CN108764013A - A kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks - Google Patents
A kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks Download PDFInfo
- Publication number
- CN108764013A CN108764013A CN201810261711.2A CN201810261711A CN108764013A CN 108764013 A CN108764013 A CN 108764013A CN 201810261711 A CN201810261711 A CN 201810261711A CN 108764013 A CN108764013 A CN 108764013A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- signal
- communication signals
- automatic communication
- 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
- G06F2218/04—Denoising
- G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
-
- 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
-
- 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
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The present invention relates to a kind of automatic Communication Signals Recognitions based on end-to-end convolutional neural networks, it is characterised in that successively executes pretreatment to original I/Q sampled datas of watch window and convolutional neural networks identify two steps.Wherein, pre-treatment step is using the original I of watch window/Q sampled datas as input, after discrete Fourier transform and data format registration process, output spectrum Waterfall plot;Frequency spectrum Waterfall plot of the convolutional neural networks identification step to pre-process gained after CNN feature extraction layers, MLP Feature Mappings layer and BR multi-tags classify layer, exports as input and characterizes the one-dimensional boolean vector that all signals to be identified whether there is.Compared with the pattern of traditional characteristic extraction+Classification and Identification, the present invention is using technology resolving ideas end to end, avoid complicated and inefficient Feature Engineering, the accuracy, robustness and intelligent level for improving signal identification are of great significance to the radio supervision of key area and occasion scene.
Description
Technical field
The invention belongs to radio monitoring technical fields, are related to a kind of based on end-to-end convolutional neural networks
The automatic Communication Signals Recognition of (Convolutional Neural Network, CNN), for solving classical wireless telecommunication number
The excessively complicated problem of the feature extraction faced in identification, improves the ability and intelligent level of radio signal identification.
Background technology
In recent years, with the rise of the development of wireless communication technique, especially Internet of Things, radio wave has become connection
The important carrier of all things on earth.However, the feature of radio wave opening, makes it be easy to be interfered and illegally utilize, cause normal
The serious problems such as the disturbed interruption of communication system, false reaction speech propagation.Therefore, wireless installation has become national security
Important component, enhanced radio monitoring and management, especially at the key areas such as airport, border and occasion scene
Radio supervision, have important practical significance and active demand.
Briefly, radio monitoring includes three tasks:It detects, identify and position.Wherein, detection is connect using radio
Receiving unit obtains basic electromagnetic data;Identification then by data processing extraction is more abstract practical feature, and with this into
Row classification;Finally, for the illegal signals identified, emission source also is searched using the method for radio-positioning, is radio
Supervision and law enforcement provide support.It can be seen that radio signal identification is the distillation of detection, and it is the basis of positioning, wireless
It is in core status in pyroelectric monitor, is always the emphasis and difficult point of radio monitoring technical research.
The radio signal of early stage identifies, is to acquire aerial radiofrequency signal using various Spectral acquisition equipment, by processing
Afterwards, it is shown with visual means such as spectrogram, Waterfall plot, sunset glow figures, then by professional's signal Analysis time-frequency characteristics
And find echo signal.This way is very high to the specialized capability requirement of operating personnel, and when monitoring time is elongated or nothing
When line electric signal is more, manual analysis efficiency and accuracy rate can also decline to a great extent.
Radio signal identification method popular at present is first by expert of the art's well-chosen signal key feature
(such as Cyclic spectrum density, order central moment function, power spectral density maximum value, the standard deviation of instantaneous phase) then utilizes biography
The signal processing algorithm of system calculates characteristic ginseng value, is finally classified again based on unalterable rules or the method for machine learning.
In general, a model needs to choose several therein even more than ten features as mode input, computation complexity is very high, very
Difficult actual deployment uses.Meanwhile being influenced by signal waveform diversity and multipath fading effect, this method is in feature selecting and sentences
Certainly also lack universality in terms of criterion.
Invention content
The purpose of the present invention is:In view of the deficiencies of the prior art, a kind of nothing based on end-to-end convolutional neural networks is proposed
Artificial intelligence and radio signal treatment technology depth integration are solved classical wireless telecommunication number by line electric signal recognition methods
The problem of the manual features extraction excessively complexity faced in identification, improves the recognition capability and intelligent level of radio signal.
