CN109245854A - A kind of end-to-end wireless communication system and method based on AI - Google Patents
A kind of end-to-end wireless communication system and method based on AI Download PDFInfo
- Publication number
- CN109245854A CN109245854A CN201810929721.9A CN201810929721A CN109245854A CN 109245854 A CN109245854 A CN 109245854A CN 201810929721 A CN201810929721 A CN 201810929721A CN 109245854 A CN109245854 A CN 109245854A
- Authority
- CN
- China
- Prior art keywords
- communication system
- module
- classifier
- coding
- code
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L1/00—Arrangements for detecting or preventing errors in the information received
- H04L1/0001—Systems modifying transmission characteristics according to link quality, e.g. power backoff
- H04L1/0015—Systems modifying transmission characteristics according to link quality, e.g. power backoff characterised by the adaptation strategy
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Mobile Radio Communication Systems (AREA)
- Detection And Prevention Of Errors In Transmission (AREA)
Abstract
The invention belongs to cordless communication network technical fields, disclose a kind of end-to-end wireless communication system and method based on AI, including the end-to-end each module coarse-grained policies matching of communication system and the end-to-end each tactful fine granularity parameter of communication system are selected.The present invention proposes optimal policy intelligences combination method towards the end-to-end high efficiency of transmission of communication system, according to the end-to-end transmission demand of user, comprehensively considers the multiple indexs of each module, realizes resource distribution.To, the sequencing for avoiding strategy matching may influence the predicament of its result, overcome the deficiency that one module of prior art single optimization makes " modules optimal not equal to global optimum ", realize the communication system overall situation according to user demand, the adaptive optimals policy selection such as channel circumstance.To sum up, the present invention can reduce tactful mismatch between each module, promote end-to-end system performance, reduce the bit error rate.
Description
Technical field
The invention belongs to cordless communication network technical field more particularly to a kind of end-to-end wireless communication systems based on AI
And method.
Background technique
Currently, the prior art is such that message sink coding is to improve the volume in transmission process for the purpose of validity in the industry
Code.The resource of communication be it is valuable, message sink coding pursue efficiency the higher the better, i.e., identical information content use least bits
Position.Channel coding is to be subtracted by improving the coding in transmission process for the purpose of reliability by introducing redundancy in by transmission data
The probability to malfunction less.Simultaneously in the case where source data rate is fixed, the demand of bandwidth is increased.In encryption algorithm, coding
The problems such as rate, code distance and code weight are all critically important parameters, while being also contemplated that development cost, flexibility.In wireless mobile
In channel, when decline and multipath conditions are very severe, designing the modulation scheme that one can resist mobile channel loss is that key is asked
One of topic.It is existing the advantages that stronger anti-channel loss ability, higher safety because of the noiseproof feature of digital modulation technique
Generation Mobile Communication System all uses digital modulation technique.One good modulation scheme can provide under conditions of low received signal to noise ratio
Small bit error rate, it is functional to anti-multipath and fading profiles, the smallest bandwidth is occupied, complexity is low, but does not modulate
Scheme can meet conditions above simultaneously.Therefore, it is necessary to trade off according to user demand to each index, digital modulation side is selected
Case.Structure is complicated for 5G communication process information source, channel-aware topology changes acutely, thus, information source and channel coding, transmission and code rate
The problems such as reduction is difficult is extremely urgent.Wherein, the coding adjusted according to real-time network data is difficult to realize in 5G cataloged procedure, and
There are the problems such as time complexity and space complexity optimization difficulty in cataloged procedure.Shannon, which defines, to be communicated on wireless channel
Capacity limitation, but it does not explain how to reach these boundaries.Part has been implemented around in these boundaries in recent years
Separated operating procedure, for example, code rate and code word adaptation, to realize efficient communication in the case where specific SNR is horizontal.Traditional algorithm
Computation complexity is high in practice, and needs expensive hardware and dsp software optimization processing.On the other hand, it is intended to by grinding
Study carefully scheduling of resource and reach shannon limit, but without universal solution.From the perspective of wireless communication, believed based on environment
