A satellite signal automatic modulation identification method and system based on matching verification

Abstract

The present invention belongs to the technical field of automatic modulation identification of satellite signals, and specifically relates to a method and system for automatic modulation identification of satellite signals based on matching verification, which obtains the satellite signal to be identified and transmits it to a feature extraction module for multi-dimensional feature parameter extraction; performs data normalization processing on the extracted multi-dimensional feature parameters; creates and trains a satellite signal classification model, inputs the satellite signal classification model to perform preliminary classification of the satellite signal to be identified, and obtains a round of identification results; sends it to a demodulation library to extract the demodulated signal constellation data to construct a constellation point diagram under different parameters as a data set; inputs the matching verification recognition model to perform a second iterative recognition based on matching verification to obtain a second round of recognition results to obtain the final satellite signal modulation recognition result. The above process greatly increases the recognition accuracy while ensuring the overall recognition speed through a first round of recognition of the classification model and a second iterative recognition of the matching verification.

Description

Technical Field

[0001] The present invention belongs to the technical field of satellite signal automatic modulation identification, and in particular relates to a satellite signal automatic modulation identification method and system based on matching verification. Background Art

[0002] Automatic modulation identification of satellite signals refers to the use of automated methods to identify the modulation method used by satellite signals in satellite communication systems. It is an important link in the fields of communication identification, electronic reconnaissance and situational awareness. It is widely used in electronic reconnaissance, electronic countermeasures, spectrum monitoring and other fields, and has important military and civilian value.

[0003] In the prior art, the modulation mode used by satellite signals is identified in the following traditional detection and identification methods: first, the system needs to detect the existence of satellite signals; then the parameters of the detected signals are estimated, including the frequency, bandwidth, modulation mode, etc. Finally, based on the estimated parameters, a specific algorithm or model is used to identify the specific modulation mode of the signal.

[0004] With the development of intelligent technology, the satellite signal automatic modulation recognition method based on neural network recognition algorithm has been rapidly developed due to its stronger feature extraction and classification capabilities. However, although the recognition algorithm based on neural network has stronger feature extraction and classification capabilities, it has problems such as poor accuracy in the recognition of high-order signals and low signal-to-noise ratio satellite signals.

[0005] Therefore, how to improve the existing satellite signal automatic modulation recognition method to improve the accuracy of the satellite signal automatic modulation recognition method is a technical problem that urgently needs to be solved. Summary of the invention

[0006] The purpose of the present invention is to provide a satellite signal automatic modulation identification method and system based on matching verification, so as to improve the existing satellite signal automatic modulation identification method and enhance the accuracy of the satellite signal automatic modulation identification method.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0008] A satellite signal automatic modulation identification method based on matching verification comprises the following steps:

[0009] S1: acquiring a satellite signal to be identified through a signal acquisition module, and transmitting the satellite signal to be identified to a feature extraction module, wherein the feature extraction module extracts multi-dimensional feature parameters from the satellite signal to be identified;

[0010] S2: Perform data normalization processing on the extracted multi-dimensional feature parameters according to the feature dimensions;

[0011] S3: Create a satellite signal classification model, input the labeled satellite signals into the satellite signal classification model to train the model, and input the preprocessed multi-dimensional feature parameters into the satellite signal classification model to perform preliminary classification of the satellite signals to be identified to obtain a round of identification results;

[0012] S4: sending the signal parameters in the one round of recognition results to the demodulation library to extract the demodulated signal constellation data, and constructing constellation point diagrams under different parameters as data sets;

[0013] S5: creating a matching verification recognition model, inputting the data sets of the constellation point diagrams under different parameters into the matching verification recognition model, and performing a second iterative recognition based on matching verification on the data sets of the constellation point diagrams under different parameters through the matching verification recognition model to obtain a second round of recognition results;

[0014] S6: Calculate the confidence of the two-round recognition results, and compare the confidence calculated from the two-round recognition results with a preset confidence threshold. When the confidence of the two-round recognition results is greater than the preset confidence threshold, the two-round recognition results are used as the final satellite signal modulation recognition results; otherwise, the one-round recognition results are used as the final satellite signal modulation recognition results.

