A signal modulation recognition algorithm with attention mechanism temporal convolutional network

Abstract

An embodiment of the present invention discloses a signal modulation recognition algorithm with a temporal convolutional network using an attention mechanism. The embodiment of the present invention obtains modulated signal data to be recognized; samples and processes the modulated signal data to determine a modulated signal sequence to be recognized; and inputs the modulated signal sequence to be recognized into a pre-trained temporal convolutional network model using an attention mechanism to determine the target modulation mode of the modulated signal sequence to be recognized. By using the above method, the target modulation mode is determined using a temporal convolutional network model using an attention mechanism, which can improve the accuracy of modulation mode recognition, reduce redundant computations, and shorten operation time.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a signal modulation recognition algorithm with a temporal convolutional network having an attention mechanism. Background Art

[0002] With the development of wireless communication technology, communication signals can be modulated using different modulation methods. In some scenarios, after receiving a modulated communication signal, it is necessary to determine its modulation method.

[0003] In the prior art, the modulation mode of the modulated communication signal is determined by a method based on maximum likelihood ratio theory, a method based on feature extraction, and a method based on deep learning. Although the above three methods can all identify the modulation method of the modulated signal, they all have certain problems. For example, the method based on maximum likelihood ratio theory requires prior knowledge of the modulated communication signal, and the establishment of the function is highly professional and computationally complex. The selection of features and classifiers in the feature extraction method fundamentally determines the classification effect. In a complex electromagnetic environment, the communication environment changes diversely, and the selection of fixed features is not applicable to all scenarios and has certain limitations. The method of converting the modulated communication signal into an image processing method based on deep learning does not take into account the features carried by the signal itself and the calculation is complex. Among the methods for directly performing end-to-end recognition on the modulated communication signal, the main ones are based on convolutional neural networks (CNN) and long short-term memory networks (LSTM). The CNN or LSTM does not fully extract the features of the modulated communication signal sequence, which will result in a low recognition rate. Although the composite network combining CNN and LSTM can extract time-frequency domain features, the computational complexity is relatively large.

[0004] In summary, how to improve the accuracy of modulation mode recognition, reduce redundant calculations and shorten the operation time are the problems that need to be solved at present. Summary of the Invention

[0005] In view of this, an embodiment of the present invention provides a signal modulation recognition algorithm with an attention mechanism temporal convolutional network, which can improve the accuracy of modulation mode recognition, reduce redundant calculations and shorten operation time.

[0006] In a first aspect, an embodiment of the present invention provides a method for identifying a signal modulation mode, the method comprising:

[0007] Acquiring modulation signal data to be identified;

[0008] Sampling the modulation signal data to be identified to determine a modulation signal sequence to be identified;

[0009] The modulation signal sequence to be identified is input into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the target modulation mode of the modulation signal sequence to be identified.

[0010] Optionally, inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine a target modulation mode of the modulation signal sequence to be identified specifically includes:

[0011] Inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value among the probability values ​​corresponding to at least one modulation mode;

[0012] The modulation mode corresponding to the maximum value is determined as the target modulation mode.

[0013] Optionally, the temporal convolutional network model of the attention mechanism includes an input layer, a batch normalization layer, a temporal convolutional network layer, an attention mechanism layer, a fully connected layer and an output layer.

[0014] Optionally, inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value among probability values ​​corresponding to at least one modulation mode specifically includes:

[0015] Inputting the modulation signal sequence to be identified into the batch normalization layer of the temporal convolutional network model combined with the attention mechanism through the input layer, performing normalization processing, and determining first output data;

[0016] Inputting the first output data into a temporal convolutional network layer, extracting shallow layer features, deep layer features, time domain features, and frequency domain features, and determining second output data of the temporal convolutional network layer;

[0017] Inputting the second output data into an attention mechanism layer to determine third output data of the attention mechanism layer;

[0018] Inputting the third output data into a fully connected layer to determine a maximum value among probability values ​​corresponding to at least one classification result;

[0019] The modulation mode corresponding to the maximum value is determined as the target modulation mode and output through the output layer.