Technical solution of the invention is as described below.
A kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks, it is characterised in that watch window
The successively execution pretreatment of original I/Q sampled datas and convolutional neural networks identify two steps, specific as shown in Figure 1.Wherein, in advance
Processing step is aligned using the original I of watch window/Q sampled datas as input by discrete Fourier transform and data format
After processing, output spectrum Waterfall plot;Frequency spectrum Waterfall plot of the convolutional neural networks identification step to pre-process gained is passed through as input
Cross CNN (Convolutional Neural Network, convolutional neural networks, abbreviation CNN) feature extraction layer, MLP (Multi-
Layer Perceptron, multilayer perceptron, abbreviation MLP) Feature Mapping layer and BR (Binary Relevance, binary association,
Abbreviation BR) after multi-tag classification layer, output characterizes the one-dimensional boolean vector that all signals to be identified whether there is.With traditional characteristic
The pattern of extraction+Classification and Identification is compared, and the present invention avoids complicated and inefficient feature using technology resolving ideas end to end
Engineering, improves the accuracy, robustness and intelligent level of signal identification, and reduces computation complexity, specifically such as Fig. 2 institutes
Show.
A kind of above-mentioned automatic Communication Signals Recognition based on end-to-end convolutional neural networks, the pretreatment, be for
Original I/Q sampled datas are transformed to a kind of form being more advantageous to Automatic Feature Extraction and identification.In order to give full play to volume
Advantage of the product neural network in terms of feature extraction, and reduce computation complexity, the present invention wish preprocessing process be it is lossless and
It is easy calculating.According to discrete signal processing theory and Modern Communication System framework, original I/Q sampled datas are exactly that number is logical
Believe the input of System Back-end processing, the digital signal processing algorithm of any complexity is all to be obtained on this basis by mathematic(al) manipulation
It arrives, therefore it has naturally contained without damage information.Mathematically, discrete Fourier transform is the lossless change to I/Q sampled datas
It changes, therefore has equally contained all information.Frequency domain presentation form of the result of discrete Fourier transform simultaneously as signal, more has
Conducive to carrying out signal identification under complex electromagnetic environment.Therefore, the present invention carries out direct computation of DFT to original I/Q sampled datas first
Leaf transformation obtains the frequency domain presentation of signal, then carries out subsequent processing again.
A kind of above-mentioned automatic Communication Signals Recognition based on end-to-end convolutional neural networks, the data format pair
Together, it is the format that the result of continuous multigroup discrete Fourier transform is combined into frequency spectrum Waterfall plot, increases the time domain of single sample
Information content, the processing convenient for follow-up convolutional neural networks and signal identification.Discrete fourier change is carried out to original I/Q sampled datas
Changing visually to be considered " line " as a result, be an one-dimensional sequence of complex numbers, it indicates that the frequency spectrum sometime put is close
Degree spectrum.But radio signal waveform always changes over time, the feature for only observing one " time point " is difficult to accurately identify letter
Number, therefore the observed result at continuous multiple time points is subjected to stacked combination, become in " face ", so that it may to express time domain simultaneously
With the information characteristics of frequency domain, signal identification rate is helped to improve.In this two dimensional surface, horizontal axis indicates that frequency, the longitudinal axis indicate
Time, element value are exactly the end value of discrete Fourier transform.If element value mapped by gray value of image, so that it may with
A visual picture is drawn, which is exactly frequency spectrum Waterfall plot, specific as shown in Figure 3.
A kind of above-mentioned automatic Communication Signals Recognition based on end-to-end convolutional neural networks, the frequency spectrum Waterfall plot,
Size is M × N × 2.Wherein, 2 indicate that picture is two channels, the respectively channels I and the channels Q;N indicates discrete Fourier transform
Points, in order to application fast fourier transform algorithm realize discrete Fourier transform, to accelerate arithmetic speed, N values need to take 2
Positive integer power;M indicates that the group number of the discrete Fourier transform result merged in a frequency spectrum Waterfall plot, size reflect one
The time span of a sample, M values are bigger, and the time span of characterization is longer, but calculate more complicated.