[0015] Preferably, the specific process of the feature extraction module in step S1 extracting multi-dimensional feature parameters from the satellite signal to be identified is as follows:

[0016] S11: Perform frequency offset correction on the input satellite signal, and then perform Fourier transform on the time domain data after the frequency offset correction to convert it into frequency domain data, and then extract frequency domain features, including: spectrum lines, double spectrum lines, quadruple spectrum lines, octave spectrum lines, double spectrum maximum value, quadruple spectrum maximum value, relative value between double spectrum maximum value and second largest value, relative value between quadruple spectrum maximum value and second largest value.

[0017] S12: After normalization, square root filtering, and gardner timing synchronization of the time domain data, the high-order cumulants, the Euclidean distance between IQ symbols, and the error between the theoretical value of the constellation ring amplitude and the actual amplitude calculated by the locked IQ symbol are calculated through the locked IQ symbol.

[0018] Preferably, the specific process of performing data normalization processing on the extracted multi-dimensional feature parameters according to the feature dimensions in step S2 is as follows:

[0019] S21: Create a mapping interval for data normalization processing [ a , b ];

[0020] S22: Mapping the extracted feature parameters of each dimension to the mapping interval [a , b ], the specific mapping formula is as follows:

[0021] X=[( xx min ) / ( x max -x min )]*( ba )+ a ;

[0022] in, a To map the data to the lower limit of the mapping interval, b To map data to the upper limit of the mapping interval, x min is the minimum value of the characteristic parameter, x max is the maximum value of the characteristic parameter, X is the characteristic parameter after mapping, x is the feature parameter before mapping.

[0023] Preferably, the satellite signal classification model is a model composed of three planes, and the three planes include a discrimination plane and two support planes.

[0024] Preferably, in step S3, the specific process of inputting the preprocessed multi-dimensional feature parameters into the satellite signal classification model to perform preliminary classification of the satellite signals to be identified to obtain a round of identification results is as follows:

[0025] S31: Let the preprocessed multi-dimensional feature parameters be linearly separable m data samples, the data samples are represented as follows:

[0026] ( x 1 (0) , x 2 (0) ,···, x n (0) , y 0),

[0027] ( x 1 (1) , x 2 (1) ,···, x n (1) , y 1),

[0028] ···

[0029] ( x 1 (m) , x 2(m) ,···, x n (m) , y m ),

[0030] in, x n for n Dimensional features, y is a binary output;

[0031] S32: constructing constraint optimization conditions, finding the vector value corresponding to the minimum through a preset algorithm, and finding the separating hyperplane parameters of the original problem;

[0032] S33: Find a specified number of support vectors, and calculate the separation hyperplane parameters corresponding to each support vector to obtain the final classification hyperplane and the final classification decision function.

[0033] Preferably, in step S4, the specific process of sending the signal parameters in the one-round identification result into the demodulation library to extract the demodulated signal constellation data is as follows:

[0034] S41: Estimating the carrier frequency and symbol rate using frequency multiplication for the identified satellite signal modulation type;

[0035] S42: down-converting the satellite signal according to the carrier frequency estimation value to obtain baseband data;

[0036] S43: designing a matched filter according to the symbol rate estimation value, filtering the baseband data, and performing timed sampling on the filtered data using the Gardner algorithm to obtain optimal sampling point data;

[0037] S44: Find the synchronization head according to the best sampling point data, phase-lock each frame of data, and obtain modulation symbols, that is, constellation data.

[0038] Preferably, the matching verification recognition model is a deep convolutional neural network model having at least five layers;

[0039] The first layer includes a convolution layer with a kernel size of 7*7 and a stride of 2 + a maximum pooling layer with a kernel size of 3*3 and a stride of 2 + a batch normalization layer + a ReLU activation function;

[0040] The second layer includes three improved bottleneck layers. Each improved bottleneck layer consists of a 1*1 convolution layer (for reducing the number of input channels), a 3*3 convolution layer (for feature extraction), and a 1*1 convolution layer (for restoring the number of channels). The number of input and output channels of each improved bottleneck layer is the same, and addition operations can be performed.