The influence of the requirements design modulation scheme optimization singal reporting code such as breath and algorithm complexity is huge.The prior art one " is based on
A kind of image transmitting based on combined channel coding is disclosed in the image transmission rate self-adapting distribution method of combined channel coding "
Rate adaptation distribution method.This method comprises the concrete steps that: each step establishes target function model and isAR4JA code is respectively adopted in second step and R4JA code carries out source coding and channel coding.The
Three steps give the entropy of current information source frame, concentrate searching to meet in maximum value boundary and are greater than all right of given current information source entropy
The rate element answered, and using minimum value therein as source coding rate.4th step gives current signal, in minimum decoding
Look for all corresponding rate elements for being less than given current signal in threshold set, and will wherein maximum value as channel coding
Rate.This method determines the major influence factors of strategy also with the method for the solution for establishing function model.The prior art
The code rate allocation method and system of a kind of channel are disclosed in two " a kind of code rate allocation methods and system of channel ".The tool of this method
Body step is: each step determines the letter of message sink coding code stream according to the convex function of the rate distortion function of Embedded source coding
The attainable maximum value of road encoder bit rate institute.Second step is compiled by the attainable maximum value of channel coding code rate institute and the first channel
Code collection is closed, and determines second channel code set.Third step determines the channel of message sink coding stream in second channel code set
First optimal rate-allocation strategy.4th step carries out the code of channel to message sink coding code stream according to the first most Rate Distribution Strategy
Rate distribution.This method reduces the Data Rate Distribution time by modeling analysis." the wireless communication based on deep learning of the prior art three
It is disclosed in Modulation Signals Recognition method " (application number 201710720483.6,107547460 A of application publication number CN) based on deep
Spend the wireless communication Modulation Signals Recognition method of study.This method comprises the concrete steps that: the first step is to the tune to be identified captured
Signal processed is sampled.Second step is normalized to obtained sample sequence is sampled, according to the sample sequence after normalization
Make the two-dimensional histogram of modulated signal.Third step constructs depth convolutional neural networks.4th step is rolled up using trained depth
Product neural network.5th step utilizes trained depth convolutional neural networks identification wireless communication modulated signal.This method utilizes
Optimization modulated signal improves recognition correct rate, also high algorithm complexity while improving discrimination.In conclusion depth
Study is roughly divided into two ways for dispatching, and one is modules are substituted using deep learning, another kind is to utilize depth
Learning management modules.
In conclusion problem of the existing technology is: deep learning is more in existing scheduling is used to assist or substitute
Some module policy algorithm, seldom modules strategy is managed using deep learning.
Solve the difficulty and meaning of above-mentioned technical problem:
Since the algorithm complexity of deep learning is higher, the time delay required in the communications is lower, and original instrument is not enough to
Meet demand.However, each module of the usual independent process of existing communication system, each module executes specific function, but each
The optimal of module is not equal to global optimum.It is preferably computational that distributed processing framework and specialized hardware have deep learning
Energy.Since most of signal processing algorithms can only substantially capture model in communication, and deep learning does not need accurate model, because
This, which can use deep learning training and selects the optimal policy combination of the end-to-end each module of communication system and determine, influences its property
The major influence factors of energy.
Summary of the invention
In view of the problems of the existing technology, the end-to-end wireless communication system and side that the present invention provides a kind of based on AI
Method.
The invention is realized in this way the end-to-end wireless communications method based on AI, described end-to-end wireless based on AI
Communication means specifically includes: the end-to-end each module coarse-grained policies matching of communication system, is the end-to-end each module choosing of communication system
Optimal algorithm out；The end-to-end each module policy fine granularity parameter of communication system is selected；By channel condition, source characteristic, Yong Huxu
It asks and its strategy combination inputs and update algorithms library in case next time uses.
