Modulation signal identification model for signal identification and model training method

Abstract

The embodiment of the invention provides a modulation signal identification model for signal identification and a model training method. The model comprises an input module, a first feature extraction module, a second feature extraction module, a first one-dimensional inverse residual module connected with the first feature extraction module, a second one-dimensional inverse residual module connected with the second feature extraction module, a superposition module, a third one-dimensional inverse residual module, a fourth one-dimensional inverse residual module and a global average pooling module. And a classifier module and an output module. According to the technical scheme provided by the embodiment of the invention, the modulation signal identification model for signal identification is constructed, a residual structure is widely used, information is effectively extracted with less parameter quantity, the model is lightened, the modulation signal identification model extracts corresponding information from high-order features and de-noising is carried out based on wavelet transform, and the identification accuracy of the modulation signal identification model is improved. The influence of noise on the signal form is reduced, and the performance of a modulation signal recognition 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 signal recognition model for signal recognition and a model training method. Background Art

[0002] With the rapid development of wireless communication technology, numerous modulation techniques have been widely integrated into various communication scenarios to adapt to the differentiated requirements of different services in terms of transmission rate, spectrum utilization rate, and anti-interference performance. Against the backdrop of this diverse communication means, receiving-end devices must accurately identify the modulation type of the received signal to ensure that correct demodulation operations can be performed, thereby restoring the original communication information.

[0003] Currently, existing technical means for determining the modulation mode of communication signals include methods based on decision theory and statistical pattern recognition techniques, etc. However, these technical paths have several limitations: on the one hand, they often rely on complex mathematical derivations and large amounts of calculations, and are prone to errors during parameter estimation; on the other hand, when there is a mismatch between the preset model and the actual channel characteristics, their performance will be significantly impaired. In addition, these methods usually require relying on additional training data sets, which increases the complexity of engineering implementation to a certain extent. More critically, in a low signal-to-noise ratio environment, their recognition accuracy will drop sharply. Summary of the Invention

[0004] Embodiments of the present disclosure provide a modulation signal recognition model for signal recognition and a model training method, which widely use residual structures to effectively extract information with fewer parameters, making the model lightweight. The modulation signal recognition model extracts corresponding information from high-order features and denoises based on wavelet transform, reducing the influence of noise on the signal form, and significantly improving the performance of the modulation signal recognition model.

[0005] In a first aspect, embodiments of the present disclosure provide a modulation signal recognition model for signal recognition, and the model includes:

[0006] According to the input-output relationship, the modulation signal recognition model includes: an input module, a first feature extraction module, a second feature extraction module, a first one-dimensional inverted residual module connected to the first feature extraction module, a second one-dimensional inverted residual module connected to the second feature extraction module, a stacking module, a third one-dimensional inverted residual module, a fourth one-dimensional inverted residual module, a global average pooling module, a classifier module, and an output module. The network architectures of the first feature extraction module and the second feature extraction module are the same;

[0007] Among them, the input module is used to receive the in-phase signal and the quadrature signal of the same signal to be recognized in different representation forms. The first feature extraction module is used to process the in-phase signal, and the second feature extraction module is used to process the quadrature signal. The feature extraction module includes a wavelet transform convolution denoising sub-module, a high-order central moment feature extraction module, and a feature extraction network. The wavelet transform denoising sub-module is used to denoise the recognized quadrature signal or in-phase signal. The high-order central moment feature extraction module is used to determine the eigenvalue data of the denoised quadrature signal or in-phase signal in multiple signal detection dimensions. The feature extraction network is used to extract the feature information in the feature data. The one-dimensional inverted residual module is used to process the input feature residual to obtain the feature to be used. The classifier module is used to determine the evaluation attributes corresponding to the signal to be recognized under different modulation methods. The output module is used to output the evaluation attributes processed by the classification module.

[0008] In a second aspect, an embodiment of the present invention further provides a method for model training, which includes:

[0009] When it is detected that the modulation signal recognition model corresponding to the current training round is obtained, a plurality of test samples are obtained, where the current training round is that a preset number of training samples have participated in the training of the modulation signal 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;

[0010] Based on the plurality of test samples, determine the accuracy rate of the modulation signal recognition model corresponding to the current training round;

[0011] 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 signal recognition model to continue training the modulation signal recognition model based on a preset number of training samples until the accuracy rates of multiple consecutive training rounds meet the preset conditions, and use the target learning rate obtained when the preset conditions are met as the learning rate used when the modulation signal recognition model participates in training, and continue to perform model training on the modulation signal recognition model.

[0012] In a third aspect, an embodiment of the present invention further provides a method for identifying the modulation methods of multiple signals, which includes:

[0013] Obtain the in-phase signal and the quadrature signal of the signal to be recognized with the modulation method to be recognized in different representation forms; among them, the in-phase signal and the quadrature signal are signals split from the in-phase quadrature signal;

[0014] Input the in-phase signal and the quadrature signal into a pre-trained modulation signal recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signal to be recognized under different modulation methods; wherein, the evaluation attributes are used to characterize the credibility under the corresponding modulation mode.

[0015] Based on the evaluation attributes, determine the target modulation method of the signal to be recognized.

[0016] In a fourth aspect, an embodiment of the present invention further provides a device for model training, which includes:

[0017] A test sample acquisition module, an accuracy determination module, and a model training module.

[0018] The test sample acquisition module is configured to obtain a plurality of test samples when detecting the modulation signal recognition model corresponding to the current training round, where the current training round is that a preset number of training samples have participated in the training of the modulation signal 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.

[0019] The accuracy determination module is configured to determine the accuracy corresponding to the modulation signal recognition model in the current training round based on the plurality of test samples.

[0020] The model training module is configured to, if the accuracies 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 signal recognition model to continue training the modulation signal recognition model based on a preset number of training samples until the accuracies of multiple consecutive training rounds meet the preset conditions, and use the target learning rate obtained when meeting the preset conditions as the learning rate used when the modulation signal recognition model participates in training, and continue to perform model training on the modulation signal recognition model.