[0041] The third layer includes a plurality of improved bottleneck layers, each of which has different numbers of input and output channels, wherein the first improved bottleneck layer includes a convolution layer with a convolution kernel size of 3*3 and a stride of 2, and the first improved bottleneck layer is used for downsampling;

[0042] The structure of the fourth layer is similar to that of the third layer, and the number of channels in the fourth layer is greater than that in the third layer;

[0043] The structure of the fifth layer is also similar to that of the third layer, except that the number of channels is greater than that of the fourth layer;

[0044] Each layer contains multiple residual modules, each residual module has several layers of neurons, and adjacent neurons inside the residual module are connected by skip connections.

[0045] Preferably, the matching verification recognition model includes an input layer, a padding ZeroPad, a two-dimensional convolution layer, a batch normalization layer, an activation function, a maximum pooling layer, a plurality of convolution blocks and a recognition block combination layer, an average pooling layer, a flattening layer, a fully connected layer and an output layer, each of the convolution blocks and the recognition block combination layer is composed of a convolution block and a recognition block, each convolution block includes two channels, one of which includes a plurality of two-dimensional convolution layers, a batch normalization layer, and an activation function, and the other channel is provided with a two-dimensional convolution layer and a batch normalization layer, each recognition block includes two channels, one of which includes a plurality of two-dimensional convolution layers, a batch normalization layer, and an activation function.

[0046] In a second aspect, a satellite signal automatic modulation identification system based on matching verification is provided, which is used to implement the satellite signal automatic modulation identification method based on matching verification, including a signal acquisition module, a feature extraction module, a normalization processing module, a model creation module, a satellite signal classification model, a demodulation library, a matching verification identification model and a confidence calculation module, wherein the signal acquisition module is connected to the feature extraction module, the feature extraction module is connected to the normalization processing module, the model creation module is connected to the satellite signal classification model and the matching verification identification model, the satellite signal classification model is connected to the demodulation library, and the matching verification identification model is connected to the confidence calculation module;

[0047] The signal acquisition module is used to acquire the satellite signal to be identified through the signal acquisition module, and transmit the satellite signal to be identified to the feature extraction module;

[0048] The feature extraction module is used to extract multi-dimensional feature parameters of the satellite signal to be identified;

[0049] The normalization processing module is used to perform data normalization processing on the extracted multi-dimensional feature parameters according to the feature dimensions;

[0050] The model creation module is used to create a satellite signal classification model and a matching verification recognition model;

[0051] The satellite signal classification model is used to perform preliminary classification on the input pre-processed multi-dimensional feature parameters to obtain a round of recognition results;

[0052] The demodulation library is used to send the signal parameters in the one-round recognition result into the demodulation library to extract the demodulated signal constellation data, and construct the constellation point diagram under different parameters as a data set;

[0053] The matching verification recognition model is used to perform a second iterative recognition based on matching verification on the data set of the constellation point diagram under different parameters to obtain a second round of recognition results;

[0054] The confidence calculation module is used to calculate the confidence of the two-round recognition results, compare the confidence calculated from the two-round recognition results with a preset confidence threshold, and when the confidence of the two-round recognition results is greater than the preset confidence threshold, use the two-round recognition results as the final satellite signal modulation recognition results; otherwise, use the one-round recognition results as the final satellite signal modulation recognition results.

[0055] The beneficial effects of the present invention include:

[0056] The satellite signal automatic modulation identification method and system based on matching verification provided by the present invention obtains the satellite signal to be identified, and transmits it to the feature extraction module to extract multi-dimensional feature parameters; performs data normalization processing on the extracted multi-dimensional feature parameters; creates and trains a satellite signal classification model, inputs the satellite signal classification model to perform preliminary classification of the satellite signal to be identified to obtain a round of identification results; sends it to the demodulation library to extract the demodulated signal constellation data to construct constellation point diagrams under different parameters as a data set; inputs the matching verification recognition model to perform secondary iterative recognition based on matching verification to obtain second rounds of recognition results to obtain the final satellite signal modulation recognition results. The above process first accurately extracts key and effective signal features, and combines the first round of recognition of the classification model and the secondary iterative recognition of matching verification to greatly increase the recognition accuracy while ensuring the overall recognition speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 The figure is a flow chart of the satellite signal automatic modulation identification method based on matching verification of the present invention.

