CN115941407A - Signal modulation identification method based on recursive convolutional network and attention mechanism - Google Patents
Signal modulation identification method based on recursive convolutional network and attention mechanism Download PDFInfo
- Publication number
- CN115941407A CN115941407A CN202211665176.XA CN202211665176A CN115941407A CN 115941407 A CN115941407 A CN 115941407A CN 202211665176 A CN202211665176 A CN 202211665176A CN 115941407 A CN115941407 A CN 115941407A
- Authority
- CN
- China
- Prior art keywords
- data
- signal modulation
- convolutional network
- recursive convolutional
- layer
- 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.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention provides a signal modulation identification method based on a recursive convolutional network and an attention mechanism, which comprises the following steps: sampling each communication signal to generate a data set; performing data enhancement on the data set; establishing a signal modulation identification model, wherein the signal modulation identification model comprises a recursive convolutional network and a multi-scale feature fusion module which are sequentially arranged, and the recursive convolutional network and the multi-scale feature fusion module are used for extracting feature information of communication signal data; training the signal modulation recognition model by adopting the data set after data enhancement, and further optimizing the signal modulation recognition model by using a residual error neural network to obtain the optimized signal modulation recognition model; and building a classifier for the optimized signal modulation recognition model. The method can fully extract the characteristic information of the signal, and can improve the calculation speed and the recognition effect of the traditional signal modulation recognition method.
Description
Technical Field
The invention relates to the technical field of signal modulation identification, in particular to a signal modulation identification method based on a recursive convolutional network and an attention mechanism.
Background
With the development of communication technology, the communication environment is more and more complex, and in order to meet the requirements of a large number of users, more and more signal modulation modes are provided. The signal modulation identification refers to identifying the modulated signal and noise, ensuring demodulation and feature extraction, and playing an important role in the fields of communication and the like. For example, in the civilian field, parameter configurations that can be used for detecting illegal radios and monitoring legal radios meet standards, and signal modulation identification also plays a crucial role in cognitive ratio.
The signal modulation identification is mainly divided into three processes: data preprocessing, feature extraction and classification decision. The data preprocessing is to estimate the carrier rate and the symbol rate of the signals after down-conversion, and provide a plurality of suitable data for subsequent operations; the feature extraction is the most important of the three processes, and directly influences the recognition effect, namely, the original data is transformed, and some features which are easy to classify are extracted. However, in the conventional signal modulation identification method, the signal characteristic information is not sufficiently extracted, the calculation speed is slow, and the identification effect is poor.
Therefore, it is necessary to provide a new signal modulation identification method, which can sufficiently extract the feature information of the signal and can improve the calculation speed and identification effect of the conventional signal modulation identification method.
Disclosure of Invention
Solves the technical problem
Aiming at the defects in the prior art, the invention provides a signal modulation identification method based on a recursive convolutional network and an attention mechanism, which can fully extract the characteristic information of a signal and can improve the calculation speed and the identification effect of the traditional signal modulation identification method.
Technical scheme
In order to achieve the purpose, the invention is realized by the following technical scheme:
the invention provides a signal modulation identification method based on a recursive convolutional network and an attention mechanism, which comprises the following steps:
s1, sampling each communication signal to generate a data set;
s2, performing data enhancement on the data set;
s3, establishing a signal modulation identification model, wherein the signal modulation identification model comprises a recursive convolutional network and a multi-scale feature fusion module which are sequentially arranged, and the recursive convolutional network and the multi-scale feature fusion module are used for extracting feature information of communication signal data;
s4, training the signal modulation recognition model by adopting the data set after data enhancement, and further optimizing the signal modulation recognition model by using a residual error neural network to obtain the optimized signal modulation recognition model;
and S5, building a classifier for the optimized signal modulation recognition model.
Further, the recursive convolutional network performs layer-by-layer convolution operation on a plurality of feature tensors in a recursive manner, and the feature tensors are fused with the branches with the channel attention mechanism to integrate the multi-scale feature information of the data.
Further, the multi-scale feature fusion module adopts a plurality of convolution operations with different convolution kernel sizes, and generates a weight vector to act on the input of the signal data.
Further, the data set includes a training set and a test set, and the data volume of the training set and the test set is 7.
