Modulation recognition model for recognizing multiple paths of signals and model training method

Abstract

The embodiment of the invention provides a modulation recognition model for recognizing multiple paths of signals and a model training method. The modulation recognition model sequentially comprises an input module, a cross-channel fusion module, at least two feature extraction modules, a global average pooling module, a classifier module and an output module according to an input and output relation. The input module is used for receiving in-phase orthogonal signals and phase amplitude signals of the same to-be-identified signal in different representation forms; the cross-channel fusion module is used for respectively processing the in-phase orthogonal signals and the phase amplitude signals and then carrying out feature fusion to obtain fusion features; the global average pooling module is used for determining global features of the to-be-used features; the classifier module is used for determining evaluation attributes of the to-be-identified signal in different modulation modes by using the global feature information; the output module is used for outputting the evaluation attributes. According to the technical scheme provided by the embodiment of the invention, the evaluation attribute is output based on the modulation identification model used for identifying multiple paths of signals, and the performance of the modulation identification model is remarkably improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the technical field of signal processing, and in particular, to a modulation recognition model for identifying multiple signals and a model training method. Background Art

[0002] With the rapid development of wireless communication technology, various modulation methods are widely used in different communication scenarios to meet the requirements of different services for transmission rate, spectral efficiency, anti-interference ability, etc. In such a diverse communication environment, the receiving end needs to accurately identify the modulation method used by the received signal in order to correctly demodulate and restore the original information.

[0003] Currently, the methods used to determine the modulation method of communication signals are mainly decision theory, statistical pattern recognition, etc. The disadvantages of these methods are that they usually require complex mathematical derivations and a large amount of calculations, there are deviations in parameter estimation, and the performance will decline when the model does not match the actual channel characteristics. In addition, these methods usually require additional training samples, are difficult to implement in engineering, and the recognition performance will drop sharply under low signal-to-noise ratio conditions. Summary of the Invention

[0004] Embodiments of the present disclosure provide a modulation recognition model for identifying multiple signals and a model training method, so as to achieve the effect of outputting an evaluation attribute based on the modulation recognition model for identifying multiple signals and significantly improving the performance of the modulation recognition model.

[0005] In a first aspect, embodiments of the present disclosure provide a modulation recognition model for identifying multiple signals. According to the input-output relationship, the modulation recognition model sequentially includes an input module, a cross-channel fusion module, at least two feature extraction modules, a global average pooling module, a classifier module, and an output module;

[0006] Among them, the input module is used to receive the in-phase quadrature signal and the phase amplitude signal of the same signal to be recognized in different representation forms; the cross-channel fusion module is used to separately process the in-phase quadrature signal and the phase amplitude signal and then perform feature fusion to obtain a fusion feature; the at least two feature extraction modules include a dot product feature extraction sub-module and an additive attention mechanism feature extraction sub-module. The dot product feature extraction sub-module is used to process the fusion feature into a conversion feature of a target dimension, and the additive attention mechanism feature extraction sub-module is used to screen and process the conversion feature to obtain a screened feature to be used; the global average pooling module is used to determine the global feature of the feature to be used; the classifier module is used to use the global feature to determine the evaluation attribute of the signal to be recognized in different modulation methods; the output module is used to output the evaluation attribute; wherein, the evaluation attribute represents the credibility under the corresponding modulation mode.

[0007] In a second aspect, an embodiment of the present invention further provides a method for model training, the method comprising:

[0008] When it is detected that the modulation recognition model corresponding to the current training round is obtained, a plurality of test samples are acquired, where the current training round is when a preset number of training samples have participated in the training of the modulation recognition model, and the test samples and the training samples include signals to be recognized and corresponding modulation manners; wherein, the signal to be recognized is an in-phase quadrature signal;

[0009] Based on the plurality of test samples, determine the accuracy rate of the modulation recognition model corresponding to the current training round;

[0010] If the accuracy rates of the current training round and the preset number of polling times before the current training round do not meet the preset conditions, then adjust the learning rate of the modulation recognition model, and continue to train the modulation recognition model with the adjusted learning rate based on a preset number of training samples until the accuracy rates of consecutive multiple training rounds all meet the preset conditions, and use the learning rate obtained when the preset conditions are met as the target learning rate of the modulation recognition model, and continue to perform model training based on the modulation recognition model with the target learning rate.

[0011] In a third aspect, an embodiment of the present invention further provides a method for identifying the modulation manners of multiple signals, the method comprising:

[0012] Acquire the signals to be recognized with the modulation manners to be recognized, the in-phase quadrature signals and the phase amplitude signals in different representation forms; wherein, the phase amplitude signal is a signal obtained by converting the in-phase quadrature signal;

[0013] Input the in-phase quadrature signal and the phase amplitude signal into a pre-trained modulation manner recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signals to be recognized under different modulation manners;

[0014] Based on the evaluation attributes, determine the target modulation manner of the signal to be recognized.

[0015] In a fourth aspect, an embodiment of the present invention further provides an apparatus for model training, the apparatus comprising:

[0016] A test sample acquisition module, an accuracy rate determination module, and a model training module.

[0017] A test sample acquisition module, configured to obtain a plurality of test samples when detecting that the modulation recognition model corresponding to the current training round is obtained, where the current training round is that a preset number of training samples have participated in the training of the modulation recognition model, and the test samples and the training samples include signals to be recognized and corresponding modulation methods; wherein, the signal to be recognized is an in-phase quadrature signal;

[0018] An accuracy rate determination module, configured to determine the accuracy rate corresponding to the modulation recognition model in the current training round based on the plurality of test samples;

[0019] A model training module, configured to, if the accuracy rates in the current training round and the preset number of polling times before the current training round do not meet the preset conditions, adjust the learning rate of the modulation recognition model, and continue to train the modulation recognition model with the adjusted learning rate based on a preset number of training samples until the accuracy rates in multiple consecutive training rounds all meet the preset conditions, and use the learning rate obtained when meeting the preset conditions as the target learning rate of the modulation recognition model, and continue to perform model training based on the modulation recognition model with the target learning rate.

[0020] In a fifth aspect, an embodiment of the present invention further provides a device for identifying modulation methods of multiple signals, and the device includes:

[0021] A signal to be recognized acquisition module, an evaluation attribute determination module, and a target modulation method determination module.

[0022] A signal to be recognized acquisition module, configured to obtain signals to be recognized with modulation methods to be recognized, in-phase quadrature signals and phase amplitude signals in different representation forms; wherein, the phase amplitude signal is a signal obtained by converting the in-phase quadrature signal;

[0023] An evaluation attribute determination module, configured to input the in-phase quadrature signal and the phase amplitude signal into a pre-trained modulation method recognition model for modulation mode recognition, and output evaluation attributes corresponding to the signal to be recognized under different modulation methods;

[0024] A target modulation method determination module, configured to determine the target modulation method of the signal to be recognized based on the evaluation attributes.

