Radar interference identification method based on LSTM and variational auto-encoder

Abstract

The invention belongs to the technical field of radar interference identification, and discloses a radar interference identification method based on LSTM and a variational auto-encoder. According to the method, a joint network of the LSTM and the variational auto-encoder is constructed to encode the time sequence signal to obtain a corresponding feature vector, then the feature vectors of the emission signal and the echo signal are spliced to obtain a combined vector capable of representing the interference type, and then classification learning is performed on the combined vector through the convolutional neural network, so that the interference type can be represented. And identification of interference signals is realized. By means of the capability of the LSTM network to solve the long-term dependence problem and the advantages of the variational auto-encoder in dimension reduction and feature extraction, time sequence data can be effectively processed, and complex distribution of data can be learned; and meanwhile, vectors before and after the interference process are spliced to represent a change relation before and after the signal, so that interference pattern identification under any change can be adapted.

Description

Technical Field

[0001] The present invention mainly relates to the technical field of radar interference recognition, and in particular to a radar interference recognition method based on LSTM and variational autoencoder. Background Art

[0002] Electromagnetic interference has a serious impact on radar detection. Based on the interference pattern, selecting appropriate waveforms and signal processing methods are common means of anti-electromagnetic interference. The performance of interference pattern recognition plays an important role in subsequent anti-interference, so interference pattern recognition has received widespread attention.

[0003] In the research related to interference identification, a commonly used method is to obtain a time-frequency image by time-frequency analysis of the received signal, and transform the interference identification problem into an image identification problem. However, this type of method is not effective in time, and its performance depends largely on the preprocessing effect of the time-frequency image, including frequency resolution, start and end time jumps, etc. Another commonly used method is to directly process the time series signal, such as directly classifying and identifying the original time domain sequence of the interference signal, but the recognition performance is poor under low interference-to-noise ratio.

[0004] In recent years, with the rise of machine learning and deep learning, interference identification technology has also entered the stage of intelligent development. After feature extraction, the decision tree method, back propagation neural network (BP) and support vector machine (SVM) can be used to identify the type of active interference signals, and good performance has been achieved. At the same time, the convolutional neural network (CNN) model can also be used to identify radar interference signals. However, these methods usually assume that the radar transmission waveform is fixed, that is, the radar interference signal perception and recognition network is all recognized under a certain specific transmission waveform. For cognitive radar, the radar transmission signal will change with the transmission, and the interference pattern will also change with the transmission waveform. Due to the large transmission waveform parameter space and the large number of interference patterns, the interference signal parameter space is huge, making it difficult for conventional interference identification networks to adapt to the interference pattern recognition under arbitrary changes, that is, it is not robust and universal for multiple transmission signals. Summary of the invention

[0005] The purpose of the present invention is to provide a radar interference identification method based on LSTM and variational autoencoder. Aiming at the problem that the existing interference identification network cannot adapt to the interference pattern recognition under arbitrary changes, a method is proposed which is not limited by the change of the transmitted signal and can also identify the interference type when the transmitted signal changes.

[0006] To achieve the above object, the present invention provides a radar interference identification method based on LSTM and variational autoencoder, comprising the following steps:

[0007] Step 1: Train the feature extraction network:

[0008] A feature extraction network is created using a long short-term memory network (LSTM) and a variational autoencoder (VAE) network. The feature extraction network includes an encoder, a fully connected layer, and a decoder.

[0009] The encoder uses the long short-term memory network LSTM to process the input sequence and extract the hidden state of the feature extraction network as the feature vector. The input sequence includes the transmitted signal s(t) and the interfered echo signal j(t). The extracted feature vectors are the transmitted feature vector and the interference feature vector.

[0010] The fully connected layers include linear layers, which construct latent variables based on the parameter distribution of the hidden states of the feature extraction network;

[0011] The decoder uses a long short-term memory network (LSTM) to decode the latent variables into a reconstructed output sequence.

[0012] Step 2: Train the classifier:

[0013] Create a classifier based on a convolutional neural network. The classifier includes one or more convolution modules consisting of a two-dimensional convolution kernel, a batch normalization layer, an activation function, and a pooling layer. The bottom layer of the classifier is a fully connected layer.