Further, the coarseness matching for realizing the end-to-end global modules optimal policy of communication system is further wrapped
It includes:
(1) central classifier is that message sink coding module establishes algorithms library；
Message sink coding is original Morse code, II yard of ASC and telegraph code etc..Modern common source coding method has
Huffman coding, L-Z coding, three kinds of lossless codings of arithmetic coding.In reality scene, the situation of different scene information sources has very
It is a variety of, information source have it is known, it is unknown, have a memory, the features such as memoryless are needed when formulating source coding scheme
Consider user demand and the feature of information source etc..Huffman is encoded, in L-Z coding and arithmetic coding difference typing algorithms library.
(2) central classifier is that channel coding module establishes algorithms library；
Channel coding is the coding for the purpose of the reliability for improving information transmission.There are many classical volumes for channel coding
Code, such as RS code, BCH code and convolutional code etc., existing frequently-used channel coding generally use random coded method and combine iteration soft
Decoding approaches the performance of maximum likelihood method.When formulating channel coding schemes, need to consider channel conditions, user is to the bit error rate
Requirement the problems such as, main method has Polar code, Turbo code and LDPC code etc..When formulating channel coding schemes, need
Consider channel conditions, bit error rate minimum requirements of user etc..Polar code, Turbo code and LDPC code are distinguished into typing channel coding
Algorithms library in.
(3) central classifier is that digital modulation module establishes algorithms library；
The final purpose of modulation technique is exactly with quality as well as possible while to occupy least bandwidth in wireless channel
To transmit signal.In digital communication system design, when selecting modulation scheme, often in bandwidth efficiency, power efficiency, error code
It trades off between the indexs such as rate, channel loss situation is also the key factor of determining modulation scheme.When formulating digital modulation scheme, need
Consider that system is more likely to the reliability of communication or bandwidth efficiency determines the method for increasing information signal Error Control；System
System is more likely to power efficiency or bandwidth efficiency determines the modulation scheme of multi-system.What cellular communication at present was taken is that number is logical
It is strong and safer to do channel loss for letter technology, digital modulation good in anti-interference performance.Therefore we study digital modulation.It will be common
Ditital modulation method ASK, FSK, PSK, GFSK, GMSK, QAM, DPSK, mQAM, mPSK, TCM, VSB, OFDM typing number tune
Algorithms library processed.
(4) central classifier is that digital demodulating block establishes algorithms library
The algorithm of the digital demodulation of receiver designs its demodulating algorithm according to transmitter section digital modulation is corresponding.For example,
If in transmitter digital modulation part be 16QAM modulator approach, receiver section also use 16QAM solution
Tune method.
(5) central classifier is that channel decoding module establishes algorithms library
The decoded algorithm of the channel of receiver designs its decoding algorithm according to transmitter section channel coding is corresponding.For example,
If channel coding portions is Polar code in transmitter, Polar code is also used in the channel decoding part of receiver
Decoder.
(6) central classifier is that source coding module establishes algorithms library
The algorithm of the source coding of receiver designs its decoding algorithm according to transmitter section message sink coding is corresponding.For example,
If message sink coding part is Huffman code in transmitter, also used in the channel decoding part of receiver
Huffman decoding.Huffman tree is gradually constructed by comparing weight, then encoding and decoding are carried out by Huffman tree.
(7) for central classifier using intensified learning network by user information, channel environmental information etc. is defeated as neural network
Enter, uniformly selects each module optimal algorithm combination, and it is logical to be subjected to model library update in central classifier for loss function value feedback
Cross that central classifier establishes loss function using intensified learning and the present situation of network carries out coarseness matching.
Further, fine granularity parameter is selected uniformly determines each modular algorithm by neural metwork training by central classifier
Variable factor；
(1) according to the feature of information source, user demand and information source environment setting message sink coding module and source coding module
Policing algorithm needs the variable controlled.Message sink coding module and source coding module need the variable controlled to have a code length, code rate and
Code weight.
(2) the requirement setting letter set and for the bit error rate is reseted according to the code length in message sink coding module, code rate and code
The variable that road coding module and channel decoding module need to control.Channel coding and channel decoding need the variable controlled to have code
It is long, code rate and code weight.