[0020] Optionally, the temporal convolutional network layer is composed of a stack of ResBlock layers. For each of the ResBlock layers, the first output data is input into the temporal convolutional network layer, shallow features, deep features, time domain features, and frequency domain features are extracted, and the second output data of the temporal convolutional network layer is determined, specifically including:

[0021] Extract shallow features through the first convolutional layer, where the convolution kernel size of the first convolutional layer is 1;

[0022] Extract deep features, time domain features, and frequency domain features through the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, and the fifth convolutional layer, wherein the convolution kernel size of the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, and the fifth convolutional layer is 3, and the corresponding expansion coefficients are 1, 2, 4, and 8 respectively;

[0023] The shallow features extracted by the first convolutional layer and the deep features, time domain features and frequency domain features extracted by the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer are added to determine the second output data of the temporal convolutional network layer.

[0024] Optionally, inputting the second output data into an attention mechanism layer to determine third output data of the attention mechanism layer specifically includes:

[0025] Inputting the second output data into an attention mechanism layer, and determining important features among the shallow features, deep features, time domain features, and frequency domain features through the attention mechanism layer;

[0026] Different weights are assigned to the important features, and weighted summation is performed to determine the third output data of the attention mechanism layer.

[0027] Optionally, the training process of the temporal convolutional network model combined with the attention mechanism includes:

[0028] Acquire historical modulation signal data and a modulation mode corresponding to the historical modulation signal data;

[0029] Sampling the historical modulation signal data to determine a historical modulation signal sequence;

[0030] The temporal convolutional network model combined with the attention mechanism is trained according to the historical modulation signal sequence and the modulation mode corresponding to the historical modulation signal data.

[0031] In a second aspect, an embodiment of the present invention provides a device for identifying a signal modulation mode, characterized in that the device includes:

[0032] An acquisition unit, configured to acquire the modulated signal data to be identified;

[0033] A first determining unit is configured to sample and process the modulation signal data to be identified to determine a modulation signal sequence to be identified;

[0034] The second determination unit is used to input the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine a target modulation mode of the modulation signal sequence to be identified.

[0035] Optionally, the second determining unit is specifically configured to:

[0036] Inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value among the probability values ​​corresponding to at least one modulation mode;

[0037] The modulation mode corresponding to the maximum value is determined as the target modulation mode.

[0038] Optionally, the temporal convolutional network model of the attention mechanism includes an input layer, a batch normalization layer, a temporal convolutional network layer, an attention mechanism layer, a fully connected layer and an output layer.

[0039] Optionally, the second determining unit is specifically configured to:

[0040] Inputting the modulation signal sequence to be identified into the batch normalization layer of the temporal convolutional network model combined with the attention mechanism through the input layer, performing normalization processing, and determining first output data;

[0041] Inputting the first output data into a temporal convolutional network layer, extracting shallow layer features, deep layer features, time domain features, and frequency domain features, and determining second output data of the temporal convolutional network layer;

[0042] Inputting the second output data into an attention mechanism layer to determine third output data of the attention mechanism layer;

[0043] Inputting the third output data into a fully connected layer to determine a maximum value among probability values ​​corresponding to at least one classification result;

[0044] The modulation mode corresponding to the maximum value is determined as the target modulation mode and output through the output layer.

[0045] Optionally, the temporal convolutional network layer is composed of a stack of ResBlock layers, and for each of the ResBlock layers, the second determining unit is specifically configured to:

[0046] Extract shallow features through the first convolutional layer, where the convolution kernel size of the first convolutional layer is 1;

[0047] Extract deep features, time domain features, and frequency domain features through the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, and the fifth convolutional layer, wherein the convolution kernel size of the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, and the fifth convolutional layer is 3, and the corresponding expansion coefficients are 1, 2, 4, and 8 respectively;

[0048] The shallow features extracted by the first convolutional layer and the deep features, time domain features and frequency domain features extracted by the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer are added to determine the second output data of the temporal convolutional network layer.

[0049] Optionally, the second determining unit is specifically configured to:

[0050] Inputting the second output data into an attention mechanism layer, and determining important features among the shallow features, deep features, time domain features, and frequency domain features through the attention mechanism layer;

[0051] Different weights are assigned to the important features, and weighted summation is performed to determine the third output data of the attention mechanism layer.

[0052] Optionally, the acquisition unit is further configured to: acquire historical modulation signal data and a modulation mode corresponding to the historical modulation signal data;

[0053] The first determining unit is further configured to: perform sampling processing on the historical modulation signal data to determine a historical modulation signal sequence;

[0054] The device also includes a training unit for training the temporal convolutional network model combined with the attention mechanism according to the historical modulation signal sequence and the modulation mode corresponding to the historical modulation signal data.