[0014] Use the softmax function as a classifier to output the interference style classification results;

[0015] Step 3: Input the transmitted signal s(t) to be identified and the interfered echo signal j(t), and use the trained feature extraction network and classifier to identify the interference pattern:

[0016] S1: Encode the transmitted signal s(t) and the received echo signal j(t) that is interfered with respectively, obtain the hidden state of the encoded feature extraction network as the feature vector of the signal, and the obtained feature vector is the transmitted feature vector and the interference feature vector;

[0017] S2: concatenate the two feature vectors in step S1 to obtain a concatenated vector y that can characterize the relationship between the transmitted signal s(t) and the interfered echo signal j(t);

[0018] S3: Perform a one-dimensional convolution operation on the concatenated vector y to transform its dimension and obtain the tensor Y;

[0019] S4: Use the classifier to classify the transformed tensor Y and output the interference style classification result.

[0020] Furthermore, the input data size of the convolution module is (128, 1, 256, 1024), where 128 represents 128 input sequences per batch; the number of channels 1 represents a single channel, 256 is the dimension after dimensionality transformation, and the vector length is 1024.

[0021] Furthermore, the loss function in the classifier uses the cross entropy loss function, assuming c i is the type of radar interference signal, p(c i ) is the target distribution, q(c i ) is the predicted distribution, then the cross entropy of p and q is expressed as

[0022] Furthermore, the parameters of the hidden state of the feature extraction network in step 1 are the mean and logarithmic variance of the emission feature vector and the interference feature vector.

[0023] Furthermore, the training method of the feature extraction network in step 1 is centralized training:

[0024] S5: Combine the transmitted signal s(t) and the echo signals j(t) with different patterns to construct an input sequence;

[0025] S6: The input sequence is passed through the long short-term memory network LSTM encoder to obtain the hidden state, the mean and log variance are extracted from the hidden state, and the latent variables are generated by reparameterization from the mean and log variance;

[0026] S7: The latent variable is decoded into a reconstructed output sequence through a long short-term memory network LSTM decoder;

[0027] S8: Use the mean of all signal losses between the reconstructed output sequence and the input sequence as the total loss function to train the entire feature extraction network.

[0028] Furthermore, the training method of training the feature extraction network in step 1 is separate training:

[0029] S9: Combine the transmitted signal s(t) and each pattern of the interfered echo signal j(t) as separate input sequences;

[0030] S10: The input sequence is passed through the long short-term memory network LSTM encoder to obtain the hidden state, the mean and logarithmic variance are extracted from the hidden state, and the latent variables are generated by reparameterization from the mean and logarithmic variance:

[0031] S11: The latent variable is decoded into a reconstructed output sequence through a long short-term memory network LSTM decoder;

[0032] S12: Using the mean signal loss between the reconstructed output sequence and the input sequence in each combination as the corresponding loss function, train the feature extraction network of the corresponding signal.

[0033] Furthermore, the training process of the feature extraction network includes:

[0034] S13: Let the input sequence be x(t), and pass through the long short-term memory network through the LSTM encoder to obtain the hidden state h_enc of the feature extraction network, h_enc = Encoder(x(t)), where Encoder() represents the encoder function;

[0035] S14: Use μ, logσ through a linear layer 2 = ExtractMeanAndLogVar(h_enc) Extract the mean μ and logarithmic variance logσ of the hidden state of the feature extraction network 2 ;

[0036] S15: by mean μ and logarithmic variance logσ 2 Generate latent variable z by reparameterization, z = μ + noise * σ, where noise is a random noise sampled from a standard normal distribution and σ represents the standard deviation;

[0037] S16: Decode the latent variables into a reconstructed output sequence x′(t), x′(t)=Decoder(z), where Decoder() represents a decoder function.

[0038] Furthermore, the tensor Y is Y=conv(y), where conv() represents a convolution function.

[0039] Furthermore, the loss function of the feature extraction network includes the mean square error loss MSE(x, x′) and KL divergence loss between the output sequence reconstructed by the feature extraction network and the input sequence. The KL divergence loss is The loss function is

[0040] loss=MSE(x,x′)+kld_weight*KL(N(μ,σ2),N(0,1)).