[0025] In a sixth aspect, an embodiment of the present invention further provides an electronic device, and the electronic device includes:

[0026] One or more processors;

[0027] A storage device, configured to store one or more programs,

[0028] When the one or more programs are executed by the one or more processors, the one or more processors implement the model training method and the method for identifying the modulation mode of a multiplex signal as described in any one of the embodiments of the present invention.

[0029] In a seventh aspect, an embodiment of the present invention further provides a storage medium including computer-executable instructions, and the computer-executable instructions are used to execute the model training method and the method for identifying the modulation mode of a multiplex signal as described in any one of the embodiments of the present invention when executed by a computer processor.

[0030] In an eighth aspect, an embodiment of the present invention further provides a computer program product, including a computer program, characterized in that the computer program implements the model training method and the method for identifying the modulation mode of a multiplex signal as described in any one of the embodiments of the present invention when executed by a processor.

[0031] The technical solution of the embodiment of the present disclosure, according to the input-output relationship, the modulation recognition model for identifying multiplex signals sequentially includes an input module, a cross-channel fusion module, at least two feature extraction modules, a global average pooling module, a classifier module, and an output module, which solves the problems in the prior art that when determining the modulation mode of a communication signal according to methods such as decision theory and statistical pattern recognition, complex mathematical derivations and a large amount of calculations are required, and the performance will decline when the model does not match the actual channel characteristics. The embodiment of the present disclosure realizes outputting an evaluation attribute based on the modulation recognition model for identifying multiplex signals, and significantly improves the performance of the modulation recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the introduced drawings are only the drawings of a part of the embodiments to be described by the present invention, rather than all the drawings. For those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts.

[0033] Figure 1 is a schematic diagram of a modulation recognition model for identifying multiplex signals provided by an embodiment of the present disclosure;

[0034] Figure 2 is a flowchart of a modulation recognition model for identifying multiplex signals provided by an embodiment of the present disclosure;

[0035] Figure 3 is a flowchart of the cross-channel fusion module provided by an embodiment of the present disclosure;

[0036] Figure 4 is a flowchart of the dot product feature extraction sub-module provided by an embodiment of the present disclosure;

[0037] Figure 5 It is a schematic flow chart of the additive attention mechanism feature extraction sub-module provided by an embodiment of the present disclosure;

[0038] Figure 6 It is a schematic flow chart of the model training method provided by an embodiment of the present disclosure;

[0039] Figure 7 It is a schematic flow chart of the method for identifying the modulation mode of multi-channel signals provided by an embodiment of the present disclosure;

[0040] Figure 8 It is a schematic structural diagram of the model training device provided by an embodiment of the present disclosure;

[0041] Figure 9 It is a schematic structural diagram of the device for identifying the modulation mode of multi-channel signals provided by an embodiment of the present disclosure;

[0042] Figure 10 It is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. Detailed implementation manners

[0043] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present invention, rather than limiting the present invention. Additionally, it should be noted that for the sake of description, only parts related to the present invention rather than all structures are shown in the accompanying drawings.

[0044] Before introducing the technical solutions provided by the embodiments of the present disclosure, an exemplary description of the application scenario can be given first. The technical solutions provided by the embodiments of the present disclosure can be applied in the scenario of determining the modulation mode of a signal to be recognized. For example, in the scenario of constructing a modulation recognition model for identifying multi-channel signals based on the input-output relationship and obtaining the modulation mode of the signal to be recognized based on the constructed and trained modulation recognition model for identifying multi-channel signals. Based on the technical solutions of the embodiments of the present disclosure, the evaluation attributes are output based on the modulation recognition model for identifying multi-channel signals, significantly improving the performance of the modulation recognition model.

[0045] Embodiment 1

[0046] Figure 1 It is a schematic diagram of the modulation recognition model for identifying multi-channel signals provided by an embodiment of the present disclosure. The embodiments of the present disclosure are applicable to the situation of determining the modulation mode of a signal to be recognized.

[0047] As Figure 1 shown, it is possible to understand the model architecture of the modulation recognition model for identifying multi-channel signals provided by the embodiments of the present invention based on Figure 1 ​

[0048] According to the input-output relationship, the modulation recognition model sequentially includes an input module 10, a cross-channel fusion module 20, at least two feature extraction modules 30 and 40, a global average pooling module 50, a classifier module 60, and an output module 70.

[0049] Among them, the modulation recognition model refers to a model used to automatically identify the modulation method of wireless communication signals. In wireless communication, signals can be transmitted through different modulation methods. Through the constructed modulation recognition model, the specific modulation method of the signal can be inferred from the received signal. It should be noted that the modulation method determines the encoding method of wireless communication signals. Only by knowing the modulation method can the signal be correctly demodulated and the original information be restored.

[0050] Among them, the input module is used to receive the in-phase quadrature signal and the phase-amplitude signal of the same signal to be recognized in different representation forms; the cross-channel fusion module is used to separately process the in-phase quadrature signal and the phase-amplitude signal and then perform feature fusion to obtain fused features; at least two feature extraction modules include a dot product feature extraction sub-module and an additive attention mechanism feature extraction sub-module. The dot product feature extraction sub-module is used to process the fused features into transformed features of the target dimension, and the additive attention mechanism feature extraction sub-module is used to screen and process the transformed features to obtain the features to be used after screening; the global average pooling module is used to determine the global features of the features to be used; the classifier module is used to use the global features to determine the evaluation attributes of the signal to be recognized in different modulation methods; the output module is used to output the evaluation attributes; among them, the evaluation attribute represents the credibility under the corresponding modulation mode. It should be noted that the cross-channel fusion module is crosschannel fusion, the dot product feature extraction sub-module is star channel, the additive attention mechanism feature extraction sub-module is addition attention encoder, the global average pooling module is globalAvgpool, and the classifier module is classifier.

[0051] Optionally, the in-phase quadrature signal corresponding to the signal to be recognized is an I / Q signal, and the phase-amplitude signal corresponding to the signal to be recognized is an A / P signal.