[0055] In a third aspect, an embodiment of the present invention provides an electronic device comprising a memory and a processor, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement a method as described in the first aspect or any possible embodiment of the first aspect.

[0056] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, the program being executed by a processor to implement a method as described in the first aspect or any possible embodiment of the first aspect.

[0057] This embodiment of the present invention obtains modulation signal data to be identified; samples and processes the modulation signal data to determine a modulation signal sequence to be identified; and then inputs the modulation signal sequence to be identified into a pre-trained temporal convolutional network model incorporating an attention mechanism to determine the target modulation mode of the modulation signal sequence to be identified. This method, which utilizes a temporal convolutional network model incorporating an attention mechanism to determine the target modulation mode, can improve modulation mode recognition accuracy, reduce redundant computations, and shorten computation time. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The above and other objects, features and advantages of the present invention will become more apparent through the following description of the embodiments of the present invention with reference to the accompanying drawings, in which:

[0059] Figure 1 This is a flow chart of a method for identifying a signal modulation mode according to an embodiment of the present invention;

[0060] Figure 2 Schematic diagram of a temporal convolutional network model of an attention mechanism according to an embodiment of the present invention;

[0061] Figure 3 is a flow chart of another method for identifying a signal modulation mode according to an embodiment of the present invention;

[0062] Figure 4 Schematic diagram of a temporal convolutional network layer structure according to an embodiment of the present invention;

[0063] Figure 5 This is a training flow chart of a temporal convolutional network model combined with an attention mechanism according to an embodiment of the present invention;

[0064] Figure 6 It is a schematic diagram of a signal modulation mode identification device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] The present disclosure is described below based on examples, but the present disclosure is not limited to these examples. Certain specific details are described in detail in the following detailed description of the present disclosure. A person skilled in the art can fully understand the present disclosure without these details. To avoid obscuring the essence of the present disclosure, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0066] Furthermore, persons of ordinary skill in the art will appreciate that the figures provided herein are for illustration purposes only and are not necessarily drawn to scale.

[0067] Unless the context clearly requires otherwise, words like “include”, “comprising” and the like throughout this application should be interpreted as including rather than exclusive or exhaustive; that is, as meaning “including but not limited to”.

[0068] In the description of the present disclosure, it should be understood that the terms "first," "second," etc. are used for descriptive purposes only and are not to be understood as indicating or implying relative importance. In addition, in the description of the present disclosure, unless otherwise specified, "plurality" means two or more.

[0069] In the prior art, the modulation mode of the modulated communication signal is determined by a method based on the maximum likelihood ratio theory, a method based on feature extraction, and a method based on deep learning. Specifically, the method based on the maximum likelihood ratio theory is guided by Bayesian theory, and designs a likelihood function through the prior information of the modulated communication signal to obtain multiple likelihood ratio values ​​of the modulated communication signal, and determines that the modulation mode corresponding to the maximum value of the likelihood ratio value is the modulation mode adopted by the modulated communication signal. The method based on the maximum likelihood ratio theory converts the modulation mode classification problem into a multiple composite hypothesis testing problem with maximum likelihood as the criterion for processing, with high recognition accuracy and good reliability. Although the modulation mode can be determined by the above method, there are certain problems. Specifically, the method based on the maximum likelihood ratio theory requires prior knowledge of the modulated communication signal, and the establishment of the function is highly professional and the calculation is complex. The method based on feature extraction is firstly manually designed. Extract the signal features that can distinguish different modulation modes, and then select the appropriate classifier that matches the features to realize the recognition and classification of the modulation modes. For some scenarios, the most matching features and classifiers are selected to realize the recognition of the modulation mode. The selection of features and classifiers based on the feature extraction method fundamentally determines the classification effect. In a complex electromagnetic environment, the communication environment changes variously. The selection of fixed features is not applicable to all scenarios and has certain limitations. The method based on deep learning converts the modulated communication signal into image processing or directly modulated communication signal, and realizes the end-to-end signal modulation mode recognition through the powerful representation ability of neural network. The method of converting the modulated communication signal into image processing in the modulation recognition method based on deep learning does not take into account the features carried by the signal itself and the calculation is complex. The method of directly performing end-to-end recognition on the modulated communication signal is mainly based on Convolutional Neural Network (CNN). Neural Network (CNN) and Long Short-Term Memory (LSTM) networks, however, do not fully extract features from modulated communication signal sequences, resulting in low recognition rates. While a composite network combining CNN and LSTM can extract time-frequency domain features, it requires a high level of computation. Therefore, improving modulation recognition accuracy, reducing redundant computations, and shortening computation time are currently pressing challenges.