A kind of above-mentioned automatic Communication Signals Recognition based on end-to-end convolutional neural networks, the convolutional neural networks
Identification, including CNN feature extraction layers, MLP Feature Mappings layer and BR multi-tags classification layer.Wherein, CNN feature extraction layers are using more
Level convolutional neural networks structure, while from the angle extraction signal high dimensional feature of time domain and frequency domain;MLP Feature Mapping layers use
The multilayer perceptron structure connected entirely carries out dimensionality reduction mapping to the signal characteristic extracted;BR multi-tags classify layer using binary
(Binary Relevance) method of association carries out multi-tag classification to receiving data, when can rule out this simultaneously/frequency observation window in
All signals to be identified whether there is.Wherein, when/size of frequency observation window determined by the frequency spectrum Waterfall plot inputted, frequency spectrum waterfall
The span of Butut horizontal axis is exactly frequency watch window, and the span of the longitudinal axis is exactly time watch window.
A kind of above-mentioned automatic Communication Signals Recognition based on end-to-end convolutional neural networks, BR multi-tags classification
Layer carries out multi-tag classification using binary correlation method, is made of multiple parallel single label single classifiers, each single point of label of list
Class device judges that one of which signal whether there is, and the output result of all list label single classifiers is combined, and constitutes an energy
Enough characterize the one-dimensional boolean vector that all signals to be identified whether there is.Wherein vectorial length is equal to signal type to be identified
Sum when vectorial i-th of element value is True, indicates that i-th of signal exists, otherwise be False, indicates i-th of signal
It is not present.Under complex electromagnetic environment, it is difficult to observe single, pure signal, it is substantially the aliasing of multiple signals, because
This just should not only export a label as a result, and should use a multi-tag knot in adjudicating the identification of aliasing signal
Fruit indicates.By comparative analysis, the present invention carries out multi-tag classification using binary correlation method.Binary correlation method ties up a m more
Labeling problem is converted into m single labeling problem, can be with when type identification if necessary to increase m+1 dimensional signals
It completely dispenses with and changes existing model training as a result, directly carrying out the training of the m+1 model parameter, be very easy to realize
Incremental learning, this is extremely important in practical applications, and other multi-tag sorting techniques such as classifier chains method, label power set be legal
Do not have the advantage then.
The advantages of the present invention over the prior art are that:
(1) present invention is avoided using the structure radio signal identification model of thinking end to end compared with traditional mode
Complicated feature selecting and pattern measurement link, automatically extracts signal high dimensional feature using convolutional neural networks, improves signal
The accuracy and robustness of identification, and reduce computation complexity.
(2) present invention uses input of the frequency spectrum Waterfall plot as convolutional neural networks, with computation complexity is low, information is complete
It is whole without lose, when/advantages such as frequency domain character ability to express is strong, and frequency spectrum Waterfall plot is more easy to as a kind of two dimensional data layout
Being fruitful for field of image recognition is used for reference, Project Realization is facilitated.
(3) present invention is using multi-tag classification results as the output of convolutional neural networks, when can rule out this simultaneously/frequency
All signals to be identified whether there is in observation window, have unique advantage in the identification of aliasing signal under complex electromagnetic environment.
Wherein, used binary is associated with multi-tag classification, have realize simple, training data easily mark, can the advantages such as incremental learning.
Description of the drawings
Fig. 1 is a kind of radio signal identification model general block diagram based on end-to-end convolutional neural networks in the present invention;
Fig. 2 is the contrast schematic diagram of end-to-end system mentality of designing and conventional thought in the present invention;
Fig. 3 is that the frequency spectrum Waterfall plot of preprocessing module in the present invention synthesizes schematic diagram;
Fig. 4 is a kind of example of Gnuradio signal acquisition flow graphs in the present invention;
Fig. 5 is a kind of structure chart of the real-time radio electric signal identifying system based on the present invention.