[0052] Among them, the I / Q signal is the real and imaginary components of a complex signal in the rectangular coordinate system, used to represent the amplitude and phase information of the signal. In this embodiment, the complex signal can be a modulated signal. The I signal is the component in phase with the reference carrier, reflecting the amplitude change of the signal. The Q signal is the component orthogonal (phase difference of 90°) to the reference carrier, reflecting the phase change of the signal. Any modulated signal can be uniquely represented by the I and Q components. The A / P signal is a direct representation of the amplitude (A) and phase (P) information of the signal in the polar coordinate system. The A signal represents the instantaneous amplitude of the signal, reflecting the signal strength. The P signal represents the instantaneous phase of the signal, reflecting the phase change of the signal. The I / Q signal, as the rectangular coordinate representation of the complex signal, and the A / P signal, as the polar coordinate representation of the complex signal, can be converted into each other, and the information after conversion is completely equivalent. When the I / Q signal is known, When the A / P signal is known, I = A · cos(P), Q = A · sin(P)

[0053] It should be noted that referring to Figure 2 , after receiving the in-phase quadrature signal and the phase amplitude signal, first, feature extraction can be performed on the in-phase quadrature signal and the phase amplitude signal respectively based on the cross-channel fusion module to obtain their respective feature vectors. For example, features such as amplitude, phase, and frequency are extracted from the I / Q signal; features such as amplitude change rate and phase change rate are extracted from the A / P signal. Then, the two feature vectors of the in-phase quadrature signal and the phase amplitude signal can be fused to generate a fused feature. The fused feature is usually a high-dimensional vector, containing the feature information of both the I / Q signal and the A / P signal. For example, if 3 features are extracted from the I / Q signal and 2 features are extracted from the A / P signal, a 5-dimensional or higher-dimensional feature vector, that is, the fused feature, may be obtained after fusion. When performing feature fusion of the in-phase quadrature signal and the phase amplitude signal, the two feature vectors of the in-phase quadrature signal and the phase amplitude signal can be directly concatenated to form a longer vector, or the two feature vectors can be weighted and summed according to the importance of the features to obtain the fused feature. For the fused feature, the dimension of the fused feature depends on the number of features of the I / Q signal and the A / P signal and the fusion method. Moreover, the fused feature contains the information of two different signals, the I / Q signal and the A / P signal, and has multi-modal characteristics.

[0054] Among them, the target dimension refers to the dimensionality of the expected feature space to which the input features, i.e., the fused features, are mapped, which is usually determined by task requirements or model design. For example, assume that the fused feature is a 100-dimensional vector, but the subsequent task of the multiplicative attention mechanism feature extraction sub-module only requires 10-dimensional features, then the target dimension is 10. The target dimension can be a fixed value or can be automatically and dynamically adjusted according to the length of the input features. The transformed feature is the output of the dot product feature extraction sub-module and is the feature vector obtained by mapping the fused features to the target dimension after a linear transformation. The filtered features to be used are the output of the multiplicative attention mechanism feature extraction sub-module and are the feature vectors obtained by applying attention weighting to the transformed features. It should be noted that the dot product feature extraction sub-module realizes feature dimensionality reduction through a linear transformation, has a low computational complexity, and the weight matrix is learnable to adapt to different task requirements. The attention weights of the multiplicative attention mechanism feature extraction sub-module can be dynamically adjusted according to the input data, have strong adaptability, can suppress noise features, and improve the quality of features. The dot product sub-module realizes the transformation of the feature space, and the multiplicative attention sub-module realizes feature screening. Through the synergistic effect of the two, efficient feature extraction and optimization can be achieved.

[0055] It should also be noted that, referring to Figure 2 , the global feature of the features to be used refers to the overall statistical information extracted from the features to be used through the operation of the global average pooling module, which is usually used to capture the global distribution characteristics in the spatial or temporal dimension of the features. The global average pooling module can perform an average calculation on the spatial dimension of the feature vector output by the multiplicative attention mechanism feature extraction sub-module to obtain the global feature of the features to be used. When using the global average pooling module to determine the global feature of the features to be used, the features at each frequency point (or time frame) can be aggregated to obtain the global feature at each frequency point (or time frame). Through the global average pooling module, a high-dimensional feature matrix can be compressed into a one-dimensional vector, reducing the computational complexity. By capturing the overall distribution characteristics of the signal and ignoring local details, the influence of noise on the features is reduced. The signal to be recognized refers to the signal for which it is necessary to determine the evaluation attributes in different modulation modes. Different modulation modes refer to different modulation types of the signal. The evaluation attribute refers to the performance index or classification result generated by the classifier module based on the global feature for evaluating the signal to be recognized in different modulation modes. Through the classifier module, it can be determined which modulation mode the signal belongs to.

[0056] Optionally, the cross-channel fusion module includes a first convolutional network and a second convolutional network with the same network structure, a first feature extraction network connected to the first convolutional network, a second feature extraction network connected to the second convolutional network, a similar feature determination unit, a feature multiplication unit, a feature addition unit, and a fusion unit, and the first feature extraction network and the second feature extraction network have the same network structure.

[0057] Among them, the first convolutional network is used to process the in-phase quadrature signal to obtain a first feature; the second convolutional network is used to process the phase-amplitude signal to obtain a second feature; the first feature extraction network is used to extract features from the first feature to obtain a third feature; the second feature extraction network is used to extract features from the second feature to obtain a fourth feature; the similar feature determination unit is used to determine the similarity between the third feature and the fourth feature and output it; the feature multiplication unit is respectively used to determine a first result between the third feature and the similarity, and a second result between the fourth feature and the similarity; the first feature addition unit corresponding to the first feature extraction network is used to process the second result and the third feature to obtain a first fused feature; the second feature superposition unit corresponding to the second feature extraction network is used to process the first result and the fourth feature to obtain a second fused feature; the fusion unit is used to perform a fusion process on the first fused feature and the second fused feature to obtain a fused feature.

[0058] It should be noted that referring to Figure 3 , the inputs of the first convolutional network and the second convolutional network are the in-phase quadrature signal and the phase-amplitude signal respectively. The first convolutional network can extract features such as the time-frequency characteristics, phase change, and amplitude change of the in-phase quadrature signal. The first feature can reflect the complex characteristics of the I / Q signal, such as the symbol rate, carrier frequency offset, etc. The second convolutional network can extract features such as the phase distribution, amplitude change, and noise characteristics of the phase-amplitude signal. The second feature can reflect the amplitude and phase characteristics of the A / P signal, such as the modulation index, amplitude modulation depth, and phase noise of the signal. The first convolutional network and the second convolutional network extract different aspects of features of the I / Q signal and the A / P signal through the same processing method, providing support for subsequent processing tasks. The input of the first feature extraction network is the first feature, and the output is the third feature. The input of the second feature extraction network is the second feature, and the output is the fourth feature. The similarity refers to the cosine similarity between the third feature and the fourth feature. The inputs of the similar feature determination unit are the third feature and the fourth feature, and the output result is the similarity between the third feature and the fourth feature. For example, the cosine similarity between the third feature and the fourth feature can be calculated based on the similar feature determination unit. The cosine similarity is an index to measure the similarity of the directions of two vectors, with a value range of [-1, 1], indicating the similarity degree between the third feature and the fourth feature. It can be based on The numerator is the vector representation of the third feature and the fourth feature, and the denominator is the L2 norm of the vectors of the third feature and the fourth feature. For example, when the cosine similarity of the third feature and the fourth feature calculated based on the third feature and the fourth feature is 0.9, it means that the third feature and the fourth feature are very similar. The first result refers to the result obtained by multiplying the third feature by the similarity. The second result refers to the result obtained by multiplying the fourth feature by the similarity. The first fused feature refers to the feature obtained by performing feature superposition processing on the second result and the third feature. The second fused feature refers to the feature obtained by performing feature superposition processing on the first result and the fourth feature. The fused feature refers to the feature obtained by performing feature superposition processing on the first fused feature and the second fused feature.

