Radar signal denoising method based on multi-channel network

Abstract

The invention provides a radar signal denoising method and device based on a multi-channel network, and the method comprises the following steps: employing a simulation radar data set as a training data set, and dividing the training data set into a training data set and a verification data set according to a proportion; secondly, preprocessing each training sample, randomly disrupting training sample data, and normalizing the data to obtain a training set and a verification set; then, constructing a deep m-channel neural network, training the constructed deep multi-channel neural network by adopting the training set, and adjusting hyper-parameters by adopting the verification set; and adopting a mean square error as a loss function of regression prediction. And finally, obtaining a radar signal to be classified, denoising the radar signal to be classified by using the trained radar signal denoising network, and then classifying the denoised data by using the trained radar signal classification network to obtain a radar signal classification result.

Description

Technical Field

[0001] The invention belongs to the field of radar technology, and in particular is a radar signal denoising method based on a multi-channel network. Background Art

[0002] Radar is an electronic instrument that uses electromagnetic waves to detect the relative distance between objects and other characteristics. Reconnaissance and interference are two important functions of radar. Among them, radar reconnaissance tasks can be divided into: intelligence reconnaissance, support reconnaissance, radar warning, guidance interference, etc. according to actual conditions. Radar signal modulation classification and parameter recognition are important functions of reconnaissance and one of the main contents of radar detection and reconnaissance signal analysis and processing. Based on classification, instruments and equipment can analyze the types and properties of different radar modulation signals. Therefore, the classification and identification of different signals is a key technology in radar military confrontation.

[0003] In actual situations, the radar signal environment includes targets, environmental echoes, and noise interference caused by active and passive interference. The signal received by ground equipment from space cannot contain only useful signals. During the transmission process, various noises will be mixed with useful signals. The coexistence of noise and signal in the channel is harmful to transmission. Its existence causes analog signal distortion, digital signal bit errors, and data transmission rate is also affected. Therefore, it is particularly important to suppress the clutter noise signal in the radar signal for the correct decoding of the radar signal.

[0004] At present, the ability to sort signals has become an important symbol of radar reconnaissance systems in the modern military field. Before deep learning was introduced into the radar field, radar signal modulation recognition technology was based on various characteristic parameters of the signal, such as radar parameters such as high-order moments, to calculate, analyze and compare the characteristics of different modulation types, and then classify the signal. This method is relatively complex to implement and requires a high level of radar expertise. Deep learning has penetrated into various fields, and through computer imitation of human learning methods, it can eventually replace human thinking patterns. How to effectively combine deep learning so that it can have radar signal recognition capabilities is an effective way for radar to have artificial intelligence, and it is also one of the key research contents of radar signal sorting. Summary of the invention

[0005] In order to overcome the defects of complexity and high cost described in the above-mentioned prior art, the present invention proposes to construct a deep multi-channel neural network based on a deep learning algorithm, and provides a method that can effectively remove noise from radar signals.

[0006] The technical solution of the present invention: a radar signal denoising method based on a multi-channel network comprises the following steps:

[0007] Step 1: Use the simulated radar data set as the training data set, and create a noisy radar signal data set based on the training data set, and divide it into a training data set and a verification data set in proportion;

[0008] Step 2: Preprocess each training sample, randomly shuffle the training sample data, normalize the data, map all features of the data to the same scale, and obtain the training set and validation set;

[0009] Step 3: construct a deep multi-channel neural network, including a preliminary feature extraction part, a deep feature extraction part and an upsampling part; build a radar signal classification network based on the convolutional neural network, and train the radar signal classification network with the training data set in step 2;

[0010] Step 4: Use the training set to train the constructed deep multi-channel neural network, and use the validation set to adjust the hyperparameters; use the mean square error as the loss function of regression prediction, use the callback function to check the internal state of the training model, and automatically reduce the learning rate when the evaluation index no longer improves. When the loss function no longer decreases, save the model parameters to obtain the trained deep multi-channel neural network;

[0011] Step 5: Obtain the radar signal to be classified, use the trained radar signal denoising network to denoise the radar signal to be classified, and then use the trained radar signal classification network to classify the denoised data to obtain the radar signal classification result.

[0012] The present invention has the following beneficial effects:

[0013] The present invention proposes a radar signal denoising method based on a multi-channel network. First, the trained radar signal denoising is used for preprocessing to reduce the impact of noise on the classification results. Then, the denoised data is classified using the trained radar signal classification network, which can effectively improve the accuracy of radar signal recognition and sorting. The deep neural network can extract the deep features of the radar signal, and multiple channels can fully integrate the learned different signal features to reduce feature loss. The network has excellent denoising performance and meets the needs of noise suppression for radar signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings:

[0015] Figure 1 A flowchart of a radar signal denoising method based on a multi-channel network according to the present invention;

[0016] Figure 2 Schematic diagram of the multi-channel neural network model structure. DETAILED DESCRIPTION

[0017] The drawings are for illustrative purposes only and should not be construed as limiting the present patent;

[0018] In order to better illustrate the present embodiment, some parts in the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product;

[0019] It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0020] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0021] Example

[0022] The embodiment of the present invention provides a method and device for implementing radar signal denoising in a multi-channel network, such as Figures 1 to 2 shown.

[0023] This example includes the following steps:

[0024] Step 1: Use the simulated radar dataset as the training dataset and create a noisy radar signal dataset based on the training dataset.

