Dual-stream modulation signal classification method and system based on multi-feature fusion

Abstract

The present disclosure provides a dual-stream modulation signal classification method and system based on multi-feature fusion, which relates to the technical field of radio modulation signal classification, obtains I / Q data of the original modulation signal, and divides the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream; inputs the high signal-to-noise ratio stream and the low signal-to-noise ratio stream into a dual-stream CABLD model, the dual-stream CABLD model includes two CABLD classification channels, each CABLD classification channel includes a first CABLD model and a second CABLD model, after the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively, firstly perform multi-signal classification through the first CABLD model, obtain an AW-type signal synthesized by AM‑DSB and WBFM signals, then extract the frequency domain amplitude and phase features of the AW-type signal respectively and input them into the second CABLD model, perform binary classification on the AM‑DSB and WBFM signals in the AW-type signal, and obtain the classification results of the AM‑DSB and WBFM signals. The present disclosure improves the accuracy of signal classification.

Description

Technical Field

[0001] The present disclosure relates to the technical field of radio modulation signal classification, and in particular to a dual-stream modulation signal classification method and system based on multi-feature fusion. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] In recent years, the role of communication technology in the 5G era has become increasingly apparent, and spatial cognitive radio technology is particularly important. Among them, Automatic Modulation Recognition (AMR) is a key technology in spatial cognitive radio, which has practical applications in wireless communication information processing. Establishing a high-performance automatic modulation recognition model can improve the classification accuracy of signals. However, many signal recognition studies are based on a single network, resulting in many signals being unable to be effectively distinguished, and the application of multi-domain fusion processing and joint modulation recognition methods is relatively rare.

[0004] Traditional AMR methods are mainly divided into maximum likelihood-based AMR and feature extraction-based AMR. Traditional methods not only require a lot of labor costs when extracting signal features, but also have unsatisfactory signal classification accuracy. Neural network-based methods can automatically extract features from a large number of signal samples, solving the shortcomings of traditional methods. Among them, there is a spatiotemporal multi-channel learning framework (MCLDNN) for automatic modulation recognition. This network can extract features more effectively from the perspective of time and space. Under high signal-to-noise ratio, the network can well distinguish QAM16 and I QAM64, but more than half of WBFM signals are still mistakenly identified as AM-DSB, which limits the overall recognition accuracy of the network.

[0005] Although the above existing studies have made some progress in automatic modulation recognition, they still have some problems:

[0006] First, current electromagnetic signal processing usually only uses a single type of signal representation, ignoring other potentially important feature information and the complementarity between features in different domains, which makes it easy for modulation types with similar single features to be confused; second, with the rapid development of wireless communication technology, modulation methods have become more complex and diversified, and factors such as noise will also affect the accuracy of modulation classification. Enhancing the classification robustness of the model in low signal-to-noise ratio communication scenarios has also become one of the problems to be solved.

[0007] In order to solve the above problems, the present invention proposes a dual-stream modulation signal classification method and system based on multi-feature fusion, which is based on the Convolutional Attention Bidirectional Long Short-Term Memory Deep Network (CABLD) and builds a dual-stream structure. AM-DSB (double-sideband amplitude modulation) and WBFM (wideband frequency modulation) signals are synthesized into a type of AW, and the other modulation signal types remain unchanged. First, multiple types of modulation signals are classified, and the frequency domain amplitude and phase features of the classified AW signals are extracted and then binary classified, which effectively improves the modulation classification accuracy in a low signal-to-noise ratio environment.

[0008] According to some embodiments, the present disclosure adopts the following technical solutions:

[0009] The dual-stream modulation signal classification method based on multi-feature fusion includes:

[0010] Acquire I / Q data of an original modulated signal, and divide the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream;

[0011] The high SNR stream and the low SNR stream are input into the dual-stream CABLD model for multi-signal classification and binary classification, and finally the binary classification results of AM-DSB and WBFM signals are obtained;

[0012] Among them, the dual-stream CABLD model includes two CABLD classification channels, each CABLD classification channel includes a first CABLD model and a second CABLD model. After the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively. First, multi-signal classification is performed by the first CABLD model to obtain an AW-type signal synthesized by AM-DSB and WBFM signals, and then the frequency domain amplitude and phase characteristics of the AW-type signal are extracted respectively and input into the second CABLD model, and the AM-DSB and WBFM signals in the AW-type signal are classified into two categories to obtain the classification results of the AM-DSB and WBFM signals.