[0021] In a fifth aspect, an embodiment of the present invention further provides a device for identifying the modulation methods of multiple signals, which includes:

[0022] A signal to be recognized acquisition module, an evaluation attribute output module, and a target modulation method determination module.

[0023] The signal to be recognized acquisition module is configured to obtain the signal to be recognized with the modulation method to be recognized, the in-phase signal and the quadrature signal in different representation forms; wherein, the in-phase signal and the quadrature signal are signals split from the in-phase quadrature signal.

[0024] An evaluation attribute output module, configured to input the in-phase signal and the quadrature signal into a pre-trained modulation signal recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signal to be recognized under different modulation methods; wherein, the evaluation attributes are used to characterize the credibility under the corresponding modulation mode.

[0025] A target modulation method determination module, configured to determine the target modulation method of the signal to be recognized based on the evaluation attributes. In a sixth aspect, an embodiment of the present invention further provides an electronic device, where 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 containing 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] According to the input-output relationship, the technical solution of the embodiment of the present disclosure includes: an input module, a first feature extraction module, a second feature extraction module, a first one-dimensional inverted residual module connected to the first feature extraction module, a second one-dimensional inverted residual module connected to the second feature extraction module, a stacking module, a third one-dimensional inverted residual module, a fourth one-dimensional inverted residual module, a global average pooling module, a classifier module, and an output module. The network architectures of the first feature extraction module and the second feature extraction module are the same. When determining the signal modulation type based on methods such as decision theory and statistical pattern recognition, it depends on complex mathematical derivations and large amounts of calculations, and errors are likely to occur during the parameter estimation process. When there is a mismatch between the preset model and the actual channel characteristics, the performance will be significantly impaired, and it depends on additional training data sets. In a low signal-to-noise ratio environment, the recognition accuracy will decrease sharply. The embodiment of the present invention constructs a modulation signal recognition model for signal recognition, widely uses the residual structure, effectively extracts information with fewer parameters, makes the model lightweight, the modulation signal recognition model extracts corresponding information from high-order features and denoises based on wavelet transform, reduces the influence of noise on the signal form, and significantly improves the performance of the modulation signal 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, without creative efforts, other drawings can also be obtained based on these drawings

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

[0034] Figure 2 is a schematic flowchart of the modulation signal recognition model for signal recognition provided by the embodiment of the present disclosure;

[0035] Figure 3 is a schematic flowchart of the wavelet transform convolution denoising sub-module provided by the embodiment of the present disclosure;

[0036] Figure 4 is a schematic flowchart of the convolution denoising unit provided by the embodiment of the present disclosure;

[0037] Figure 5 is a schematic flowchart of the one-dimensional inverted residual module provided by the embodiment of the present disclosure;

[0038] Figure 6 is a schematic flowchart of the training of the modulation signal recognition model provided by the embodiment of the present disclosure;

[0039] Figure 7 It is a schematic flow chart of the application of the modulation signal recognition model provided by the embodiments of the present disclosure;

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

[0041] Figure 9 It is a schematic structural diagram of the device for identifying the modulation modes of multiple signals provided by the embodiments of the present disclosure;

[0042] Figure 10 It is a schematic structural diagram of an electronic device provided by the embodiments 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 drawings.

[0044] Before introducing the technical solutions provided by the embodiments of the present disclosure, the application scenarios can be exemplarily described. The technical solutions provided by the embodiments of the present disclosure can be applied to the scenario of determining the modulation mode of a signal to be recognized. For example, in the scenario of constructing a modulation signal recognition model for signal recognition based on the input-output relationship and obtaining the modulation mode of the signal to be recognized based on the constructed and trained modulation signal recognition model for signal recognition. Based on the technical solutions of the embodiments of the present disclosure, the residual structure is widely used, and information can be effectively extracted with fewer parameters, making the model lightweight. The modulation signal recognition model extracts corresponding information from high-order features and denoises based on wavelet transform, reducing the influence of noise on the signal form and significantly improving the performance of the modulation signal recognition model.

[0045] Embodiment 1

[0046] Figure 1 It is a schematic diagram of the modulation signal recognition model for signal recognition provided by the embodiments 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 Figure 1 understand the model architecture of the modulation signal recognition model for signal recognition provided by the embodiments of the present invention.

[0048] According to the input-output relationship, the modulation signal recognition model includes: an input module 10, a first feature extraction module 20, a second feature extraction module 30, a first one-dimensional inverted residual module 40 connected to the first feature extraction module 20, a second one-dimensional inverted residual module 50 connected to the second feature extraction module 30, a stacking module 60, a third one-dimensional inverted residual module 70, a fourth one-dimensional inverted residual module 80, a global average pooling module 90, a classifier module 100, and an output module 110. The network architectures of the first feature extraction module 20 and the second feature extraction module 30 are the same.

[0049] Among them, the input module is used to receive the in-phase signal and the quadrature signal of the same signal to be recognized in different representation forms. The first feature extraction module is used to process the in-phase signal, and the second feature extraction module is used to process the quadrature signal. The feature extraction module includes a wavelet transform convolutional denoising sub-module, a high-order central moment feature extraction module, and a feature extraction network. The wavelet transform convolutional denoising sub-module is used to denoise the recognized quadrature signal or in-phase signal. The high-order central moment feature extraction module is used to determine the eigenvalue data of the denoised quadrature signal or in-phase signal in multiple signal detection dimensions. The feature extraction network is used to extract the feature information in the feature data. The one-dimensional inverted residual module is used to process the input feature residuals to obtain the features to be used. The classifier module is used to determine the evaluation attributes corresponding to the signal to be recognized under different modulation methods. The output module is used to output the evaluation attributes processed by the classification module.

[0050] It should be noted that the signal to be recognized can be an in-phase quadrature signal, that is, an I / Q signal. The in-phase quadrature signal consists of an in-phase component and a quadrature component. The in-phase component is the baseband signal component that is in phase with the reference carrier signal. The quadrature component is the baseband signal component that is 90° out of phase with the reference carrier signal. The I / Q signal is represented by a two-dimensional coordinate system (I-Q plane) and can completely describe the amplitude and phase information of the signal, which is widely used in the field of wireless communication. The in-phase signal usually refers to the in-phase component (I), that is, the baseband signal component that is in phase with the reference carrier signal. The in-phase signal only contains the amplitude information of the signal and has no phase shift. The quadrature signal usually refers to the quadrature component (Q), that is, the baseband signal component that is 90° out of phase with the reference carrier signal. The quadrature signal is orthogonal to the in-phase signal and is used to describe the phase information of the signal.