[0058] Figure 2 It is a structural diagram of the matching verification recognition model of the present invention. DETAILED DESCRIPTION

[0059] The following is combined with Figure 1~Figure 2 The present invention is further described in detail:

[0060] Example 1

[0061] See attached Figure 1 As shown, a satellite signal automatic modulation identification method based on matching verification includes the following steps:

[0062] S1: The satellite signal to be identified is acquired through the signal acquisition module, and the satellite signal to be identified is transmitted to the feature extraction module. The feature extraction module extracts multi-dimensional feature parameters of the satellite signal to be identified, and obtains multi-dimensional feature parameters of the satellite signal, thereby providing an accurate data basis for subsequent automatic modulation identification of satellite signals and improving the efficiency and accuracy of subsequent modulation identification.

[0063] S2: Perform data normalization on the extracted multi-dimensional feature parameters according to the feature dimensions. Through the data normalization process, the dimensional differences between the feature parameters of each dimension are eliminated to ensure that each feature parameter has the same measurement scale when input into the subsequent satellite signal classification model, so that the numerical ranges of different features are consistent, thereby avoiding analysis bias caused by dimensional differences and ensuring that the satellite signal classification model will not be unbalanced due to scale differences during the learning and training process, thereby improving the performance and stability of the model.

[0064] S3: Create a satellite signal classification model, input the labeled satellite signals into the satellite signal classification model to train the model, and input the preprocessed multi-dimensional feature parameters into the satellite signal classification model to perform preliminary classification of the satellite signals to be identified to obtain a round of identification results.

[0065] S4: Send the signal parameters in the said round of recognition results to the demodulation library to extract the demodulated signal constellation data, and construct the constellation point diagram under different parameters as a data set. By obtaining the constellation point diagram data set under different parameters, a data basis is provided for the subsequent matching verification recognition model to realize secondary iterative recognition through a matching verification process.

[0066] S5: Create a matching verification recognition model, input the data set of the constellation point diagram under different parameters into the matching verification recognition model, and use the matching verification recognition model to perform a second iterative recognition based on matching verification on the data set of the constellation point diagram under different parameters to obtain a second round of recognition results. In this process, the onnxruntime inference engine is used. The onnxruntime inference engine provides an efficient execution environment that enables machine learning models to be quickly executed on various hardware. The onnxruntime inference engine improves the inference efficiency by optimizing the network structure and underlying operations of the model.

[0067] S6: Calculate the confidence of the two-round recognition results, and compare the confidence calculated from the two-round recognition results with a preset confidence threshold. When the confidence of the two-round recognition results is greater than the preset confidence threshold, the two-round recognition results are used as the final satellite signal modulation recognition results; otherwise, the one-round recognition results are used as the final satellite signal modulation recognition results.

[0068] In the prior art, compared with the most traditional detection and recognition methods, the recognition algorithms based on neural networks developed later have stronger feature extraction and classification capabilities, but there are still problems such as poor recognition accuracy under high-order signals and low signal-to-noise ratios. The present invention obtains the satellite signal to be identified and transmits it to the feature extraction module for multi-dimensional feature parameter extraction; performs data normalization processing on the extracted multi-dimensional feature parameters; creates and trains a satellite signal classification model, inputs the satellite signal classification model to perform preliminary classification of the satellite signal to be identified to obtain a round of recognition results; sends it to the demodulation library to extract the demodulated signal constellation data to construct constellation point diagrams under different parameters as a data set; inputs the matching verification recognition model to perform secondary iterative recognition based on matching verification to obtain the second round of recognition results to obtain the final satellite signal modulation recognition results. The above process first accurately extracts key and effective signal features, and combines the first round of recognition of the classification model and the secondary iterative recognition of the matching verification, while greatly increasing the modulation recognition accuracy, ensuring the overall recognition speed.

[0069] In this embodiment, the specific process of the feature extraction module in step S1 extracting multi-dimensional feature parameters from the satellite signal to be identified is as follows:

[0070] S11: Perform frequency offset correction on the input satellite signal, and then perform Fourier transform on the time domain data after the frequency offset correction to convert it into frequency domain data, and then extract frequency domain features. The specific frequency domain features are: spectrum lines, double spectrum lines, quadruple spectrum lines, octave spectrum lines, double spectrum maximum value, quadruple spectrum maximum value, relative value between double spectrum maximum value and second largest value, relative value between quadruple spectrum maximum value and second largest value.