[0013] According to some embodiments, the present disclosure adopts the following technical solutions:

[0014] Dual-stream modulation signal classification system based on multi-feature fusion, including:

[0015] A signal acquisition module, used to acquire I / Q data of the original modulated signal, and divide the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream;

[0016] The classification module is used to input the high signal-to-noise ratio stream and the low signal-to-noise ratio stream into the dual-stream CABLD model to perform multi-signal classification and binary classification, and finally obtain the binary classification results of AM-DSB and WBFM signals;

[0017] Among them, the dual-stream CABLD model includes two CABLD classification channels, each CABLD classification channel includes a first CABLD model and a second CABLD model. After the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively. First, multi-signal classification is performed by the first CABLD model to obtain an AW-type signal synthesized by AM-DSB and WBFM signals, and then the frequency domain amplitude and phase characteristics of the AW-type signal are extracted respectively and input into the second CABLD model, and the AM-DSB and WBFM signals in the AW-type signal are classified into two categories to obtain the classification results of the AM-DSB and WBFM signals.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions:

[0019] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the dual-stream modulated signal classification method based on multi-feature fusion is implemented.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device implements the dual-stream modulated signal classification method based on multi-feature fusion.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The disclosed dual-stream modulated signal classification method based on multi-feature fusion constructs a convolutional attention bidirectional long short-term memory deep network, referred to as a CABLD model, and builds a dual-stream architecture based on the CABLD model. Each stream is composed of two CABLD models. The first CABLD model is responsible for multi-classification of modulated signals, and the second CABLD model is responsible for reclassification of AM-DSB and WBFM signals, which solves the problems of difficulty in distinguishing AM-DSB and WBFM signals and low classification accuracy of low signal-to-noise ratio signals, and improves the accuracy of signal components.

[0024] The dual-stream modulation signal classification method based on multi-feature fusion disclosed in the present invention inputs the modulation signal into the signal-to-noise ratio classifier for classification, inputs the I / Q data of the classified high signal-to-noise ratio signal into the high signal-to-noise ratio stream, extracts the power spectrum data of the classified low signal-to-noise ratio signal into the low signal-to-noise ratio stream, synthesizes the AM-DSB and WBFM signals into a type of AW, and keeps the other modulation signal types unchanged. The multi-type modulation signals are classified through the first CABLD model, and the frequency domain amplitude and phase features of the classified AW signals are extracted and input into the second CABLD model for binary classification of AM-DSB and WBFM. The similarities and differences between the time domain data and the frequency domain data of different modulation signals are fully considered, and the serious confusion problem of signal classification of AM-DSB, WBFM and other signals with significant time domain similarity and significant frequency domain difference can be effectively solved, and the modulation classification accuracy in a low signal-to-noise ratio environment can also be effectively improved.

[0025] The disclosed dual-stream modulated signal classification method based on multi-feature fusion constructs a convolutional attention bidirectional long short-term memory deep network, referred to as a CABLD model, which includes a convolutional neural network (CNN) module, an efficient channel attention (ECA) module, two bidirectional CUDA deep neural network long short-term memory (BiCuDNNLSTM) modules and a deep neural network (DNN) module. The CNN module fully extracts the features of the data through the combination of convolutional layers and pooling layers, and performs effective spatial hierarchical representation; the ECA module realizes the capture of the dependency relationship between channels through global average pooling and one-dimensional convolution kernels, thereby improving the model performance; the BiCuDNNLSTM module captures the forward and backward dependencies in the sequence data through forward and backward learning, thereby improving the ability to understand the sequence data; the DNN module can learn the complex patterns of the data and prevent the model from overfitting through the connection of the fully connected layer and the Dropout layer, and finally realize the classification of the signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.