[0059] Optionally, the first convolutional network and the second convolutional network include a batch normalization layer, a one-dimensional convolution, and an activation function; the feature extraction network includes a one-dimensional depth convolutional layer, a batch normalization layer, a one-dimensional convolutional layer, and an activation function layer.

[0060] Among them, the batch normalization layer is a technique that accelerates the training of neural networks by normalizing the input feature distribution. In the batch normalization layer, the mean and variance of each dimension can be calculated for the input features of the current batch, and then the input features are normalized to zero mean and unit variance. Finally, learnable parameters can be introduced to restore the expressive ability of the features. Through the batch normalization layer, the internal covariate shift can be reduced, making the gradient more stable. Also, through normalization, the activation value distribution can be made more concentrated, reducing the sensitivity to the initial weights. The one-dimensional convolution is a convolution operation performed on one-dimensional signals, and local features are extracted by sliding the convolution kernel. Through the one-dimensional convolution, local patterns in the signal can be captured, and the signal length can also be reduced by sliding the convolution kernel to extract high-level features. The activation function, as a non-linear function, is used to introduce the non-linear ability of the neural network so that it can learn complex patterns. Through the activation function, the calculation can be simple, and the gradient vanishing problem can be alleviated. In the embodiments of the present invention, an appropriate activation function can be selected according to the task. The one-dimensional depth convolutional layer is a lightweight convolution operation, and each channel is convolved independently without mixing between channels. Through the one-dimensional depth convolutional layer, the number of parameters can be reduced, and local features can be extracted.

[0061] It should be noted that the batch normalization layer is BN, the one-dimensional convolution is Conv1D, the activation function is ReLU, and the one-dimensional depth convolution is DWConv1D.

[0062] Optionally, the dot product feature extraction sub-module according to the input-output relationship sequentially includes a one-dimensional depth convolution layer, a batch normalization layer, a first fully connected layer, a second fully connected layer, a first activation function layer connected to the first fully connected layer, a third fully connected layer, a batch normalization layer, a one-dimensional depth convolution, a random path dropout layer, a one-dimensional average pooling layer, a fourth fully connected layer, a second activation function layer, a fifth fully connected layer, and an activation function layer; wherein, the fused feature is processed by the one-dimensional depth convolution layer and the batch normalization layer in sequence to obtain a normalized feature; the normalized feature is processed by the first fully connected layer and the second fully connected layer respectively to obtain a fifth feature corresponding to the first fully connected layer and a sixth feature corresponding to the second fully connected layer; the fifth feature is processed by the first activation function layer to obtain a first dot product feature, and the sixth feature and the first dot product feature are dot-multiplied to obtain a seventh feature to be input to the third fully connected layer; the seventh feature is processed by the third fully connected layer to obtain an eighth feature; the eighth feature is processed by the batch normalization layer, the one-dimensional depth convolution layer, and the random path dropout layer in sequence to obtain a superimposed feature, and the superimposed feature is fused with the fused feature to obtain a feature to be processed; the feature to be processed is processed by the one-dimensional average pooling layer, the fourth fully connected layer, the second activation function layer, the fifth fully connected layer, and the activation function layer in sequence to obtain a second dot product feature; the second dot product feature and the feature to be processed are dot-multiplied to obtain a conversion feature to be input to the additive attention mechanism feature extraction sub-module.

[0063] Among them, referring to Figure 4 , the fully connected layer is a basic layer structure in a neural network, where each neuron is connected to all neurons in the previous layer. The fully connected layer can map the input features to the output space to achieve feature integration and classification. The random path dropout layer, as a regularization technique, is used to prevent the neural network from overfitting. During training, some neurons are randomly discarded to reduce the dependence between neurons, improve the generalization ability of the model, and reduce the risk of overfitting.

[0064] It should be noted that the normalized features refer to the features output after the fused features pass through a one-dimensional depth convolutional layer and a batch normalization layer. The fifth feature refers to the features output after the normalized features pass through the first fully connected layer. The sixth feature refers to the features output after the normalized features pass through the second fully connected layer. The first dot product feature refers to the features output after the fifth feature passes through the first activation function layer. The seventh feature refers to the features obtained after the sixth feature and the first dot product feature are multiplied. The eighth feature refers to the features obtained after the seventh feature passes through the third fully connected layer. The stacked features refer to the features obtained after the eighth feature passes through a batch normalization layer, a one-dimensional depth convolutional layer, and a stochastic depth layer. The features to be processed refer to the features obtained after the stacked features and the fused features are added. The second dot product feature refers to the features obtained after the features to be processed pass through a one-dimensional average pooling layer, the fourth fully connected layer, an activation function layer, the fourth fully connected layer, and an activation function layer. The transformed features refer to the features obtained after the second dot product feature and the features to be processed are dot product processed.

[0065] It should be noted that the fully connected layer is FC, the stochastic depth layer is DropPath, the one-dimensional average pooling layer is AcgPoolld, and the activation function layer is Hsigmoid.

[0066] In this embodiment, the additive attention mechanism feature extraction sub-module includes an efficient additive attention mechanism unit, a first addition and normalization processing unit, a feature extraction sub-network, and a second addition and normalization processing unit.

[0067] Among them, the efficient additive attention mechanism unit is used to process the transformed features to obtain attention features. It should be noted that, referring to Figure 5 , in the efficient additive attention mechanism unit, after the input transformed features pass through multiple linear layers on the left side of the figure, multiple results can be directly added to obtain the added result. After the input transformed features pass through multiple linear layers on the right side of the figure, the multiple output results can be multiplied by the added result obtained above. After the result of the multiplication process passes through another linear layer, it is added to the result of the normalization process after passing through multiple linear layers on the left side of the figure, and then the result of the addition process passes through a linear layer to obtain the attention features.

[0068] The first addition and normalization processing unit is used to perform addition and normalization processing on the attention feature and the transformation feature to obtain the feature to be extracted to be input into the feature extraction sub-network. It should be noted that in a neural network, the addition operation refers to linearly combining the attention features. This can be achieved through linear transformation of a weight matrix and a bias vector. Specifically, for the feature to be extracted, they can be multiplied by corresponding weights and added with biases, and then the results are added together to obtain a new output. This addition operation helps to fuse information from different sources, thereby enhancing the expressive ability of the network. The normalization processing unit is used to perform standardization processing on the output or intermediate results of the neural network. The purpose of normalization is to ensure that the data has a unified distribution characteristic in subsequent network layers, which helps to stabilize the training process and improve the convergence speed and generalization ability of the model. In the first addition and normalization processing unit, common normalization methods include batch normalization, layer normalization, etc. These normalization methods adjust the mean and variance of the data so that the output data conforms to a specific distribution (such as a normal distribution), thereby improving the training effect of the network. In a neural network, the addition and normalization processing units are often used together. The addition operation can be performed on the output to fuse the information of multiple features, and then the normalization processing unit is applied to perform standardization processing on the result. This combined use can significantly improve the performance and stability of the network, enabling the model to better handle complex tasks and data.