Further, step S1 further includes:
performing data normalization on the data in the data set:
where x is the raw data, μ is the mean of the data set samples, σ is the standard deviation of the data set samples, x * The data is standardized;
sample alignment of the data set: translating data of the data set to have its center of gravity g 1 Move to
Nearby, wherein the data center of gravity g 1 The calculation method of (2) is as follows:
further, step S2 specifically includes: and adopting supervised single-sample data to enhance and process the data in the data set, and performing transposition and merging operation on the characteristic vectors of partial signal data to expand the data set.
Further, the data processing process of the recursive convolutional network specifically includes:
performing pooling layer operations on the signal data: dividing an input signal data characteristic tensor into a plurality of non-overlapping square areas with the size of 2 multiplied by 2, and calculating 4 numerical values in each area to obtain an average value;
performing the operation of the pooling layer four times to obtain four identical feature tensors c i Then, a convolution operation of a recursive structure is performed while introducing another branch which passes the input through a channel attention mechanism and outputs using a ReLU activation function and a batch normalization processing layer.
Further, the data processing process of the multi-scale feature fusion module specifically includes:
firstly, carrying out batch normalization processing on input, then respectively carrying out convolution operation on each channel by using three convolutions with convolution kernel sizes of 3, 5 and 7, ensuring that two-dimensional sizes of output are the same by using a zero filling method, then adding the output and carrying out convolution operation of 1 multiplied by 1 to obtain a weight vector, and finally multiplying the weight vector with the input to act on a characteristic tensor to obtain the output.
Further, the residual neural network comprises a convolution input layer and a residual convolution layer, and outputs through a full connection layer.
Further, the classifier comprises a full connection layer and a Softmax layer, wherein the full connection layer utilizes a matrix vector multiplication mode to carry out local feature information extracted by the network layer before feature space conversion and integration, and classifies through the Softmax layer after multiplying a weight matrix and an input vector and offsetting the weight matrix.
Advantageous effects
The invention provides a signal modulation identification method based on a recursive convolutional network and an attention mechanism, which can fully extract the characteristic information of a signal and improve the calculation speed and identification effect of the traditional signal modulation identification method; furthermore, the method can be used for effectively mining deep-level feature information in a recursive mode in a layer-by-layer progressive mode during signal processing; furthermore, the method uses a channel attention mechanism to fuse channel information, and is beneficial to feature extraction; finally, the method carries out additional preprocessing on the data in the pre-training stage so as to increase the diversity of the communication signal samples, and the preprocessed model is not limited by a large number of training samples any more, so that the method has practical significance in practical application.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. It is obvious that the drawings in the following description are only some embodiments of the invention, and that for a person skilled in the art, other drawings can be derived from them without inventive effort.
Fig. 1 is a schematic diagram illustrating steps of a signal modulation identification method based on a recursive convolutional network and an attention mechanism according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a model framework of a signal modulation identification method based on a recursive convolutional network and an attention mechanism according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be obtained by a person skilled in the art without inventive step based on the embodiments of the present invention, are within the scope of protection of the present invention.
Referring to fig. 1 and fig. 2, an embodiment of the present invention provides a signal modulation identification method based on a recursive convolutional network and an attention mechanism, including the following steps:
s1, sampling each communication signal to generate a data set;
s2, performing data enhancement on the data set;
s3, establishing a signal modulation identification model, wherein the signal modulation identification model comprises a recursive convolutional network and a multi-scale feature fusion module which are sequentially arranged, and the recursive convolutional network and the multi-scale feature fusion module are used for extracting feature information of communication signal data;
s4, training the signal modulation recognition model by adopting the data set after data enhancement, and further optimizing the signal modulation recognition model by using a residual error neural network to obtain the optimized signal modulation recognition model;
and S5, building a classifier for the optimized signal modulation recognition model.
In this embodiment, for step S1, first, the data set is divided into a training set and a test set, the ratio of the training set to the test set is maintained at 7:
data normalization of data in the dataset:
where x is the raw data, μ is the mean of the data set samples, σ is the standard deviation of the data set samples, x * The data is subjected to standardization processing;
sample aligning the data set: translating data of the data set to have its center of gravity g 1 Move to
Nearby, wherein the data center of gravity g 1 The calculation method of (2) is as follows: />
In this embodiment, referring to fig. 2, in order to increase the structural diversity of the data sample, the data set is extended to avoid an overfitting phenomenon in the model training process, and step S2 performs data enhancement on the data set, specifically including performing data enhancement processing on the data with supervised single sample data, transposing and merging the characteristic vectors of part of signal data, and so on, to achieve the purpose of extending the data set, thereby avoiding overfitting in the training process, and maximally using useful information in the data set.