[0027] Figure 1 is a training flow chart of an embodiment of the present disclosure;

[0028] Figure 2 A schematic diagram of the structure of a convolutional attention bidirectional long short-term memory deep network according to an embodiment of the present disclosure;

[0029] Figure 3 Schematic diagram of the structure of an efficient channel attention module according to an embodiment of the present disclosure;

[0030] Figure 4It is a schematic diagram of the structure of two bidirectional CUDA deep neural network long short-term memory models in an embodiment of the present disclosure;

[0031] Figure 5 It is a structural schematic diagram of a multi-feature fusion dual-stream architecture based on a convolutional attention bidirectional long short-term memory deep network according to an embodiment of the present disclosure;

[0032] Figure 6 A schematic diagram of the network structure of a signal-to-noise ratio classifier according to an embodiment of the present disclosure;

[0033] Figure 7 A schematic diagram of a confusion matrix of a dual-stream CABLD network at -12dB in an embodiment of the present disclosure;

[0034] Figure 8 A schematic diagram of the confusion matrix of the MCLDNN network at -12dB in an embodiment of the present disclosure;

[0035] Fig. 9 A schematic diagram of a confusion matrix of a dual-stream CABLD network at 12 dB in an embodiment of the present disclosure;

[0036] Fig.10 This is a schematic diagram of the confusion matrix of the MCLDNN network at 12dB according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0037] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0038] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0039] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0040] Example 1

[0041] In one embodiment of the present disclosure, a dual-stream modulated signal classification method based on multi-feature fusion is provided, including:

[0042] Step 1: Obtain I / Q data of the original modulated signal, and divide the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream;

[0043] Step 2: Input the high SNR stream and the low SNR stream into the dual-stream CABLD model to perform multi-signal classification and binary classification, and finally obtain the binary classification results of AM-DSB and WBFM signals;

[0044] Among them, the dual-stream CABLD model includes two CABLD classification channels, each CABLD classification channel includes a first CABLD model and a second CABLD model. After the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively. First, multi-signal classification is performed by the first CABLD model to obtain an AW-type signal synthesized by AM-DSB and WBFM signals, and then the frequency domain amplitude and phase characteristics of the AW-type signal are extracted respectively and input into the second CABLD model, and the AM-DSB and WBFM signals in the AW-type signal are classified into two categories to obtain the classification results of the AM-DSB and WBFM signals.

[0045] As an embodiment, the present disclosure solves the problem that AM-DSB and WBFM signals are difficult to distinguish and the classification accuracy of low signal-to-noise ratio signals is not high, and constructs a dual-stream CABLD model, that is, a dual-stream CABLD model architecture is built based on a convolutional attention bidirectional long short-term memory deep network (referred to as a CABLD model). A dual-stream architecture is built on the basis of the CABLD model, and each stream is composed of two CABLD models. The first CABLD model is responsible for multi-classification of modulated signals, and the second CABLD model is responsible for reclassification of AM-DSB and WBFM signals. The specific training process of the dual-stream CABLD model is as follows:

[0046] Step 1: Obtain the I / Q data of the modulated signal in the existing data set and divide the data set into a training set, a validation set, and a test set;

[0047] Specifically, the original I / Q data of the modulated signal in the existing data set is obtained, and is divided into a training set, a validation set, and a test set in a ratio of 6:2:2.

[0048] Step 2: Construct a two-stream CABLD model based on the convolutional attention bidirectional long short-term memory deep network (CABLD model for short);

[0049] Among them, Figure 5 As shown, the dual-stream CABLD model includes two CABLD classification channels, each of which is composed of two CABLD models. Each CABLD classification channel includes a first CABLD model and a second CABLD model. The first CABLD model is responsible for multi-classification of modulated signals, and the second CABLD model is responsible for binary classification of AM-DSB and WBFM signals.

[0050] Furthermore, the structures of the first CABLD model and the second CABLD model are the same, such as Figure 2 As shown, each CABLD model includes a CNN module, an efficient channel attention module, two bidirectional CUDA deep neural network long short-term memory modules, and a DNN module.

[0051] Among them, the CNN module includes five convolution layers, and the input data is two-dimensional data. First, the original two-dimensional data is split into two one-dimensional data, and the two one-dimensional data are convolved separately. Then, the two convolved data are spliced ​​into a two-dimensional data, and then a two-dimensional convolution is performed again. A maximum pooling layer is added after the convolution layer to highlight the features, and a Dropout layer is used to prevent overfitting; at the same time, the original two-dimensional input data is two-dimensionally convolved, and also passes through the maximum pooling layer and the Dropout layer; the two-dimensional data after the two Dropout layers are spliced ​​into a new two-dimensional data, and then a layer of two-dimensional convolution is performed to obtain the output of the CNN part.