[0070] In an embodiment of the present invention, in order to solve the problem of reducing redundant calculations and shortening operation time while improving the accuracy of modulation mode recognition, a method for identifying a signal modulation mode is proposed. Figure 1 As shown, Figure 1 This is a flow chart of a method for identifying a signal modulation mode according to an embodiment of the present invention. Specifically, it includes:

[0071] Step S100: Acquire the modulated signal data to be identified.

[0072] Specifically, the modulation signal to be identified is in-phase (I) / quadrature (Q) signal data, wherein there are 11 possible modulation modes of the modulation signal data to be identified, including 8 digital modulation modes and 3 analog modulation modes. The 8 digital modulation modes are 8PSK, BPSK, CPFSK, GFSK, PAM4, 16QAM, 64QAM and QPSK, and the 3 analog modulation modes are AM-DSB, AM-SSB and WBFM.

[0073] Step S101: Sampling the modulation signal data to be identified to determine a modulation signal sequence to be identified.

[0074] In the embodiment of the present invention, assuming that the sampling point set when sampling the modulated signal data to be identified is N, the length of the modulated signal sequence to be identified is N, and the data format is [2, N], where 2 represents I and Q signal data.

[0075] Step S102: input the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine a target modulation mode of the modulation signal sequence to be identified.

[0076] In the embodiment of the present invention, the temporal convolutional network model of the attention mechanism is as follows Figure 2 As shown, it includes an input layer 201, a batch normalization layer 202, a temporal convolutional network layer 203, an attention mechanism layer 204, a fully connected layer 205 and an output layer 206.

[0077] In one possible implementation, inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine a target modulation mode of the modulation signal sequence to be identified specifically includes:

[0078] The modulation signal sequence to be identified is input into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value of the probability values ​​corresponding to at least one modulation mode; and the modulation mode corresponding to the maximum value is determined as the target modulation mode.

[0079] In the embodiment of the present invention, combined with Figure 2 Each layer in the temporal convolutional network model of the attention mechanism processes the modulation signal sequence to be identified. The specific processing flow is as follows: Figure 3 As shown, the following steps are included:

[0080] Step S300: Input the modulated signal sequence to be identified into the batch normalization (BN) layer of the temporal convolutional network model combined with the attention mechanism through the input layer, perform normalization processing, and determine the first output data.

[0081] Specifically, the data format of the modulation signal sequence to be identified is [2, N], the input batch size is X, that is, X [2, N] modulation signal sequences to be identified, and the batch normalization layer is used to speed up the training and convergence of the network. The expression of the batch normalization layer is as follows: yy

[0082]

[0083] Among them, the x (k) is the modulation signal sequence to be identified input through the input layer, and the y (k) is the first output data after normalization, the β (k) is a learnable translation parameter, the γ (k) is the scaling parameter, the μ (k) is the mean value of the input modulation signal sequence to be identified, the σ (k) is the standard deviation of the input modulation signal sequence to be identified, the superscript k represents the kth dimension of the data, and ε is a value to prevent the denominator from being zero.

[0084] Step S301: input the first output data into a temporal convolutional network layer, extract shallow features, deep features, time domain features, and frequency domain features, and determine the second output data of the temporal convolutional network layer.

[0085] Specifically, shallow features are extracted through the first convolutional layer, wherein the convolution kernel size of the first convolutional layer is 1; deep features, time domain features and frequency domain features are extracted through the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer, wherein the convolution kernel size of the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer is 3, and the corresponding expansion coefficients are 1, 2, 4 and 8 respectively; the shallow features extracted by the first convolutional layer and the deep features, time domain features and frequency domain features extracted by the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer are added to determine the second output data of the temporal convolutional network layer.