In this embodiment, the recursive convolutional network recursively performs layer-by-layer convolution operations on a plurality of feature tensors, and fuses with branches with channel attention mechanisms to integrate multi-scale feature information of data. The data processing process specifically comprises the following steps:
1) Performing pooling layer operation on the signal data: the input signal data characteristic tensor is divided into a plurality of non-overlapping square areas with the size of 2 multiplied by 2, and 4 values in each area are calculated to obtain an average value.
The average value is generally calculated by the following formula:
through the operation of the pooling layer, the feature information can be further extracted, the size of the feature map is reduced, and the requirements on the calculated amount and the memory are reduced to a certain extent.
2) Performing the operation of the pooling layer four times to obtain four identical feature tensors c i Then, a convolution operation of a recursive structure is performed while introducing another branch which passes the input through a channel attention mechanism and outputs using a ReLU activation function and a batch normalization processing layer.
Specifically, the method comprises the following steps:
firstly, two of the feature tensors are subjected to convolution operation, the sizes of convolution kernels are divided into 1 and 3, filling is not used, the obtained two-dimensional feature tensor with large size is subjected to pooling processing to enable the sizes of the two feature tensors to be the same, and then addition operation is carried out to output the feature tensor. Then, selecting one feature tensor to perform convolution operation with the convolution kernel size of 5, performing pooling processing on the previous output after performing convolution operation with the convolution kernel size of 1, and performing addition on the two to obtain a new feature tensor, so as to execute a fourth tensor block, wherein the simplified expression of the output C is as follows:
C＝BN{Conv(BN{Conv(BN{Conv(c 1 ,c 2 )},c 3 )},c 4 )}
each convolution layer is followed by a ReLU activation function and batch normalization processing layer, and feature information is well integrated.
3) The recursive convolutional network also introduces another branch, the branch outputs the input through a channel attention mechanism and also uses a ReLU activation function and a batch normalization processing layer, the information correlation among channels of the characteristic tensor is enhanced, the nonlinear transformation capability of the model is enhanced, and the output of the recursive convolutional network are added and then input to the following self-attention mechanism layer for subsequent operation.
Specifically, the model can better learn the correlation of different parts in a plurality of feature vectors through a self-attention mechanism, and each input vector a is multiplied by threeA coefficient w q ，w k ，w v Three values of q, k and v are obtained:
q i ＝w q ·a i ，k i ＝w k ·a i ，v i ＝w v ·a i
calculating the correlation alpha of any two vectors by using a point multiplication method i,j 。
αi ,j ＝qi·kj
Will be alpha i,j The formed matrix A is subjected to a ReLU activation function and then is compared with v i The constructed matrix V calculates an output vector b corresponding to the input vector a.
In this embodiment, the data processing process of the multi-scale feature fusion module specifically includes:
firstly, carrying out batch normalization processing on input, then respectively carrying out convolution operation on each channel by using three convolutions with convolution kernel sizes of 3, 5 and 7, ensuring that two-dimensional sizes of output are the same by using a zero filling method, then adding the output and carrying out convolution operation of 1 multiplied by 1 to obtain a weight vector, and finally multiplying the weight vector with the input to act on a characteristic tensor to obtain the output.
In this embodiment, the residual neural network includes a convolution input layer and a residual convolution layer, and outputs through one fully-connected layer. The convolution input layer ensures the integrity of the structure and information of data and is beneficial to extracting feature information subsequently, the residual error layer extracts effective feature information in the enhanced signal by using convolution layers with convolution kernel size of 3 multiplied by 3 generally, and each residual error convolution layer can also use a ReLU activation function to enhance the generalization capability of the model.
In this embodiment, the classifier includes a full connection layer and a Softmax layer, where the full connection layer performs local feature information extracted by the network layer before feature space transformation and integration by means of matrix vector multiplication, and classifies the local feature information by the Softmax layer after multiplying a weight matrix by an input vector and offsetting the weight matrix.
Specifically, the full connection layer performs local feature information extracted by the network layer before feature space conversion and integration by means of matrix vector multiplication, multiplies a weight matrix by an input vector, biases the weight matrix, classifies the weight matrix by the Softmax layer, and tests a sample X if the total number of targets contained in the training set is C test The probability corresponding to the ith class of objects in the set of objects is expressed as:
where x is the input to the full link layer, W T Is a weight matrix, b is an offset,
probability of output for the-Softmax layer.