[0071] S12: After normalization, square root filtering, and gardner timing synchronization of the time domain data, the high-order cumulants, the Euclidean distance between IQ symbols, and the error between the theoretical value of the constellation ring amplitude and the actual amplitude calculated by the locked IQ symbol are calculated through the locked IQ symbol.

[0072] Example 2

[0073] On the basis of Example 1, the specific process of performing data normalization processing on the extracted multi-dimensional feature parameters according to the feature dimensions in step S2 is as follows:

[0074] S21: Create a mapping interval for data normalization processing [a , b ];

[0075] S22: Mapping the extracted feature parameters of each dimension to the mapping interval [ a , b ], the specific mapping formula is as follows:

[0076] X=[( xx min ) / ( x max -x min )]*( ba )+ a ;

[0077] in, a To map the data to the lower limit of the mapping interval, b To map data to the upper limit of the mapping interval, x min is the minimum value of the characteristic parameter, x max is the maximum value of the characteristic parameter, X is the characteristic parameter after mapping, x is the feature parameter before mapping.

[0078] In this embodiment, the satellite signal classification model is a model composed of three planes, including a discriminant plane and two support planes. The discriminant plane is the core of the satellite signal classification model and is used to divide the data set into different categories. The discriminant plane is a linear segmentation line in a multidimensional space, which is used to separate two categories. In three-dimensional space, it is a plane, and in a higher-dimensional space, the discriminant plane is a subspace. The two support planes are the training sample points closest to the discriminant plane, which determine the position and direction of the decision boundary. They are the points that have the greatest impact on the classification decision boundary. By maximizing the distance between the support vector and the decision boundary, the model can obtain the best generalization ability.

[0079] In the present invention, the specific process of inputting the pre-processed multi-dimensional feature parameters into the satellite signal classification model to perform preliminary classification of the satellite signals to be identified to obtain a round of identification results in step S3 is as follows:

[0080] S31: Let the preprocessed multi-dimensional feature parameters be linearly separable m data samples, the data samples are represented as follows:

[0081] ( x 1 (0) , x 2 (0) ,···, xn (0) , y 0),

[0082] ( x 1 (1) , x 2 (1) ,···, x n (1) , y 1),

[0083] ···

[0084] ( x 1 (m) , x 2 (m) ,···, x n (m) , y m ),

[0085] in, x n for n Dimensional features, y is a binary output;

[0086] S32: constructing constraint optimization conditions, finding the vector value corresponding to the minimum through a preset algorithm, and finding the separating hyperplane parameters of the original problem;

[0087] S33: Find a specified number of support vectors, and calculate the separation hyperplane parameters corresponding to each support vector to obtain the final classification hyperplane and the final classification decision function.

[0088] The specific process of sending the signal parameters in the round of recognition results to the demodulation library to extract the demodulated signal constellation data in step S4 is as follows:

[0089] S41: Estimating the carrier frequency and symbol rate using frequency multiplication for the identified satellite signal modulation type;

[0090] S42: down-converting the satellite signal according to the carrier frequency estimation value to obtain baseband data;

[0091] S43: designing a matched filter according to the symbol rate estimation value, filtering the baseband data, and performing timed sampling on the filtered data using the Gardner algorithm to obtain optimal sampling point data;

[0092] S44: Find the synchronization head according to the best sampling point data, phase-lock each frame of data, and obtain modulation symbols, that is, constellation data.

[0093] Example 3

[0094] On the basis of Example 1 or Example 2, the matching verification recognition model is a deep convolutional neural network model having at least five layers;

[0095] The first layer includes a convolution layer with a kernel size of 7*7 and a stride of 2 + a maximum pooling layer with a kernel size of 3*3 and a stride of 2 + a batch normalization layer + a ReLU activation function;

[0096] The second layer includes three improved bottleneck layers. Each improved bottleneck layer consists of a 1*1 convolution layer (for reducing the number of input channels), a 3*3 convolution layer (for feature extraction), and a 1*1 convolution layer (for restoring the number of channels). The number of input and output channels of each improved bottleneck layer is the same, and addition operations can be performed.