[0041] Beneficial effects:

[0042] The present invention provides a radar interference recognition method based on LSTM and variational autoencoder. In view of the problem that the conventional radar interference signal recognition network cannot adapt to the interference pattern recognition under arbitrary changes, a joint network of LSTM and variational autoencoder is constructed to encode the time series signal to obtain the corresponding feature vector, and then the feature vectors of the transmission signal and the echo signal are spliced ​​to obtain a combination vector that can represent the interference type, and then the combination vector is classified and learned through a convolutional neural network to realize the recognition of the interference signal. The present invention can effectively process time series data and learn the complex distribution of data by means of the ability of LSTM network to solve long-term dependency problems and the advantages of variational autoencoder in dimensionality reduction and feature extraction; at the same time, the vectors before and after the interference process are spliced ​​to represent the relationship between the changes before and after the signal, which is helpful to adapt to the interference pattern recognition under arbitrary changes. Simulation results show that in the experiment of eight common radar transmission signals and four common interference combinations, the overall recognition accuracy of the method of the present invention reaches 90%, realizing effective classification and recognition; at the same time, the overall recognition rate on untrained samples is improved by 6.5% compared with the conventional method, reflecting the effectiveness and robustness of the proposed method. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a diagram of a received signal under an SMP interference pattern according to an embodiment of the present invention;

[0044] Figure 2 is a diagram of a received signal under an SRJ interference pattern according to an embodiment of the present invention;

[0045] Figure 3 It is a flow chart of a radar interference identification method based on LSTM and variational autoencoder involved in an embodiment of the present invention;

[0046] Figure 4 It is a schematic diagram of the feature extraction network model;

[0047] Figure 5 is a diagram showing the recognition effects of five interference patterns according to an embodiment of the present invention;

[0048] Figure 6 This is a comparison chart of the recognition accuracy of the embodiment of the present invention and the conventional method.

[0049] Description of reference numerals:

[0050] x(t) is the input sequence; h_enc is the hidden state of the feature extraction network, h_enc = Encoder(x(t)); Encoder() is the encoder function; μ is the mean of the potential distribution of the variational autoencoder; logσ 2is the logarithmic variance; ExtractMeanAndLogVar() is the function for extracting the mean and logarithmic variance; z is the latent variable; noise is a random noise sampled from a standard normal distribution; σ is the standard deviation of the latent distribution of the variational autoencoder network; x′(t) is the reconstructed output sequence: Decoder() is the decoder function; s(t) is the transmitted signal; j(t) is the interfered echo signal; y is the concatenated vector; Y is a tensor; conv() is the convolution function; MSE(x, x′) is the mean square error loss between the output sequence reconstructed by the feature extraction network and the input sequence; KL divergence loss is KL(N(μ, σ 2 ), N(0, 1)); kld_weight is the KL divergence loss weight; loss is the loss function of the feature extraction network; c i is the category of radar interference signal; p(c i ) is the target distribution; q(c i ) is the predicted distribution. DETAILED DESCRIPTION

[0051] The preferred mechanism and implementation method of the present invention are further described below in conjunction with the accompanying drawings and specific implementation methods.

[0052] With the continuous development of DRFM (Digital Radio Frequency Memory) transponders, active mainlobe interference has gradually become the most mainstream type of interference in electronic countermeasures. In the embodiment of the present invention, several main types of mainlobe interference patterns are identified, but it should be understood that this is only an example and not a limitation of the present invention. The method of the present invention is generally applicable to the identification of other radar interference patterns.

[0053] In the conventional radar reception and transmission and signal processing process, when the recognition network senses that the echo signal has been interfered with, game theory methods are used to find out whether there are suitable active and passive anti-interference processing methods; if it can be suppressed by active anti-interference processing methods, there is no need to modify the transmission waveform; if it cannot be suppressed, it is necessary to design a targeted transmission waveform through active anti-mainlobe interference methods, thereby modifying the transmission waveform and re-entering the processing process.

[0054] It can be seen that the radar transmission signal will change, and the interference pattern will also change with the transmission signal. Then the relationship between the interference signal and the transmission signal can be expressed as:

[0055] j(t)=f[s(t)] (Formula 1)

[0056] Among them, s(t) is the transmitted signal, h(T) is the echo signal affected by interference, and f represents a certain relationship between the two.