Will test sample X by maximum posterior probability test Probability of classification to maximum object c 0 The method comprises the following steps:
c 0 ＝argmaxP(i│xtest)
the loss function of the classifier is generally designed as a cross entropy, the parameters are learned by calculating the gradient of the loss function with respect to the parameters using training data, and the learned parameters are fixed when the model converges. The invention adopts a cost function based on cross entropy, which can be expressed as:
wherein N represents the number of training samples in a batch, z (i) is used to represent the ith training sample, P (i-x) train ) Representing the probability that the training sample corresponds to the ith target.
The method has the advantages that the characteristic information of the signal can be fully extracted, and the calculation speed and the recognition effect of the traditional signal modulation recognition method can be improved; furthermore, the method can be used for effectively mining deep-level feature information in a recursive mode in a layer-by-layer progressive mode during signal processing; furthermore, the method uses a channel attention mechanism to fuse channel information, and is beneficial to feature extraction; finally, the method performs additional preprocessing on the data in the pre-training stage to increase the diversity of the communication signal samples, and the preprocessed model is not limited by a large number of training samples any more, so that the method has practical significance in practical application.
The above examples are only intended to illustrate the technical solution of the present invention, and not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and the modifications or the substitutions do not depart from the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.
Claims (10)
1. A signal modulation identification method based on a recursive convolutional network and an attention mechanism is characterized by comprising the following steps:
s1, sampling each communication signal to generate a data set;
s2, performing data enhancement on the data set;
s3, establishing a signal modulation identification model, wherein the signal modulation identification model comprises a recursive convolutional network and a multi-scale feature fusion module which are sequentially arranged, and the recursive convolutional network and the multi-scale feature fusion module are used for extracting feature information of communication signal data;
s4, training the signal modulation recognition model by adopting the data set after data enhancement, and further optimizing the signal modulation recognition model by using a residual error neural network to obtain the optimized signal modulation recognition model;
and S5, building a classifier for the optimized signal modulation recognition model.
2. The signal modulation identification method based on the recursive convolutional network and the attention mechanism as claimed in claim 1, wherein the recursive convolutional network recursively performs layer-by-layer convolution operation on a plurality of feature tensors and fuses with the branch having the channel attention mechanism to integrate the multi-scale feature information of the data.
3. The signal modulation identification method based on the recursive convolutional network and the attention mechanism as claimed in claim 2, wherein the multi-scale feature fusion module adopts a plurality of convolution operations with different convolution kernel sizes and generates a weight vector to act on the input of the signal data.
4. The signal modulation identification method based on the recursive convolutional network and the attention mechanism as claimed in claim 1, wherein the data set comprises a training set and a test set, and the data volume of the training set and the test set is 7.
5. The method for identifying signal modulation based on the recursive convolutional network and the attention mechanism as claimed in claim 1, wherein step S1 further comprises:
data normalization of data in the dataset:
where x is the raw data, μ is the mean of the data set samples, σ is the standard deviation of the data set samples, x * The data is standardized;
sample alignment of the data set: translating data of the data set to have its center of gravity g 1 Move to
Nearby, wherein the data center of gravity g 1 The calculation method of (2) is as follows:
6. the method for identifying signal modulation based on the recursive convolutional network and the attention mechanism as claimed in claim 1, wherein the step S2 specifically comprises: and adopting supervised single sample data to enhance and process the data in the data set, and performing transposition and merging operations on the characteristic vectors of part of signal data to expand the data set.
7. The signal modulation identification method based on the recursive convolutional network and the attention mechanism as claimed in claim 3, wherein the data processing procedure of the recursive convolutional network is specifically as follows:
performing pooling layer operations on the signal data: dividing an input signal data characteristic tensor into a plurality of non-overlapping square areas with the size of 2 multiplied by 2, and calculating 4 numerical values in each area to obtain an average value;
performing four times of the pooling layer operation to obtain four identical feature tensors c i Then, a convolution operation of a recursive structure is performed while introducing another branch which passes the input through a channel attention mechanism and outputs using a ReLU activation function and a batch normalization processing layer.
8. The signal modulation identification method based on the recursive convolutional network and the attention mechanism as claimed in claim 7, wherein the data processing process of the multi-scale feature fusion module specifically comprises:
firstly, carrying out batch normalization processing on input, then respectively carrying out convolution operation on each channel by using three convolutions with convolution kernel sizes of 3, 5 and 7, ensuring that two-dimensional sizes of output are the same by using a zero filling method, then adding the output and carrying out convolution operation of 1 multiplied by 1 to obtain a weight vector, and finally multiplying the weight vector with the input to act on a characteristic tensor to obtain the output.