[0086] In a possible implementation, the temporal convolutional network layer is composed of a series of stacked ResBlock layers. The structure of each ResBlock layer is as follows: Figure 4As shown, the main branch includes a first convolutional layer 401, a second convolutional layer 402, a third convolutional layer 403, a fourth convolutional layer 404 and a fifth convolutional layer 405, wherein the first convolutional layer 401 includes a convolution with a channel number of 64 and a convolution kernel size of 1, expressed as Conv(64, 1) and an activation function Relu; the second convolutional layer 402 includes a convolution with a channel number of 64, a convolution kernel size of 3, a dilation coefficient of 1, and a filling method of causal (indicating the causality of the time series), expressed as Causal Conv(64, 3, 1) and an activation function Relu; the third convolutional layer 403 includes a convolution with a channel number of 64, a convolution kernel size of 3, a dilation coefficient of 2, and a filling method of causal, expressed as Causal Conv(64, 3, 2) and activation function Relu; the fourth convolutional layer 404 includes a convolution with 64 channels, a convolution kernel size of 3, a dilation coefficient of 4, and a causal filling method, expressed as Causal Conv(64, 3, 4) and an activation function Relu; the fifth convolutional layer 405 includes a convolution with 64 channels, a convolution kernel size of 3, a dilation coefficient of 8, and a causal filling method, expressed as Causal Conv(64, 3, 8) and an activation function Relu; the Figure 4 It also includes an Add module for adding the shallow features and the deep features, time domain features and frequency domain features. Figure 4 It also includes a convolution with a channel number of 64 and a convolution kernel size of 1 on the right branch, which is expressed as Conv(64, 1) to unify the number of channels; Figure 4 It also includes a left branch, which is used to add residual structure and multi-scale fusion features to improve the expressiveness of data information, fuse shallow features with deep features, and avoid overfitting.

[0087] In the implementation of the present invention, by setting the expansion coefficient, convolution can obtain longer dependencies without increasing the amount of calculation, have a flexible receptive field, and extract deep features, time domain features, and frequency domain features.

[0088] Step S302: Input the second output data into the attention mechanism layer to determine the third output data of the attention mechanism layer.

[0089] Specifically, the second output data is input into the attention mechanism layer, and the important features among the shallow features, deep features, time domain features and frequency domain features are determined by the attention mechanism layer; different weights are assigned to the important features, and weighted summation is performed to determine the third output data of the attention mechanism layer.

[0090] In the embodiment of the present invention, the attention mechanism layer is used to determine important features among the shallow features, deep features, time domain features, and frequency domain features, reduce unimportant redundant information, and perform weighted summation on the second output data of each channel input, which is expressed as follows:

[0091] μ i =tanh(Wx i +b)

[0092]

[0093] S i =∑αx i

[0094] Among them, x i is the output of the temporal convolutional network, i.e., the second output data, W and b are shared parameters of the attention mechanism, μ i Represents the nonlinear mapping of data, μ w is the random value for initialization, α is the attention mechanism weight for each data, S i A weighted representation of multiple data using an attention mechanism.

[0095] In an embodiment of the present invention, the third output data of the final output of the attention mechanism layer is a set of one-dimensional data. The length of the one-dimensional data is the same as the number of channels of the two-dimensional data output by the temporal convolutional network. The introduction of the attention mechanism replaces the data flattening operation of the Flatten layer, and the subsequent full connection operation can be performed directly, which greatly reduces the amount of calculation.

[0096] Step S303: Input the third output data into a fully connected layer to determine the maximum value of the probability values ​​corresponding to at least one classification result.

[0097] Specifically, the fully connected layer maps the important feature representations learned by the attention mechanism to the sample label space to achieve classification. The classification result, namely the modulation mode, is assumed to be divided into 11 categories, corresponding to 11 modulation modes. The probability value of each modulation mode is determined respectively, and the maximum probability value is determined.

[0098] Step S304: Determine the modulation mode corresponding to the maximum value as the target modulation mode and output it through the output layer.

[0099] In one possible implementation, the probability values ​​corresponding to the modulation modes 8PSK, BPSK, CPFSK, GFSK, PAM4, 16QAM, 64QAM and QPSK, AM-DSB, AM-SSB and WBFM are 0.6, 0.3, 0.2, 0.4, 0.5, 0.7, 0.8 and 0.9 respectively, and the maximum value is 0.9, that is, the target modulation mode corresponding to the maximum value 0.9 is WBFM.