Specific implementation mode
Below in conjunction with attached drawing and implementation example, the present invention is described in further detail.Pass through description detailed enough
These implement example so that skilled artisans appreciate that and the practice present invention.In the purport and model for not departing from the present invention
In the case of enclosing, logic, realize and others change can be made to implementation.Therefore, described further below should not be by
It is interpreted as limited significance, the scope of the present invention is only defined solely by the appended claims.
Foregoing invention content is from the level of core ideas and algorithm, to a kind of based on the wireless of end-to-end convolutional neural networks
Electric signal recognition methods is described in detail.But during actual development, other than realizing present disclosure, also need
Want some extra works.A kind of specific implementation mode of the present invention will be introduced step by step below.
1. building hardware and software development environment
Since the present invention relates to two subject directions of radio monitoring and artificial intelligence, required development environment is also relatively more mixed
It is miscellaneous, it is specific as shown in table 1.Wherein, general software radio platform (USRP) is completed for acquiring radio signal, desktop PC machine
The storage of radio signal, the training of model and the operation of model.Due to being related to training and the big data of deep neural network
The storage of amount, it is higher to the configuration requirement of PC machine.
1 hardware and software development environment of table
2. gathered data
It needs to use USRP equipment and PC machine when gathered data.First, signal acquisition stream is built using Gnuradio softwares
Figure, a kind of feasible flow graph design scheme develop study course as shown in figure 4, specific construction method please voluntarily refers to Gnuradio.Stream
Figure is the concept in Gnuradio software architectures, and essence is exactly to utilize Gnuradio built-in tool packets, a description of structure
The python codes of information flow and processing unit function.Then, PC machine is connect with USRP equipment, runs the flow graph, can incited somebody to action
Radio signal is stored in PC machine hard disk in real time with the format of I/Q sampled datas.
3. modelling
End-to-end convolutional neural networks model is built according to the description of Fig. 1 and foregoing invention content.Assuming that needing to identify m kinds
Signal, one group of feasible model parameter configuration are as shown in the table.
A kind of end-to-end convolutional neural networks model of table 2
4. data preparation and mark
The arrangement of data and annotation process are comparatively laborious, mainly by being accomplished manually.First, continuous 262144 samplings are intercepted
Point, as a sample data.Then any feasible method can be used to be labeled the type of contained signal in the sample,
Annotation results format sample is as follows, is indicated with Y.Y is the one-dimensional boolean vector that a length is m, wherein m indicates this model one
It can recognize that m kind signal types altogether.If i-th kind of signal exists, by YiIt is assigned a value of True, it is on the contrary then be assigned a value of False.
Y={ True, False, True, False ... ..., False, True }m
5. model training and test
The training of model only needs to use PC machine with test.Utilize what is carried in 1.4 or more versions of tensorflow
Keras increases income image recognition tool packet, the model parameter defined in step 3, with python language write model training and
The code of test specifically refers to keras study courses.If test result is bad, it can be used and increase data volume, modification model parameter
The methods of carry out
Adjustment repeats this iterative process, until meeting index request.
6. model uses
After the completion of model training, it is also necessary to which actual deployment uses.A kind of real-time radio electric signal identification based on the present invention
The structure of system is as shown in Figure 5.Wherein, USRP is changed as goods shelf products without any software and hardware;Needed in PC machine 3 from
Software module is defined, is Gnuradio signal acquisition modules, preprocessing module and convolutional neural networks identification module respectively.
Gnuradio signal acquisition modules are similar with the Gnuradio signal acquisition flow graphs mentioned in step 2, are used for controlling
USRP equipment processed, and complete the acquisition and storage of original I/Q sampled datas.But it in order to coordinate other module cooperatives to work, needs
Change details at two:
● framing storage is carried out to original I/Q sampled datas, every 262144 continuous sampling point is divided into a frame, is then stored in
One individual file, is placed in hard disk, as the archive of initial data and the buffering of subsequent processing.