[0069] The feature extraction sub-network is used to extract the feature to be extracted for feature extraction to obtain the feature to be normalized. Among them, the feature extraction sub-network includes a one-dimensional convolutional layer, an activation function layer, a dropout layer, and a one-dimensional convolutional layer. It should be noted that the dropout layer, as a regularization technique for preventing overfitting of the neural network, will randomly set the outputs of a part of neurons to zero during the training process, thereby forcing the network to learn more robust feature representations. The second addition and normalization processing unit is used to perform feature addition and normalization processing on the feature to be normalized and the feature to be extracted to obtain the feature to be used. Among them, the feature to be used refers to the feature obtained by performing feature addition and normalization processing on the feature to be normalized and the feature to be extracted.

[0070] It should be noted that the efficient additive attention mechanism unit is EfficientAdditiveAttention, the addition and normalization processing unit is Add&Norm, the dropout layer is Dropout, the linear layer is Linear, and the normalization is Norm.

[0071] Specifically, a modulation recognition model for identifying multi-channel signals is constructed, including an input module, a cross-channel fusion module, at least two feature extraction modules, a global average pooling module, a classifier module, and an output module, so as to perform subsequent signal processing based on the modulation recognition model.

[0072] The technical solution of the embodiment of the present disclosure includes, in sequence according to the input-output relationship, an input module, a cross-channel fusion module, at least two feature extraction modules, a global average pooling module, a classifier module, and an output module for the modulation recognition model, which solves the problems in the prior art that when determining the modulation mode of a communication signal based on methods such as decision theory and statistical pattern recognition, complex mathematical derivations and a large amount of calculations are required, and the performance will decline when the model does not match the actual channel characteristics. The embodiment of the present disclosure realizes outputting evaluation attributes based on the modulation recognition model for identifying multiplex signals, significantly improving the performance of the modulation recognition model.

[0073] Embodiment 2

[0074] Figure 6 It is a schematic flowchart of the model training method provided by the embodiment of the present invention. On the basis of the foregoing embodiment, the training of the modulation recognition model is described in detail. The training method of the modulation recognition model can be executed by the training device of the modulation recognition model. The device can be implemented in the form of software and / or hardware. The hardware can be a mobile electronic device, and the electronic device can execute the training method of the modulation recognition model provided by the present technical solution. The specific implementation manner can refer to the technical solution of this embodiment. Among them, the same or corresponding technical terms as those in the above embodiment will not be described in detail here.

[0075] As Figure 6 shown, the method specifically includes the following steps:

[0076] S210. When it is detected that the modulation recognition model corresponding to the current training round is obtained, obtain a plurality of test samples.

[0077] Among them, the current training round means that a preset number of training samples have participated in the training of the modulation recognition model. The test samples and the training samples include the signals to be recognized and the corresponding modulation modes. Among them, the signal to be recognized is an in-phase quadrature signal.

[0078] Optionally, the modulation modes include double-sideband amplitude modulation, single-sideband amplitude modulation, binary phase shift keying, quadrature phase shift keying, 8-phase shift keying, 8-quadrature amplitude modulation, 16-quadrature amplitude modulation, 16-phase shift keying, 64-quadrature amplitude modulation, Gaussian minimum frequency shift keying, and orthogonal frequency division multiplexing.

[0079] It should be noted that during the training process of the modulation recognition model, the modulation recognition model can be optimized through multiple training rounds. In multiple training rounds, each training round will update the model parameters using the training samples. The training samples are the samples used for training the modulation recognition model. The test samples are the samples used for testing the accuracy of the modulation recognition model after each training round. The signal to be recognized can refer to an in-phase quadrature signal.

[0080] It should also be noted that double - sideband amplitude modulation means that the carrier and the upper and lower sidebands are transmitted simultaneously, and the information is contained in the sidebands; single - sideband amplitude modulation means that only one sideband (upper sideband or lower sideband) is transmitted, suppressing the carrier and the other sideband; binary phase - shift keying means that binary data is represented by changing the carrier phase (0° or 180°); quadrature phase - shift keying means that the 360° phase is divided into 4 states (0°, 90°, 180°, 270°), and each symbol carries 2 bits; 8 - phase shift keying means that the 360° phase is divided into 8 states, and each symbol carries 3 bits; 8 - quadrature amplitude modulation means that data is encoded into 8 different symbols by modulating the amplitude and phase of the signal; 16 - quadrature amplitude modulation means that the 360° phase and amplitude are combined into 16 symbols, and each symbol carries 4 bits; 16 - phase shift keying means that 16 different phase states are used to represent binary data, and the symbol set includes four orthogonal double - amplitude phase - shift sequences; 64 - quadrature amplitude modulation means that the 360° phase and amplitude are combined into 64 symbols, and each symbol carries 6 bits; Gaussian minimum - shift keying means that the signal is pre - processed by a Gaussian filter and then minimum - shift keying modulation is performed; orthogonal frequency - division multiplexing means that a high - speed data stream is divided into multiple low - speed sub - carriers, each sub - carrier is independently modulated, and the sub - carriers are orthogonal to each other.

[0081] Specifically, when training the modulation recognition model, at the current training round, multiple test samples are obtained. Moreover, the test samples and training samples include the signal to be recognized and the modulation method corresponding to the signal to be recognized.

[0082] S220. Determine the accuracy rate corresponding to the modulation recognition model at the current training round based on multiple test samples.

[0083] Among them, the accuracy rate is the core index for evaluating the performance of the modulation recognition model, indicating the proportion of the number of samples correctly predicted by the modulation recognition model to the total number of samples.

[0084] Specifically, after inputting multiple test samples into the modulation recognition model, the accuracy rate corresponding to the modulation recognition model at the current training round can be determined based on the output result and the modulation method corresponding to the signal to be recognized.

[0085] S230. If the accuracy rates of the current training round and the preset polling times before the current training round do not meet the preset conditions, then adjust the learning rate of the modulation recognition model, and continue to train the modulation recognition model with the adjusted learning rate based on a preset number of training samples until the accuracy rates of consecutive multiple training rounds all meet the preset conditions, and use the learning rate obtained when meeting the preset conditions as the target learning rate of the modulation recognition model, and continue to perform model training based on the modulation recognition model with the target learning rate.