[0100] In the implementation of the present invention, a temporal convolutional network model combined with an attention mechanism is used to determine the target modulation mode, which can improve the accuracy of modulation mode recognition, reduce redundant calculations, and reduce operation time.

[0101] In one possible implementation, the training process of the temporal convolutional network model combined with the attention mechanism is as follows: Figure 5 The specific steps are as follows:

[0102] Step S500: Acquire historical modulation signal data and a modulation mode corresponding to the historical modulation signal data.

[0103] Step S501: Sampling the historical modulation signal data to determine a historical modulation signal sequence.

[0104] In an embodiment of the present invention, the historical modulation signal sequence can be randomly divided, and the historical modulation signal sequence can be randomly and non-repetitively divided into a training set and a test set in proportion. The training set is used for the subsequent training of the temporal convolutional network model of the attention mechanism, and the test set is used to verify the temporal convolutional network model of the attention mechanism.

[0105] Step S502: training the temporal convolutional network model combined with the attention mechanism according to the historical modulation signal sequence and the modulation mode corresponding to the historical modulation signal data.

[0106] In a possible implementation, after training the temporal convolutional network model combined with the attention mechanism through the above steps, the newly acquired modulation signal data to be identified can be classified to determine the target modulation mode.

[0107] Figure 6 FIG. 1 is a schematic diagram of a signal modulation method identification device according to an embodiment of the present invention. Figure 6 As shown, the apparatus of this embodiment includes an acquiring unit 601 , a first determining unit 602 , and a second determining unit 603 .

[0108] Among them, the acquisition unit 601 is used to obtain the modulated signal data to be identified; the first determination unit 602 is used to sample and process the modulated signal data to be identified to determine the modulated signal sequence to be identified; the second determination unit 603 is used to input the modulated signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the target modulation mode of the modulated signal sequence to be identified.

[0109] Optionally, the second determining unit is specifically configured to:

[0110] Inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value among the probability values ​​corresponding to at least one modulation mode;

[0111] The modulation mode corresponding to the maximum value is determined as the target modulation mode.

[0112] Optionally, the temporal convolutional network model of the attention mechanism includes an input layer, a batch normalization layer, a temporal convolutional network layer, an attention mechanism layer, a fully connected layer and an output layer.

[0113] Optionally, the second determining unit is specifically configured to:

[0114] Inputting the modulation signal sequence to be identified into the batch normalization layer of the temporal convolutional network model combined with the attention mechanism through the input layer, performing normalization processing, and determining first output data;

[0115] Inputting the first output data into a temporal convolutional network layer, extracting shallow layer features, deep layer features, time domain features, and frequency domain features, and determining second output data of the temporal convolutional network layer;

[0116] Inputting the second output data into an attention mechanism layer to determine third output data of the attention mechanism layer;

[0117] Inputting the third output data into a fully connected layer to determine a maximum value among probability values ​​corresponding to at least one classification result;

[0118] The modulation mode corresponding to the maximum value is determined as the target modulation mode and output through the output layer.

[0119] Optionally, the temporal convolutional network layer is composed of a stack of ResBlock layers, and for each of the ResBlock layers, the second determining unit is specifically configured to:

[0120] Extract shallow features through the first convolutional layer, where the convolution kernel size of the first convolutional layer is 1;

[0121] Extract deep features, time domain features, and frequency domain features through the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, and the fifth convolutional layer, wherein the convolution kernel size of the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer, and the fifth convolutional layer is 3, and the corresponding expansion coefficients are 1, 2, 4, and 8 respectively;

[0122] The shallow features extracted by the first convolutional layer and the deep features, time domain features and frequency domain features extracted by the second convolutional layer, the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer are added to determine the second output data of the temporal convolutional network layer.

[0123] Optionally, the second determining unit is specifically configured to:

[0124] Inputting the second output data into an attention mechanism layer, and determining important features among the shallow features, deep features, time domain features, and frequency domain features through the attention mechanism layer;

[0125] Different weights are assigned to the important features, and weighted summation is performed to determine the third output data of the attention mechanism layer.