● after the storage for often completing a frame data, " data frame arriving signal " is sent to preprocessing module, informs pretreatment
Module has new data frame arrival, can start subsequent arithmetic.The format and sending method of the signal are not intended to be limited in any, it can be with
Using modes of intelligence transmission such as socket, message queue, semaphores.
Preprocessing module and the preprocessing module function defined in step 5 are essentially identical, are used for reading original I/Q samplings
Data, and by agreement transformation, synthesis frequency spectrum Waterfall plot.Only difference is that need to receive " data frame arriving signal " herein,
And when start subsequent arithmetic according to the signal deciding.
Convolutional neural networks identification module and the code of step 5 model training stage differ greatly, and are embodied in:
● since step 5 has completed model training, and optimum model parameter is deposited automatically, therefore, has only been needed herein
It is loaded directly into the model, without training again.Specifically, being exactly to call myModel=in code
Keras.models.load_model (filepath) function, the best model that wherein filepath is arranged when being model training
Storing path.
● the frequency spectrum Waterfall plot of input is predicted using existing model, Y=myModel.predict can be directly invoked
(X) function, wherein X are the frequency spectrum Waterfall plots of input, and Y is the prediction result of output.
● it is last, according to application requirement, need to export prediction result.Or it issues this PC machine other information and shows mould
Block, or other PC machine shown dedicated for information are issued by physical link, or even it is sent to internet cloud end, for remote
Journey user checks.The present invention does not have this any restrictions.
Non-elaborated part of the present invention belongs to techniques known.
Claims (6)
1. a kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks, it is characterised in that:To watch window
Original I/Q sampled datas successively carry out pretreatment and convolutional neural networks identify two steps, wherein:
Pre-treatment step:Using the original I of watch window/Q sampled datas as input, by discrete Fourier transform and data lattice
After formula registration process, output spectrum Waterfall plot;
Convolutional neural networks identification step:Using pretreatment gained frequency spectrum Waterfall plot as input, by CNN feature extraction layers,
After MLP Feature Mappings layer and BR multi-tags classification layer, output characterizes the one-dimensional boolean vector that all signals to be identified whether there is.
2. a kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks according to claim 1, special
Sign is:The frequency spectrum Waterfall plot, size are M × N × 2;Wherein, 2 indicate that picture is two channels, the respectively channels I and Q
Channel;N indicates that the points of discrete Fourier transform, N values take 2 positive integer power;M indicates merging in a frequency spectrum Waterfall plot
The group number of discrete Fourier transform result, M values take positive integer.
3. a kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks according to claim 1, special
Sign is:The one-dimensional boolean vector, length indicate the sum of signal to be identified;The value of i-th of element of the vector is
It when True, indicates that i-th of signal exists, otherwise be False, indicates that i-th of signal is not present.
4. a kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks according to claim 1, special
Sign is:The CNN feature extraction layers, sample multi-level convolutional neural networks structure, while from the angle of time domain and frequency domain
Extract signal high dimensional feature.
5. a kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks according to claim 1, special
Sign is:The MLP Feature Mapping layers carry out the signal characteristic extracted using the multilayer perceptron structure connected entirely
Dimensionality reduction maps.