[0097] The third layer includes a plurality of improved bottleneck layers, each of which has different numbers of input and output channels, wherein the first improved bottleneck layer includes a convolution layer with a convolution kernel size of 3*3 and a stride of 2, and the first improved bottleneck layer is used for downsampling;

[0098] The structure of the fourth layer is similar to that of the third layer, and the number of channels in the fourth layer is greater than that in the third layer;

[0099] The structure of the fifth layer is also similar to that of the third layer, except that the number of channels is greater than that of the fourth layer;

[0100] Each layer contains multiple residual modules, each residual module has several layers of neurons, and adjacent neurons inside the residual module are connected by skip connections.

[0101] In another implementation of this embodiment, see Figure 2 The matching verification recognition model includes an input layer, a padding ZeroPad, a two-dimensional convolution layer, a batch normalization layer, an activation function, a maximum pooling layer, a plurality of convolution blocks and a recognition block combination layer, an average pooling layer, a flattening layer, a fully connected layer and an output layer. Each of the convolution blocks and the recognition block combination layer is composed of a convolution block and a recognition block. Each convolution block includes two channels, one of which includes a plurality of two-dimensional convolution layers, a batch normalization layer, and an activation function, and the other channel is provided with a two-dimensional convolution layer and a batch normalization layer. Each recognition block includes two channels, one of which includes a plurality of two-dimensional convolution layers, a batch normalization layer, and an activation function.

[0102] A satellite signal automatic modulation identification system based on matching verification is used to implement the satellite signal automatic modulation identification method based on matching verification, comprising a signal acquisition module, a feature extraction module, a normalization processing module, a model creation module, a satellite signal classification model, a demodulation library, a matching verification identification model and a confidence calculation module, wherein the signal acquisition module is connected to the feature extraction module, the feature extraction module is connected to the normalization processing module, the model creation module is connected to the satellite signal classification model and the matching verification identification model, the satellite signal classification model is connected to the demodulation library, and the matching verification identification model is connected to the confidence calculation module.

[0103] The signal acquisition module is used to acquire the satellite signal to be identified through the signal acquisition module, and transmit the satellite signal to be identified to the feature extraction module. The feature extraction module is used to extract multi-dimensional feature parameters of the satellite signal to be identified; the normalization processing module is used to perform data normalization processing on the extracted multi-dimensional feature parameters according to the feature dimensions; the model creation module is used to create a satellite signal classification model and a matching verification recognition model; the satellite signal classification model is used to perform preliminary classification on the input pre-processed multi-dimensional feature parameters to obtain a first-round recognition result; the demodulation library is used to send the signal parameters in the first-round recognition result to the demodulation library to extract the demodulated signal constellation data and construct the constellation point diagram under different parameters as a data set; the matching verification recognition model is used to perform secondary iterative recognition based on matching verification on the data set of the constellation point diagram under different parameters to obtain a second-round recognition result; the confidence calculation module is used to calculate the confidence of the second-round recognition result, compare the confidence calculated by the second-round recognition result with a preset confidence threshold, and when the confidence of the second-round recognition result is greater than the preset confidence threshold, the second-round recognition result is used as the final satellite signal modulation recognition result, otherwise the first-round recognition result is used as the final satellite signal modulation recognition result.

[0104] In summary, the satellite signal automatic modulation identification method and system based on matching verification provided by the present invention obtains the satellite signal to be identified and transmits it to the feature extraction module to extract multi-dimensional feature parameters; performs data normalization processing on the extracted multi-dimensional feature parameters; creates and trains a satellite signal classification model, inputs the satellite signal classification model to perform preliminary classification of the satellite signal to be identified to obtain a round of identification results; sends it to the demodulation library to extract the demodulated signal constellation data to construct a constellation point diagram under different parameters as a data set; inputs the matching verification recognition model to perform secondary iterative recognition based on matching verification to obtain second-round recognition results to obtain the final satellite signal modulation recognition result, thereby achieving accurate extraction of key and effective signal features, and combining one-round recognition of the classification model and secondary iterative recognition of the matching verification, while greatly increasing the recognition accuracy, ensuring the overall recognition speed.

