CN113343924A - Modulation signal identification method based on multi-scale cyclic spectrum feature and self-attention generation countermeasure network - Google Patents
Modulation signal identification method based on multi-scale cyclic spectrum feature and self-attention generation countermeasure network Download PDFInfo
- Publication number
- CN113343924A CN113343924A CN202110746817.3A CN202110746817A CN113343924A CN 113343924 A CN113343924 A CN 113343924A CN 202110746817 A CN202110746817 A CN 202110746817A CN 113343924 A CN113343924 A CN 113343924A
- Authority
- CN
- China
- Prior art keywords
- attention
- self
- network
- modulation signal
- scale
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
- G06F17/141—Discrete Fourier transforms
- G06F17/142—Fast Fourier transforms, e.g. using a Cooley-Tukey type algorithm
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Discrete Mathematics (AREA)
- Signal Processing (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a modulation signal identification method for generating an antagonistic network based on multi-scale cyclic spectrum characteristics and self attention, and a modulation signal identification model is established based on the multi-scale cyclic spectrum characteristics. The model adopts self-attention to generate an antagonistic network, and the self-attention module adopts a recurrent neural network to generate visual attention, so that the network can focus on local information of the cyclic spectrum. The context automatic encoder is used for combining the original drawing and the attention drawing, and a multi-scale loss function is set, so that the features can be extracted in different decoding layers, and the information of different scales is fused to obtain richer feature information. In order to increase the reliability of data, the value of the data is improved by using an AutoAutoAutoAutoAutoAutoAutoAutoAutoaugmentation data enhancement algorithm, and the generalization capability of the self-attention generation countermeasure network is enhanced by combining with the multi-scale cyclic spectrum feature. The method is suitable for various communication systems, can effectively improve the identification accuracy of the modulation signal, and is more in line with the development trend of future intelligent communication.
Description
Technical Field
The invention relates to the field of intelligent communication, in particular to a modulation signal identification method for generating an antagonistic network based on multi-scale cyclic spectrum characteristics and self attention.
Background
Under the situation that the current social communication industry is developing day by day, the communication environment changes complicatedly, and in order to improve the utilization rate of frequency bands, different modulation modes have to be adopted for different communication signals. Therefore, the research on how to effectively detect and identify these signals is conducted by countless people in both military and civilian fields. Especially for the military field of non-cooperative communication, the method has important strategic significance for efficient detection and identification of signals.
In recent years, artificial intelligence technology has come to the peak, and the recent research direction thereof is deep learning technology which has shown remarkable performances in a plurality of fields, such as image processing and voice recognition, and the combination of artificial intelligence and radio communication to improve the performance of the radio communication is a new direction of future research. Modulation identification of communication signals plays a very important role in the whole communication process, and the exploration of deep learning schemes in the modulation identification technology becomes one of potential directions for improving the performance of a wireless communication system.
Disclosure of Invention
The invention solves the problems of low identification accuracy and unstable performance in the existing modulation signal identification method, and provides a modulation signal identification method for generating an antagonistic network based on multi-scale cyclic spectrum characteristics and self attention.