[0086] Among them, the preset polling times refer to the fixed evaluation intervals preset during the training process of the modulation recognition model for evaluating the performance of the modulation recognition model. The polling times refer to evaluating the performance of the modulation recognition model on the test set every N training iterations. By regularly evaluating, observe whether the performance of the modulation recognition model improves with the number of training rounds. When the accuracy of the current training round and the preset polling times before the current training round do not meet the preset conditions, the learning rate of the modulation recognition model can be adjusted. For example, the learning rate of the modulation recognition model can be multiplied by 0.8.

[0087] In this embodiment, obtain a preset number of training samples corresponding to the current training round; for the training samples, input the training samples into the modulation recognition model for modulation mode recognition to obtain an output result; determine the loss value according to the output result and the modulation mode in the training samples, and correct the parameters of the modulation recognition model based on the loss value; when all the preset number of training samples have participated in training the modulation recognition model, obtain the modulation recognition model corresponding to the current training round.

[0088] Among them, in order to improve the accuracy of model training, training samples of different modulation modes can be obtained. It should be noted that when the training samples are input into the modulation recognition model for modulation mode recognition, the model parameters in the modulation recognition model do not meet the expected requirements. Therefore, there is a certain difference between the modulation mode corresponding to the actual evaluation attribute output based on the model parameters at this time and the theoretical modulation mode. Therefore, based on the modulation mode corresponding to the actual evaluation attribute of each training sample and the theoretical modulation mode, the corresponding error loss value can be determined. It should be noted that the training parameters can be set to default values before the model is trained. When training the modulation signal recognition model, the training parameters in the model can be corrected based on the output result. That is to say, the applicable modulation signal recognition model can be obtained by correcting the loss function in the model. Each training sample has a corresponding loss value, which is determined based on the modulation mode corresponding to the actual evaluation attribute of the training sample and the theoretical modulation mode. Specifically, the training error of the loss function, that is, the loss parameter, can be used as the condition for detecting whether the current loss function has converged, such as whether the training error is less than the preset error or whether the error change trend tends to be stable, or whether the current number of iterations is equal to the preset number. If the convergence condition is detected, such as the training error of the loss function reaches less than the preset error or the error change tends to be stable, it indicates that the training of the modulation signal recognition model is completed, and the iterative training can be stopped at this time.

[0089] Specifically, if the accuracy rate in the current training round and the accuracy rates in the preset number of polling rounds before the current training round do not meet the preset conditions, that is, when the accuracy rate does not meet the standard, the learning rate of the modulation recognition model can be adjusted. For the modulation recognition model after adjusting the learning rate, the modulation recognition model can be continuously trained based on a preset number of training samples. When the accuracy rates in multiple consecutive training rounds all meet the preset conditions, the modulation recognition model obtained when the preset conditions are met is used as the modulation recognition model for use.

[0090] In the technical solution of the embodiment of the present disclosure, when detecting the modulation recognition model corresponding to the current training round, a plurality of test samples are obtained. Then, based on the plurality of test samples, the accuracy rate corresponding to the modulation recognition model in the current training round is determined. Finally, if the accuracy rates in the current training round and the preset number of polling rounds before the current training round do not meet the preset conditions, the learning rate of the modulation recognition model is adjusted to continuously train the modulation recognition model based on a preset number of training samples until the accuracy rates in multiple consecutive training rounds all meet the preset conditions, and the modulation recognition model obtained when the preset conditions are met is used as the modulation recognition model for use, dynamically adjusting the learning rate, improving the training efficiency of the modulation recognition model, and providing a high-quality and high-stability model for the modulation recognition task.

[0091] Embodiment III

[0092] Figure 7 It is a schematic flowchart of a method for identifying the modulation mode of a multi-channel signal provided by an embodiment of the present invention. On the basis of the foregoing embodiment, a detailed description is given of determining the target modulation mode of the signal to be recognized. The specific implementation manner can refer to the technical solution of this embodiment. Among them, the same or corresponding technical terms as those in the above embodiment are not described herein again.

[0093] As Figure 7 shown, the method specifically includes the following steps:

[0094] S310. Obtain the in-phase quadrature signal and the phase-amplitude signal of the signal to be recognized for the modulation mode to be recognized in different representation forms.

[0095] Among them, the phase-amplitude signal is a signal obtained by converting the in-phase quadrature signal. The signal to be recognized is a signal for which the modulation mode needs to be determined.

[0096] Specifically, when determining the modulation mode to be recognized of the in-phase quadrature signal, the phase-amplitude signal corresponding to the in-phase quadrature signal can be first obtained based on the in-phase quadrature signal.

[0097] S320. Input the in-phase quadrature signal and the phase-amplitude signal into the pre-trained modulation mode recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signal to be recognized under different modulation modes.

[0098] S330. Determine the target modulation mode of the signal to be recognized based on the evaluation attributes.

[0099] In the technical solution of the embodiments of the present disclosure, the signals to be recognized with the modulation modes to be recognized are obtained, including the in-phase quadrature signals and the phase amplitude signals in different representation forms. Then, the in-phase quadrature signals and the phase amplitude signals are input into a pre-trained modulation mode recognition model for modulation mode recognition, and the evaluation attributes corresponding to the signals to be recognized under different modulation modes are output. Finally, based on the evaluation attributes, the target modulation mode of the signal to be recognized is determined. At the same time, the I / Q signals and the A / P signals are used as inputs to obtain relevant information from the rich inputs, so as to achieve a higher accuracy rate of modulation mode recognition.

[0100] Embodiment 4

[0101] Figure 8 is a schematic structural diagram of the model training device provided by the embodiments of the present disclosure, as Figure 8 shown. The device includes a test sample acquisition module, an accuracy rate determination module, and a model training module.

[0102] The test sample acquisition module is configured to acquire a plurality of test samples when detecting the modulation recognition model corresponding to the current training round. Wherein, the current training round means that a preset number of training samples have participated in the training of the modulation recognition model. The test samples and the training samples include the signals to be recognized and the corresponding modulation modes. Wherein, the signal to be recognized is an in-phase quadrature signal.

[0103] The accuracy rate determination module is configured to determine the accuracy rate corresponding to the modulation recognition model in the current training round based on the plurality of test samples.

[0104] The model training module is configured to, if the accuracy rates in the current training round and the preset number of polling times before the current training round do not meet the preset conditions, adjust the learning rate of the modulation recognition model, and continue to train the modulation recognition model with the adjusted learning rate based on a preset number of training samples until the accuracy rates in multiple consecutive training rounds meet the preset conditions, and use the learning rate obtained when meeting the preset conditions as the target learning rate of the modulation recognition model, and continue to perform model training based on the modulation recognition model with the target learning rate.

[0105] Figure 9 is a schematic structural diagram of the device for recognizing the modulation modes of multiple signals provided by the embodiments of the present disclosure, as Figure 9 shown. The device includes a signal to be recognized acquisition module, an evaluation attribute determination module, and a target modulation mode determination module.