6. a kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks according to claim 1, special
Sign is:The BR multi-tags classification layer, carries out multi-tag classification, by multiple parallel using binary correlation method to receiving data
Single label single classifier composition, each list label single classifier judges that one of which signal whether there is, all single label lists
The output result of grader is combined, constitute one can characterize one-dimensional boolean that all signals to be identified whether there is to
Amount.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810261711.2A CN108764013A (en) | 2018-03-28 | 2018-03-28 | A kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810261711.2A CN108764013A (en) | 2018-03-28 | 2018-03-28 | A kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108764013A true CN108764013A (en) | 2018-11-06 |
Family
ID=63980440
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810261711.2A Pending CN108764013A (en) | 2018-03-28 | 2018-03-28 | A kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108764013A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109756367A (en) * | 2018-12-24 | 2019-05-14 | 云南大学 | A kind of radio monitoring system and method based on edge calculations |
CN110996343A (en) * | 2019-12-18 | 2020-04-10 | 中国人民解放军陆军工程大学 | Interference recognition model based on deep convolutional neural network and intelligent recognition algorithm |
CN111382803A (en) * | 2020-03-18 | 2020-07-07 | 电子科技大学 | Feature fusion method based on deep learning |
CN111404852A (en) * | 2020-03-03 | 2020-07-10 | 西安电子科技大学 | Modulation mode identification method based on amplitude and spectral amplitude characteristics |
CN111474955A (en) * | 2020-04-22 | 2020-07-31 | 上海特金信息科技有限公司 | Unmanned aerial vehicle image signal system identification method, device, equipment and storage medium |
CN111507299A (en) * | 2020-04-24 | 2020-08-07 | 中国人民解放军海军航空大学 | Method for identifying STBC (space time Block coding) signal on frequency domain by using convolutional neural network |
CN111652108A (en) * | 2020-05-28 | 2020-09-11 | 中国人民解放军32802部队 | Anti-interference signal identification method and device, computer equipment and storage medium |
CN111985567A (en) * | 2020-08-21 | 2020-11-24 | 河北先河环保科技股份有限公司 | Automatic pollution source type identification method based on machine learning |
CN112818891A (en) * | 2021-02-10 | 2021-05-18 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Intelligent identification method for communication interference signal type |
CN113037411A (en) * | 2019-12-24 | 2021-06-25 | 清华大学 | Multi-user signal detection method and device based on deep learning |
CN113311366A (en) * | 2020-02-27 | 2021-08-27 | Aptiv技术有限公司 | Harness testing device and method for verifying connection during assembly of harness |
CN113542180A (en) * | 2021-06-30 | 2021-10-22 | 北京频谱视觉科技有限公司 | Frequency domain identification method of radio signal |
CN113780106A (en) * | 2021-08-24 | 2021-12-10 | 电信科学技术第五研究所有限公司 | Deep learning signal detection method based on radio waveform data input |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107122738A (en) * | 2017-04-26 | 2017-09-01 | 成都蓝色起源科技有限公司 | Automatic Communication Signals Recognition based on deep learning model and its realize system |
-
2018
- 2018-03-28 CN CN201810261711.2A patent/CN108764013A/en active Pending
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107122738A (en) * | 2017-04-26 | 2017-09-01 | 成都蓝色起源科技有限公司 | Automatic Communication Signals Recognition based on deep learning model and its realize system |
Non-Patent Citations (3)
Title |
---|
ANDREW MAXWELL等: "Deep learning architectures for multi-label classification of intelligent health risk prediction", 《BMC BIOINFORMATICS》 * |
孙松涛等: "基于CNN特征空间的微博多标签情感分类", 《工程科学与技术》 * |
梁微等: "地面无线广播电视频谱检测方法", 《广播电视信息》 * |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109756367A (en) * | 2018-12-24 | 2019-05-14 | 云南大学 | A kind of radio monitoring system and method based on edge calculations |
CN110996343A (en) * | 2019-12-18 | 2020-04-10 | 中国人民解放军陆军工程大学 | Interference recognition model based on deep convolutional neural network and intelligent recognition algorithm |
CN110996343B (en) * | 2019-12-18 | 2023-06-30 | 中国人民解放军陆军工程大学 | Intelligent recognition system and recognition method of interference recognition model based on deep convolutional neural network |
CN113037411A (en) * | 2019-12-24 | 2021-06-25 | 清华大学 | Multi-user signal detection method and device based on deep learning |
CN113037411B (en) * | 2019-12-24 | 2022-02-15 | 清华大学 | Multi-user signal detection method and device based on deep learning |
CN113311366A (en) * | 2020-02-27 | 2021-08-27 | Aptiv技术有限公司 | Harness testing device and method for verifying connection during assembly of harness |