The technical scheme adopted by the invention is that a modulated signal identification method for generating an antagonistic network based on multi-scale cyclic spectrum characteristics and self attention comprises the following steps:
step 1: and generating a three-dimensional cyclic spectrogram, such as QPSK, 8PSK, 16QAM and 64QAM, under different modulation signals by adopting a time domain smoothing FFT accumulation algorithm. And then processing the generated three-dimensional cycle spectrogram along a certain equal height distance in the Z-axis direction into a two-dimensional image to generate a data set required by the self-attention generating countermeasure network. Generating characteristic information of the countermeasure network data set from self attention, wherein the characteristic information comes from a multi-scale two-dimensional cycle spectrogram, and labels are distributed according to a modulation mode;
step 2: randomly disorganizing the sample data of the self-attention generation countermeasure network in the step 1, wherein 5000 pictures are used for forming a training set, 1000 pictures are used for forming a verification set, 1000 pictures are used for forming a test set, and the pictures are processed into 64 multiplied by 64 pixels;
and step 3: based on the data set in the step 2, the value of the data is improved by using an AutoAutoAutoAutoAutoAutoAutoAutoaugmentation data enhancement algorithm, and the generalization capability of the self-attention to the anti-network is enhanced by combining with multi-scale feature information;
and 4, step 4: the self-attention module for generating the self-attention of the countermeasure network in the step 3 can make the network focus on the local information of the cyclic spectrum. The self-attention module adopts a recurrent neural network to generate visual attention, wherein the recurrent neural network consists of two residual error layers, two LSTM layers and a convolutional layer for generating attention;
and 5: the method comprises the steps that a context automatic encoder is adopted in a generation network of a generation countermeasure network for combination of an original picture and an attention diagram, a multi-scale loss function is set for the context automatic encoder, features can be extracted in different decoding layers to form output with different sizes, and then information with different scales is fused to obtain richer feature information;
step 6: the discrimination network for generating the countermeasure network is mainly realized by adopting a five-layer convolutional neural network, the discrimination network aims to distinguish a cycle spectrogram generated by the generation network from a real cycle spectrogram as much as possible, and the output of the discrimination network is the probability that a picture is in a certain type of modulation mode;
and 7: the identification of the modulated signal is done according to the model established by step 1, step 2, step 3, step 4, step 5 and step 6.
Preferably, in the step 1, the three-dimensional cyclic spectrogram is generated under four conditions of signal-to-noise ratio of 0dB, 5dB, 10dB and 15dB, and the data set is generated by converting the three-dimensional cyclic spectrogram into a two-dimensional image through a contour line at a distance of 10 from the XOY plane in the Z-axis direction. The labels that self-attentively generate a counterpoise network are QPSK, 8PSK, 16QAM, and 64QAM modulation classes.
Preferably, in the step 2, the picture is processed to a size of 64 × 64 pixels by using a resize function of OpenCV.
Preferably, in step 3, the basic idea of the automation data enhancement algorithm is to use enhancement learning to find an optimal image transformation strategy from the data itself, and learn different enhancement methods for different tasks to increase the adaptivity of the model. The implementation comprises the following steps:
step 3.1: setting 10 (changeable number) data enhancement processing methods;
step 3.2: selecting 5 (variable number) methods from 10, randomly generating probabilities and corresponding amplitudes for using the methods, each method obtaining a corresponding sub-strategy;
step 3.3: processing samples in the training process by randomly adopting one of 5 sub-strategies, and verifying the effectiveness of the sub-strategy through data on a verification set;
step 3.4: obtaining an effective ground sub-strategy scheme after about 100 training periods;
step 3.5: finishing data enhancement by adopting a trained strategy;
preferably, in the step 4, the learned attention map at each time step is a matrix from 0 to 1, and the higher the attention degree is, the larger the value is, and the brighter the display is in the picture; at each time step, the current generated self-attention map is concatenated with the original image and then input into the next block of the recursive network, which in turn loops.
Preferably, during training, the data batch is set to 100, and the RMSprop algorithm is used to optimize the update of the learning rate. The initialization parameter of the self-attention map is set to 0.4, and as the time step increases, the network focuses more and more on the peak region of the circular spectrum.
Preferably, in step 5, the multi-scale context automatic encoder adopts 20 conversion blocks, and adds a skip connection to prevent a blurred output of the image. The penalty function in each recursive block is defined as the mean square error between the output self-attention map for that time step and the image binary mask.
In step 5, the multi-scale context automatic encoder training comprises the following steps:
step 5.1: training the finally generated self-attention map and an original two-dimensional cycle spectrogram together as the input of a context automatic encoder, wherein the encoder is used for reducing the dimension of an image, and a decoder is used for reconstructing the original image;
step 5.2: and disturbing the sequence of the cyclic spectrograms of different classes in the verification set, and verifying the performance of the established context automatic encoder.