Claims

1. A satellite signal automatic modulation identification method based on matching verification, characterized in that: The following steps are involved: S1: acquiring a satellite signal to be identified through a signal acquisition module, and transmitting the satellite signal to be identified to a feature extraction module, wherein the feature extraction module extracts multi-dimensional feature parameters from the satellite signal to be identified; S2: Perform data normalization processing on the extracted multi-dimensional feature parameters according to the feature dimensions; S3: Create a satellite signal classification model, input the labeled satellite signals into the satellite signal classification model to train the model, and input the preprocessed multi-dimensional feature parameters into the satellite signal classification model to perform preliminary classification of the satellite signals to be identified to obtain a round of identification results; S4: sending the signal parameters in the one round of recognition results to the demodulation library to extract the demodulated signal constellation data, and constructing constellation point diagrams under different parameters as data sets; S5: creating a matching verification recognition model, inputting the data sets of the constellation point diagrams under different parameters into the matching verification recognition model, and performing a second iterative recognition based on matching verification on the data sets of the constellation point diagrams under different parameters through the matching verification recognition model to obtain a second round of recognition results; S6: Calculate the confidence of the two-round recognition results, and compare the confidence calculated from the two-round recognition results with a preset confidence threshold. When the confidence of the two-round recognition results is greater than the preset confidence threshold, the two-round recognition results are used as the final satellite signal modulation recognition results; otherwise, the one-round recognition results are used as the final satellite signal modulation recognition results.

2. The satellite signal automatic modulation identification method based on matching verification according to claim 1, characterized in that: The specific process of the feature extraction module in step S1 extracting multi-dimensional feature parameters from the satellite signal to be identified is as follows: S11: performing frequency offset correction on the input satellite signal, and then performing Fourier transform on the time domain data after the frequency offset correction to convert it into frequency domain data, and then extracting frequency domain features, wherein the frequency domain features include spectrum lines, double spectrum lines, quadruple spectrum lines, octuple spectrum lines, double spectrum maximum value, quadruple spectrum maximum value, relative value between double spectrum maximum value and second largest value, and relative value between quadruple spectrum maximum value and second largest value; S12: After normalization, square root filtering, and gardner timing synchronization of the time domain data, the high-order cumulants, the Euclidean distance between IQ symbols, and the error between the theoretical value of the constellation ring amplitude and the actual amplitude calculated by the locked IQ symbol are calculated through the locked IQ symbol.

3. The satellite signal automatic modulation identification method based on matching verification according to claim 1, characterized in that: The specific process of performing data normalization processing on the extracted multi-dimensional feature parameters according to the feature dimensions in step S2 is as follows: S21: Create a mapping interval for data normalization processing [ a , b ]; S22: Mapping the extracted feature parameters of each dimension to the mapping interval [ a , b ], the specific mapping formula is as follows: X=[( xx min ) / ( x max -x min )]*( ba )+ a ； in, a To map the data to the lower limit of the mapping interval, b To map data to the upper limit of the mapping interval, x min is the minimum value of the characteristic parameter, x max is the maximum value of the characteristic parameter, X is the characteristic parameter after mapping, x is the feature parameter before mapping.

4. The satellite signal automatic modulation identification method based on matching verification according to claim 1, characterized in that: The satellite signal classification model is a model composed of three planes, and the three planes include a discrimination plane and two support planes.

5. The satellite signal automatic modulation identification method based on matching verification according to claim 4 is characterized in that: The specific process of inputting the pre-processed multi-dimensional feature parameters into the satellite signal classification model to perform preliminary classification of the satellite signals to be identified and obtain a round of identification results in step S3 is as follows: S31: Let the preprocessed multi-dimensional feature parameters be linearly separable m data samples, the data samples are represented as follows: ( x 1 (0) ， x 2 (0) ，···， x n (0) ， y 0)， ( x 1 (1) ， x 2 (1) ，···， x n (1) ， y 1)， ··· ( x 1 (m) ， x 2 (m) ，···， x n (m) ， y m )， in, x n for n Dimensional features, y is a binary output; S32: constructing constraint optimization conditions, finding the vector value corresponding to the minimum through a preset algorithm, and finding the separating hyperplane parameters of the original problem; S33: Find a specified number of support vectors, and calculate the separation hyperplane parameters corresponding to each support vector to obtain the final classification hyperplane and the final classification decision function.