[0052] The Efficiency Channel Attention (ECA) module uses the output of the CNN part as the input of the ECA part. The structure of ECA is as follows: Figure 3 As shown in the figure, it mainly includes three parts: Global Average Pooling (GAP), one-dimensional convolution layer and activation function layer. GAP obtains the global context information of each channel by averaging the input features, the one-dimensional convolution layer is used for cross-channel interaction, the activation function is used to obtain the attention weight between channels, and finally, the attention weight is applied to the original input features; ECA promotes cross-channel interaction through a weighted method. It uses a learnable channel weight matrix to learn the attention between channels, and then applies it to the channels of the output features, thereby improving the performance of the CNN convolution layer.

[0053] Two bidirectional CUDA deep neural network long short-term memory models BiCuDNNLSTM. LSTM can learn data for a long time, overcoming the long-term dependency problem. Bidirectional LSTM is a special architecture that enhances LSTM performance by running input sequences in both directions to make judgments to extract meaningful features. CuDNNLSTM is an efficient LSTM implementation of CuDNN, a GPU-accelerated deep neural network primitive library that provides robust and simple multi-threading for LSTM networks.

[0054] BiCuDNNLSTM is an effective technique for identifying short-term and long-term dependencies in input time series datasets. The BiCuDNNLSTM module takes the output of the ECA module as the input of the BiCuDNNLSTM part. Figure 4As shown, the forward CuDNNLSTM layer accepts the time series data sequence input in the forward direction (from the past to the present), that is, the feature sequence output from the ECA module is learned in the order from the beginning to the end, and the reverse CuDNNLSTM layer accepts the time series data sequence input in the reverse direction (from the present to the past), that is, the feature sequence output from the ECA module is learned in the order from the end to the beginning. That is, for the feature sequence output by the ECA module, BiCuDNNLSTM learns in two ways, from the beginning to the end and from the end to the beginning. For each data, the relationship with the previous and next data in both directions is fully explored and combined, so as to learn deeper feature information. Compared with the unidirectional method, it can increase the amount of available information provided to the neural network so as to extract valuable content from the information of time dependency and memory blocks. It can also perform additional training by traversing the input time series data in both directions, thereby improving the accuracy of the prediction model by extracting further important patterns from the same input data set.

[0055] The DNN module consists of two Dense (fully connected) layers and two Droupout layers overlapping each other. The DNN module uses the output of the BiCuDNNLSTM module as the input of the DNN part. The Dense layer converts the input data into a new feature space, and continuously adjusts the weights and biases of the model during training to minimize the loss function; the role of the Droupout layer is to randomly discard (i.e. set to zero) the output of a part of the neurons in the network during training to prevent the model from overfitting and force the network to learn a more generalized feature representation.

[0056] A Dense layer is connected after the DNN module to map the output features of the DNN module to the feature space of the category, thus completing the classification of the modulation type.

[0057] Step 3: Input the training set into the SNR classifier, input the I / Q data of the classified high SNR signal into the high SNR stream, extract the power spectrum data of the classified low SNR signal and input it into the low SNR stream;

[0058] Step 4: AM-DSB and WBFM signals are synthesized into one type of AW, and the other modulation signal types remain unchanged. The first CABLD model is used to classify the multi-type modulation signals. The frequency domain amplitude and phase features of the classified AW signals are extracted and input into the second CABLD model for binary classification of AM-DSB and WBFM.

[0059] Step 5: Input the validation set data to adjust the parameters of the model;

[0060] Step 6: Input the test set data for testing. If it is classified as not an AW signal after the first CABLD model, stop early. If it is classified as an AW signal, enter the second CABLD signal for binary classification of AM-DSB and WBFM.

[0061] As an embodiment, the specific implementation process of the dual-stream modulation signal classification method based on multi-feature fusion disclosed in the present invention is as follows:

[0062] Step S1: obtaining I / Q data of an original modulated signal, and dividing the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream;

[0063] Specifically, the data modulation method is I / Q modulation. The I / Q vector consists of two data vectors, an in-phase component and an orthogonal component. The original time features are extracted from the in-phase component and the orthogonal component, which is expressed by formula (1):

[0064] (1)

[0065] in, is the I / Q vector of the data, is the original data, is the in-phase component of the data, is the orthogonal component of the data, where n is .

[0066] Furthermore, the I / Q data is divided into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream, including:

[0067] The I / Q data is input into the signal-to-noise ratio classifier, such as Figure 6 As shown, the SNR classifier consists of an input layer, a convolutional neural network layer, a fully connected layer, and an output layer. The signal samples are trained with the SNR as the output label. In this example, the signal with a SNR greater than -10dB is classified as a high SNR signal, otherwise it is a SNR signal; then the I / Q data of the high SNR signal is divided into a high SNR stream; the power spectrum data of the low SNR signal is divided into a low SNR stream.

[0068] The power spectrum data is the energy distribution of the modulated signal in the frequency domain, which is a measure of the change of signal power with frequency. The acquisition process is: first obtain the modulated signal representation, then calculate the autocorrelation function of the modulated signal, and then perform Fourier transform on the autocorrelation function to obtain the power spectrum density of the signal. Specifically, assuming that the modulated signal can be expressed as,

[0069] (2)

[0070] in, and are the in-phase and quadrature components of the signal respectively.

[0071] Calculate the autocorrelation function of the modulated signal ,

[0072] (3)

[0073] in, n is the sample index, is the time delay, represents the conjugate of the signal.

[0074] Then the autocorrelation function is Fourier transformed to obtain the power spectral density of the signal ,

[0075] (4)

[0076] in, Represents a Fourier transform operation.

[0077] Step S2: Input the high SNR stream and the low SNR stream into the dual-stream CABLD model to perform multi-signal classification and binary classification, and finally obtain the binary classification results of AM-DSB and WBFM signals.

[0078] Specifically, the dual-stream CABLD model includes two CABLD classification channels, each of which includes a first CABLD model and a second CABLD model. After the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively. First, multi-signal classification is performed by the first CABLD model to obtain an AW-type signal synthesized from AM-DSB and WBFM signals. Then, the frequency domain amplitude and phase characteristics of the AW-type signal are extracted respectively and input into the second CABLD model. The AM-DSB and WBFM signals in the AW-type signal are classified into two categories to obtain the classification results of the AM-DSB and WBFM signals.

[0079] Multi-signal classification includes: first inputting a two-dimensional modulated signal into the CNN module, splitting the two-dimensional modulated signal into two one-dimensional data, convolving the two one-dimensional data separately, and then splicing the two convolved data into one two-dimensional data, and then performing a two-dimensional convolution again, and adding a maximum pooling layer after the convolution layer, and using a Dropout layer to prevent overfitting; at the same time, the original two-dimensional input two-dimensional modulated signal is two-dimensionally convolved, and also passes through the maximum pooling layer and the Dropout layer; the two-dimensional data after the two Dropout layers are spliced ​​into a new two-dimensional data, and then a layer of two-dimensional convolution is performed to obtain the output of the CNN module.

[0080] Furthermore, the output of the CNN module is used as the input of the efficient channel attention module. ECA promotes cross-channel interaction through a weighted method, which adopts a learnable channel weight matrix To learn the attention between channels, and then apply it to the channels of the output features, thereby improving the performance of the CNN convolutional layer, as expressed by formula (2),

[0081] (5)

[0082] in, is the output of CNN convolution, is the feature output after ECA weighting, is the channel weight matrix.

[0083] Global average pooling obtains the global context information of each channel by averaging the input features. The one-dimensional convolution layer is used for cross-channel interaction, and the activation function is used to obtain the attention weights between channels. Finally, the attention weights are applied to the original input features to obtain the attention weights between channels. The process of obtaining the attention weights between channels can be described as:

[0084] (6)

[0085] in, represents the GAP process, Represents a one-dimensional convolution process.