(3) variable for needing to control according to the limitation of complexity, the requirement setting digital modulation of the bit error rate and digital demodulation.
Digital modulation and digital demodulation need the variable controlled to have modulation type and order of modulation.
(4) it is selected that loss function, parameter are established.
For using Block Error Rate as loss function: allowable loss function consider optimization Block Error Rate loss function L=(l, r,
O ...), wherein l indicate code length, r indicate code rate, o indicate order of modulation, l, r, o ... for loss function independent variable by center
Classifier unified management, solution is sent as an envoy to the smallest variable of loss function value, and is exported.By its channel circumstance and its is corresponding optimal
Strategy inputs central classifier and updates algorithms library.
Another object of the present invention is to provide a kind of using the above-mentioned end-to-end integrated design of the communication system based on AI
The wireless communication system of method.
In conclusion advantages of the present invention and good effect are as follows: the present invention is built by the investigation common algorithm of modules
The algorithms library of vertical modules, overcomes the problem of the cured algorithm of modules in the prior art, so that faster carrying out
Selection reduces and calculates the time.The present invention is kept away using the end-to-end global modules strategy of neural network unified management communication system
The sequencing for having exempted from strategy matching may influence the predicament of its result, and overcoming existing one module of single optimization in the art makes
The deficiency for obtaining " the optimal of modules are not equal to global optimum ", realizes the communication system overall situation according to user demand, channel circumstance
Etc. adaptive optimals policy selection.
The present invention selects the major influence factors of each modular algorithm, is managed collectively by central classifier, allowable loss
Function, the independent variable of loss function are the major influence factors of each modular algorithm, overcome the compromise of existing index each in the art
Problem.The present invention is being selected after optimal policy every time by user information, the message feedbacks such as channel circumstance to central classifier, and
Algorithms library is updated, the challenge of prior art poor robustness is overcome.
Detailed description of the invention
Fig. 1 is a kind of end-to-end wireless communications method design flow diagram based on AI provided in an embodiment of the present invention.
Fig. 2 is the end-to-end wireless communication system figure provided in an embodiment of the present invention based on AI.
Fig. 3 is a kind of end-to-end wireless communications method flow chart based on AI provided in an embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to embodiments, to the present invention
It is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to
Limit the present invention.
The present invention is managed collectively the strategy of each module using deep learning.It is specific that each module of communication system is carried out it
When the tactful mismatch between communication system module desired result will be not achieved, i.e., " the optimal of each module is not equal in function
Global is optimal ".The present invention comprehensively considers the influence factor of modules from communication system angle end to end.First
Each module carries out coarse-grained policies matching in communication system, secondly carries out the selected crucial effect for determining each module of fine granularity parameter
Factor is each module design algorithm to realize the end-to-end global best performance of communication system.
Application principle of the invention is explained in detail with reference to the accompanying drawing.
As shown in Figure 1, a kind of end-to-end wireless communications method based on AI of the embodiment of the present invention the following steps are included:
S101: the coarseness matching of the end-to-end each module optimal policy of communication system；
S102: the end-to-end each tactful fine granularity parameter of communication system is selected；
S103: by channel condition, source characteristic, user demand and optimal strategy combination etc. input and update algorithms library.
In a preferred embodiment of the invention, step S101 is specifically included:
(1) central classifier is that communication system message sink coding module establishes algorithms library；
(2) central classifier is that communication system channel coding module establishes algorithms library；
(3) central classifier is that communication system digital modulation module establishes algorithms library；
(4) central classifier is that communication system digital demodulating block establishes algorithms library；
(5) central classifier is that communication system channel decoder module establishes algorithms library；
(6) central classifier is that communication system source coding module establishes algorithms library；
(7) from communication system end to end angle using neural network be each module select its suitable strategy.