[0057] The radar receiving signal is composed of the transmitting signal s(t), the interfered echo signal j(t) and the noise signal n(t), that is,

[0058] r(t)=s(t)+j(t)+n(t) (Formula 2)

[0059] Combined with (Equation 1), it can be seen that when the transmitted signal changes, the radar's received signal will change. Figure 1 and Figure 2 As shown in Figure 2, when the pulse width and repetition frequency of the transmitted signal change, the existing spectrum dispersion interference (SMSP) is difficult to effectively cover all change modes. At this time, the interference device can choose to sample a specific pulse sequence and forward it intermittently, that is, change from spectrum dispersion interference (SMSP) to intermittent sampling and repeated forwarding interference (ISRJ) to improve the pertinence and efficiency of the interference. At the same time, combined with (Formula 2), when the transmitted signal and the interfered echo signal change, the received signal will also change accordingly. For example, the waveform of the signal will change from continuous to intermittent, making it difficult for the identification network to identify the changed interference pattern.

[0060] In view of the problem that the interference pattern changes with the transmitted signal, making it difficult for the recognition network to identify the interference pattern, such as Figures 1 to 6 As shown, an embodiment of the present invention discloses a technical solution of a radar interference identification method based on LSTM and variational autoencoder. Figure 3 The figure shows a flow chart of a radar interference identification method based on LSTM and variational autoencoder according to an embodiment of the present invention. The present invention combines the Long Short-Term Memory (LSTM) and Variational Auto-Encoder (VAE) network frameworks to construct a joint LSTM+VAE network to encode the input time series signal to obtain a feature vector, and finally constructs a convolutional neural network as a classifier to complete the identification of the feature vector.

[0061] Example 1

[0062] A radar interference identification method based on LSTM and variational autoencoder, characterized in that it includes the following steps:

[0063] Step 1: Train the feature extraction network:

[0064] A feature extraction network is created using a long short-term memory network LSTM and a variational autoencoder VAE network. The feature extraction network combines the sequence processing capabilities of LSTM and the generative model characteristics of VAE, allowing it to effectively process time series data, learn the complex distribution of data, and generate new data instances from the learned latent space. The processing of the feature extraction network model is as follows: Figure 4As shown in the figure, with the help of the encoder, latent space and decoder framework of the VAE network and the sequence processing capability of LSTM, the feature extraction network includes an encoder, a fully connected layer and a decoder.

[0065] The encoder uses the long short-term memory network LSTM to process the input sequence and extract the hidden state of the feature extraction network as the feature vector. The input sequence includes the transmitted signal s(t) and the interfered echo signal j(t). The extracted feature vectors are the transmitted feature vector and the interference feature vector.

[0066] The fully connected layer includes a linear layer, which is used to implement linear transformation. The formula is: Y = XW + b, where X is the input matrix, W is the weight matrix, b is the bias vector, and Y is the output matrix. The specific code is implemented as nn.Linear in PyTorch.

[0067] Construct latent variables based on the parameter distribution of the hidden state of the feature extraction network; that is, Figure 4 The sample vector in ;

[0068] The decoder uses a long short-term memory network (LSTM) to decode the latent variables into a reconstructed output sequence.

[0069] Step 2: Train the classifier:

[0070] A classifier is created based on a convolutional neural network, wherein the classifier includes one or more convolution modules consisting of a two-dimensional convolution kernel, a BN layer, an activation function, and a pooling layer, and the bottom layer of the classifier is a fully connected layer; the classifier in this embodiment includes three or more convolution modules consisting of a two-dimensional convolution kernel, a BN layer, an activation function, and a pooling layer, and the bottom layer of the classifier is a fully connected layer;

[0071] The fully connected layer is located at the bottom of the convolutional neural network. It is used to map the feature space calculated by the previous layer to the sample label space and integrate the feature vectors together to facilitate the final classification.

[0072] Use the softmax function as a classifier to output the interference style classification results;

[0073] The Softmax function is a normalized exponential function and an activation function in classification tasks. It has a special mathematical expression for converting a vector into a probability distribution. The sum of all elements is 1, and the number of elements is the number of categories.