[0106] A signal acquisition module to be recognized, which is used to acquire signals to be recognized with modulation methods to be recognized, in-phase and quadrature signals and phase-amplitude signals in different representation forms; wherein, the phase-amplitude signal is a signal obtained by converting the in-phase and quadrature signal.

[0107] An evaluation attribute determination module, which is used to input the in-phase and quadrature signal and the phase-amplitude signal into a pre-trained modulation mode recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signals to be recognized under different modulation methods.

[0108] A target modulation method determination module, which is used to determine the target modulation method of the signal to be recognized based on the evaluation attribute.

[0109] The model training device provided by the embodiments of the present disclosure can execute the model training method provided by any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects for executing the method.

[0110] The device for identifying the modulation method of multi-channel signals provided by the embodiments of the present disclosure can execute the method for identifying the modulation method of multi-channel signals provided by any embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects for executing the method.

[0111] It should be noted that the various units and modules included in the above device are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the embodiments of the present disclosure.

[0112] Embodiment Five

[0113] Figure 10 It is a schematic structural diagram of an electronic device provided by the embodiments of the present disclosure. The following refers to Figure 10 , which shows a schematic structural diagram of an electronic device (such as Figure 10 the terminal device or server in) 500 suitable for implementing the embodiments of the present disclosure. The terminal device in the embodiments of the present disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), in-vehicle terminals (such as in-vehicle navigation terminals), and the like. Figure 10 The electronic device shown is only an example and should not bring any limitation to the functions and usage scope of the embodiments of the present disclosure.

[0114] As Figure 10As shown, the electronic device 500 may include a processing device (such as a central processing unit, a graphics processing unit, etc.) 501, which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other through a bus 504. An editing / output (I / O) interface 505 is also connected to the bus 504.

[0115] Generally, the following devices may be connected to the I / O interface 505: an input device 506 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 507 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 508 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 509. The communication device 509 may allow the electronic device 500 to communicate with other devices wirelessly or wiredly to exchange data. Although Figure 10 the electronic device 500 with various devices is shown, it should be understood that it is not required to implement or have all the shown devices. Instead, more or fewer devices may be implemented or had.

[0116] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart may be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes program codes for performing the method shown in the flowchart. In such an embodiment, the computer program may be downloaded and installed from a network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above functions defined in the method of the embodiment of the present disclosure are executed.

[0117] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only for illustrative purposes and are not used to limit the scope of these messages or information.

[0118] The electronic device provided by the embodiment of the present disclosure and the methods for model training and for identifying the modulation mode of a multiplex signal provided by the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment may be referred to in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0119] Embodiment Six

[0120] An embodiment of the present disclosure provides a computer storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for model training and the method for identifying the modulation mode of multiplex signals provided in the above embodiments.

[0121] It should be noted that the computer-readable medium in the present disclosure may be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device. In the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium may also be any computer-readable medium other than the computer-readable storage medium, and the computer-readable signal medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0122] In some embodiments, the server can communicate using any currently known or future-developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), the Internet (e.g., the Internet), and end-to-end networks (e.g., ad hoc end-to-end networks), as well as any currently known or future-developed network.

[0123] The above computer-readable medium may be included in the above electronic device; or it may exist separately without being assembled into the electronic device.

[0124] The above computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to:

[0125] When detecting the modulation recognition model corresponding to the current training round, obtain a plurality of test samples, where the current training round is when a preset number of training samples have participated in the training of the modulation recognition model, and the test samples and the training samples include signals to be recognized and corresponding modulation methods; among them, the signal to be recognized is an in-phase quadrature signal;

[0126] Determine the accuracy rate corresponding to the modulation recognition model in the current training round based on the plurality of test samples;

[0127] If the accuracy rates of the current training round and the preset polling times before the current training round do not meet the preset conditions, adjust the learning rate of the modulation recognition model, and continue to train the modulation recognition model with the adjusted learning rate based on a preset number of training samples until the accuracy rates of consecutive multiple training rounds all meet the preset conditions, and use the learning rate obtained when meeting the preset conditions as the target learning rate of the modulation recognition model, and continue to perform model training based on the modulation recognition model with the target learning rate.

[0128] Obtain the in-phase quadrature signals and phase-amplitude signals of the signal to be recognized under different representation forms for the modulation method to be recognized; among them, the phase-amplitude signal is a signal obtained by converting the in-phase quadrature signal;

[0129] Input the in-phase quadrature signal and the phase-amplitude signal into the pre-trained modulation method recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signal to be recognized under different modulation methods;

[0130] Determine the target modulation method of the signal to be recognized based on the evaluation attributes.

[0131] Computer program code for performing the operations of this disclosure can be written in one or more programming languages or combinations thereof. The programming languages include, but are not limited to, object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed 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 case of a remote computer, the remote computer can be connected to the user's computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0132] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in the flowchart or block diagram can represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks can occur in a different order than that marked in the accompanying drawings. For example, two consecutive blocks shown can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0133] The units described in the embodiments of this disclosure can be implemented in software or in hardware. Among them, the name of the unit does not, in some cases, constitute a limitation on the unit itself.

[0134] The functions described above herein can be performed at least in part by one or more hardware logic components. For example, by way of non-limitation, exemplary types of hardware logic components that can be used include: field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), systems on a chip (SOC), complex programmable logic devices (CPLD), and so on.

[0135] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer 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 of the foregoing.

[0136] The above description is only a preferred embodiment of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the disclosure involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosure concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the present disclosure.

[0137] In addition, although the operations are depicted in a particular order, this should not be construed as requiring that the operations be performed in the particular order shown or in a sequential order. In certain circumstances, multitasking and parallel processing may be advantageous. Similarly, although a number of specific implementation details are included in the above discussion, these should not be construed as limiting the scope of the present disclosure. Certain features described in the context of separate embodiments can also be implemented in combination in a single embodiment. Conversely, the various features described in the context of a single embodiment can also be implemented separately or in any suitable sub-combination in multiple embodiments.

[0138] Although the subject matter has been described in language specific to structural features and / or methodological acts, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or acts described above. Rather, the specific features and acts described above are merely example forms for implementing the claims.

Claims

1. A modulation recognition model for identifying multi-channel signals, characterized in that Including: According to the input-output relationship, the modulation recognition model sequentially includes an input module, a cross-channel fusion module, at least two feature extraction modules, a global average pooling module, a classifier module, and an output module; Among them, the input module is used to receive the in-phase quadrature signal and the phase-amplitude signal of the same signal to be recognized in different representation forms; the cross-channel fusion module is used to process the in-phase quadrature signal and the phase-amplitude signal respectively and then perform feature fusion to obtain a fused feature; among the at least two feature extraction modules, there are a dot product feature extraction sub-module and an additive attention mechanism feature extraction sub-module. The dot product feature extraction sub-module is used to process the fused feature into a transformed feature of a target dimension, and the additive attention mechanism feature extraction sub-module is used to screen and process the transformed feature to obtain a screened feature to be used; the global average pooling module is used to determine the global feature of the feature to be used; the classifier module is used to use the global feature to determine the evaluation attribute of the signal to be recognized in different modulation methods; the output module is used to output the evaluation attribute; among them, the evaluation attribute represents the credibility under the corresponding modulation mode.

