming signals using simple natural language, is easier to operate, and provides rich data set resources for signal sorting, identification, and other tasks.

[0071] It should be noted that, as used in this document, the terms "first," "second," etc. are used only to distinguish one entity or action from another, and do not necessarily require or imply any actual relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element. The terms "connected" or "coupled" or any other variant thereof are intended to cover a physical or logical connection, whether direct or indirect, between or among components. The terms "top," "bottom," "front," "back," "left," "right," and the like in reference to an element are used to indicate a relative position of the element with respect to another element, and are not intended to denote a specific orientation of the element. The terms "on," "overlying," "underlying," "beneath," "below," "above," and the like in reference to the placement of one element on or relative to another element are used to indicate a relative position between the elements, and are not intended to denote a specific orientation of the elements.

[0072] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" and the like means that the specific feature or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, a person skilled in the art can combine and combine different embodiments or examples described in the present specification.

[0073] The above is a further detailed description of the present invention in combination with specific preferred embodiments, and cannot be considered as limiting the specific implementation of the present invention to these descriptions. For those skilled in the art, without departing from the concept of the present invention, a number of simple deductions or replacements can be made, which should be considered as falling within the protection scope of the present invention.

Claims

1. A method for radar waveform generation based on a generative adversarial network, characterized in that, The method comprises the following steps: acquiring text content, wherein the text content describes the demand for interference signals of a preset category; inputting the text content into a trained text classification model for processing to obtain a label of the interference signals of the preset category described in the text content; inputting the label into a trained generative adversarial network for processing to obtain the interference signals corresponding to the label; wherein the trained text classification model is trained by taking data of a first preset category as a training data set, wherein the data of the first preset category is text data and a label corresponding to the text data, and the text data describes the characteristics of the interference signals of the preset category; the trained generative adversarial network is trained by taking data of a second preset category as a training data set, wherein the data of the second preset category is the interference signals of the preset category and a label corresponding to the interference signals of the preset category.

2. The method of claim 1, wherein the generator is a generative adversarial network (GAN). Before training the initial generative adversarial network, the method further comprises the following steps: constructing a generative adversarial network; wherein the generative adversarial network comprises a generator and a discriminator, and the output of the generator is connected to the input of the discriminator.

3. The method of claim 2, wherein the generator is a generative adversarial network (GAN) and the discriminator is a GAN discriminator. The method of training the initial generative adversarial network comprises the following steps: acquiring the data of the second preset category to construct a training data set, wherein the training data set comprises a plurality of training samples, each training sample comprises a simulation-generated interference signal of the preset category and a label corresponding to the interference signal of the preset category; wherein the interference signal of the preset category comprises a narrowband interference signal and a linear frequency modulation interference signal; inputting the training data set into the initial generative adversarial network, so that the generator learns the corresponding relationship between the simulation-generated interference signal of the preset category and the label corresponding to the interference signal of the preset category in the training sample, and outputs the learned interference signal; inputting the interference signal learned by the generator into the discriminator, so that the discriminator connects the input interference signal with the label, and outputs a preset probability; and iteratively repeating the above steps until the loss function of the generative adversarial network reaches a minimum value within a preset number of iterations, thereby obtaining the trained generative adversarial network.

4. The method of claim 3, wherein the generator is a generative adversarial network (GAN) and the discriminator is a GAN discriminator. The expression of the loss function of the generative adversarial network is as follows: lossDiscriminator = lossReal + lossGenerated; wherein lossGenerated represents the loss of the discriminator on generated data, N represents the number of training samples, i represents the number of training batches, probGenerated i represents the output predicted probability of the discriminator on generated data, lossReal represents the loss of the discriminator on real data, probReal i represents the output predicted probability of the discriminator on real data, lossDiscriminator represents the total loss of the discriminator, lossGenerator represents the loss of the generator.

5. The method of claim 3, wherein the generator is a generative adversarial network (GAN) and the discriminator is a GAN discriminator. in the trained generative adversarial network, the output expression of the ideal generator and the ideal discriminator is as follows: where G * represents the output of the ideal generator, D * represents the output of the ideal discriminator, V(D, G) represents the value function, G represents the output of the generator, D represents the output of the discriminator, E represents expectation, x represents a data point of real data, p data represents the real data distribution, p z represents the prior distribution, D(x) represents the output of the discriminator on a real data sample x, G(z) represents the output on an input noise z.

6. The method of claim 1, wherein, the method of inputting the label into the trained generative adversarial network for processing to obtain the interference signals corresponding to the label comprises the following steps: inputting the label into the trained generative adversarial network, and the generator reshapes a noise array through a self-defined layer, converts the input label into an embedding vector, and then trains the reshaped noise array and the embedding vector through a one-dimensional transpose convolution layer to generate the interference signals corresponding to the label; inputting the label and the interference signals corresponding to the label into the discriminator, and outputting a prediction probability through a one-dimensional convolution layer and a ReLU.

7. The method of claim 1, wherein, the interference signals of the preset category comprise a narrowband interference signal, and the expression of the narrowband interference signal is as follows: where I NB (k) denotes a narrowband interference signal, k denotes the serial number of the unit of distance, N denotes the number of narrowband interferences, A n denotes the amplitude of the nth narrowband interference, f n denotes the frequency of the nth narrowband interference.

8. The generative adversarial network-based radar waveform generation method of claim 1, wherein, The preset category of interference signals includes a linear frequency modulation interference signal, and an expression of the linear frequency modulation interference signal is: where I CM (k) denotes a linear frequency modulation jamming signal, k denotes the serial number of distance unit, M denotes the number of frequency modulation jamming, A m denotes the amplitude of the mth frequency modulation jamming, f m denotes the frequency of the mth frequency modulation jamming, g m denotes the frequency modulation rate of the mth frequency modulation jamming. 9.A radar waveform generation device based on a generative adversarial network, characterized by Comprise: The text acquisition module is used for acquiring text content, and the text content describes a demand for a preset category of interference signals; The text processing module is used for inputting the text content into a trained text classification model for processing to obtain a label of a preset category of interference signals described in the text content; The label processing module is used for inputting the label into a trained generative adversarial network for processing to obtain a generated interference signal corresponding to the label; Wherein, the trained text classification model takes first preset category data as a training data set, and is obtained by training an initial text classification model, the first preset category data is text data and a label corresponding to the text data, and the text data describes a feature of a preset category of interference signals; the trained generative adversarial network takes second preset category data as a training data set, and is obtained by training an initial generative adversarial network, and the second preset category data is a preset category of interference signals and a label corresponding to the preset category of interference signals.

Images