[0074] Step 3: Input the transmitted signal s(t) to be identified and the interfered echo signal j(t), and use the trained feature extraction network and classifier to identify the interference pattern:

[0075] S1: Encode the transmitted signal s(t) and the received echo signal j(t) that is interfered with respectively, obtain the hidden state of the encoded feature extraction network as the feature vector of the signal, and the obtained feature vector is the transmitted feature vector and the interference feature vector;

[0076] S2: concatenate the two feature vectors in step S1 to obtain a concatenated vector y that can characterize the relationship between the transmitted signal s(t) and the interfered echo signal j(t);

[0077] S3: Perform a one-dimensional convolution operation on the concatenated vector y to transform its dimension and obtain the tensor Y;

[0078] S4: Use the classifier to classify the transformed tensor Y and output the interference style classification result.

[0079] In the process of model training, the present invention trains the entire LSTM+VAE model. When extracting features, the encoder part of the model is used to extract the hidden state as the feature vector of the input sequence. The present invention provides two training methods: centralized training and separate training. The two have the same process, but the data division is different. The sample transmission signal and the echo signals with different styles of interference constitute the first training data set, which is used to train the LSTM+VAE feature extraction network.

[0080] In the centralized training mode, the sample emission signal and the echo signal interfered with in different styles are combined to construct the input sequence. The input sequence is passed through the LSTM encoder to obtain the hidden state, the mean and logarithmic variance are extracted from the hidden state, the mean and logarithmic variance are used to generate latent variables through reparameterization, and the latent variables are reconstructed into the output sequence through the LSTM decoder; the mean of all signal losses between the reconstructed output sequence and the input sequence is used as the total loss function to train the entire feature extraction network.

[0081] In the separate training mode, the sample transmission signal and the echo signal interfered with in different styles are taken as separate input sequences, where each interfered echo signal forms an input sequence. These input sequences are respectively passed through the LSTM encoder to obtain the hidden state, and the mean and logarithmic variance are extracted from the hidden state. The mean and logarithmic variance are used to generate latent variables through reparameterization, and then the latent variables are reconstructed into the output sequence through the LSTM decoder. The mean of the loss between the reconstructed output sequence of each signal and the input sequence is used as the corresponding loss function to train the feature extraction network of the corresponding signal.

[0082] Specifically, let the input sequence be x(t), and the hidden state h_enc is obtained through the LSTM encoder, that is, h_enc = Encoder(x(t)), where Encoder() represents the encoder function. The encoder is constructed using an LSTM network, and the specific parameters are shown in Table 1:

[0083] Table 1 LSTM Encoder Network Parameters

[0084] input_size hidden_size latent_size num_layers 12000 512 256 1

[0085] Among them, input_size is the size of the input sequence, hidden_size is the size of the hidden layer, latent_size is the size of the latent space, and num_layers is the number of network layers.

[0086] Subsequently, the mean μ and log variance logσ are extracted from the hidden state 2 , that is:

[0087] μ, logσ 2 = ExtractMeanAndLogVar(h_enc) (Equation 3)

[0088] These two values are used to parameterize the distribution in the latent space. The latent variable z is generated by reparameterization from the mean μ and standard deviation σ, that is, z = μ + noise * σ. ExtractMeanAndLogVar() in (Equation 8) is an operation to extract the mean and log variance, which is implemented by a linear layer (fully connected layer), specifically by performing matrix multiplication on the input data and adding a bias vector. Among them, noise is a random noise sampled from the standard normal distribution.

[0089] Next, the output sequence is reconstructed from the latent variable by the decoder. That is

[0090] x'(t) = Decoder(z) (Equation 4) The decoder function Decoder() is also constructed using an LSTM network, and the network parameters are shown in Table 2:

[0091] Table 2 LSTM Decoder Network Parameters

[0092] input_size output_size hidden_size num_layers 256 12000 512 1

[0093] Among them, input_size is the size of the input sequence, output_size is the size of the output sequence, hidden_size is the size of the hidden layer, and num_layers is the number of network layers.