In a preferred embodiment of the invention, step S102 is specifically included:
(1) central classifier is that message sink coding and decoder module select its major influence factors；
(2) central classifier is that channel coding and decoder module select its major influence factors；
(3) central classifier is that digital modulation and demodulation module select its major influence factors；
(4) for central classifier according to user demand, such as bit error rate, the demands such as algorithm complexity set up loss function；
(5) central classifier is that each algorithm selects independent variable of its major influence factors as loss function；
(6) central classifier determines the value of major influence factors by minimizing loss function.
Application principle of the invention is further described with reference to the accompanying drawing.
The present invention has its corresponding algorithms library towards end-to-end communication system architecture, each module of communication system.In
Entreat classifier by neural network according to source characteristic, channel conditions, user demand is each module matching strategy and is determined each
The major influence factors of strategy.
Step 1: central classifier is that message sink coding module establishes algorithms library.
Message sink coding is original Morse code, II yard of ASC and telegraph code etc..Modern common source coding method has
Huffman coding, L-Z coding, three kinds of lossless codings of arithmetic coding.In reality scene, the situation of different scene information sources has very
It is a variety of, message sink coding have it is known, it is unknown, have a memory, the features such as memoryless, when formulating source coding scheme,
Need to consider the scheme of the feature decision message sink coding of user demand and information source.Huffman is encoded, L-Z coding and arithmetic
Coding is respectively in typing algorithms library.
Step 2: central classifier is that channel coding module establishes algorithms library.
Channel coding is the coding for the purpose of the reliability for improving information transmission.There are many classical volumes for channel coding
Code, such as RS code, BCH code and convolutional code etc., existing frequently-used channel coding generally use random coded method and combine iteration soft
Decoding approaches the performance of maximum likelihood method.When formulating channel coding schemes, need to consider channel conditions, user is to the bit error rate
Requirement the problems such as, by its main method Polar code, Turbo code and LDPC code difference typing channel coding algorithms library in.
Step 3: central classifier is that digital modulation module establishes algorithms library.
The final purpose of modulation technique is exactly with quality as well as possible while to occupy least bandwidth in wireless channel
To transmit signal.In digital communication system design, when selecting modulation scheme, in bandwidth efficiency, power efficiency, bit error rate etc.
It is traded off between index, is also the key factor of determining modulation scheme in different channel loss.It can be according to system
The reliability or bandwidth efficiency for being more likely to communication determine the method for increasing information signal Error Control；More according to system
Tend to power efficiency or bandwidth efficiency determines the modulation scheme of multi-system.What cellular communication at present was taken is digital communication skill
It is strong and safer to do channel loss for art, digital modulation good in anti-interference performance.Therefore we study digital modulation.By common number
Word modulator approach has ASK, FSK, PSK, GFSK, GMSK, QAM, DPSK, mQAM, mPSK, TCM, VSB, OFDM typing modulation algorithm
Library.
Step 4: central classifier is that digital demodulating block establishes algorithms library
The algorithm of the digital demodulation of receiver designs its demodulating algorithm according to transmitter section digital modulation is corresponding.For example,
If in transmitter digital modulation part be 16QAM modulator approach, receiver section also use 16QAM solution
Tune method.
Step 5: central classifier is that channel decoding module establishes algorithms library
The decoded algorithm of the channel of receiver designs its decoding algorithm according to transmitter section channel coding is corresponding.For example,
If channel coding portions is Polar code in transmitter, Polar code is also used in the channel decoding part of receiver
Decoder.
Step 6: central classifier is that source coding module establishes algorithms library
The algorithm of the source coding of receiver designs its decoding algorithm according to transmitter section message sink coding is corresponding.For example,
If message sink coding part is Huffman code in transmitter, also used in the channel decoding part of receiver
Huffman decoding.Huffman tree is gradually constructed by comparing weight, then encoding and decoding are carried out by Huffman tree.
Step 7: loss function, policy selection are established.
Using intensified learning network by user information, channel environmental information etc. inputs central classifier as neural network,
The algorithm of each module application is uniformly selected, and loss function value feedback is subjected to algorithms library update in central classifier.With optimization
For system Block Error Rate, it is present that loss function and network about Block Error Rate using intensified learning are inputted by central classifier
Situation, central classifier are that each strategy of each module is combined, and export loss function value, selecting keeps loss function minimum
A strategy combination.
Step 8: each module selects its important influence factor.
(1) according to source characteristic, user demand and channel circumstance setting message sink coding module and source coding module policy
Algorithm needs the variable controlled.Message sink coding module and source coding module need the variable controlled to have code length, code rate and code weight.
(2) the requirement setting letter set and for the bit error rate is reseted according to the code length in message sink coding module, code rate and code
The variable that road coding module and channel decoding module need to control.Channel coding and channel decoding need the variable controlled to have code
It is long, code rate and code weight.
(3) variable for needing to control according to the limitation of complexity, the requirement setting digital modulation of the bit error rate and digital demodulation.
Digital modulation and digital demodulation need the variable controlled to have modulation type and order of modulation.
Step 9: for optimizing the bit error rate: design considers the loss function L=(l, r, o ...) of optimization Block Error Rate,
Middle l expression code length, r expression code rate, o expression order of modulation, l, r, o ... it is unified by central classifier for loss function independent variable
Management, solution are sent as an envoy to the smallest variate-value of loss function value, and are exported.By its channel circumstance user demand and its is corresponding optimal
Strategy inputs central classifier and updates algorithms library.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (9)
1. a kind of end-to-end wireless communications method based on AI, which is characterized in that the end-to-end wireless communications method packet based on AI
It includes:
The end-to-end each module coarse-grained policies matching of communication system；
The fine granularity parameter for communicating end-to-end each module policy is selected；
By channel condition, source characteristic, user demand and its strategy combination etc. input and update algorithms library.
2. the end-to-end wireless communications method based on AI as described in claim 1, which is characterized in that the end based on AI is arrived
End wireless communications method further comprises:
(1) central classifier is that communication system message sink coding module establishes algorithms library；
(2) central classifier is that communication system channel coding module establishes algorithms library；
(3) central classifier is that communication system digital modulation module establishes algorithms library；
(4) central classifier is that communication system digital demodulating block establishes algorithms library；
(5) central classifier is that communication system channel decoder module establishes algorithms library；
(6) central classifier is that communication system source coding module establishes algorithms library；
(7) from communication system end to end angle using neural network be each module select its suitable strategy.
3. the end-to-end wireless communications method based on AI as claimed in claim 2, which is characterized in that the center classifier root
According to the method that the feature of demand and information source determines message sink coding, Huffman is encoded, L-Z coding and arithmetic coding are recorded respectively
In the algorithms library for entering communication system message sink coding module；
Polar code, Turbo code and LDPC code are distinguished the algorithm of typing communication system channel coding module by the center classifier
Library；
According to different user demands, algorithm complexity demand determine communication system digital modulation module modulation scheme, it is described in
Classifier is entreated to record Ditital modulation method ASK, FSK, PSK, GFSK, GMSK, QAM, DPSK, mQAM, mPSK, TCM, VSB, OFDM
Enter the algorithms library to the communication system digital modulation module.
4. the end-to-end wireless communications method based on AI as claimed in claim 2, which is characterized in that the communication system number
The algorithm of demodulation module, communication system channel decoder module and communication system source coding module is respectively according to transmitter part score
The corresponding design decoding algorithm of the algorithm of word modulation, channel coding and message sink coding and demodulating algorithm.
5. the end-to-end wireless communications method based on AI as claimed in claim 2, which is characterized in that described from communication system end
Angle to end selects its suitable strategy using neural network for each module are as follows:
The end-to-end each module optimal policy of communication system is found, central classifier, by user information, is believed using intensified learning network
Road environmental information is inputted as neural network, uniformly selects each module optimal algorithm combination, and by loss function value feedback in
Classifier is entreated, algorithms library update is carried out.