6. The satellite signal automatic modulation identification method based on matching verification according to claim 1, characterized in that: The specific process of sending the signal parameters in the round of recognition results to the demodulation library to extract the demodulated signal constellation data in step S4 is as follows: S41: Estimating the carrier frequency and symbol rate using frequency multiplication for the identified satellite signal modulation type; S42: down-converting the satellite signal according to the carrier frequency estimation value to obtain baseband data; S43: designing a matched filter according to the symbol rate estimation value, filtering the baseband data, and performing timed sampling on the filtered data using the Gardner algorithm to obtain optimal sampling point data; S44: Find the synchronization head according to the best sampling point data, phase-lock each frame of data, and obtain modulation symbols, that is, constellation data.

7. The satellite signal automatic modulation identification method based on matching verification according to claim 1, characterized in that: The matching verification recognition model is a deep convolutional neural network model with at least five layers; The first layer includes a convolution layer with a kernel size of 7*7 and a stride of 2 + a maximum pooling layer with a kernel size of 3*3 and a stride of 2 + a batch normalization layer + a ReLU activation function; The second layer includes three improved bottleneck layers. Each improved bottleneck layer consists of a 1*1 convolution layer, a 3*3 convolution layer, and a 1*1 convolution layer. The number of input and output channels of each improved bottleneck layer is the same, and addition operations can be performed. The third layer includes a plurality of improved bottleneck layers, each of which has different numbers of input and output channels, wherein the first improved bottleneck layer includes a convolution layer with a convolution kernel size of 3*3 and a stride of 2, and the first improved bottleneck layer is used for downsampling; The structure of the fourth layer is similar to that of the third layer, and the number of channels in the fourth layer is greater than that in the third layer; The structure of the fifth layer is also similar to that of the third layer, except that the number of channels is greater than that of the fourth layer; Each layer contains multiple residual modules, each residual module has several layers of neurons, and adjacent neurons inside the residual module are connected by skip connections.

8. The satellite signal automatic modulation identification method based on matching verification according to claim 1, characterized in that: The matching verification recognition model includes an input layer, a padding ZeroPad, a two-dimensional convolution layer, a batch normalization layer, an activation function, a maximum pooling layer, a plurality of convolution blocks and a recognition block combination layer, an average pooling layer, a flattening layer, a fully connected layer and an output layer. Each of the convolution blocks and the recognition block combination layer is composed of a convolution block and a recognition block. Each convolution block includes two channels, one of which includes a plurality of two-dimensional convolution layers, a batch normalization layer, and an activation function, and the other channel is provided with a two-dimensional convolution layer and a batch normalization layer. Each recognition block includes two channels, one of which includes a plurality of two-dimensional convolution layers, a batch normalization layer, and an activation function.

9. A satellite signal automatic modulation identification system based on matching verification, used to implement a satellite signal automatic modulation identification method based on matching verification according to any one of claims 1 to 8, characterized in that: It includes a signal acquisition module, a feature extraction module, a normalization processing module, a model creation module, a satellite signal classification model, a demodulation library, a matching verification recognition model and a confidence calculation module, wherein the signal acquisition module is connected to the feature extraction module, the feature extraction module is connected to the normalization processing module, the model creation module is connected to the satellite signal classification model and the matching verification recognition model, the satellite signal classification model is connected to the demodulation library, and the matching verification recognition model is connected to the confidence calculation module; The signal acquisition module is used to acquire the satellite signal to be identified through the signal acquisition module, and transmit the satellite signal to be identified to the feature extraction module; The feature extraction module is used to extract multi-dimensional feature parameters of the satellite signal to be identified; The normalization processing module is used to perform data normalization processing on the extracted multi-dimensional feature parameters according to the feature dimensions; The model creation module is used to create a satellite signal classification model and a matching verification recognition model; The satellite signal classification model is used to perform preliminary classification on the input pre-processed multi-dimensional feature parameters to obtain a round of recognition results; The demodulation library is used to send the signal parameters in the one-round recognition result into the demodulation library to extract the demodulated signal constellation data, and construct the constellation point diagram under different parameters as a data set; The matching verification recognition model is used to perform a second iterative recognition based on matching verification on the data set of the constellation point diagram under different parameters to obtain a second round of recognition results; The confidence calculation module is used to calculate the confidence of the two-round recognition results, compare the confidence calculated from the two-round recognition results with a preset confidence threshold, and when the confidence of the two-round recognition results is greater than the preset confidence threshold, use the two-round recognition results as the final satellite signal modulation recognition results; otherwise, use the one-round recognition results as the final satellite signal modulation recognition results.

Images