Step 5.3: the loss function employs multi-scale loss:
compared with the prior art, the modulation signal identification method based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network has the following technical effects:
the invention relates to a modulation signal identification method based on multi-scale cyclic spectrum characteristics and a self-attention generation countermeasure network. The model generates an antagonistic network by using self-attention, wherein the self-attention module generates visual attention by using a recurrent neural network, so that the network can focus on local information of the cyclic spectrum. A context automatic encoder is adopted in a generation network for combination of an original picture and an attention diagram, a multi-scale loss function is set, features can be extracted in different decoding layers, and richer feature information can be obtained by fusing different scale information. The discrimination network is realized by adopting a five-layer convolutional neural network, and the aim is to distinguish a cyclic spectrogram generated by the generated network from a real cyclic spectrogram as far as possible. In order to increase the reliability of data, the value of the data is improved by using an AutoAutoAutoAutoAutoAutoAutoAutoAutomation data enhancement algorithm, and the generalization capability of self-attention to the anti-network is enhanced by combining with multi-scale feature information. The method is suitable for various communication systems, has strong generalization, can improve the identification accuracy of the modulation signal to a greater extent, and is more suitable for the requirements of the current 5G communication and the development trend of future intelligent communication.
Drawings
Fig. 1 is a two-dimensional cyclic spectrum of a modulated signal.
Fig. 2a is a front schematic view of a self-attention generating antagonistic network architecture.
Fig. 2b is a rear schematic view of a self-attention generating countermeasure network architecture.
Fig. 3 is a block diagram of a multi-scale context auto-encoder.
Fig. 4 is a diagram of the structure of the discriminator module.
Detailed Description
The present invention is described in further detail with reference to the following examples, but the scope of the present invention is not limited thereto.
The invention relates to a modulation signal identification method based on multi-scale cyclic spectrum characteristics and a self-attention generation antagonistic network, wherein a two-dimensional cyclic spectrum and a self-attention generation antagonistic network structure diagram are respectively shown in a figure 1 and a figure 2, and the method comprises the following steps:
step 1: and generating a three-dimensional cyclic spectrogram, such as QPSK, 8PSK, 16QAM and 64QAM, under different modulation signals by adopting a time domain smoothing FFT accumulation algorithm. And then processing the generated three-dimensional cycle spectrogram into a two-dimensional image at a certain equal height in the Z-axis direction to generate a data set required by the self-attention generating countermeasure network.
In step 1, the specific implementation process of the time domain smoothing FFT accumulation algorithm includes the following steps:
step 1.1: inputting a cyclostationary signal x (t) and defining a cyclic autocorrelation function as:
where τ represents a time interval and t is a time period;
step 1.2: performing Fourier transform on the formula in the step 1.1 to obtain a cyclic spectral density function, wherein the mathematical expression can be expressed as:
meanwhile, the fourier transform of the time domain signal can be expressed as:
step 1.3: after a sliding short-time fourier transform can be expressed as:
step 1.4: by combining the above formulas, a cyclic spectrum density function estimated based on the time domain smoothing FFT accumulation algorithm can be obtained, which can be expressed as:
wherein the content of the first and second substances,
n is the data length, N1Is the length of the signal segment after windowing,
for discrete Fourier transform after windowing function processing, wnIs a windowing function.
Step 2: randomly disorganizing the sample data required by the self-attention generation and the anti-network training in the step 1, wherein 5000 pictures are used for forming a training set, 1000 pictures are used for forming a verification set, 1000 pictures are used for forming a test set, and the pictures are processed into the size of 64 multiplied by 64 pixels by adopting a resize function of OpenCV.
And step 3: based on the training data in the step 2, the value of the data is improved by using an AutoAutoAutoAutoAutoAutoAutoAutoaugmentation data enhancement algorithm, and the generalization capability of the self-attention to the anti-network is enhanced by combining with multi-scale feature information. The basic idea of the AutoAutoAutoAutoAutoAutoAutoAutoAutoAutoAutoaugmentation data enhancement algorithm is to use reinforcement learning to find the optimal image transformation strategy from the data itself, and learn different enhancement methods for different tasks, and the implementation comprises the following steps:
step 3.1: setting 10 (changeable number) data enhancement processing methods;
step 3.2: selecting 5 (variable number) methods from 10, randomly generating probabilities and corresponding amplitudes for using the methods, each method obtaining a corresponding sub-strategy;
step 3.3: processing samples in the training process by randomly adopting one of 5 sub-strategies, and verifying the effectiveness of the sub-strategy through data on a verification set;
step 3.4: obtaining an effective ground sub-strategy scheme after about 100 training periods;
step 3.5: and finishing data enhancement by adopting a trained strategy.