2. The model according to claim 1, wherein The cross-channel fusion module includes a first convolutional network and a second convolutional network with the same network structure, a first feature extraction network connected to the first convolutional network, a second feature extraction network connected to the second convolutional network, a similarity feature determination unit, a feature multiplication unit, a feature addition unit, and a fusion unit. The network structures of the first feature extraction network and the second feature extraction network are the same; Among them, the first convolutional network is used to process the in-phase quadrature signal to obtain the first feature; the second convolutional network is used to process the phase-amplitude signal to obtain the second feature; The first feature extraction network is used to extract features from the first feature to obtain a third feature; the second feature extraction network is used to extract features from the second feature to obtain a fourth feature; The similarity feature determination unit is used to determine the similarity between the third feature and the fourth feature and output it; The feature multiplication unit is respectively used to determine a first result between the third feature and the similarity, and a second result between the fourth feature and the similarity; The first feature addition unit corresponding to the first feature extraction network is used to process the second result and the third feature to obtain a first fused feature; the second feature superposition unit corresponding to the second feature extraction network is used to process the first result and the fourth feature to obtain a second fused feature; The fusion unit is used to fuse and process the first fused feature and the second fused feature to obtain the fused feature.

3. The model according to claim 2, characterized in that, The first convolutional network and the second convolutional network include a batch normalization layer, a one-dimensional convolution, and an activation function; the feature extraction network includes a one-dimensional depth convolution layer, a batch normalization layer, a one-dimensional convolution layer, and an activation function layer.

4. The model according to claim 1, characterized in that According to the input-output relationship, the dot product feature extraction sub-module sequentially includes a one-dimensional depth convolutional layer, a batch normalization layer, a first fully connected layer, a second fully connected layer, a first activation function layer connected to the first fully connected layer, a third fully connected layer, a batch normalization layer, a one-dimensional depth convolution, a stochastic depth layer, a one-dimensional average pooling layer, a fourth fully connected layer, a second activation function layer, a fifth fully connected layer, and an activation function layer; Among them, the fused feature is processed by the one-dimensional depth convolutional layer and the batch normalization layer in sequence to obtain a normalized feature; Based on the first fully connected layer and the second fully connected layer respectively processing the normalized feature, a fifth feature corresponding to the first fully connected layer and a sixth feature corresponding to the second fully connected layer are obtained; Based on the first activation function layer processing the fifth feature, a first dot product feature is obtained, and the sixth feature and the first dot product feature are dot product processed to obtain a seventh feature to be input to the third fully connected layer; Based on the third fully connected layer processing the seventh feature, an eighth feature is obtained; The eighth feature is processed by the batch normalization layer, the one-dimensional depth convolutional layer, and the stochastic depth layer in sequence to obtain a superimposed feature, and the superimposed feature and the fused feature are fused to obtain a feature to be processed; The feature to be processed is processed by the one-dimensional average pooling layer, the fourth fully connected layer, the second activation function layer, the fifth fully connected layer, and the activation function layer in sequence to obtain a second dot product feature; By performing dot product processing on the second dot product feature and the feature to be processed, a conversion feature to be input to the additive attention mechanism feature extraction sub-module is obtained.

5. The model according to claim 1, wherein The additive attention mechanism feature extraction sub-module includes an efficient additive attention mechanism unit, a first addition and normalization processing unit, a feature extraction sub-network, and a second addition and normalization processing unit; Among them, the efficient additive attention mechanism unit is used to process the conversion feature to obtain an attention feature; The first addition and normalization processing unit is used to perform addition and normalization processing on the attention feature and the conversion feature to obtain a feature to be extracted to be input to the feature extraction sub-network; The feature extraction sub-network is used to extract the feature to be extracted to obtain a feature to be normalized, where the feature extraction sub-network includes a one-dimensional convolutional layer, an activation function layer, a dropout layer, and a one-dimensional convolutional layer; The second addition and normalization processing unit is used to perform feature addition and normalization processing on the feature to be normalized and the feature to be extracted to obtain the feature to be used.

6. The model according to claim 1, wherein The in-phase and quadrature signal corresponding to the signal to be recognized is an I / Q signal, and the phase-amplitude signal corresponding to the signal to be recognized is an A / P signal.

7. A model training method, characterized in that, Including the modulation recognition model according to any one of claims 1-6, the training method includes: When the modulation recognition model corresponding to the current training round is detected, obtain a plurality of test samples, where the current training round is when a preset number of training samples have participated in the training of the modulation recognition model, and the test samples and the training samples include signals to be recognized and corresponding modulation methods; among them, the signal to be recognized is an in-phase quadrature signal; Determine the accuracy rate corresponding to the modulation recognition model in the current training round based on the plurality of test samples; If the accuracy rates in the current training round and the preset number of polling times before the current training round do not meet the preset conditions, adjust the learning rate of the modulation recognition model, and continue to train the modulation recognition model with the adjusted learning rate based on a preset number of training samples until the accuracy rates in multiple consecutive training rounds all meet the preset conditions, and use the learning rate obtained when the preset conditions are met as the target learning rate of the modulation recognition model, and continue to perform model training based on the modulation recognition model with the target learning rate.

8. The method according to claim 7, wherein The modulation methods include double-sideband amplitude modulation, single-sideband amplitude modulation, binary phase shift keying, quadrature phase shift keying, 8-phase shift keying, 8-quadrature amplitude modulation, 16-quadrature amplitude modulation, 16-phase shift keying, 64-quadrature amplitude modulation, Gaussian minimum frequency shift keying, and orthogonal frequency division multiplexing.

9. The method according to claim 7, characterized in that, The modulation recognition model corresponding to the current training round is trained based on the following method: Obtain a preset number of training samples corresponding to the current training round; For the training samples, input the training samples into the modulation recognition model for modulation method recognition to obtain an output result; Determine a loss value according to the output result and the modulation method in the training samples, and perform parameter correction on the modulation recognition model based on the loss value; When all the preset number of training samples have participated in the training of the modulation recognition model, obtain the modulation recognition model corresponding to the current training round.

10. A method for identifying the modulation mode of multi-channel signals, including the modulation recognition model according to any one of claims 1-6, characterized in that, The method further includes: Obtain the in-phase quadrature signal and the phase amplitude signal of the signal to be recognized with the modulation method to be recognized in different representation forms; among them, the phase amplitude signal is a signal obtained by converting the in-phase quadrature signal; Input the in-phase quadrature signal and the phase amplitude signal into a pre-trained modulation method recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signal to be recognized under different modulation methods; Determine the target modulation method of the signal to be recognized based on the evaluation attributes.

Images