9. The method of claim 8, wherein the residual neural network comprises a convolution input layer and a residual convolution layer, and is outputted through a fully-connected layer.
10. The signal modulation identification method based on the recursive convolutional network and the attention mechanism as claimed in claim 9, wherein the classifier comprises a fully-connected layer and a Softmax layer, wherein the fully-connected layer performs the feature space transformation integration by means of matrix vector multiplication on the local feature information extracted by the network layer before, and performs classification by the Softmax layer after multiplying the weight matrix by the input vector and then offsetting the weight matrix.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211665176.XA CN115941407A (en) | 2022-12-23 | 2022-12-23 | Signal modulation identification method based on recursive convolutional network and attention mechanism |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211665176.XA CN115941407A (en) | 2022-12-23 | 2022-12-23 | Signal modulation identification method based on recursive convolutional network and attention mechanism |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115941407A true CN115941407A (en) | 2023-04-07 |
Family
ID=86697514
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211665176.XA Pending CN115941407A (en) | 2022-12-23 | 2022-12-23 | Signal modulation identification method based on recursive convolutional network and attention mechanism |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115941407A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116593980A (en) * | 2023-04-20 | 2023-08-15 | 中国人民解放军93209部队 | Radar target recognition model training method, radar target recognition method and device |
| CN116662906A (en) * | 2023-05-22 | 2023-08-29 | 中国人民解放军93209部队 | Signal identification method based on tree-like perception fusion convolutional network and feature compression |
-
2022
- 2022-12-23 CN CN202211665176.XA patent/CN115941407A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116593980A (en) * | 2023-04-20 | 2023-08-15 | 中国人民解放军93209部队 | Radar target recognition model training method, radar target recognition method and device |
| CN116593980B (en) * | 2023-04-20 | 2023-12-12 | 中国人民解放军93209部队 | Radar target recognition model training method, radar target recognition method and device |
| CN116662906A (en) * | 2023-05-22 | 2023-08-29 | 中国人民解放军93209部队 | Signal identification method based on tree-like perception fusion convolutional network and feature compression |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Ghaderizadeh et al. | Multiscale dual-branch residual spectral–spatial network with attention for hyperspectral image classification | |
| CN115941407A (en) | Signal modulation identification method based on recursive convolutional network and attention mechanism | |
| CN111612017B (en) | Target detection method based on information enhancement | |
| CN111274869A (en) | Method for classifying hyperspectral images based on parallel attention mechanism residual error network | |
| CN108520215B (en) | Single-sample face recognition method based on multi-scale joint feature encoder | |
| US11700156B1 (en) | Intelligent data and knowledge-driven method for modulation recognition | |
| CN113076994A (en) | Open-set domain self-adaptive image classification method and system | |
| CN104700100A (en) | Feature extraction method for high spatial resolution remote sensing big data | |
| US20240054760A1 (en) | Image detection method and apparatus | |
| CN115410050A (en) | Tumor cell detection equipment based on machine vision and method thereof | |
| CN112381763A (en) | Surface defect detection method | |
| CN113221848B (en) | Hyperspectral open set field self-adaption method based on multi-classifier domain confrontation network | |
| CN114492768A (en) | Twin capsule network intrusion detection method based on small sample learning | |
| CN114896887B (en) | Frequency-using equipment radio frequency fingerprint identification method based on deep learning | |
| CN111222442A (en) | Electromagnetic signal classification method and device | |
| Patel et al. | An optimized deep learning model for flower classification using NAS-FPN and faster R-CNN | |
| CN111368648B (en) | Individual identification method and device for radar radiation source, electronic equipment and storage medium thereof | |
| CN115982613A (en) | Signal modulation identification system and method based on improved convolutional neural network | |
| CN117155706A (en) | Network abnormal behavior detection method and system | |
| CN111768792A (en) | Audio steganalysis method based on convolutional neural network and domain confrontation learning | |
| CN110995631B (en) | Communication signal modulation mode identification method and system based on LSTM and SVM | |
| Thapa et al. | Deep learning for breast cancer classification: Enhanced tangent function | |
| CN115987730A (en) | Signal modulation identification method based on tree-shaped perception fusion convolutional network | |
| CN116956105A (en) | Classification model training method, defect identification method, device and electronic equipment | |
| CN112052881B (en) | Hyperspectral image classification model device based on multi-scale near-end feature splicing |
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 |