And 4, step 4: training the network based on the data set described in step 2 and step 3, wherein the attention diagram learned at each time step is a matrix from 0 to 1, and the larger the value, the higher the attention degree. The LSTM model has three gates: forgetting gate, input gate and output gate. W and b represent weight and bias, respectively. The forgetting gate is used for determining the forgetting degree of the last unit state information, and the formula can be written as follows: f. oft＝σ(Wfg[Ht-1,Xt]+bf). The input gate is used to determine which new information is stored in the cell state. First, the sigmoid layer decides which values we want to update, denoted as it＝σ(Wi·[Ht-1,Xt]+bi). Next, the tanh layer creates a new candidate vector C (t)*Can be added to the state, represented as: c (t)*＝Tanh(Wc·[Ht-1,Xt]+ b). Adding the new and old information to obtain a new cell state, which is expressed as: c (t) ═ ft*C(t-1)+it*C(t)*. And finally, the output gate determines the output information part, which is obtained by multiplying the output result of sigmoid by C (t) after tanh action, and the formulas are respectively as follows: o ist＝σ(Wo·[Ht-1,Xt]+bo) And h (t) ═ Ot*tanh(C(t))。
In the invention, the data batch during training is set as 100, and the RMSprop algorithm is adopted to optimize the updating of the learning rate. The initialization parameter of the self-attention map is set to 0.4, and as the time step increases, the network focuses more and more on the peak region of the circular spectrum. The self-attention module comprises a plurality of time steps, the attention diagram of the previous step is used as the input of the next step, and therefore characteristic information of different steps can be obtained. Meanwhile, the attention of the attention map of the previous step is small, the attention gradually increases with the increase of the step, and the step is set to be 5 in consideration of the memory problem.
And 5: in the generation network for generating the countermeasure network, a context automatic encoder is adopted for combination of an original picture and an attention diagram, as shown in fig. 3, a multi-scale loss function is set for the context automatic encoder, features can be extracted in different decoding layers to form output with different sizes, and then information with different scales is fused to obtain richer feature information.
In step 5, the multi-scale context automatic encoder includes the following steps:
step 5.1: the context auto-encoder sets 20 transform blocks and adds skip connections to increase the sharpness of the image output.
Step 5.2: setting a multi-scale loss function, capturing more context information from different scales, and expressing the formula as follows:
step 5.3: setting global loss function, measuring automatic encoderThe global feature difference between the generated image and the corresponding real image is extracted through a DenseNet network, and the global loss function is defined as: l isP＝LMSE(DenseNet(O),DenseNet(T))。
Step 5.4: the data set is input to a context autocoder for training and testing, and the weights and biases of the network are continuously updated through an optimization algorithm.
Step 6: the discrimination network for generating the countermeasure network mainly adopts 5 convolution layers, each convolution layer adopts a RuLU activation function, a full connection layer and a Softmax layer, the output of the discrimination network is the probability that a picture is in a certain type of modulation mode, and the maximum probability type is used as a final recognition result, as shown in fig. 4.
And 7: the identification of the modulated signal is done according to the model established by step 1, step 2, step 3, step 4, step 5 and step 6.