[0025] Step 2: Preprocess the data set. Randomly shuffle the radar signal training sample data, expand the dimensions of the sample data X and label Y, and normalize the data.

[0026] Step 3: Build a deep multi-channel neural network. Figure 2 The figure is a schematic diagram of the deep m-channel neural network model structure. The neural network is constructed with the convolutional neural network CNN as the unit. First, a wide convolutional neural network CNN is connected to perform preliminary feature extraction. Then the data is input into the deep feature extraction layer of multiple branches in parallel. These branches have the same structure, consisting of n convolutional layers CONV1D and n / 2 pooling layers pooling. Pooling operation is performed every two convolutional layers to avoid excessive number of parameters and redundancy. The m branches of the output parameters are connected and fused to reduce feature loss during the processing. Since pooling shortens the time step, two upsampling layers are connected to restore the data size. Finally, a convolutional layer with 1 feature layer and a convolution kernel size of 1*1 is connected.

[0027] Step 4: Neural network training and parameter optimization. Figure 1The following is a flowchart of radar signal processing based on a deep multi-channel neural network. At the beginning of training, the network parameters are initialized. The training set and validation set of the radar signal data set are input into the deep m-channel neural network constructed in the third step. The validation set is used to adjust the hyperparameters. The mean square error is used as the loss function for regression prediction. The callback function is used to check the internal state of the training model. When the evaluation index is no longer improved, the learning rate is automatically reduced. When the number of epochs where the loss function is no longer reduced exceeds 5, stop training and save the model parameters.

[0028] Step 5: Test data timing signal prediction. Input the test data into the network model saved in step 4, and output the radar signal that has been noise-suppressed and predicted by the deep m-channel neural network.

[0029] The simulated radar data set in the present invention includes 24 signals of different modulation types, namely OOK signal, 4ASK signal, 8ASK signal, BPSK signal, QPSK signal, 8PSK signal, 16PSK signal, 32PSK signal, 16APSK signal, 32APSK signal, 64APSK signal, 128APSK signal, 16QAM signal, 32QAM signal, 64QAM signal, 128QAM signal, 256QAM signal, AM-SSB-WC signal, AM-SSB-SC signal, AM-DSB-WC signal, AM-DSB-SC signal, FM signal, GMSK signal and QPSK signal. There are 26 signal-to-noise ratios under each signal type, for example, from -20dB to 30dB, the number of single signal-to-noise ratio samples of each modulation type is 3000, and the radar signal storage form in the data set is IQ two-way signal, the single sample length is 1024 points, and the sample dimension is 1024×2. Tests show that the present invention has good denoising performance, can accurately predict the secondary radar response signal, and ensure the correct decoding of the radar signal.

[0030] The terms used in the drawings to describe positional relationships are only used for illustrative purposes and should not be construed as limiting this patent;

[0031] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. The present invention proposes a radar signal denoising method based on a multi-channel network, comprising the following steps: Step 1: Use the simulated radar data set as the training data set, and create a noisy radar signal data set based on the training data set, and divide it into a training data set and a verification data set in proportion; Step 2: Preprocess each training sample, randomly shuffle the training sample data, normalize the data, map all features of the data to the same scale, and obtain the training set and validation set; Step 3: construct a deep multi-channel neural network, including a preliminary feature extraction part, a deep feature extraction part and an upsampling part; build a radar signal classification network based on the convolutional neural network, and train the radar signal classification network with the training data set in step 2; Step 4: Use the training set to train the constructed deep multi-channel neural network, and use the validation set to adjust the hyperparameters; use the mean square error as the loss function of regression prediction, use the callback function to check the internal state of the training model, and automatically reduce the learning rate when the evaluation index no longer improves. When the loss function no longer decreases, save the model parameters to obtain the trained deep multi-channel neural network; Step 5: Obtain the radar signal to be classified, use the trained radar signal denoising network to denoise the radar signal to be classified, and then use the trained radar signal classification network to classify the denoised data to obtain the radar signal classification result.

2. The radar signal denoising method according to claim 1, characterized in that: The simulated radar data set includes 24 signals of different modulation types, namely OOK signal, 4ASK signal, 8ASK signal, BPSK signal, QPSK signal, 8PSK signal, 16PSK signal, 32PSK signal, 16APSK signal, 32APSK signal, 64APSK signal, 128APSK signal, 16QAM signal, 32QAM signal, 64QAM signal, 128QAM signal, 256QAM signal, AM-SSB-WC signal, AM-SSB-SC signal, AM-DSB-WC signal, AM-DSB-SC signal, FM signal, GMSK signal and QPSK signal.

3. The radar signal denoising method according to claim 2, characterized in that: There are 26 types of signal-to-noise ratios for each signal type, for example, from -20dB to 30dB, and the number of single signal-to-noise ratio samples for each modulation type is 3000. The radar signals in the data set are stored in the form of IQ two-way signals, with a single sample length of 1024 points and a sample dimension of 1024×2.

4. The radar signal denoising method according to claim 1, characterized in that: The deep multi-channel neural network includes 1 to m parallel channels, and the number of parallel channels can be dynamically set according to actual transmission requirements.

5. The radar signal denoising method according to claim 1, characterized in that: The convolutional neural network has no limit on the number of feature layers and the size of the convolution kernel, which can be dynamically set according to actual transmission requirements.

Images