CN113705787B - Digital modulation signal identification method based on deep collaborative training - Google Patents
Digital modulation signal identification method based on deep collaborative training Download PDFInfo
- Publication number
- CN113705787B CN113705787B CN202110995377.5A CN202110995377A CN113705787B CN 113705787 B CN113705787 B CN 113705787B CN 202110995377 A CN202110995377 A CN 202110995377A CN 113705787 B CN113705787 B CN 113705787B
- Authority
- CN
- China
- Prior art keywords
- data set
- modulation signal
- digital modulation
- network
- training
- 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.)
- Active
Links
Images
Classifications
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
Abstract
The invention belongs to the technical field of deep learning, and particularly relates to a digital modulation signal identification method based on deep collaborative training. The traditional digital modulation signal identification method based on deep learning needs a large amount of marked data samples for training. In actual communication activities, only a large number of unlabeled signal samples are readily available, and obtaining the manual labels is costly and inefficient. According to the invention, two CLDNN networks are set up for collaborative training, the differentiation of views is realized by using the generated confrontation, a large amount of unmarked samples are fully utilized to assist the learning under a small amount of marked samples, the identification accuracy is effectively improved, and the feasibility and the practicability of deep learning in a digital modulation signal identification task are enhanced.
Description
Technical Field
The invention belongs to the technical field of deep learning, and particularly relates to a digital modulation signal identification method based on deep collaborative training.
Background
Digital signal modulation identification is an important technology in the field of wireless communication, and has wide application in both civil and military fields. The task of digital signal modulation identification is mainly to identify the modulation type of a signal under the condition of only a small amount of a priori information or no a priori information, and is an important step between signal detection and demodulation. With the rapid development of wireless communication, the modulation modes of signals become various, and the cost of modulation identification also increases.
Modulation identification methods can generally be classified into two categories: likelihood ratio based modulation identification methods and feature based modulation identification methods. The modulation recognition method based on the likelihood ratio can obtain the optimal result under the meaning of the Bayesian criterion, but needs a large amount of prior information, and the algorithm has high complexity, is very sensitive to parameter setting and has poor robustness. The modulation recognition method based on the characteristics mainly utilizes a known modulation signal sample to train a characteristic extractor, automatically learns and extracts the signal characteristics, and classifies unknown signals according to the extracted characteristics. Commonly used feature extractors are SVM, decision trees, LSTM, CNN, etc.
The existing modulation recognition method based on the characteristics usually needs a large number of known marked samples to train the characteristic extractor, and then modulation recognition tasks are carried out after the optimal characteristic extractor is obtained. In practical situations, only a large amount of unlabeled samples are often available, and obtaining labeled samples is difficult and costly. Typical feature-based modulation identification methods do not take full advantage of the large number of unlabeled samples. Therefore, how to fully utilize a large amount of unmarked samples to realize digital signal modulation identification under a small amount of marked samples becomes an urgent problem to be solved in the field of radio modulation identification.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a digital signal modulation identification method based on deep collaborative training. The core of the invention is mainly divided into four parts: firstly, preprocessing known signal data to realize data expansion; secondly, generating a countermeasure sample through a countermeasure algorithm; thirdly, completing the cooperative training by using the confrontation sample; and fourthly, completing type confirmation of the unknown modulation signal based on the trained feature extractor.
The technical scheme of the invention is as follows:
a method for recognizing a digital modulation signal based on deep collaborative training, as shown in fig. 1, the recognition includes the following steps:
s1, acquiring digital modulation signal data and labels, recording the digital modulation signal data and the labels as a data set D, and preprocessing the data set D, wherein the preprocessing comprises the following steps: partitioning the data set, including a marked data set S and a unmarked data set U, the marked data set being marked
Unmarked data set is marked as->
D = S U and M < N, x is marked data, y is the mark corresponding to x, x u The data are unmarked data, M is the number of marked data sets, and N is the number of unmarked data sets; expanding the marked data set;
s2, constructing an identification network comprising two CLDNN networks with the same structure, wherein the CLDNN network comprises a convolution part, an LSTM layer and two full-connection layers as shown in figure 2; the convolution part comprises three convolution blocks, each convolution block consists of a convolution layer, a batch normalization layer and a ReLU activation function, and a uniform pooling layer with the kernel size of 2 is arranged between every two convolution blocks; two CLDNN networks are used as feature extractors and are respectively marked as f 1 (. And f) 2 (. The) the data set obtained in the step S1 is adopted to carry out cooperative training on the recognition network, wherein the cooperative training specifically comprises the following steps:
generating a countermeasure sample data set D 'by adopting a countermeasure algorithm according to the data set D, wherein D' = { g (x) | x ∈ D }, and g (·) is generated into the countermeasure algorithm; the cross-entropy loss of two CLDNN networks on a tagged dataset S is:
L sup ＝H(y，f 1 (x))+H(y，f 2 (x))
where H (-) is a cross entropy loss function;
the JS divergence loss for the two CLDNN networks on the unmarked dataset D was:
L cot ＝JS(f 1 (x u )||f 2 (x u ))
wherein JS (-) is JS divergence;
CLDNN network f 1 (. Output on the original data set D) and the network f 2 (.) a cross-entropy loss function of the output on the challenge sample data set D', and a network f 2 (. Output on the original data set D) and the network f 1 (.) the cross-entropy loss function of the output on the challenge sample data set D' is:
L dif ＝H(f 1 (x)，f 2 (g 1 (x)))+H(f 2 (x)，f 1 (g 2 (x)))
the joint loss function to obtain the training result is:
L all ＝L sup +λ cot L cot +λ dif L dif
wherein λ cot And λ dif Are two hyperparameters, controlling the loss function L separately cot And L dif Specific gravity in the joint loss function:
wherein λ max Is λ cot And λ dif Maximum value of (d); α is a growth coefficient; exp (·) is an exponential function based on a natural constant e; t is the current number of training rounds; t is a unit of stable Is a hyperparameter of lambda cot And λ dif Number of training rounds when stable;
updating the feature extractor through reverse gradient propagation to obtain a trained recognition network;
and S3, acquiring a digital modulation signal, preprocessing the digital modulation signal by adopting the method of the step S1, and inputting the preprocessed digital modulation signal into a trained recognition network to finish recognition of the modulation signal.
The invention has the beneficial effects that the invention provides the digital signal modulation recognition method based on the deep collaborative training, and the confrontation sample is generated by utilizing the generation confrontation algorithm to meet the collaborative training assumption, so that the performance of the feature extractor under the training of a large number of samples can be approached under the condition of a very small number of samples. And through the preprocessing of the signal data, the generalization capability of the feature extractor is improved, so that the robustness of the feature extractor is stronger. The deep cooperative training algorithm is applied to the digital signal modulation recognition, so that the defects of the existing method are overcome, and the signal modulation recognition algorithm is more practical and reliable.
Drawings
FIG. 1 is a flow chart of a method for recognizing a digital modulation signal based on deep cooperative training according to the present invention;
fig. 2 is a schematic diagram of the CLDNN network model of the present invention.
Detailed Description
The technical scheme of the invention is described in detail in the summary of the invention, and the method of the invention extracts and utilizes a large amount of information without marked data on the basis of fully utilizing a small amount of information with marked data, so that the characteristic extractor can obtain excellent performance under the condition of a small amount of marked samples, and the practicability of the digital signal modulation identification method is improved.
According to the digital signal modulation recognition method based on deep collaborative training, firstly, data set expansion and network model generalization are carried out through signal preprocessing, and generation of countermeasure data is achieved through a countermeasure algorithm. The challenge data may be considered a different view of the signal samples than the original samples, and the combination of the generated samples and the original samples may satisfy the assumption of co-training. In the training process, the optimization direction of the feature extractor is constrained through a specific loss function, and the emphasis bias of the training is controlled by using coefficient weighting. The two feature extractors can learn each other under the assumption of cooperative training, and optimize each other until the optimal condition is reached. And the modulation type of the unknown signal can be determined by utilizing the trained feature extractor.
Claims (1)
1. A digital modulation signal identification method based on deep collaborative training is characterized by comprising the following steps:
s1, acquiring digital modulation signal data and labels, recording the digital modulation signal data and the labels as a data set D, and preprocessing the data set D, wherein the preprocessing comprises the following steps: partitioning the data set, including a marked data set S and a unmarked data set U, the marked data set being marked
Unmarked data set denoted>
And D = S ^ U and M < N, x is a number of markedAccording to y is the mark corresponding to x, x u The data are unmarked data, M is the number of marked data sets, and N is the number of unmarked data sets; expanding the marked data set;
s2, constructing an identification network, wherein the identification network comprises two CLDNN networks with the same structure, and each CLDNN network comprises a convolution part, an LSTM layer and two full-connection layers; the convolution part comprises three convolution blocks, each convolution block consists of a convolution layer, a batch normalization layer and a ReLU activation function, and a uniform pooling layer with the kernel size of 2 is arranged between every two convolution blocks; two CLDNN networks are used as feature extractors and are respectively marked as f 1 (. Cndot.) and f 2 (. The) the data set obtained in the step S1 is adopted to carry out cooperative training on the recognition network, wherein the cooperative training specifically comprises the following steps:
generating an antagonistic sample data set D ', D' = { g (x) | x ∈ D }, and g (·) is generated into an antagonistic algorithm according to the data set D by adopting an antagonistic generating algorithm; the cross entropy loss of two CLDNN networks on a labeled data set S is:
L sup ＝H(y，f 1 (x))+H(y，f 2 (x))
where H (-) is a cross entropy loss function;
JS divergence loss on unmarked data set U for both CLDNN networks was:
L cot ＝JS(f 1 (x u )||f 2 (x u ))
wherein JS (-) is JS divergence;
CLDNN network f 1 (. Output on the original data set D) and the network f 2 (.) a cross-entropy loss function of the output on the challenge sample data set D', and a network f 2 (. Output on the original data set D) and the network f 1 (. O) the cross entropy loss function of the output on the challenge sample data set D' is:
L dif ＝H(f 1 (x)，f 2 (g 1 (x)))+H(f 2 (x)，f 1 (g 2 (x)))
the joint loss function to obtain the training result is:
L all ＝L sup +λ cot L cot +λ dif L dif
wherein λ cot And λ dif Are two hyperparameters, controlling the loss function L separately cot And L dif Specific gravity in the joint loss function:
wherein λ is max Is λ cot And λ dif Maximum value of (d); α is a growth coefficient; exp (·) is an exponential function based on a natural constant e; t is the current number of training rounds; t is stable Is a hyperparameter λ cot And λ dif Number of training rounds when stable;
updating the feature extractor through reverse gradient propagation to obtain a trained recognition network;
and S3, acquiring a digital modulation signal, preprocessing the digital modulation signal by adopting the method of the step S1, and inputting the preprocessed digital modulation signal into a trained recognition network to finish recognition of the modulation signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110995377.5A CN113705787B (en) | 2021-08-27 | 2021-08-27 | Digital modulation signal identification method based on deep collaborative training |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110995377.5A CN113705787B (en) | 2021-08-27 | 2021-08-27 | Digital modulation signal identification method based on deep collaborative training |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113705787A CN113705787A (en) | 2021-11-26 |
| CN113705787B true CN113705787B (en) | 2023-04-07 |
Family
ID=78655956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110995377.5A Active CN113705787B (en) | 2021-08-27 | 2021-08-27 | Digital modulation signal identification method based on deep collaborative training |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113705787B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114584441A (en) * | 2022-03-02 | 2022-06-03 | 中国人民解放军军事科学院战争研究院 | Digital signal modulation identification method based on deep learning |
| CN114611671B (en) * | 2022-03-15 | 2023-04-07 | 电子科技大学 | Method for automatically modulating and identifying dimension of amplified data based on self-encoder |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108718288A (en) * | 2018-03-30 | 2018-10-30 | 电子科技大学 | Recognition of digital modulation schemes method based on convolutional neural networks |
| CN112115821A (en) * | 2020-09-04 | 2020-12-22 | 西北工业大学 | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107124381B (en) * | 2017-03-27 | 2020-05-22 | 华南理工大学 | Automatic identification method for digital communication signal modulation mode |
| CN110598530A (en) * | 2019-07-30 | 2019-12-20 | 浙江工业大学 | Small sample radio signal enhanced identification method based on ACGAN |
| CN112115963B (en) * | 2020-07-30 | 2024-02-20 | 浙江工业大学 | Method for generating unbiased deep learning model based on transfer learning |
-
2021
- 2021-08-27 CN CN202110995377.5A patent/CN113705787B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108718288A (en) * | 2018-03-30 | 2018-10-30 | 电子科技大学 | Recognition of digital modulation schemes method based on convolutional neural networks |
| CN112115821A (en) * | 2020-09-04 | 2020-12-22 | 西北工业大学 | Multi-signal intelligent modulation mode identification method based on wavelet approximate coefficient entropy |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113705787A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112731309B (en) | Active interference identification method based on bilinear efficient neural network | |
| CN113705787B (en) | Digital modulation signal identification method based on deep collaborative training | |
| CN114157539B (en) | Data-aware dual-drive modulation intelligent identification method | |
| CN113392731B (en) | Modulation signal classification method and system based on graph neural network | |
| CN113657607B (en) | Continuous learning method for federal learning | |
| Lu et al. | Boosting few-shot image recognition via domain alignment prototypical networks | |
| Shu et al. | A SVM multi-class image classification method based on DE and KNN in smart city management | |
| CN111291705B (en) | Pedestrian re-identification method crossing multiple target domains | |
| CN113920472A (en) | Unsupervised target re-identification method and system based on attention mechanism | |
| CN116403058A (en) | Remote sensing cross-scene multispectral laser radar point cloud classification method | |
| CN114722926A (en) | Convolution clustering method for scale vortex time sequence diagram in towed sensor array | |
| CN116524197A (en) | Point cloud segmentation method, device and equipment combining edge points and depth network | |
| CN114189416B (en) | Digital modulation signal identification method based on consistency regularization | |
| Zhang et al. | Heterogeneous deep model fusion for automatic modulation classification | |
| CN114168782B (en) | Deep hash image retrieval method based on triplet network | |
| CN109871850A (en) | A kind of classification method of the mobile lidar data based on neural network model | |
| CN111125395B (en) | CAD drawing retrieval method and system based on double-branch deep learning | |
| CN111428674B (en) | Multi-loss joint training method for keeping consistency of multi-metric space | |
| CN114372538A (en) | Method for convolution classification of scale vortex time series in towed sensor array | |
| CN113239788A (en) | Mask R-CNN-based wireless communication modulation mode identification method | |
| Zhang et al. | Class-based Core Feature Extraction Network for Few-shot Classification | |
| Liu et al. | Gallery based k-reciprocal-like re-ranking for heavy cross-camera discrepancy in person re-identification | |
| CN113300788B (en) | Blind receiver method and device based on Capsule network | |
| CN117635218B (en) | Business district flow prediction method based on six-degree separation theory and graph annotation network | |
| CN114037866B (en) | Generalized zero sample image classification method based on distinguishable pseudo-feature synthesis |
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 |