[0094] The loss function of this network is the loss function of VAE, which consists of the reconstruction loss and the KL divergence loss. The specific expression is:

[0095] loss=MSE(x,x′)+kld_weight*KL(N(μ,σ 2 ), N(0, 1)) (Formula 5)

[0096] Among them, the reconstruction loss MSE(x, x′) is the mean square error loss between the output sequence reconstructed by the feature extraction network and the input sequence, which measures the ability to reconstruct the input data; kld_weight is the KL divergence loss weight, KL(N(μ,σ 2 ), N(0, 1)) is the KL divergence loss, and the potential distribution used for learning is close to the standard normal distribution, and its mathematical expression is:

[0097]

[0098] Among them, μ is the mean of the potential distribution of the VAE network, and σ is the standard deviation of the potential distribution of the VAE network.

[0099] Step 2: Train the classifier:

[0100] The present invention creates a classifier based on a convolutional neural network. The classifier consists of three convolution modules consisting of a two-dimensional convolution kernel, a BN layer, an activation function, and a pooling layer, and the last layer is a fully connected layer. Finally, softmax is used as the classifier to output the interference style classification result. The classifier network model parameters are shown in Table 3 below:

[0101] Table 3 Classifier network parameters

[0102]

[0103]

[0104] The first layer is the input data, and the input data size is (128, 1, 256, 1024), 128 represents 128 input sequences per batch; the number of channels 1 represents a single channel, 256 is the dimension after dimensionality transformation, and the vector length is 1024. The subsequent convolution modules are similar in structure, all of which are composed of convolution layer + BN + activation function + pooling layer, but the convolution kernel size is different.

[0105] The training set data of the classifier is formed by splicing and convolving the feature vector of the transmitted signal extracted by the feature extraction network encoder part with the feature vector of the echo signal interfered with in different styles. The present invention refers to it as the second training data set, that is, the tensor formed after the splicing and convolution is used to train the classifier.

[0106] With s(t) representing the transmitted signal, j(t) representing the interfered echo signal, the feature vectors extracted by the feature extraction network encoder are respectively represented as y1=h_enc(s(t)), y2=h_enc(j(t)), h_enc() represents the process of obtaining the hidden state after being processed by the feature extraction network encoder. The relationship between these two feature vectors largely reflects the relationship characteristics between the transmitted signal and the interfered echo signal, which are respectively represented as:

[0107] h_enc(s(t))=Encoder(s(t)) (Formula 7)

[0108] h_enc(j(t))=Encoder(j(t)) (Formula 8)

[0109] Wherein Encoder() represents an encoder function, which is the encoder part of the feature extraction network based on LSTM-VAE constructed by the present invention.

[0110] Concatenate the feature vectors y1 and y2 to obtain the concatenated vector y:

[0111] y=y1+y2 (Formula 9)

[0112] The vector y is transformed by a one-dimensional convolution operation to obtain the final tensor Y:

[0113] Y = conv(y) (Equation 10)

[0114] Among them, conv() represents the convolution function. Its parameter settings are shown in Table 4 below:

[0115] Table 4 One-dimensional convolution parameters

[0116] Input size Convolution kernel size Output size (128,1,1024) Size=(1,256,3),step=1,padding='same' (128,256,1024)

[0117] The loss function used in the classifier network training process is the cross entropy loss function. The cross entropy loss function is generally used in classification tasks to measure the difference between two probability distributions. Let c i is the type of radar interference signal, p(c i ) is the target distribution, q(c i ) is the predicted distribution, then the cross entropy of p and q can be expressed as:

[0118] H(p,q)=-∑ i p(c i )logq(c i ) (Formula 11)

[0119] Step 3: Use the trained feature extraction network and classifier to identify interference patterns.

[0120] After training, the feature extraction network model and the classifier network are obtained. Finally, the overall network model is applied. First, the original transmission signal s(t) and the received interfered echo signal j(t) are encoded respectively, and the hidden state of the encoded network is obtained as the feature vector of the signal. Then, the two feature vectors are spliced ​​to obtain the spliced ​​vector y that can characterize the relationship between the transmission signal and the interfered echo signal. Then, the spliced ​​vector y is transformed by a one-dimensional convolution operation to obtain a tensor Y. Finally, the classifier is used to classify and learn the transformed tensor, and the interference pattern classification result is output.

[0121] In order to verify the performance of the method of the present invention, the recognition effect of the proposed recognition network is experimentally tested below.