The design of the technical scheme is based on a multi-scale cyclic spectrum characteristic and a method for recognizing the modulation signal of the self-attention generation countermeasure network, an innovative structural design is adopted, and a modulation signal recognition model is established based on the multi-scale cyclic spectrum characteristic. The model adopts self-attention to generate an antagonistic network, the self-attention module adopts a recurrent neural network to generate visual attention, the network can focus on local information of a cyclic spectrum, and the self-attention module comprises residual error connection, an LSTM network and a convolutional layer and generates an attention diagram. In the generation network, a context automatic encoder is adopted for combination of an original drawing and an attention drawing, a multi-scale loss function is set, features can be extracted in different decoding layers, and information of different scales is fused to obtain richer feature information. Setting a global loss function, and extracting global characteristic difference between a generated image of the automatic encoder and a corresponding real image through a DenseNet network. In order to increase the reliability of data, the value of the data is improved by using an AutoAutoAutoAutoAutoAutoAutoAutoAutomation data enhancement algorithm, and the generalization capability of self-attention to the anti-network is enhanced by combining with multi-scale feature information. The method has stronger generalization on the basis of improving the identification accuracy of the modulation signal, is suitable for various wireless communication systems, and better conforms to the development trend of future intelligent communication.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.
Claims (9)
1. A method for identifying a modulated signal based on multi-scale cyclic spectral features and a self-attention generating countermeasure network, the method comprising the steps of:
step 1: and generating a three-dimensional cyclic spectrogram, such as QPSK, 8PSK, 16QAM and 64QAM, under different modulation signals by adopting a time domain smoothing FFT accumulation algorithm. And then processing the generated three-dimensional cycle spectrogram along a certain equal height distance in the Z-axis direction into a two-dimensional image to generate a data set required by the self-attention generating countermeasure network. Generating characteristic information of the countermeasure network data set from attention, wherein the characteristic information comes from a multi-scale two-dimensional cycle spectrogram, and the labels are distributed according to a modulation mode corresponding to the cycle spectrogram;
step 2: randomly disorganizing the sample data of the self-attention generation countermeasure network in the step 1, wherein 5000 pictures are used for forming a training set, 1000 pictures are used for forming a verification set, 1000 pictures are used for forming a test set, and the pictures are processed into 64 multiplied by 64 pixels;
and step 3: based on the data set in the step 2, the value of the data is improved by using an AutoAutoAutoAutoAutoAutoAutoAutoaugmentation data enhancement algorithm, and the generalization capability of the self-attention to the anti-network is enhanced by combining with multi-scale feature information;
and 4, step 4: the self-attention module for generating the self-attention of the countermeasure network in the step 3 can make the network focus on the local information of the cyclic spectrum. The self-attention module adopts a recurrent neural network to generate visual attention, wherein the recurrent neural network consists of three residual error layers, two LSTM layers and a convolutional layer for generating attention;
and 5: the method comprises the steps that a context automatic encoder is adopted in a generation network of a generation countermeasure network for combination of an original picture and an attention diagram, a multi-scale loss function is set for the context automatic encoder, image features can be extracted in different decoding layers to form output with different sizes, and then information with different scales is fused to obtain richer feature information;
step 6: the discrimination network for generating the countermeasure network is mainly realized by adopting a five-layer convolutional neural network, the discrimination network aims to distinguish a cycle spectrogram generated by the generation network from a real cycle spectrogram as much as possible, and the output of the discrimination network is the probability that a picture is in a certain type of modulation mode;
and 7: the identification of the modulated signal is done according to the model established by step 1, step 2, step 3, step 4, step 5 and step 6.
2. The method for identifying the modulation signal based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network according to claim 1, wherein the method comprises the following steps: in the step 2, the picture is processed to 64 × 64 pixels by using a resize function of OpenCV.
3. The method for identifying the modulation signal based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network according to claim 1, wherein the method comprises the following steps: in step 3, the basic idea of the automation data enhancement algorithm is to use enhancement learning to find the optimal image transformation strategy from the data itself, and for learning different enhancement methods for different tasks, the implementation includes the following steps:
step 3.1: setting 10 (changeable number) data enhancement processing methods;
step 3.2: selecting 5 (variable number) methods from 10, randomly generating probabilities and corresponding amplitudes for using the methods, each method obtaining a corresponding sub-strategy;
step 3.3: processing samples in the training process by randomly adopting one of 5 sub-strategies, and verifying the effectiveness of the sub-strategy through data on a verification set;
step 3.4: obtaining an effective ground sub-strategy scheme after about 100 training periods;
step 3.5: and finishing data enhancement by adopting a trained strategy.