6. the end-to-end fine granularity parameter method for selecting of communication system as described in claim 1 further comprises:
A. central classifier is that message sink coding and decoder module select its major influence factors；
B. central classifier is that channel coding and decoder module select its major influence factors；
C. central classifier is that digital modulation and demodulation module select its major influence factors；
D. for central classifier according to user demand, such as bit error rate, the demands such as algorithm complexity set up loss function；
E. central classifier is that each algorithm selects independent variable of its major influence factors as loss function；
F. central classifier determines the value of major influence factors by minimizing loss function.
7. realizing the selected method of the end-to-end fine granularity parameter of communication system as claimed in claim 6, which is characterized in that described
Central classifier is according to source characteristic, user demand and information source environment setting message sink coding module and source coding module policy
Algorithm needs the variable controlled, and message sink coding module and source coding module need the variable controlled to have code length, code rate and code weight；
The center classifier resets the requirement set and for the bit error rate according to the code length in message sink coding module, code rate and code
The variable that channel coding module and channel decoding module need to control, the variable that channel coding and channel decoding need to control are set
There are code length, code rate and code weight；
According to the limitation of complexity, the requirement modulation digital modulation of the bit error rate and digital demodulation need to control the center classifier
Variable, digital modulation and digital demodulation need the variable controlled to have modulation type and order of modulation.
8. realizing the selected method of the end-to-end fine granularity parameter of communication system as claimed in claim 6, which is characterized in that described
Central classifier formulates loss function according to user demand, optimizes communication system end to end performance, by the main influence of each module
Factor extracts the independent variable as loss function out, and the value of key factor is solved by minimizing loss function.
9. a kind of wireless communication system using the end-to-end wireless communications method described in claim 1~8 any one based on AI
System.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810929721.9A CN109245854B (en) | 2018-08-15 | 2018-08-15 | AI-based end-to-end wireless communication system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810929721.9A CN109245854B (en) | 2018-08-15 | 2018-08-15 | AI-based end-to-end wireless communication system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109245854A true CN109245854A (en) | 2019-01-18 |
| CN109245854B CN109245854B (en) | 2022-03-01 |
Family
ID=65070664
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810929721.9A Active CN109245854B (en) | 2018-08-15 | 2018-08-15 | AI-based end-to-end wireless communication system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109245854B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110719112A (en) * | 2019-09-12 | 2020-01-21 | 天津大学 | Deep learning-based parameter adaptive RS code decoding method |
| CN111769903A (en) * | 2020-06-09 | 2020-10-13 | 国家数字交换系统工程技术研究中心 | Network security defense method applied to network security defense system and related device |
| CN117714244A (en) * | 2024-02-05 | 2024-03-15 | 丝路梵天(甘肃)通信技术有限公司 | Wireless signal reconstruction method, device, receiver and system based on neural network |
| CN117714244B (en) * | 2024-02-05 | 2024-05-31 | 丝路梵天(甘肃)通信技术有限公司 | Wireless signal reconstruction method, device, receiver and system based on neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101005469A (en) * | 2006-12-28 | 2007-07-25 | 北京航空航天大学 | Joint coding and decoding device and method for source channel |
| CN104010199A (en) * | 2014-05-22 | 2014-08-27 | 哈尔滨工业大学 | Signal superposition multiplexing video transmission method based on group decoding technology |
| CN107547460A (en) * | 2017-08-21 | 2018-01-05 | 西安电子科技大学 | Radio communication Modulation Signals Recognition method based on deep learning |
| US20180198520A1 (en) * | 2016-01-18 | 2018-07-12 | Qoscience, Inc. | Method and apparatus for the detection of distortion or corruption of cellular communication signals |