[0086] Furthermore, through two bidirectional CUDA deep neural network long short-term memory models BiCuDNNLSTM, such as Figure 4 As shown, the forward CuDNNLSTM layer accepts a forward (from the past to the present) input time series data sequence, and the reverse CuDNNLSTM layer accepts a reverse input time series data sequence (from the present back to the past). Compared with the unidirectional method, it can increase the amount of available information provided to the neural network to extract valuable content from the information of time dependencies and memory blocks. It can also perform additional training by traversing the input time series data in both directions, thereby improving the accuracy of the prediction model by extracting further important patterns from the same input data set.

[0087] Finally, a Dense layer is connected after the DNN module to map the output features of the DNN module to the feature space of the category, thus completing the classification of the modulation type.

[0088] The first CABLD model after the above process is responsible for multiple classification of the modulated signal, which is ten categories in this example, including AM-SSB, AW, BPSK, QPSK, 8PSK, 16QAM, 64QAM, CPFSK, GFSK, and PAM4, among which AW is a synthesis of AM-DSB and WBFM signals. If it is classified as not an AW signal after the first CABLD model, it stops early. If it is classified as an AW signal, it enters the second CABLD signal for binary classification of AM-DSB and WBFM.

[0089] Furthermore, if it is classified as an AW signal, it enters the second CABLD signal for binary classification of AM-DSB and WBFM, that is, the frequency domain amplitude and phase characteristics of the AW signal are extracted respectively and input into the second CABLD model, and the AM-DSB and WBFM signals in the AW signal are binary classified to obtain the classification results of the AM-DSB and WBFM signals.

[0090] The frequency domain amplitude and phase characteristics of the modulation signal are as follows:

[0091] Assume that the modulated signal is expressed as,

[0092] (7)

[0093] in, is the I / Q vector of the data, is the original data, is the in-phase component of the data, is the orthogonal component of the data, where n is .

[0094] After Fourier transform, the frequency domain amplitude and phase information of the signal are obtained:

[0095] (8)

[0096] (9)

[0097] in, , In-phase components The amplitude and phase after Fourier transformation, , They are orthogonal components Amplitude and phase after Fourier transformation.

[0098] Therefore, the new input data format is as follows:

[0099] (10)

[0100] Among them, hstack means horizontal splicing, vstack means vertical splicing, express The i-th row of data in express The i-th row of data in .

[0101] Will The second CABLD model is used as input to implement binary classification of AM-DSB and WBFM.

[0102] Simulation experiment

[0103] 1) Experimental conditions:

[0104] The software and hardware environment of the experiment is shown in Table 1;

[0105] The detailed parameters of the dataset used are shown in Table 2;

[0106] Table 1: Experimental hardware and software environment

[0107]

[0108] Table 2 Detailed parameters of the dataset used

[0109]

[0110] 2) Experimental content:

[0111] Use the dataset given in the experimental conditions, use categorical cross entropy as the loss function, and use the Adam algorithm as the optimizer. Set the number of iterations to 200, the number of batch samples to 400, and the initial learning rate to 0.001. If the validation loss does not decrease within 5 epochs, the learning rate will be halved. If the validation loss remains stable within 50 epochs, the training process stops.

[0112] The experimental results named the method of the present invention DCABLD, and the confusion matrix of the present invention at -12dB is as follows: Figure 7 As shown, the confusion matrix of the MCLDNN network at -12dB is as follows Figure 8 As shown, at -12dB, the accuracy of the present disclosure is 27.91%, and the accuracy of the MCLDNN network is 17.91%, which is ten percentage points higher than the previous one. This shows that the classification effect of the present disclosure on low signal-to-noise ratio signals is greatly improved.

[0113] The confusion matrix of the present disclosure at -12dB is as follows Figure 7 As shown in Figure 2, the classification accuracy of AM-DSB and WBFM is 29.5%; the confusion matrix of the MCLDNN network at -12dB is shown in Figure 2. Figure 8As shown, the classification accuracy of AM-DSB and WBFM is 20%, which is 9.5 percentage points higher than that of the previous one. The confusion matrix of the present disclosure at 12 dB is shown in Fig. 9 As shown, the classification accuracy of AM-DSB and WBFM is 85%; the confusion matrix of the MCLDNN network at 12dB is shown in Fig.10 As shown in the figure, the classification accuracy of AM-DSB and WBFM is 73.5%, which is 11.5 percentage points higher than that of WBFM. From the classification accuracy of the two categories under high and low signal-to-noise ratios, it can be seen that the confusion problem of AM-DSB and WBFM has been solved. The above experimental analysis proves the correctness and effectiveness of the method proposed in the present disclosure.