[0122] For the data set setting, in the simulation experiment, it is assumed that the optional space of the transmission waveform consists of eight common low intercept waveforms, namely: linear frequency modulation signal (LFM), Barker code phase coded signal, Frank code phase coded signal, P1 code phase coded signal, P2 code phase coded signal, P3 code phase coded signal, P4 code phase coded signal, Costas frequency modulation coded signal. Considering the interference pattern recognition of each frame of radar signal, the jammer interference pattern is set to four common DRFM interferences and no interference, a total of five: no interference (Unjammed), smart noise convolution interference (NCJ), intermittent sampling repetitive forwarding interference (ISRJ), velocity deception interference (VDJ), and distance deception interference (RDJ).

[0123] During the experimental training, there are 2000 training data for each transmission signal and each interference, and the total recognition accuracy on the training data set reaches 97%. There are 400 test data for each transmission signal and each interference on the test set. The following is the recognition accuracy of each transmission signal and each interference type on the test data set when the signal-to-noise ratio is 10dB. Figure 5 As shown in the figure, it can be seen from the recognition results that only when the transmission signal is P1 and the interference signal is intermittent sampling repeated forwarding interference (ISRJ), the recognition rate is low, but the overall recognition accuracy is about 90%, which has a good recognition effect.

[0124] Furthermore, in order to verify the ability of the proposed method to recognize interference patterns under arbitrary changes compared with the conventional method, the following comparative experiment is set up: four transmission signals in the data set are used for training, and another four signals are used for testing, namely, linear frequency modulation signal (LFM), Barker code phase coded signal, P1 code phase coded signal, P2 code phase coded signal are used for training, and Frank code phase coded signal, P3 code phase coded signal, P4 code phase coded signal, Costas frequency modulation coded signal are used for testing. Under the condition of a signal-to-noise ratio of 10 dB, compared with the conventional method of classifying time-frequency graphs using a CNN network, the average recognition rate curve on each interference signal is obtained as shown in the figure below: Figure 6 shown.

[0125] from Figure 6 It can be seen that compared with the conventional method, the recognition effect of this method for intermittent sampling repeated forwarding jamming (ISRJ) and RDJ is poor, but the effect for unjammed (Unjammed), smart noise convolution jamming (NCJ) and velocity deception jamming (VDJ) is better than the conventional method, and the overall average recognition rate is 6.5% higher than the conventional method, which reflects the effectiveness and robustness of the proposed method.

[0126] The present invention provides a radar interference recognition method based on LSTM and variational autoencoder. Aiming at the problem that the existing interference recognition network cannot adapt to the interference pattern recognition under arbitrary changes, a method is proposed which is not limited by the change of the transmitted signal and can also recognize the interference type when the transmitted signal changes.

[0127] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices (systems), computer equipment or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0128] The present invention is described with reference to a flowchart of a method according to an embodiment of the present invention. It should be understood that each process in the flowchart and a combination of processes in the flowchart can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the process in the flowchart. Figure 1 A device that specifies functions in a process or multiple processes.

[0129] Finally, it should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. However, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A radar interference identification method based on LSTM and variational autoencoder, characterized in that: The following steps are involved: Step 1: Train the feature extraction network: A feature extraction network is created using a long short-term memory network (LSTM) and a variational autoencoder (VAE) network. The feature extraction network includes an encoder, a fully connected layer, and a decoder. The encoder uses the long short-term memory network LSTM to process the input sequence and extract the hidden state of the feature extraction network as the feature vector. The input sequence includes the transmitted signal s(t) and the interfered echo signal j(t). The extracted feature vectors are the transmitted feature vector and the interference feature vector. The fully connected layers include linear layers, which construct latent variables based on the parameter distribution of the hidden states of the feature extraction network; The decoder uses a long short-term memory network (LSTM) to decode the latent variables into a reconstructed output sequence. Step 2: Train the classifier: Create a classifier based on a convolutional neural network. The classifier includes one or more convolution modules consisting of a two-dimensional convolution kernel, a batch normalization layer, an activation function, and a pooling layer. The bottom layer of the classifier is a fully connected layer. Use the softmax function as a classifier to output the interference style classification results; Step 3: Input the transmitted signal s(t) to be identified and the interfered echo signal j(t), and use the trained feature extraction network and classifier to identify the interference pattern: S1: Encode the transmitted signal s(t) and the received echo signal j(t) that is interfered with respectively, obtain the hidden state of the encoded feature extraction network as the feature vector of the signal, and the obtained feature vector is the transmitted feature vector and the interference feature vector; S2: concatenate the two feature vectors in step S1 to obtain a concatenated vector y that can characterize the relationship between the transmitted signal s(t) and the interfered echo signal j(t); S3: Perform a one-dimensional convolution operation on the concatenated vector y to transform its dimension and obtain the tensor Y; S4: Use the classifier to classify the transformed tensor Y and output the interference style classification result.

