CN109255313A - A kind of method of promotion signal recognition correct rate - Google Patents
A kind of method of promotion signal recognition correct rate Download PDFInfo
- Publication number
- CN109255313A CN109255313A CN201811000707.7A CN201811000707A CN109255313A CN 109255313 A CN109255313 A CN 109255313A CN 201811000707 A CN201811000707 A CN 201811000707A CN 109255313 A CN109255313 A CN 109255313A
- Authority
- CN
- China
- Prior art keywords
- dimension
- dimensionality reduction
- accuracy
- signal
- correct rate
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention discloses a kind of methods of promotion signal recognition correct rate, it is characterized in that, the method comprises the following steps: the pretreatment of (1) data uses CUDA to program so that bispectrum transformation can be realized in GPU platform using the identical ABPSK actual signal acquisition data of lead code；(2) dimensionality reduction is carried out to signal characteristic using t-SNE algorithm；(3) use Adam as the optimizer of t-SNE algorithm, the method of the present invention improves the speed of feature extraction, alleviate because of the excessively high bring dimension disaster problem of characteristic dimension, and have preferable effect to the tagsort after dimensionality reduction, accuracy and speed are all increased.
Description
Technical field
The invention belongs to communication signal recognition field, in particular to a kind of method of promotion signal recognition correct rate.
Background technique
In Individual Identification in Communication Signals, parameter Estimation and feature extraction are the extremely important problems of two classes, accurately into
Row target parameter estimation and feature extraction can guarantee communication countermeasure and oppose anti-effective implementation.Signal individual identification refers to logical
It crosses and collects signal by centainly handling the process for identifying signal source, identify that signal source individual has in electronic countermeasure
Important meaning, grasp other side's radar running parameter be equal to grasped confrontation initiative.
ABPSK (Aeronautical Binary Phase Shift Keying) is the one of DBPSK (2 phase differential keying)
Kind special shape, is to have used for reference the principle of QPSK to a kind of common improved modulation system of BPSK, utilized special binary system
Differential encoding and orthogonal modulation technique convert 180 ° of phase changes in DBPSK to the ABPSK of 90 ° of phase changes, due to same
ABPSK signal under communication system suffers from identical lead code, belongs to and communicates the knowledge of radiation source individual under identical operating mode
Other problem just can only distinguish signal source by fine feature difference between lead code after getting rid of energy response, pass through
The development in many years is crossed, these, which extract the reason of characterization method derives many mutation, generally causes difference, has inside signal source
Component unstability, performance parameter have non-linear, and common fine feature extracting mode is known as fingerprint characteristic, however this
Category feature is generally unable to reach theoretical performance in practical applications, after getting rid of energy response, can not by lead code it
Between fine feature difference come the problem of distinguishing signal source, and there are accuracy and speed be not high to single signal tagsort
Defect, traditional method can not be solved the problems, such as because of the excessively high bring dimension disaster of characteristic dimension.
Summary of the invention
In order to solve the above-mentioned technical problem, the present invention proposes following technical scheme:
A kind of method of promotion signal recognition correct rate, method and step are as follows:
(1) pretreatment of data is made using the identical ABPSK actual signal acquisition data of lead code using CUDA programming
Obtaining bispectrum transformation can realize in GPU platform；
(2) dimensionality reduction is carried out to signal characteristic using t-SNE algorithm, reduces the dimension because of bispectrum transformation results in step (1)；
(3) use Adam as the optimizer of t-SNE algorithm.
Further scheme is,
T-SNE algorithm is introduced as when SVM ties up higher kernel function using VC and classified in the step (2).
Further scheme is,
Dimensionality reduction in the step (2) is manifold dimension-reducing.
Beneficial effect by adopting the above technical scheme is:
The method of the present invention improves the speed of feature extraction, alleviates because of the excessively high bring dimension disaster problem of characteristic dimension,
And there is preferable effect to the tagsort after dimensionality reduction, accuracy and speed are all increased.