[0114] Example 2

[0115] In one embodiment of the present disclosure, a dual-stream modulation signal classification system based on multi-feature fusion is provided, including:

[0116] A signal acquisition module, used to acquire I / Q data of the original modulated signal, and divide the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream;

[0117] The classification module is used to input the high signal-to-noise ratio stream and the low signal-to-noise ratio stream into the dual-stream CABLD model to perform multi-signal classification and binary classification, and finally obtain the binary classification results of AM-DSB and WBFM signals;

[0118] Among them, the dual-stream CABLD model includes two CABLD classification channels, each CABLD classification channel includes a first CABLD model and a second CABLD model. After the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively. First, multi-signal classification is performed by the first CABLD model to obtain an AW-type signal synthesized by AM-DSB and WBFM signals, and then the frequency domain amplitude and phase characteristics of the AW-type signal are extracted respectively and input into the second CABLD model, and the AM-DSB and WBFM signals in the AW-type signal are classified into two categories to obtain the classification results of the AM-DSB and WBFM signals.

[0119] Example 3

[0120] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the dual-stream modulated signal classification method based on multi-feature fusion is implemented.

[0121] Example 4

[0122] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the dual-stream modulated signal classification method based on multi-feature fusion.

[0123] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, 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 generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0124] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0125] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A dual-stream modulation signal classification method based on multi-feature fusion, characterized in that: include: Acquire I / Q data of an original modulated signal, and divide the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream; The high SNR stream and the low SNR stream are input into the dual-stream CABLD model for multi-signal classification and binary classification, and finally the binary classification results of AM-DSB and WBFM signals are obtained; The dual-stream CABLD model includes two CABLD classification channels, each of which includes a first CABLD model and a second CABLD model. After the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively, and firstly perform multi-signal classification through the first CABLD model to obtain an AW-type signal synthesized from AM-DSB and WBFM signals, and then extract the frequency domain amplitude and phase characteristics of the AW-type signal respectively and input them into the second CABLD model, and perform binary classification on the AM-DSB and WBFM signals in the AW-type signal to obtain the classification results of the AM-DSB and WBFM signals; The dual-stream CABLD model includes two CABLD classification channels, each of which includes a first CABLD model and a second CABLD model. Each CABLD model includes a CNN module, an efficient channel attention module, two bidirectional CUDA deep neural network long short-term memory modules and a DNN module. First, a two-dimensional modulated signal is input into the CNN module, the two-dimensional modulated signal is split into two one-dimensional data, the two one-dimensional data are convolved separately, and then the two convolved data are spliced ​​into a two-dimensional data, and then a two-dimensional convolution is performed again. A maximum pooling layer is added after the convolution layer, and a Dropout layer is used to prevent overfitting; at the same time, the original two-dimensional input two-dimensional modulated signal is two-dimensionally convolved, and also passes through the maximum pooling layer and the Dropout layer; the two-dimensional data after the two Dropout layers are spliced ​​into a new two-dimensional data, and then a layer of two-dimensional convolution is performed to obtain the output of the CNN module.

2. The dual-stream modulation signal classification method based on multi-feature fusion according to claim 1 is characterized in that: The method of obtaining the I / Q data of the original modulated signal includes: the data modulation mode is I / Q modulation, the I / Q vector is composed of two data vectors of an in-phase component and an orthogonal component, and the original time feature is extracted from the in-phase component and the orthogonal component, which is expressed as: (1); in, is the I / Q vector of I / Q data, is the original data, is the in-phase component of the data, is the orthogonal component of the data, where n is .

3. The dual-stream modulation signal classification method based on multi-feature fusion according to claim 1 is characterized in that: Dividing the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream, comprising: The I / Q data is input into a signal-to-noise ratio classifier, which consists of an input layer, a convolutional neural network layer, a fully connected layer and an output layer. The signal-to-noise ratio is used as the output label, and signals with a signal-to-noise ratio greater than -10dB are classified as high signal-to-noise ratio signals, otherwise they are signal-to-noise ratio signals; then the I / Q data of the high signal-to-noise ratio signal is classified into a high signal-to-noise ratio stream; and the power spectrum data of the low signal-to-noise ratio signal is classified into a low signal-to-noise ratio stream.