2. The radar interference identification method based on LSTM and variational autoencoder according to claim 1 is characterized in that: The input data size of the convolution module is (128, 1, 256, 1024), where 128 represents 128 input sequences per batch; the number of channels 1 represents a single channel, 256 is the dimension after dimensionality transformation, and the vector length is 1024.

3. The radar interference identification method based on LSTM and variational autoencoder according to claim 1 is characterized in that: The loss function in the classifier uses the cross entropy loss function, assuming c i is the type of radar interference signal, p(c i ) is the target distribution, q(c i ) is the predicted distribution, then the cross entropy of p and q is expressed as 4. The radar interference identification method based on LSTM and variational autoencoder according to claim 1 is characterized in that: The parameters of the hidden state of the feature extraction network in step 1 are the mean and logarithmic variance of the transmitted feature vector and the interference feature vector.

5. The radar interference identification method based on LSTM and variational autoencoder according to claim 4 is characterized in that: The training method for training the feature extraction network in step 1 is centralized training: S5: Combine the transmitted signal s(t) and the echo signals j(t) with different patterns to construct an input sequence; S6: The input sequence is passed through the long short-term memory network LSTM encoder to obtain the hidden state, the mean and log variance are extracted from the hidden state, and the latent variables are generated by reparameterization from the mean and log variance; S7: The latent variable is decoded into a reconstructed output sequence through a long short-term memory network LSTM decoder; S8: Use the mean of all signal losses between the reconstructed output sequence and the input sequence as the total loss function to train the entire feature extraction network.

6. The radar interference identification method based on LSTM and variational autoencoder according to claim 4 is characterized in that: The training method for training the feature extraction network in step 1 is separate training: S9: Combine the transmitted signal s(t) and each pattern of the interfered echo signal j(t) as separate input sequences; S10: The input sequence is passed through the long short-term memory network LSTM encoder to obtain the hidden state, the mean and logarithmic variance are extracted from the hidden state, and the latent variables are generated by reparameterization from the mean and logarithmic variance: S11: The latent variable is decoded into a reconstructed output sequence through a long short-term memory network LSTM decoder; S12: Using the mean signal loss between the reconstructed output sequence and the input sequence in each combination as the corresponding loss function, train the feature extraction network of the corresponding signal.

7. The radar interference identification method based on LSTM and variational autoencoder according to claim 4, characterized in that: The training process of the feature extraction network includes: S13: Let the input sequence be x(t), and pass through the long short-term memory network through the LSTM encoder to obtain the hidden state h_enc of the feature extraction network, h_enc = Encoder(x(t)), where Encoder() represents the encoder function; S14: Use μ, logσ through a linear layer 2 =ExtractMeanAndLogVar(h - enc) Extract the mean μ and logarithmic variance logσ of the hidden state of the feature extraction network 2 ; S15: by mean μ and logarithmic variance logσ 2 Generate latent variable z by reparameterization, z = μ + noise * σ, where noise is a random noise sampled from a standard normal distribution and σ represents the standard deviation; S16: Decode the latent variables into a reconstructed output sequence x′(t), x′(t)=Decoder(z), where Decoder() represents a decoder function.

8. The radar interference identification method based on LSTM and variational autoencoder according to claim 4, characterized in that: The tensor Y is Y=conv(y), where conv() represents the convolution function.

9. The radar interference identification method based on LSTM and variational autoencoder according to claim 5 or 6, characterized in that: The loss function of the feature extraction network includes the mean square error loss MSE(x, x′) between the output sequence reconstructed by the feature extraction network and the input sequence and the KL divergence loss. The KL divergence loss is The loss function is loss = MSE (x, x') + kld_weight * KL (N (μ, σ 2 ), N(0,1)).

Images