-
2018
- 2018-08-15 CN CN201810929721.9A patent/CN109245854B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101005469A (en) * | 2006-12-28 | 2007-07-25 | 北京航空航天大学 | Joint coding and decoding device and method for source channel |
| CN104010199A (en) * | 2014-05-22 | 2014-08-27 | 哈尔滨工业大学 | Signal superposition multiplexing video transmission method based on group decoding technology |
| US20180198520A1 (en) * | 2016-01-18 | 2018-07-12 | Qoscience, Inc. | Method and apparatus for the detection of distortion or corruption of cellular communication signals |
| CN107547460A (en) * | 2017-08-21 | 2018-01-05 | 西安电子科技大学 | Radio communication Modulation Signals Recognition method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| SEBASTIAN DÖRNER等: ""Deep Learning Based Communication Over the Air"", 《IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING ( VOLUME: 12, ISSUE: 1, FEB. 2018)》 * |
| 李先锋等: ""基于深度学习的通信系统设计与研究"", 《轻工科技》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110719112A (en) * | 2019-09-12 | 2020-01-21 | 天津大学 | Deep learning-based parameter adaptive RS code decoding method |
| CN110719112B (en) * | 2019-09-12 | 2023-08-29 | 天津大学 | Parameter self-adaptive RS code decoding method based on deep learning |
| CN111769903A (en) * | 2020-06-09 | 2020-10-13 | 国家数字交换系统工程技术研究中心 | Network security defense method applied to network security defense system and related device |
| CN117714244A (en) * | 2024-02-05 | 2024-03-15 | 丝路梵天(甘肃)通信技术有限公司 | Wireless signal reconstruction method, device, receiver and system based on neural network |
| CN117714244B (en) * | 2024-02-05 | 2024-05-31 | 丝路梵天(甘肃)通信技术有限公司 | Wireless signal reconstruction method, device, receiver and system based on neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109245854B (en) | 2022-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Stark et al. | Joint learning of geometric and probabilistic constellation shaping | |
| Tung et al. | DeepJSCC-Q: Constellation constrained deep joint source-channel coding | |
| Wang et al. | A novel tone reservation scheme based on deep learning for PAPR reduction in OFDM systems | |
| CN109194425A (en) | One kind being based on the end-to-end information transmission system of artificial intelligence and method | |
| CN106533620B (en) | The adaptive demodulation method of high-order QAM modulation symbol based on log-likelihood ratio thresholding | |
| CN114337933B (en) | High-speed visible light communication system based on SCMA and self-adaptive coding and decoding method | |
| Xu et al. | Deep joint source-channel coding for semantic communications | |
| CN109450594B (en) | Rate-free code degree distribution optimization method for uplink of cloud access network | |
| CN110336639A (en) | A kind of polarization code multi-layer coding modulator approach and application based on capacity distribution | |
| CN109245854A (en) | A kind of end-to-end wireless communication system and method based on AI | |
| Jankowski et al. | AirNet: Neural network transmission over the air | |
| DelVecchio et al. | Effects of forward error correction on communications aware evasion attacks | |
| CN109889266A (en) | The Polarization Coding method and system of visible light communication channel based on Gaussian approximation | |
| Kim et al. | Deep learning-assisted multi-dimensional modulation and resource mapping for advanced OFDM systems | |
| Gupta et al. | End-to-end learning-based framework for amplify-and-forward relay networks | |
| CN111641473A (en) | Novel method for shortening polarization code, communication method and system | |
| Wang et al. | Constellation design for deep joint source-channel coding | |
| Rajapaksha et al. | Low complexity autoencoder based end-to-end learning of coded communications systems | |
| CN107370564A (en) | Polarization code processing method and processing device, node | |
| CN104506278B (en) | The Joint iteration receiver linear optimization method of LDPC modulating systems | |
| CN106899388B (en) | Joint-detection and coding/decoding method of the LDPC code under mimo channel | |
| Krishna Reddy et al. | Enhancing PAPR performance in MIMO-OFDM system using hybrid optimal MMSE-MLSE equalizers | |
| Lu et al. | A learning approach to cooperative communication system design | |
| Xie et al. | Machine learning assisted adaptive LDPC coded system design and analysis | |
| Mheich et al. | Achievable rates optimization for broadcast channels using finite size constellations under transmission constraints |
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 |