4. The dual-stream modulation signal classification method based on multi-feature fusion according to claim 3 is characterized in that: The power spectrum data is the energy distribution of the modulated signal in the frequency domain, and is a measure of how the signal power changes with frequency. The acquisition process is: first, the modulated signal representation is acquired, then the autocorrelation function of the modulated signal is calculated, and then the autocorrelation function is Fourier transformed to obtain the power spectrum density of the signal.

5. The dual-stream modulation signal classification method based on multi-feature fusion according to claim 1 is characterized in that: The efficient channel attention module includes global average pooling, a one-dimensional convolution layer and an activation function layer. The output of the CNN module is used as the input of the efficient channel attention module. Global average pooling obtains the global context information of each channel by averaging the input features. The one-dimensional convolution layer is used for cross-channel interaction. The activation function is used to obtain the attention weight between channels. Finally, the attention weight is applied to the original input features to obtain the attention weight between channels.

6. The dual-stream modulation signal classification method based on multi-feature fusion according to claim 1 is characterized in that: In each CABLD classification channel, the first CABLD model performs multi-classification of modulated signals. The multi-type signals include AM-SSB, AW, BPSK, QPSK, 8PSK, 16QAM, 64QAM, CPFSK, GFSK, and PAM4, among which AW is a synthesis of AM-DSB and WBFM signals. The second CABLD model performs binary classification of AM-DSB and WBFM signals. If the signal is not classified as an AW signal after the first CABLD model, the signal stops early. If the signal is classified as an AW signal after the first CABLD model, the signal enters the second CABLD model for binary classification of AM-DSB and WBFM.

7. A dual-stream modulation signal classification system based on multi-feature fusion, characterized in that: include: A signal acquisition module, used to acquire I / Q data of the original modulated signal, and divide the I / Q data into a high signal-to-noise ratio stream and a low signal-to-noise ratio stream; The classification module is used to input the high signal-to-noise ratio stream and the low signal-to-noise ratio stream into the dual-stream CABLD model to perform multi-signal classification and binary classification, and finally obtain the binary classification results of AM-DSB and WBFM signals; The dual-stream CABLD model includes two CABLD classification channels, each of which includes a first CABLD model and a second CABLD model. After the high signal-to-noise ratio stream and the low signal-to-noise ratio stream are input into the dual-stream CABLD model, they enter a CABLD classification channel respectively, and firstly perform multi-signal classification through the first CABLD model to obtain an AW-type signal synthesized from AM-DSB and WBFM signals, and then extract the frequency domain amplitude and phase characteristics of the AW-type signal respectively and input them into the second CABLD model, and perform binary classification on the AM-DSB and WBFM signals in the AW-type signal to obtain the classification results of the AM-DSB and WBFM signals; The dual-stream CABLD model includes two CABLD classification channels, each of which includes a first CABLD model and a second CABLD model. Each CABLD model includes a CNN module, an efficient channel attention module, two bidirectional CUDA deep neural network long short-term memory modules and a DNN module. First, a two-dimensional modulated signal is input into the CNN module, the two-dimensional modulated signal is split into two one-dimensional data, the two one-dimensional data are convolved separately, and then the two convolved data are spliced ​​into a two-dimensional data, and then a two-dimensional convolution is performed again. A maximum pooling layer is added after the convolution layer, and a Dropout layer is used to prevent overfitting; at the same time, the original two-dimensional input two-dimensional modulated signal is two-dimensionally convolved, and also passes through the maximum pooling layer and the Dropout layer; the two-dimensional data after the two Dropout layers are spliced ​​into a new two-dimensional data, and then a layer of two-dimensional convolution is performed to obtain the output of the CNN module.

8. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the dual-stream modulated signal classification method based on multi-feature fusion as described in any one of claims 1-6 is implemented.

9. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the dual-stream modulated signal classification method based on multi-feature fusion as described in any one of claims 1-6.

Images