[0126] Optionally, the acquisition unit is further configured to: acquire historical modulation signal data and a modulation mode corresponding to the historical modulation signal data;

[0127] The first determining unit is further configured to: perform sampling processing on the historical modulation signal data to determine a historical modulation signal sequence;

[0128] The device also includes a training unit for training the temporal convolutional network model combined with the attention mechanism according to the historical modulation signal sequence and the modulation mode corresponding to the historical modulation signal data.

[0129] As will be appreciated by those skilled in the art, various aspects of embodiments of the present invention may be implemented as systems, methods, or computer program products. Thus, various aspects of embodiments of the present invention may take the form of a complete hardware implementation, a complete software implementation (including firmware, resident software, microcode, etc.), or an implementation combining software aspects with hardware aspects, which may all be generally referred to herein as a "circuit," "module," or "system." Additionally, various aspects of embodiments of the present invention may take the form of a computer program product implemented in one or more computer-readable media having computer-readable program code implemented thereon.

[0130] Any combination of one or more computer-readable media can be utilized. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination thereof. More specific examples (non-exhaustive enumeration) of computer-readable storage media will include the following: an electrical connection with one or more wires, a portable computer floppy disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the context of an embodiment of the present invention, a computer-readable storage medium can be any tangible medium that can contain or store a program used by an instruction execution system, device, or apparatus, or a program used in conjunction with an instruction execution system, device, or apparatus.

[0131] A computer-readable signal medium may include a propagated data signal having computer-readable program code embodied therein, such as in baseband or as part of a carrier wave. Such a propagated signal may take any of a variety of forms, including but not limited to electromagnetic, optical, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium that is not a computer-readable storage medium and that can communicate, propagate, or transmit a program for use by or in conjunction with an instruction execution system, device, or apparatus.

[0132] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0133] The computer program code for performing the operations for various aspects of the embodiments of the present invention can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, Smalltalk, C++, etc., and conventional procedural programming languages ​​such as "C" programming language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer as a stand-alone software package; partially on the user's computer and partially on a remote computer; or entirely on a remote computer or server. In the latter case, the remote computer can be connected to the user's computer through any type of network including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0134] The flowchart legends and / or block diagrams of the methods, devices (systems) and computer program products according to embodiments of the present invention described above describe various aspects of embodiments of the present invention. It will be understood that each block of the flowchart legends and / or block diagrams and the combination of blocks in the flowchart legends and / or block diagrams 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 or other programmable data processing device to produce a machine so that the instructions (executed by the processor of the computer or other programmable data processing device) create a device for implementing the function / action specified in the flowchart and / or block diagram block or block.

[0135] These computer program instructions may also be stored in a computer-readable medium that can direct a computer, other programmable data processing device, or other apparatus to operate in a particular manner, so that the instructions stored in the computer-readable medium produce an article of manufacture that includes instructions for implementing the functions / actions specified in the flowchart and / or block diagram blocks or blocks.

[0136] The computer program instructions may also be loaded onto a computer, other programmable data processing device or other apparatus to cause a series of operable steps to be performed on the computer, other programmable device or other apparatus to produce a computer-implemented process, such that the instructions executed on the computer or other programmable device provide a process for implementing the functions / actions specified in the flowchart and / or block diagram block or blocks.

[0137] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A method for identifying a signal modulation mode, characterized in that: The method includes: Acquiring modulation signal data to be identified; Sampling the modulation signal data to be identified to determine a modulation signal sequence to be identified; Inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine a target modulation mode of the modulation signal sequence to be identified; The step of inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine a target modulation mode of the modulation signal sequence to be identified specifically includes: Inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value among the probability values ​​corresponding to at least one modulation mode; Determining a modulation mode corresponding to the maximum value as a target modulation mode; The step of inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value of probability values ​​corresponding to at least one modulation mode includes: Inputting the modulation signal sequence to be identified into the batch normalization layer of the temporal convolutional network model combined with the attention mechanism through the input layer, performing normalization processing, and determining first output data; Inputting the first output data into a temporal convolutional network layer, extracting shallow layer features, deep layer features, time domain features, and frequency domain features, and determining second output data of the temporal convolutional network layer; The temporal convolutional network layer is composed of a stack of ResBlock layers. For each ResBlock layer, the first output data is input into the temporal convolutional network layer, shallow features, deep features, time domain features, and frequency domain features are extracted, and the second output data of the temporal convolutional network layer is determined, specifically including: Extract shallow features through the first convolutional layer, where the convolution kernel size of the first convolutional layer is 1; Deep features, time domain features, and frequency domain features are extracted through the second, third, fourth, and fifth convolutional layers, where the convolution kernel size of the second, third, fourth, and fifth convolutional layers is 3, and the corresponding expansion coefficients are 1, 2, 4, and 8, respectively; Adding the shallow features extracted by the first convolutional layer and the deep features, time domain features, and frequency domain features extracted by the second, third, fourth, and fifth convolutional layers to determine second output data of the temporal convolutional network layer; Among them, the filling method of the second convolution kernel, the third convolution kernel, the fourth convolution kernel and the fifth convolution kernel is causal.