CN111404852A (en) * | 2020-03-03 | 2020-07-10 | 西安电子科技大学 | Modulation mode identification method based on amplitude and spectral amplitude characteristics |
CN111404852B (en) * | 2020-03-03 | 2021-04-13 | 西安电子科技大学 | Modulation mode identification method based on amplitude and spectral amplitude characteristics |
CN111382803A (en) * | 2020-03-18 | 2020-07-07 | 电子科技大学 | Feature fusion method based on deep learning |
CN111382803B (en) * | 2020-03-18 | 2022-06-03 | 电子科技大学 | Communication signal feature fusion method based on deep learning |
CN111474955A (en) * | 2020-04-22 | 2020-07-31 | 上海特金信息科技有限公司 | Unmanned aerial vehicle image signal system identification method, device, equipment and storage medium |
CN111507299A (en) * | 2020-04-24 | 2020-08-07 | 中国人民解放军海军航空大学 | Method for identifying STBC (space time Block coding) signal on frequency domain by using convolutional neural network |
CN111652108A (en) * | 2020-05-28 | 2020-09-11 | 中国人民解放军32802部队 | Anti-interference signal identification method and device, computer equipment and storage medium |
CN111985567A (en) * | 2020-08-21 | 2020-11-24 | 河北先河环保科技股份有限公司 | Automatic pollution source type identification method based on machine learning |
CN112818891A (en) * | 2021-02-10 | 2021-05-18 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Intelligent identification method for communication interference signal type |
CN112818891B (en) * | 2021-02-10 | 2022-09-02 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Intelligent identification method for communication interference signal type |
CN113542180A (en) * | 2021-06-30 | 2021-10-22 | 北京频谱视觉科技有限公司 | Frequency domain identification method of radio signal |
CN113780106A (en) * | 2021-08-24 | 2021-12-10 | 电信科学技术第五研究所有限公司 | Deep learning signal detection method based on radio waveform data input |
CN113780106B (en) * | 2021-08-24 | 2024-02-27 | 电信科学技术第五研究所有限公司 | Deep learning signal detection method based on radio waveform data input |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108764013A (en) | A kind of automatic Communication Signals Recognition based on end-to-end convolutional neural networks | |
CN105022835B (en) | A kind of intelligent perception big data public safety recognition methods and system | |
CN108416290A (en) | Radar signal feature method based on residual error deep learning | |
CN114564982B (en) | Automatic identification method for radar signal modulation type | |
CN111178260A (en) | Modulation signal time-frequency diagram classification system based on generation countermeasure network and operation method thereof | |
US20220036231A1 (en) | Method and device for processing quantum data | |
Wei et al. | PRI modulation recognition based on squeeze-and-excitation networks | |
CN111050315B (en) | Wireless transmitter identification method based on multi-core two-way network | |
CN110148425A (en) | A kind of camouflage speech detection method based on complete local binary pattern | |
CN112749633B (en) | Separate and reconstructed individual radiation source identification method | |
Li et al. | Automatic modulation classification based on bispectrum and CNN | |
Ocquaye et al. | Dual exclusive attentive transfer for unsupervised deep convolutional domain adaptation in speech emotion recognition | |
CN114980122A (en) | Small sample radio frequency fingerprint intelligent identification system and method | |
CN105184236A (en) | Robot-based face identification system | |
CN110309774A (en) | Iris segmentation method, apparatus, storage medium and electronic equipment | |
Wang et al. | Specific emitter identification based on deep adversarial domain adaptation | |
CN117131436A (en) | Radiation source individual identification method oriented to open environment | |
Xu et al. | A jamming recognition algorithm based on deep neural network in satellite navigation system | |
Zheng et al. | Boundary adjusted network based on cosine similarity for temporal action proposal generation | |
CN115600101A (en) | Unmanned aerial vehicle signal intelligent detection method and device based on priori knowledge | |
CN112946567B (en) | Moving target fingerprint indoor positioning method based on domain antagonism neural network | |
Zhang et al. | Interference signal recognition based on multi-modal deep learning | |
Inani et al. | Machine Learning based framework for Drone Detection and Identification using RF signals | |
Yang et al. | Conventional neural network-based radio frequency fingerprint identification using raw I/Q data | |
CN115238749B (en) | Modulation recognition method based on feature fusion of transducer |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181106 |
|
WD01 | Invention patent application deemed withdrawn after publication |