[0051] It should also be noted that when obtaining the in-phase signal from the in-phase quadrature signal, the in-phase component in the in-phase quadrature signal can be directly extracted, that is, the real part of the in-phase quadrature signal; when obtaining the quadrature signal from the in-phase quadrature signal, the quadrature component in the in-phase quadrature signal can be directly extracted, that is, the imaginary part of the in-phase quadrature signal.

[0052] In this embodiment, see Figure 2, the first one-dimensional inverted residual module, the second one-dimensional inverted residual module, the third one-dimensional inverted residual module, and the fourth one-dimensional inverted residual module are IR1D blocks. The wavelet transform convolution denoising sub-module is WTConvdenoise. The high-order central moment feature extraction module is High-order moments compute.

[0053] It should be noted that multiple signal detection dimensions include the feature mean dimension, the feature variance dimension, the feature skewness dimension, and the feature kurtosis dimension. The eigenvalue data refers to the data obtained after the high-order central moment feature extraction module processes the denoised orthogonal signal or in-phase signal. The features to be used refer to the features obtained after residual processing of the input features. The modulation methods can 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 shift keying, and orthogonal frequency division multiplexing. Among them, 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 (the upper sideband or the 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 the data is encoded into 8 different symbols by adjusting 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 binary data is represented by using 16 different phase states, 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 preprocessed by a Gaussian filter and then minimum shift keying modulation is performed; orthogonal frequency division multiplexing means that the high-speed data stream is divided into multiple low-speed subcarriers, each subcarrier is independently modulated, and the subcarriers are orthogonal to each other. The evaluation attribute refers to the credibility used to characterize the corresponding modulation mode. In this embodiment, the evaluation attribute is often used to characterize the probability of the signal under the corresponding modulation mode. Its essence is a probabilistic measure, reflecting the probability values of the signal being different modulation modes.

[0054] Optionally, the wavelet transform convolution denoising sub-module includes: a discrete wavelet transform unit, a plurality of convolution denoising units, and an inverse discrete wavelet transform unit. The model structures of the plurality of convolution denoising units are the same; the corresponding components are denoised based on the convolution denoising units corresponding to each low-frequency component and high-frequency component, and the features to be inverse-processed to be input into the inverse discrete wavelet transform unit are obtained; the inverse discrete wavelet transform unit is used to perform inverse transform processing on all the features to be inverse-processed to obtain the features to be center-transformed to be input into the high-order central moment feature extraction module.

[0055] Among them, the discrete wavelet transform unit is used to process the signal to be recognized into a first preset number of low-frequency components and a second preset number of high-frequency components. It should be noted that the discrete wavelet transform unit decomposes the signal to be recognized into sub-band signals of different frequency bands through a filter bank (low-pass filter and high-pass filter). The sub-band signals include low-frequency components and high-frequency components. The low-frequency components contain the approximate information of the signal, such as trends and contours. The high-frequency components contain the detailed information of the signal, such as edges and mutations. The signal to be recognized passes through a low-pass filter (generating low-frequency components) and a high-pass filter (generating high-frequency components), and then downsampling (skipping sampling) is performed to reduce the amount of data. The first preset number refers to the number of low-frequency components. The first preset number is determined by the decomposition level. If the decomposition level is L, then L low-frequency components are generated. The low-frequency components are used to extract the global features of the signal. The second preset number refers to the number of high-frequency components. Each layer of decomposition generates one high-frequency component. If the decomposition level is L, then L high-frequency components are generated. The high-frequency components are used to extract the local features of the signal. In this embodiment, through the discrete wavelet transform unit, the signal to be recognized can be processed into one low-frequency component and multiple high-frequency components.

[0056] It should be noted that, referring to Figure 3, the convolutional denoising unit can be used to denoise the low-frequency component and the high-frequency component. The low-frequency component and the high-frequency component after passing through are called the features to be inverse-processed. The inverse discrete wavelet transform unit is used to reconstruct the low-frequency and high-frequency components after wavelet decomposition and convolutional denoising into the original signal or specified features. The inverse discrete wavelet transform unit can not only recover the original signal from the low-frequency component and the high-frequency component, but also inverse-transform the decomposed features into the input format required by the high-order central moment feature extraction module. The inverse discrete wavelet transform unit is the inverse operation of the discrete wavelet transform unit, and synthesizes the low-frequency and high-frequency components into the original signal through upsampling and convolutional filtering. For example, after the inverse discrete wavelet transform unit receives the low-frequency component (such as A3) and the high-frequency components (such as D3, D2, D1) decomposed by the discrete wavelet transform unit, starting from the highest layer, the low-frequency component and the high-frequency components are reconstructed into the next-layer low-frequency component through the inverse filter bank. This process is repeated until the original signal or the target feature is reconstructed. After obtaining the reconstructed signal or feature, after the central transformation of the feature, that is, the reconstructed signal or feature is used as the input of the subsequent high-order central moment feature extraction module. When the inverse discrete wavelet transform unit is processing, first, for the downsampling step, the data length can be extended to twice the original. Then, the upsampled components are convolved using the reconstruction filters h~ and g~, where h~ and g~ are the time-domain inverted versions of the decomposition filters h and g. Further, the convolution results of the low-frequency component and the high-frequency component can be added and combined to obtain the next-layer low-frequency component.

[0057] In this embodiment, referring to Figure 3 , the discrete wavelet transform unit is DWT, the convolutional denoising unit is Conv_denoise, and the inverse discrete wavelet transform unit is IDWT.

[0058] In this embodiment, according to the input-output relationship, the convolutional denoising unit includes a one-dimensional convolutional sub-unit with the first convolutional kernel size, a first activation function layer, a one-dimensional convolutional sub-unit with the second convolutional kernel size, a one-dimensional convolutional sub-unit with the first convolutional kernel size, a second activation function layer, and a splicing processing layer for splicing the components output by the discrete wavelet transform and the features output by the second activation function.

[0059] Among them, the size of the convolution kernel with the first convolution kernel size can be 1. The size of the convolution kernel with the second convolution kernel size can be 7. The one-dimensional convolution sub-unit with the first convolution kernel size can be used to extract the local features of the input signal. The first activation function layer can introduce non-linearity to enhance the model's expression ability. The one-dimensional convolution sub-unit with the second convolution kernel size can further extract the more advanced features in the signal. The second one-dimensional convolution sub-unit with the first convolution kernel size can restore or adjust the feature dimension to prepare for subsequent splicing. The second activation function layer enhances the feature expression ability by introducing non-linearity again. The splicing processing layer can splice the components output by the discrete wavelet transform and the features output by the second activation function. During the splicing process, the components output by the discrete wavelet transform and the features output by the second activation function can be merged in the channel dimension, and the resulting feature to be center-transformed not only has the features of the components output by the discrete wavelet transform but also includes the features output by the second activation function.

[0060] In this embodiment, referring to Figure 4 , the one-dimensional convolution sub-unit with the first convolution kernel size is Conv1D, the first activation function layer is ReLU, the one-dimensional convolution sub-unit with the second convolution kernel size is Conv1D, the one-dimensional convolution sub-unit with the first convolution kernel size is Conv1D, and the second activation function layer is ReLU.

[0061] Optionally, the high-order central moment feature extraction module extracts the feature information of the feature to be center-transformed under at least one signal detection dimension; wherein, the feature to be center-transformed is the feature output by the wavelet transform convolution denoising sub-module.

[0062] In this embodiment, at least one signal detection dimension includes the feature mean dimension, the feature variance dimension, the feature skewness dimension, and the feature kurtosis dimension. Determining the feature information under at least one signal detection dimension based on the high-order central moment feature extraction module includes: dividing the feature to be center-transformed into multiple groups of features to be processed according to the preset sliding window length and window adjustment step size, and respectively calculating the target feature values corresponding to the feature values in each group of features to be processed under each signal detection dimension; based on the target feature values of each group of features to be processed under each signal detection dimension and the preset feature length information, determining the matrix of features to be extracted corresponding to the feature to be center-transformed.

[0063] It should be noted that the high-order central moment feature extraction module is a statistic that describes the shape of the probability distribution. Feature information can be calculated based on the feature to be center-transformed, the mean of the feature to be center-transformed, the preset sliding window length, and the order of the high-order central moment. The feature information in the feature mean dimension represents the first-order central moment, and the feature information in the feature mean dimension is always 0; the feature information in the feature variance dimension represents the second-order central moment, which describes the degree of dispersion of the feature to be center-transformed; the feature information in the feature skewness dimension represents the third-order central moment, which describes the symmetry of the distribution of the feature to be center-transformed; the feature information in the feature kurtosis dimension represents the fourth-order central moment, which describes the steepness of the distribution of the feature to be center-transformed.

[0064] It should also be noted that when calculating the feature information of the feature to be center-transformed in the four signal detection dimensions, the feature information in different signal detection dimensions at multiple certain signal lengths can be calculated. The preset sliding window length refers to the number of features included in each window when calculating the feature information in different signal detection dimensions each time, that is, the length of the amount of data calculated. For example, the preset sliding window length can be 7. The window adjustment step size refers to the number of features that the window moves each time. For example, the window adjustment step size can be 1. Starting from the starting position of the feature to be center-transformed, the feature values of the preset sliding window length can be extracted, and the feature values of the preset sliding window length are called a group of features to be processed. According to the preset sliding window length and the window adjustment step size, the feature to be center-transformed can be divided into multiple groups of features to be processed. For example, the first to seventh feature values in the feature to be center-transformed can be the first group of features to be processed, and the second to eighth feature values in the feature to be center-transformed can be the second group of features to be processed. After obtaining multiple groups of features to be processed according to the preset sliding window length and the window adjustment step size, the target feature value refers to the result corresponding to the feature values in each group of features to be processed in each signal detection dimension. The preset feature length information refers to the number of feature values in the feature to be center-transformed. After determining the target feature values of each group of features to be processed in each signal detection dimension, the number of feature values in the target feature values can be obtained. After calculating the difference between the number of feature values in the feature to be center-transformed and the number of feature values in the target feature values, the number of missing feature values of the target feature values relative to the number of feature values in the feature to be center-transformed can be determined. The number of missing feature values in the target feature values can be supplemented with 0, and the number of feature values in the target feature values can be restored to the number of feature values in the feature to be center-transformed. Dividing the feature to be center-transformed into multiple groups of features to be processed according to the preset sliding window length and the window adjustment step size helps to perform segmented analysis on the feature to be center-transformed, capture local features, and provide input for subsequent statistical calculations, feature extraction, or machine learning model training.

[0065] Among them, the matrix to be feature-extracted is an M×N matrix, where M is consistent with the number of at least one signal detection dimension, and N is consistent with the preset feature length information. For example, when the signal detection dimensions include the feature mean dimension, the feature variance dimension, the feature skewness dimension, and the feature kurtosis dimension, M in the matrix to be feature-extracted is 4. When the preset feature length information is 128, then N in the matrix to be feature-extracted is 128.

[0066] In this embodiment, the feature extraction network based on the input-output relationship includes a one-dimensional convolution with a second convolution kernel size, a first batch normalization layer, and a third activation function layer.

[0067] It should be noted that in the one-dimensional convolution with the second convolution kernel size, the convolution kernel size can be 7, the padding length can be 3, and the stride can be 2. The convolution kernel is a set of learnable weight parameters used to slide on the input sequence and extract local features. A larger convolution kernel can capture more extensive context information but also increases the computational complexity. The padding length is to add additional elements (usually 0) at both ends of the input sequence to control the length of the output sequence. When the padding length is 3, it means adding 3 zeros at both ends of the input sequence. A larger padding length can retain more edge information but may introduce noise. The stride represents the step size at which the convolution kernel slides on the input sequence. When the stride is 2, it means the convolution kernel slides 2 elements each time. A larger stride reduces the number of convolution operations, thereby reducing the computational complexity. The larger the stride, the shorter the output sequence length. And the one-dimensional convolution with the second convolution kernel size in the feature extraction network can further extract more advanced features in the matrix to be feature-extracted. The first batch normalization layer is mainly used to accelerate the training process of the neural network, improve the stability and generalization ability of the model. The batch normalization layer normalizes the distribution of each small batch of data so that the mean of the data is 0 and the variance is 1. Its core idea is to reduce the "internal covariate shift", that is, the phenomenon that the input distribution of each layer in the network changes during training. The specific implementation steps of the batch normalization layer include: calculating the mean and variance of each small batch of data. Then, subtracting the mean from the data and dividing by the variance to standardize the data distribution. Finally, introducing learnable parameters γ and β to scale and shift the normalized data to restore the expressive power of the network. The third activation function layer enables the network to learn complex feature representations. In this embodiment, the third activation function layer can be ReLU. ReLU has the advantages of simplicity and efficiency and can alleviate the gradient vanishing problem.

[0068] Optionally, according to the input-output relationship, the one-dimensional inverted residual module includes a first processing unit, a second processing unit for processing the output result of the first processing unit, a third processing unit for performing dot product processing on the output results of the first processing unit and the second processing unit, a fourth unit for processing the output result of the third processing unit, and a fifth processing unit for adding the features input to the first processing unit and the features output by the fourth unit.

[0069] Optionally, according to the input-output relationship, the first processing unit includes a one-dimensional convolution of a second convolution size, a second batch normalization layer, a fourth activation function layer, a one-dimensional depth convolution of a first convolution size, and a third batch normalization layer; the second processing unit includes a first linear layer, a fifth activation function layer, a second linear layer, and a sixth activation function layer; the fourth unit includes a one-dimensional convolution of a second convolution kernel size and at least two fourth batch normalization layers.

[0070] In this embodiment, referring to Figure 5 , the one-dimensional convolution of the second convolution size in the first processing unit is Conv1D, the second batch normalization layer is BN, the fourth activation function layer is ReLU, the one-dimensional depth convolution of the first convolution size is DWConv1D, and the third batch normalization layer is BN. The first linear layer in the second processing unit is Linear, the fifth activation function layer is ReLU, the second linear layer is Linear, and the sixth activation function layer is Hsigmoid. The one-dimensional convolution of the second convolution kernel size in the fourth unit is Conv1D, and the at least two fourth batch normalization layers are BN.

[0071] It should be noted that in the first convolution size of the one-dimensional depth convolution, the convolution kernel size can be 1, the padding length can be 3, and the stride can be 2. It should also be noted that the one-dimensional depth convolution is a special convolution operation, where each input channel is convolved independently without mixing information from different channels. Through the one-dimensional depth convolution, the computational efficiency can be improved by reducing the number of parameters and the amount of computation, while retaining the feature extraction ability. The linear layer, as a basic component in the neural network, is used to perform a linear transformation on the input data. The linear layer maps the input features to the output feature space by learning a set of weights and biases. The Hsigmoid, as an activation function, is implemented by a piecewise linear function and has the characteristic of high computational efficiency.

[0072] Specifically, a modulation signal recognition model for signal recognition is constructed. The constructed modulation signal recognition model includes an input module, a first feature extraction module, a second feature extraction module, a first 1D inverted residual module connected to the first feature extraction module, a second 1D inverted residual module connected to the second feature extraction module, a superimposition module, a third 1D inverted residual module, a fourth 1D inverted residual module, a global average pooling module, a classifier module, and an output module. Based on the constructed modulation signal recognition model for signal recognition, subsequent model training and model application can be performed to determine the modulation method of the signal to be recognized.

[0073] According to the technical solution of the embodiment of the present disclosure, based on the input-output relationship, the modulation signal recognition model includes: an input module, a first feature extraction module, a second feature extraction module, a first 1D inverted residual module connected to the first feature extraction module, a second 1D inverted residual module connected to the second feature extraction module, a superimposition module, a third 1D inverted residual module, a fourth 1D inverted residual module, a global average pooling module, a classifier module, and an output module. The network architectures of the first feature extraction module and the second feature extraction module are the same, which solves the problems that when determining the signal modulation type based on methods such as decision theory and statistical pattern recognition, it depends on complex mathematical derivations and huge computational amounts, and errors are likely to occur during the parameter estimation process. When there is a mismatch between the preset model and the actual channel characteristics, the performance will be significantly impaired, and it depends on an additional training data set. In a low signal-to-noise ratio environment, the recognition accuracy will drop sharply. In the embodiment of the present invention, a modulation signal recognition model for signal recognition is constructed, and the residual structure is widely used to effectively extract information with fewer parameters, making the model lightweight. The modulation signal recognition model extracts corresponding information from high-order features and denoises based on wavelet transform to reduce the influence of noise on the signal form, and significantly improves the performance of the modulation signal recognition model.

[0074] Embodiment 2

[0075] Figure 6 It is a schematic flowchart of the training of the modulation signal recognition model provided by the embodiment of the present invention. On the basis of the foregoing embodiment, the training of the modulation signal recognition model is described in detail. The training method of the modulation signal recognition model can be executed by a training device of the modulation signal 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 signal recognition model provided by the technical solution of the present application. 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.

[0076] As Figure 6 shown, the method specifically includes the following steps:

[0077] S210. When the modulation signal recognition model corresponding to the current training round is detected, obtain a plurality of test samples.

[0078] Among them, the current training round means that a preset number of training samples have participated in the training of the modulation signal recognition model. The test samples and the training samples include the signal to be recognized and the corresponding modulation method. The signal to be recognized is an in-phase quadrature signal.

[0079] It should be noted that during the training process of the modulation signal recognition model, the modulation signal recognition model can be optimized through multiple training rounds. The training round refers to the process in which the entire training data set is completely traversed once by the modulation signal recognition model. Each complete traversal can be regarded as the modulation signal recognition model learning all the training samples and gradually adjusting the parameters to optimize the objective function. Under multiple training rounds, each training round will update the model parameters in the modulation signal recognition model using the training samples. The current training round refers to the training round of the current training modulation signal recognition model. The training samples are the samples used for training the modulation signal recognition model. The test samples are the samples used for testing the accuracy of the modulation signal recognition model after each training round. The preset number refers to the total number of samples in the training data set, or the number of samples in a specific subset, such as the batch size or the number of mini-batch samples.

[0080] Specifically, when training the modulation signal recognition model, at the current training round, obtain a plurality of test samples. Moreover, the test samples and the training samples include in-phase quadrature signals and the modulation methods corresponding to the in-phase quadrature signals.

[0081] S220. Determine the accuracy rate corresponding to the modulation signal recognition model at the current training round based on the plurality of test samples.

[0082] Among them, the accuracy rate is the core index for evaluating the performance of the modulation signal recognition model, indicating the proportion of the number of samples correctly predicted by the modulation signal recognition model to the total number of samples.

[0083] Specifically, after inputting the plurality of test samples into the modulation signal recognition model, the accuracy rate corresponding to the modulation signal 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.

[0084] 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, adjust the learning rate of the modulation signal recognition model, and continue to train the modulation signal recognition model based on a preset number of training samples until the accuracy rates of multiple consecutive training rounds meet the preset conditions. Then, use the target learning rate obtained when the preset conditions are met as the learning rate for the modulation signal recognition model during training, and continue to perform model training on the modulation signal recognition model.

[0085] Among them, the preset polling times refer to the fixed evaluation intervals preset during the training of the modulation signal recognition model for evaluating the performance of the modulation signal recognition model. The polling times mean that every N training iterations, the performance of the modulation signal recognition model on the test set is evaluated once. Through regular evaluation, observe whether the performance of the modulation signal recognition model improves with the training rounds. When the accuracy rates of the current training round and the preset polling times before the current training round do not reach the preset conditions, the learning rate of the modulation signal recognition model can be adjusted. For example, 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, the learning rate of the modulation signal recognition model can be multiplied by 0.8.

[0086] Optionally, the modulation recognition model corresponding to the current training round is trained in the following manner: 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.

[0087] Among them, in order to improve the accuracy of model training, training samples of different modulation methods can be obtained. It should be noted that when the training samples are input into the modulation signal recognition model for modulation method recognition, the model parameters in the modulation recognition model do not meet the expected requirements. Therefore, there is a certain difference between the modulation method corresponding to the actual evaluation attribute output based on the model parameters at this time and the theoretical modulation method. Therefore, based on the modulation method corresponding to the actual evaluation attribute of each training sample and the theoretical modulation method, the corresponding error loss value can be determined. It should be noted that the training parameters can be set to default values before training the model. 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, an 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 method corresponding to the actual evaluation attribute of the training sample and the theoretical modulation method. Specifically, the training error of the loss function, that is, the loss parameter, can be used as the condition for detecting whether the loss function has reached convergence, 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 detection reaches the convergence condition, 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.

[0088] In the technical solution of this embodiment of the present disclosure, when it is detected that the modulation signal recognition model corresponding to the current training round is obtained, a plurality of test samples are acquired. Then, based on the plurality of test samples, the accuracy rate corresponding to the modulation signal recognition model in the current training round is determined. Finally, 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, the learning rate of the modulation signal recognition model is adjusted to continue training the modulation signal recognition model based on a preset number of training samples until the accuracy rates of multiple consecutive training rounds meet the preset conditions, and the target learning rate obtained when the preset conditions are met is used as the learning rate used when the modulation signal recognition model participates in training, and the modulation signal recognition model is continuously trained to improve the training efficiency of the modulation signal recognition model and provide a high-quality and high-stability model for the modulation recognition task.

[0089] Embodiment III

[0090] Figure 7 It is a schematic flowchart of the application of the modulation signal recognition model provided by the embodiment of the present invention. On the basis of the foregoing embodiments, the application of the modulation signal recognition model is described in detail, and 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 embodiments will not be described again here.

[0091] As Figure 7 shown, the method specifically includes the following steps:

[0092] S310. Obtain the signal to be recognized for the modulation method to be recognized, and the in-phase signal and quadrature signal in different representation forms.

[0093] Among them, the in-phase signal and quadrature signal are signals obtained by splitting the in-phase and quadrature signal.

[0094] Specifically, when determining the adjustment method to be recognized for the in-phase and quadrature signal, the in-phase signal and quadrature signal corresponding to the in-phase and quadrature signal can be obtained based on the in-phase and quadrature signal first.

[0095] S320. Input the in-phase signal and quadrature signal into the pre-trained modulation signal recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signal to be recognized under different modulation methods.

[0096] Among them, the evaluation attribute is used to characterize the credibility under the corresponding modulation mode.

[0097] Specifically, based on the pre-trained modulation signal recognition model, the in-phase signal and quadrature signal corresponding to the in-phase and quadrature signal for which the modulation method needs to be determined are input into the modulation signal recognition model, and the probabilities corresponding to the signal to be recognized under different modulation methods are determined.

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

[0099] It should be noted that the maximum probability corresponding to the signal to be recognized under different modulation methods is the target modulation method of the signal to be recognized.

[0100] The technical solution of the embodiment of the present disclosure obtains the signal to be recognized for the modulation method to be recognized, and the in-phase signal and quadrature signal in different representation forms. Then, the in-phase signal and quadrature signal are input into the pre-trained modulation signal recognition model for modulation mode recognition, and the evaluation attributes corresponding to the signal to be recognized under different modulation methods are output. Finally, based on the evaluation attribute, the target modulation method of the signal to be recognized is determined, so as to achieve a higher modulation method recognition accuracy.

[0101] Embodiment 4

[0102] Figure 8 is a schematic structural diagram of the model training device provided by the embodiment of the present disclosure. As Figure 8 shown, the device includes a test sample acquisition module, an accuracy determination module, and a model training module.

[0103] A test sample acquisition module, configured to obtain a plurality of test samples when it is detected that the modulation signal 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 signal 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;

[0104] An accuracy rate determination module, configured to determine the accuracy rate corresponding to the modulation signal recognition model in the current training round based on the plurality of test samples;

[0105] A model training module, configured to, if the accuracy rates of the current training round and a preset number of polling times before the current training round do not meet the preset conditions, adjust the learning rate of the modulation signal recognition model, so as to continue training the modulation signal recognition model based on a preset number of training samples until the accuracy rates of multiple consecutive training rounds all meet the preset conditions, and use the target learning rate obtained when the preset conditions are met as the learning rate used when the modulation signal recognition model participates in training, and continue to perform model training on the modulation signal recognition model.

[0106] Figure 9 It is a schematic structural diagram of a device for identifying modulation methods of multiple signals provided by an embodiment of the present disclosure. As Figure 9 shown, the device includes a signal to be recognized acquisition module, an evaluation attribute output module, and a target modulation method determination module.

[0107] A signal to be recognized acquisition module, configured to obtain signals to be recognized, in-phase signals and quadrature signals of the modulation method to be recognized in different representation forms; wherein, the in-phase signal and the quadrature signal are signals split from the in-phase quadrature signal;

[0108] An evaluation attribute output module, configured to input the in-phase signal and the quadrature signal into a pre-trained modulation signal recognition model for modulation mode recognition, and output evaluation attributes corresponding to the signal to be recognized under different modulation methods; wherein, the evaluation attribute is used to characterize the credibility under the corresponding modulation mode;

[0109] A target modulation method determination module, configured to determine the target modulation method of the signal to be recognized based on the evaluation attribute.

[0110] The model training device provided by the embodiment of the present disclosure can execute the training method of the modulation signal recognition model provided by any embodiment of the present disclosure. The device for identifying modulation methods of multiple signals provided by the embodiment of the present disclosure can execute the method for identifying modulation methods of multiple signals provided by any embodiment of the present disclosure, and has 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 realized; 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 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. The following refers to Figure 10 , which shows a schematic structural diagram of an electronic device 500 suitable for implementing the embodiments of the present disclosure (such as Figure 10 the terminal device or server in). 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 impose any limitations on the functions and usage scope of the embodiments of the present disclosure.

[0114] As Figure 10 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 the program stored in the read-only memory (ROM) 502 or the program loaded from the storage device 508 into the 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. The 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 touch pad, 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 shows the electronic device 500 having various devices, it should be understood that it is not required to implement or have all the shown devices. More or fewer devices may be alternatively implemented or had.

[0116] In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, an embodiment of the present disclosure includes a computer program product that includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes program code for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by a processing device 501, the above functions defined in the methods of the embodiments of the present disclosure are performed.

[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 in the embodiment of the present disclosure and the model training method and the method for identifying the modulation mode of a multiplexed signal provided in the above embodiment belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0119] Embodiment Six

[0120] The embodiment of the present disclosure provides a computer storage medium, on which a computer program is stored, and when the program is executed by a processor, the model training method and the method for identifying the modulation mode of a multiplexed signal provided in the above embodiment are implemented.

[0121] It should be noted that the computer-readable medium described above can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can 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 can include, but are not limited to: an electrical connection with 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 can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present disclosure, a computer-readable signal medium can 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 can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using 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 ("LANs"), wide area networks ("WANs"), the Internet (e.g., the Internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

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

[0124] The above computer-readable medium carries one or more programs, and when the one or more programs are executed by the electronic device, the electronic device is caused to:

[0125] Determine the accuracy rate corresponding to the modulation signal recognition model in the current training round based on the multiple test samples;

[0126] If the accuracy rates in 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 signal recognition model, and continue to train the modulation signal recognition model based on a preset number of training samples until the accuracy rates in multiple consecutive training rounds all meet the preset conditions. Then, use the target learning rate obtained when the preset conditions are met as the learning rate used when the modulation signal recognition model participates in training, and continue to perform model training on the modulation signal recognition model.

[0127] Obtain the in-phase signal and quadrature signal of the signal to be recognized for the modulation method to be recognized in different representation forms; wherein, the in-phase signal and the quadrature signal are signals obtained by splitting the in-phase quadrature signal.

[0128] Input the in-phase signal and the quadrature signal into the pre-trained modulation signal recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signal to be recognized under different modulation methods; wherein, the evaluation attributes are used to characterize the credibility under the corresponding modulation mode.

[0129] Based on the evaluation attributes, determine the target modulation method of the signal to be recognized.

[0130] Computer program code for performing the operations of the present disclosure can be written in one or more programming languages or combinations thereof. The above-mentioned 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 an independent 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 type of network - including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0131] 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 the present disclosure. In this regard, each block in the flowchart or block diagram may 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 noted in the blocks may occur in a different order than noted in the accompanying drawings. For example, two consecutively represented blocks may actually be executed substantially in parallel, or they may 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 that performs the specified functions or operations, or by a combination of dedicated hardware and computer instructions.

[0132] The units described in the embodiments of the present disclosure can be implemented in software or in hardware. In some cases, the name of the unit does not constitute a limitation on the unit itself.

[0133] The functions described above herein can be performed, at least in part, by one or more hardware logic components. By way of example and not limitation, exemplary types of hardware logic components that can be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0134] In the context of the present disclosure, a machine-readable medium may 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 may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may 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 diskette, 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.

[0135] 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 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.

[0136] 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 sequential order. In certain environments, 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 may also be implemented combinatorially in a single embodiment. Conversely, the various features described in the context of a single embodiment may also be implemented separately or in any suitable sub-combination in multiple embodiments.

[0137] Although the subject matter has been described in language specific to structural features and / or methodological logical 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 of implementing the claims.

Claims

1. A modulation signal recognition model for signal recognition, characterized in that, Including: According to the input-output relationship, the modulation signal recognition model includes: an input module, a first feature extraction module, a second feature extraction module, a first one-dimensional inverted residual module connected to the first feature extraction module, a second one-dimensional inverted residual module connected to the second feature extraction module, a superimposing module, a third one-dimensional inverted residual module, a fourth one-dimensional inverted residual module, a global average pooling module, a classifier module, and an output module. The network architectures of the first feature extraction module and the second feature extraction module are the same. Among them, the input module is used to receive the in-phase signal and the quadrature signal of the same signal to be recognized in different representation forms. The first feature extraction module is used to process the in-phase signal, and the second feature extraction module is used to process the quadrature signal. The feature extraction module includes a wavelet transform convolutional denoising sub-module, a high-order central moment feature extraction module, and a feature extraction network. The wavelet transform denoising sub-module is used to denoise the recognized quadrature signal or in-phase signal. The high-order central moment feature extraction module is used to determine the eigenvalue data of the denoised quadrature signal or in-phase signal in multiple signal detection dimensions. The feature extraction network is used to extract the feature information in the feature data. The one-dimensional inverted residual module is used to perform residual processing on the input features to obtain the features to be used. The classifier module is used to determine the evaluation attributes corresponding to the signal to be recognized under different modulation methods. The output module is used to output the evaluation attributes processed by the classification module.

2. The model according to claim 1, wherein The wavelet transform convolutional denoising sub-module includes: a discrete wavelet transform unit, a plurality of convolutional denoising units, and an inverse discrete wavelet transform unit. The model structures of the plurality of convolutional denoising units are the same. Among them, the discrete wavelet transform unit is used to process the signal to be recognized into a first preset number of low-frequency components and a second preset number of high-frequency components. Based on the convolutional denoising units corresponding to each low-frequency component and high-frequency component, perform denoising processing on the corresponding components to obtain the features to be inversely processed to be input into the inverse discrete wavelet transform unit. The inverse discrete wavelet transform unit is used to perform inverse transform processing on all the features to be inversely processed to obtain the features to be centrally transformed to be input into the high-order central moment feature extraction module.

3. The model according to claim 2, characterized in that, According to the input-output relationship, the convolutional denoising unit includes a one-dimensional convolutional sub-unit with a first convolution kernel size, a first activation function layer, a one-dimensional convolutional sub-unit with a second convolution kernel size, a one-dimensional convolutional sub-unit with a first convolution kernel size, a second activation function layer, and a splicing processing layer for splicing the components output by the discrete wavelet transform and the features output by the second activation function.

4. The model according to claim 1, wherein The high-order central moment feature extraction module extracts the feature information of the features to be centrally transformed in at least one signal detection dimension. Among them, the features to be centrally transformed are the features output by the wavelet transform convolutional denoising sub-module.

5. The model according to claim 4, wherein The at least one signal detection dimension includes a feature mean dimension, a feature variance dimension, a feature skewness dimension, and a feature kurtosis dimension. Based on the high-order central moment feature extraction module, the feature information in the at least one signal detection dimension is determined, including: According to a preset sliding window length and a window adjustment step size, the to-be-centered-transformed feature is divided into multiple groups of to-be-processed feature groups, and the target feature values corresponding to the feature values in each to-be-processed feature group in each signal detection dimension are calculated respectively; Based on the target feature values of each to-be-processed feature group in each signal detection dimension and the preset feature length information, the to-be-feature-extracted matrix corresponding to the to-be-centered-transformed feature is determined; Wherein, the to-be-feature-extracted matrix is a matrix of order M×N, M is consistent with the number of the at least one signal detection dimension, and N is consistent with the preset feature length information.

6. The model according to claim 1, characterized in that, According to the input-output relationship, the feature extraction network includes a one-dimensional convolution with a second convolution kernel size, a first batch normalization layer, and a third activation function layer.

7. The model according to claim 1, characterized in that, According to the input-output relationship, the one-dimensional inverted residual module includes a first processing unit, a second processing unit for processing the output result of the first processing unit, a third processing unit for performing dot product processing on the output results of the first processing unit and the second processing unit, a fourth unit for processing the output result of the third processing unit, and a fifth processing unit for performing feature addition on the feature input to the first processing unit and the feature output by the fourth unit.

8. The model according to claim 7, characterized in that, According to the input-output relationship, the first processing unit includes a one-dimensional convolution with a second convolution size, a second batch normalization layer, a fourth activation function layer, a one-dimensional depth convolution with a first convolution size, and a third batch normalization layer; the second processing unit includes a first linear layer, a fifth activation function layer, a second linear layer, and a sixth activation function layer; the fourth unit includes a one-dimensional convolution with a second convolution kernel size and at least two fourth batch normalization layers.

9. A model training method, characterized in that, Including the modulation signal recognition model according to any one of claims 1-8, the training method includes: When it is detected that the modulation signal recognition model corresponding to the current training round is obtained, a plurality of test samples are acquired, wherein the current training round is that a preset number of training samples have participated in the training of the modulation signal recognition model, and the test samples and the training samples include signals to be recognized and corresponding modulation methods; wherein, the signals to be recognized are in-phase quadrature signals; Based on the plurality of test samples, the accuracy rate of the modulation signal recognition model corresponding to the current training round is determined; 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, the learning rate of the modulation signal recognition model is adjusted to continue training the modulation signal recognition model based on a preset number of training samples until the accuracy rates of consecutive multiple training rounds all meet the preset conditions, and the target learning rate obtained when the preset conditions are met is used as the learning rate used when the modulation signal recognition model participates in training, and the model training of the modulation signal recognition model is continued.

10. A method for identifying the modulation mode of multi-channel signals, including the modulation signal identification model according to any one of claims 1-8, characterized in that, The method further includes: Obtain the signal to be recognized for the modulation method to be recognized, the in-phase signal and the quadrature signal in different representation forms; wherein, the in-phase signal and the quadrature signal are signals obtained by splitting the in-phase quadrature signal; Input the in-phase signal and the quadrature signal into a pre-trained modulation signal recognition model for modulation mode recognition, and output the evaluation attributes corresponding to the signal to be recognized under different modulation methods; wherein, the evaluation attributes are used to characterize the credibility under the corresponding modulation mode; Based on the evaluation attributes, determine the target modulation method of the signal to be recognized.

Images