4. The method for identifying the modulation signal based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network according to claim 1, wherein the method comprises the following steps: in the step 4, the attention map learned at each time step is a matrix from 0 to 1, and the higher the value is, the higher the attention degree is; at each time step, the current self-attention map is concatenated with the input images and then input into the next block of the recursive network.
5. The method for identifying the modulation signal based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network as claimed in claim 4, wherein: during training, the initialization parameter of the self-attention map is set to 0.4, and as the time step increases, the network focuses more and more on the peak region of the circular spectrum.
6. The method for identifying the modulation signal based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network according to claim 1, wherein the method comprises the following steps: in step 5, the loss function in each recursive block is defined as the mean square error between the output self-attention map for the time step and the image binary mask.
7. The method for identifying the modulation signal based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network according to claim 1, wherein the method comprises the following steps: the context auto-encoder employs 20 transform blocks and adds a skip connection to prevent a blurred output of the image.
8. The method for identifying the modulation signal based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network according to claim 1, wherein the method comprises the following steps: and finally, the probability of which modulation mode the modulation signal belongs to is output through a Softmax layer.
9. The method for identifying the modulation signal based on the multi-scale cyclic spectrum characteristic and the self-attention generation countermeasure network according to claim 1, wherein the method comprises the following steps: and the classification accuracy of the network is used as an index for evaluating the identification performance of the network modulation signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110746817.3A CN113343924B (en) | 2021-07-01 | 2021-07-01 | Modulation signal identification method based on cyclic spectrum characteristics and generation countermeasure network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110746817.3A CN113343924B (en) | 2021-07-01 | 2021-07-01 | Modulation signal identification method based on cyclic spectrum characteristics and generation countermeasure network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113343924A true CN113343924A (en) | 2021-09-03 |
| CN113343924B CN113343924B (en) | 2022-05-17 |
Family
ID=77482100
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110746817.3A Active CN113343924B (en) | 2021-07-01 | 2021-07-01 | Modulation signal identification method based on cyclic spectrum characteristics and generation countermeasure network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113343924B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116486464A (en) * | 2023-06-20 | 2023-07-25 | 齐鲁工业大学(山东省科学院) | Attention mechanism-based face counterfeiting detection method for convolution countermeasure network |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140080225A1 (en) * | 2012-09-18 | 2014-03-20 | The Royal Institution For The Advancement Of Learning / Mcgill University | Methods and devices for xerogel based sensors |
| CN109523994A (en) * | 2018-11-13 | 2019-03-26 | 四川大学 | A kind of multitask method of speech classification based on capsule neural network |
| CN110084121A (en) * | 2019-03-27 | 2019-08-02 | 南京邮电大学 | Implementation method based on the human face expression migration for composing normalized circulation production confrontation network |
| CN110739002A (en) * | 2019-10-16 | 2020-01-31 | 中山大学 | Complex domain speech enhancement method, system and medium based on generation countermeasure network |
| CN111179228A (en) * | 2019-12-16 | 2020-05-19 | 浙江大学 | Single-energy CT energy spectrum imaging method based on deep learning |
| CN111259798A (en) * | 2020-01-16 | 2020-06-09 | 西安电子科技大学 | Modulation signal identification method based on deep learning |
| CN111990989A (en) * | 2020-08-31 | 2020-11-27 | 武汉大学 | Electrocardiosignal identification method based on generation countermeasure and convolution cyclic network |
| CN112767958A (en) * | 2021-02-26 | 2021-05-07 | 华南理工大学 | Zero-learning-based cross-language tone conversion system and method |
| CN112967730A (en) * | 2021-01-29 | 2021-06-15 | 北京达佳互联信息技术有限公司 | Voice signal processing method and device, electronic equipment and storage medium |
-
2021