2. The method according to claim 1, wherein The temporal convolutional network model of the attention mechanism includes an input layer, a batch normalization layer, a temporal convolutional network layer, an attention mechanism layer, a fully connected layer and an output layer.

3. The method according to claim 1, wherein Inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value of probability values ​​corresponding to at least one modulation mode specifically includes: Inputting the second output data into an attention mechanism layer to determine third output data of the attention mechanism layer; Inputting the third output data into a fully connected layer to determine a maximum value among probability values ​​corresponding to at least one classification result; The modulation mode corresponding to the maximum value is determined as the target modulation mode and output through the output layer.

4. The method according to claim 3, wherein Inputting the second output data into the attention mechanism layer to determine the third output data of the attention mechanism layer specifically includes: Inputting the second output data into an attention mechanism layer, and determining important features among the shallow features, deep features, time domain features, and frequency domain features through the attention mechanism layer; Different weights are assigned to the important features, and weighted summation is performed to determine the third output data of the attention mechanism layer.

5. The method according to claim 1, wherein The training process of the temporal convolutional network model combined with the attention mechanism includes: Acquire historical modulation signal data and a modulation mode corresponding to the historical modulation signal data; Sampling the historical modulation signal data to determine a historical modulation signal sequence; The temporal convolutional network model combined with the attention mechanism is trained according to the historical modulation signal sequence and the modulation mode corresponding to the historical modulation signal data.

6. A signal modulation mode identification device, characterized in that: The device includes: An acquisition unit, configured to acquire the modulated signal data to be identified; A first determining unit is configured to sample and process the modulation signal data to be identified to determine a modulation signal sequence to be identified; A second determination unit is configured to input the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine a target modulation mode of the modulation signal sequence to be identified; The second determining unit is specifically configured to: Inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value among the probability values ​​corresponding to at least one modulation mode; Determining a modulation mode corresponding to the maximum value as a target modulation mode; The step of inputting the modulation signal sequence to be identified into a pre-trained temporal convolutional network model combined with an attention mechanism to determine the maximum value of probability values ​​corresponding to at least one modulation mode includes: Inputting the modulation signal sequence to be identified into the batch normalization layer of the temporal convolutional network model combined with the attention mechanism through the input layer, performing normalization processing, and determining first output data; Inputting the first output data into a temporal convolutional network layer, extracting shallow layer features, deep layer features, time domain features, and frequency domain features, and determining second output data of the temporal convolutional network layer; The temporal convolutional network layer is composed of a stack of ResBlock layers. For each ResBlock layer, the first output data is input into the temporal convolutional network layer, shallow features, deep features, time domain features, and frequency domain features are extracted, and the second output data of the temporal convolutional network layer is determined, specifically including: Extract shallow features through the first convolutional layer, where the convolution kernel size of the first convolutional layer is 1; Deep features, time domain features, and frequency domain features are extracted through the second, third, fourth, and fifth convolutional layers, where the convolution kernel size of the second, third, fourth, and fifth convolutional layers is 3, and the corresponding expansion coefficients are 1, 2, 4, and 8, respectively; Adding the shallow features extracted by the first convolutional layer and the deep features, time domain features, and frequency domain features extracted by the second, third, fourth, and fifth convolutional layers to determine second output data of the temporal convolutional network layer; Among them, the filling method of the second convolution kernel, the third convolution kernel, the fourth convolution kernel and the fifth convolution kernel is causal.

7. An electronic device comprising a memory and a processor, characterized in that: The memory is configured to store one or more computer instructions, wherein the one or more computer instructions are executed by a processor to implement the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method according to any one of claims 1 to 5.

Images