Detailed description of the invention
Fig. 1 .ABPSK Signal Pretreatment flow chart；
The road Fig. 2 .I lead code is down to two-dimensional result；
The road Fig. 3 .Q lead code is down to two-dimensional result；
Fig. 4 bispectrum map of magnitudes is down to two-dimensional result；
Average value of the tri- kinds of features of Fig. 5 in each dimension lower linear core classification accuracy rate；
The average value of tri- kinds of feature polynomial kernel classification accuracy rates under each dimension of Fig. 6；
The average value of tri- kinds of features of Fig. 7 RBF core classification accuracy rate under each dimension
Specific embodiment
A kind of method of promotion signal recognition correct rate, method and step are as follows:
(1) pretreatment of data is made using the identical ABPSK actual signal acquisition data of lead code using CUDA programming
Obtaining bispectrum transformation can realize in GPU platform；
The pre-treatment step of data is sampled as shown in Figure 1, using the identical ABPSK actual signal acquisition data of lead code
Rate is 6000, and by manually marking label, 235 samples of totally 10 class, are the Different Individual of same modulation system, and signal-to-noise ratio exists
Between 5dB-20dB.
(2) when SVM, which ties up higher kernel function using VC, to be classified, signal characteristic is flowed using t-SNE algorithm
Shape dimensionality reduction reduces the dimension because of bispectrum transformation results in step (1)；
The lead code for the IQ two-way for having been removed energy response and bispectrum map of magnitudes are down to 2 dimensions respectively, after dimensionality reduction
Result it is as shown in Figure 2 and Figure 3, number in figure is that the class number of signal can will become apparent from signal after being down to 2 dimensions
Situation is clustered, distance of the sample between two-dimensional space reflects sample distance of distance between luv space, it is shown that
T-SNE powerful performance in terms of signal data visualization.
It can be seen that being obviously divided into two groups after lead code dimensionality reduction from Fig. 2,3,4, and organize interior sample and be not easily distinguishable, passes through
After bispectrum transformation and t-SNE dimensionality reduction, sample is divided into three groups, and organizes interior sample area and be easier to distinguish than lead code.
(3) use Adam as the optimizer of t-SNE algorithm.
Experimental example 1
This experimental example is influence of the analysing energy to classification results, upsets sequence into row stochastic division to 235 samples,
165 training samples, 70 test samples, respectively with the spy after with the normalized feature of non-energy and energy normalized
Sign, accuracy are the average value of ten random selection test sample accuracy, are classified using SVM, test has used three kinds
Various Mercer kernel functions are tested, and feature includes that IQ two-way lead code and bispectrum map of magnitudes, test result are as shown in table 1:
Influence of the table 1 whether there is or not energy response for classification accuracy rate
As can be seen from the table, energy influences the accuracy of classification especially big, and classification difficulty is bright later for energy normalized
Aobvious to become larger, traditional recognition method all can be the parameter (such as centre frequency, code frequency) and the normalized feature of non-energy of signal
Fusion Features are carried out, classification accuracy rate is generally all very high, when using single features after energy normalized, how to extract
Effective information becomes to extract the key of fine feature.
Experimental example 2
This experimental example is to analyze dimension to influence the performance of classification results, is randomly assigned 165 training samples and 70 surveys
Sample sheet carries out dimensionality reduction using t-SNE dimensionality reduction mode, drops to different dimensions and attempt to look for out optimal dimension, accuracy is
The average value of ten accuracy, specific as shown in table 2:
Accuracy promotes comparison after 2 energy normalized Feature Dimension Reduction of table
Dimensionality reduction has been obviously improved the accuracy of classification in the case where SVM kernel function VC ties up higher situation as can be seen from Table 2,
It alleviates because of the excessively high bring overfitting problem of dimension.When SVM is classified using linear kernel, IQ two-way lead code and bispectrum amplitude
Classification accuracy rate improves 11.34%, 13.72% and 5.71% than original signal respectively after figure dimensionality reduction, and all features are in 200 dimensions
When classify average accuracy highest.When SVM is classified using polynomial kernel, bispectrum map of magnitudes dimensionality reduction accuracy in 5 dimension
Highest improves 21.34%, Q two-way lead code all dimensionality reductions accuracy highest in 3 dimension than original signal, than original signal point
16.43% and 24.15% are not improved.When SVM is classified using RBF core, bispectrum map of magnitudes dimensionality reduction is classified just in 6 dimension
True rate highest, improving the best dimension of 50.86%, IQ two-way lead code dimensionality reduction than initial data accuracy is respectively 5 peacekeepings 4
21.86% and 27.28% has been respectively increased than original signal in dimension, accuracy.
It, can from Fig. 5 such as Fig. 5,6,7 it can be seen that bispectrum map of magnitudes is obviously more preferable than the lead code classifying quality of original signal
To find out, when SVM is classified using linear kernel function, dimension more mitigates the accuracy trend of classification, in preceding 200 dimension with dimension
Degree rises accuracy and is slowly promoted, all features accuracy highest in 200 dimension.Using polynomial kernel as can be seen from Figure 6
Bispectrum map of magnitudes accuracy when dimensionality reduction is less than 200 dimension does not change significantly when function category, has when dimension is greater than 200 dimension
Apparent accuracy decline, the accuracy highest in 5 dimension.IQ two-way lead code dimensionality reduction is gradual with the rising accuracy of dimension
Decline, all the accuracy highest in 3 dimension.In bispectrum map of magnitudes and original letter when being classified as can be seen from Figure 7 using RBF kernel function
Accuracy is more steady when number lead code dimensionality reduction is less than 6 dimension, and significantly decrease trend when more than 6 dimension, and bispectrum map of magnitudes is in 6 dimension
Classification accuracy rate highest, the best dimension of IQ two-way lead code are respectively 5 peacekeepings 4 dimension.
Although hereinbefore having been made with reference to some embodiments, present invention is described, of the invention not departing from
In the case where range, it can be carried out various improvement and can with equivalent without replacement technical point therein, especially, as long as
There is no technical contradiction, the various features in the various embodiments of institute's careless mistake of the present invention can be combined by either type and be made
It is only in omitting length and economize on resources with, the description for not carrying out exhaustive row to the case where these combinations in the present invention
Consider.Therefore, the invention is not limited to specific embodiments disclosed herein, and including falling into claim.
Claims (3)
1. a kind of method of promotion signal recognition correct rate, which is characterized in that the method comprises the following steps:
(1) pretreatment of data uses CUDA programming so that double using the identical ABPSK actual signal acquisition data of lead code
Spectral transformation can be realized in GPU platform；
(2) dimensionality reduction is carried out to signal characteristic using t-SNE algorithm, reduces the dimension because of bispectrum transformation results in step (1)；
(3) use Adam as the optimizer of t-SNE algorithm.
2. a kind of method of promotion signal recognition correct rate according to claim 1, which is characterized in that the step (2)
Middle t-SNE algorithm is introduced as when SVM ties up higher kernel function using VC and classified.
3. a kind of method of promotion signal recognition correct rate according to claim 1, which is characterized in that the step (2)
In dimensionality reduction be manifold dimension-reducing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811000707.7A CN109255313A (en) | 2018-08-30 | 2018-08-30 | A kind of method of promotion signal recognition correct rate |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811000707.7A CN109255313A (en) | 2018-08-30 | 2018-08-30 | A kind of method of promotion signal recognition correct rate |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109255313A true CN109255313A (en) | 2019-01-22 |
Family
ID=65049558
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811000707.7A Pending CN109255313A (en) | 2018-08-30 | 2018-08-30 | A kind of method of promotion signal recognition correct rate |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109255313A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102868653A (en) * | 2012-09-10 | 2013-01-09 | 电子科技大学 | Digital modulation signal classification method based on bispectrum and sparse matrix |
| US20170140417A1 (en) * | 2015-11-18 | 2017-05-18 | Adobe Systems Incorporated | Campaign Effectiveness Determination using Dimension Reduction |
| CN107886085A (en) * | 2017-11-29 | 2018-04-06 | 华北电力大学（保定） | A kind of electrical energy power quality disturbance feature extracting method based on t SNE |
| CN108197581A (en) * | 2018-01-10 | 2018-06-22 | 厦门大学 | A kind of unmanned plane signal identification detection algorithm based on improvement AC-WGANs |
| CN108427966A (en) * | 2018-03-12 | 2018-08-21 | 成都信息工程大学 | A kind of magic magiscan and method based on PCA-LDA |
-
2018
- 2018-08-30 CN CN201811000707.7A patent/CN109255313A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102868653A (en) * | 2012-09-10 | 2013-01-09 | 电子科技大学 | Digital modulation signal classification method based on bispectrum and sparse matrix |
| US20170140417A1 (en) * | 2015-11-18 | 2017-05-18 | Adobe Systems Incorporated | Campaign Effectiveness Determination using Dimension Reduction |
| CN107886085A (en) * | 2017-11-29 | 2018-04-06 | 华北电力大学（保定） | A kind of electrical energy power quality disturbance feature extracting method based on t SNE |
| CN108197581A (en) * | 2018-01-10 | 2018-06-22 | 厦门大学 | A kind of unmanned plane signal identification detection algorithm based on improvement AC-WGANs |
| CN108427966A (en) * | 2018-03-12 | 2018-08-21 | 成都信息工程大学 | A kind of magic magiscan and method based on PCA-LDA |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101917369B (en) | Method for identifying modulation mode of communication signal | |
| Linder et al. | Identification of tumor epithelium and stroma in tissue microarrays using texture analysis | |
| CN107682109B (en) | A kind of interference signal classifying identification method suitable for UAV Communication system | |
| CN102982349A (en) | Image recognition method and device | |
| CN103839078A (en) | Hyperspectral image classifying method based on active learning | |
| CN107798351B (en) | Deep learning neural network-based identity recognition method and system | |
| CN104040561A (en) | Method for identifying microorganisms via mass spectrometry and score normalisation | |
| Wang et al. | Automatic modulation classification based on joint feature map and convolutional neural network | |
| Song et al. | Comparative study of part-based handwritten character recognition methods | |
| CN102110323B (en) | Method and device for examining money | |
| Biswas et al. | Writer identification of Bangla handwritings by radon transform projection profile | |
| CN110717540A (en) | Method and device for identifying new radar source individuals | |
| CN115294615A (en) | Radio frequency fingerprint identification method based on machine learning | |
| CN104268557B (en) | Polarization SAR sorting technique based on coorinated training and depth S VM | |
| CN113128584B (en) | Mode-level unsupervised sorting method of multifunctional radar pulse sequence | |
| CN108809874B (en) | Radar and communication multi-signal classification method based on circulation support vector machine | |
| CN103246877A (en) | Image contour based novel human face recognition method | |
| CN112598084B (en) | Vehicle type identification method and terminal based on image processing | |
| CN111428064B (en) | Small-area fingerprint image fast indexing method, device, equipment and storage medium | |
| CN109255313A (en) | A kind of method of promotion signal recognition correct rate | |
| CN102902984A (en) | Remote-sensing image semi-supervised projection dimension reducing method based on local consistency | |
| CN112183300A (en) | AIS radiation source identification method and system based on multi-level sparse representation | |
| Niels et al. | Dynamic time warping applied to Tamil character recognition | |
| CN103235030B (en) | Distillate spirit brand identification method based on support vector machine and time-of-flight mass spectrometry | |
| CN114727293A (en) | Identity counterfeit attack identification method based on radio frequency fingerprint and unsupervised learning |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190122 |