- 2021-07-01 CN CN202110746817.3A patent/CN113343924B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140080225A1 (en) * | 2012-09-18 | 2014-03-20 | The Royal Institution For The Advancement Of Learning / Mcgill University | Methods and devices for xerogel based sensors |
| CN109523994A (en) * | 2018-11-13 | 2019-03-26 | 四川大学 | A kind of multitask method of speech classification based on capsule neural network |
| CN110084121A (en) * | 2019-03-27 | 2019-08-02 | 南京邮电大学 | Implementation method based on the human face expression migration for composing normalized circulation production confrontation network |
| CN110739002A (en) * | 2019-10-16 | 2020-01-31 | 中山大学 | Complex domain speech enhancement method, system and medium based on generation countermeasure network |
| CN111179228A (en) * | 2019-12-16 | 2020-05-19 | 浙江大学 | Single-energy CT energy spectrum imaging method based on deep learning |
| CN111259798A (en) * | 2020-01-16 | 2020-06-09 | 西安电子科技大学 | Modulation signal identification method based on deep learning |
| CN111990989A (en) * | 2020-08-31 | 2020-11-27 | 武汉大学 | Electrocardiosignal identification method based on generation countermeasure and convolution cyclic network |
| CN112967730A (en) * | 2021-01-29 | 2021-06-15 | 北京达佳互联信息技术有限公司 | Voice signal processing method and device, electronic equipment and storage medium |
| CN112767958A (en) * | 2021-02-26 | 2021-05-07 | 华南理工大学 | Zero-learning-based cross-language tone conversion system and method |
Non-Patent Citations (1)
| Title |
|---|
| 张少康,田德艳: "水下声目标的梅尔倒谱系数智能分类方法", 《应用声学》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116486464A (en) * | 2023-06-20 | 2023-07-25 | 齐鲁工业大学(山东省科学院) | Attention mechanism-based face counterfeiting detection method for convolution countermeasure network |
| CN116486464B (en) * | 2023-06-20 | 2023-09-01 | 齐鲁工业大学(山东省科学院) | Attention mechanism-based face counterfeiting detection method for convolution countermeasure network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113343924B (en) | 2022-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113688723B (en) | Infrared image pedestrian target detection method based on improved YOLOv5 | |
| CN110781924B (en) | Side-scan sonar image feature extraction method based on full convolution neural network | |
| Chen et al. | Automatic modulation classification scheme based on LSTM with random erasing and attention mechanism | |
| CN108764298B (en) | Electric power image environment influence identification method based on single classifier | |
| CN113505792B (en) | Multi-scale semantic segmentation method and model for unbalanced remote sensing image | |
| CN110598530A (en) | Small sample radio signal enhanced identification method based on ACGAN | |
| CN110555841B (en) | SAR image change detection method based on self-attention image fusion and DEC | |
| Emam et al. | A comparative study between CNN, LSTM, and CLDNN models in the context of radio modulation classification | |
| CN114595732B (en) | Radar radiation source sorting method based on depth clustering | |
| CN113359212B (en) | Reservoir characteristic prediction method and model based on deep learning | |
| CN114726692B (en) | SERESESESENet-LSTM-based radiation source modulation mode identification method | |
| CN116343045B (en) | Lightweight SAR image ship target detection method based on YOLO v5 | |
| CN116206185A (en) | Lightweight small target detection method based on improved YOLOv7 | |
| CN114973011A (en) | High-resolution remote sensing image building extraction method based on deep learning | |
| CN115659254A (en) | Power quality disturbance analysis method for power distribution network with bimodal feature fusion | |
| CN113343924B (en) | Modulation signal identification method based on cyclic spectrum characteristics and generation countermeasure network | |
| CN113902095A (en) | Automatic modulation identification method, device and system for wireless communication | |
| CN116340846A (en) | Aliasing modulation signal identification method for multi-example multi-label learning under weak supervision | |
| CN114595770B (en) | Long time sequence prediction method for ship track | |
| CN115810129A (en) | Object classification method based on lightweight network | |
| CN114724245B (en) | Incremental learning human body action recognition method based on CSI | |
| CN115329821A (en) | Ship noise identification method based on pairing coding network and comparison learning | |
| CN115712867A (en) | Multi-component radar signal modulation identification method | |
| CN114998731A (en) | Intelligent terminal navigation scene perception identification method | |
| CN115223033A (en) | Synthetic aperture sonar image target classification method and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |