CN109740690A - The KNN equalization algorithm based on Feature Engineering for short distance optic communication - Google Patents
The KNN equalization algorithm based on Feature Engineering for short distance optic communication Download PDFInfo
- Publication number
- CN109740690A CN109740690A CN201910023077.3A CN201910023077A CN109740690A CN 109740690 A CN109740690 A CN 109740690A CN 201910023077 A CN201910023077 A CN 201910023077A CN 109740690 A CN109740690 A CN 109740690A
- Authority
- CN
- China
- Prior art keywords
- feature
- data
- knn
- training sequence
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Abstract
The present invention discloses a kind of KNN equalization algorithm based on Feature Engineering for short distance optic communication, includes the following steps: step 1: the data comprising training sequence after receiving end is sampled are input to Feature Engineering module, and construction feature vector simultaneously carries out characteristic processing；Step 2: training sequence generator regenerates training sequence, as label corresponding with the feature vector of training sequence, composing training collection；Step 3: its feature vector being obtained using identical Feature Engineering to valid data, by the feature vector of training set and valid data as the input of KNN classifier；Step 4: according in feature space, the classification of k nearest training data of each valid data, the result of KNN classification is balanced output.The present invention solves the problems, such as that intersymbol interference, distorted signals etc. cause the performance of system to decline in inexpensive short distance optical communication system, compared to traditional DFE, FFE balanced device, improve equalization performance, reduce the tap coefficient of filter.
Description
Technical field
The present invention relates to a kind of the equal of k- neighbour (FE-KNN) based on Feature Engineering for short distance optical communication system
Account method.
Background technique
In recent years, since high bandwidth requires the universal of application program, such as high definition television, virtual reality and based on video
Social networking system etc., based on the Transmission system of short distance optical fiber, such as 5G X-haul, data center interconnection (DCI) and passive
Optical-fiber network (PON) is faced with the pressure for improving rate now.Therefore, the target of next-generation short distance optical fiber telecommunications system provides single
Wavelength is more than the transmission rate of 25-Gb/s.But simultaneously because these systems are very sensitive to cost, use the optical device of high bandwidth
All it is not easy reality to increase baud rate either in cost or in technology.Therefore, IEEE 802.3ca working group wishes
Continue to use previous generation commercial 10GHz optical transmitter and receiver.It is most of at present to be greater than about Single wavelength baud rate
The system of 25-Gb/s rate, and realized in the optical device based on 10G bandwidth.And signal is limited and optical fiber by transceiver bandwidth
The influence of dispersion will generate serious intersymbol interference, therefore need to be compensated using balancing technique to this problem.
Balancing technique compensates signal, can be divided into the equilibrium on optical signal and the equilibrium two on electric signal
Kind, wherein based on the Equilibrium Research on the electric signal based on Digital Signal Processing (DSP).Tradition is based on DSP technology to bandwidth
The research of ISI compensation, is based primarily upon feed forward equalizer (FFE) and decision feedback equalizer (DFE) is changed caused by limitation
Into, such as " the Intensity directed equalizer that Kuo Zhang was delivered in Optics Express in 2017
for the mitigation of DML chirp induced distortion in dispersion unmanaged C-
Band PAM transmission ", according to the different intensity ranks of PAM signal, it is next improved to use multiple groups tap coefficient
DFE balanced device realizes the record in 43km standard single-mode fiber (SSMF) transmission of C-band.But the program is due to using
Multiple groups tap coefficient causes balanced complexity to be substantially improved, while which employs a large amount of tap coefficient, therefore balanced fortune
It is very high to calculate complexity, low efficiency.In addition the technologies such as tradition DFE and FFE need training sequence and adaptive algorithm to update tap system
The problems such as there is training spending greatly in number, locally optimal solution.
At the same time, with the fast development of artificial intelligence in recent years, machine learning have become many research fields most by
One of research direction of welcome.Therefore, also there are many scholars at present and replace above-mentioned traditional equalization algorithm using machine learning.
What Shanghai Communications University scholar Guoyao Chen was delivered on Journal of Lightwave Technology in 2018
《Nonlinear Distortion Mitigation by Machine learning of SVM classification
For PAM-4and PAM-8modulated optical interconnection " use support vector machines (SVM) classifier
Complete the equilibrium of PAM-4 and PAM-8.It proposes a kind of algorithm of binary tree, and the SVM for combining multiple two classification is completed
Multi-class classification task.But the algorithm having proposed at present, it is all to finally obtain nonlinear decision boundaries to carry out equalization data,
These ways can not all handle the case where data caused by serious intersymbol interference are largely overlapped, therefore still need to improve.
Summary of the invention
For the defects in the prior art, the present invention provide it is a kind of for short distance optic communication based on Feature Engineering
KNN equalization algorithm, to improve the equalization algorithm performance in electrical domain.
K nearest neighbor (KNN) is a kind of machine learning algorithm based on Statistical Learning Theory.KNN passes through in feature space.KNN
Algorithm biggest advantage be its principle very, Yi Shixian, and can get outstanding classification capacity.Simultaneously as KNN is inherently
It is a multi-categorizer, when completing the equilibrium to multiclass level signal such as PAM-4/PAM-8, does not need as support vector machines etc. two
Classifier constructs multiple two classifiers to complete task.In addition, classifier can be helped to reach in its performance using Feature Engineering
Limit.
Based on this, the present invention is adopted the following technical scheme that:
For the KNN equalization algorithm based on Feature Engineering of short distance optic communication, include the following steps:
Step 1: the data comprising training sequence after receiving end is sampled are input to Feature Engineering module, Feature Engineering mould
Block is data each in training sequence by Feature Engineering, and construction feature vector simultaneously carries out characteristic processing；
Step 2: training sequence generator regenerates training sequence, as corresponding with the feature vector of training sequence
Label, composing training collection；
Step 3: its feature vector being obtained using identical Feature Engineering to valid data, by training set and valid data
Input of the feature vector as KNN classifier；
Step 4: according in feature space, the classification of k nearest training data of each valid data is determined effective
The classification of data, the result of KNN classification are balanced output.
Preferably, in step 1: the Feature Engineering that training sequence uses includes 1) feature construction method are as follows: prolonged by tap
When device obtain characteristic value of the sampled value as each data of current data and its several front and back data, construction feature to
Amount, the length are determined by the length of channel intersymbol interference.2) characteristic processing method are as follows: assign and weighing in centre tapped characteristic value
Weight, increase its calculate apart from when generation influence, obtain final feature vector.The weight parameter can take 1~1.5
Between fixed value.
Preferably, in step 2: the label of each data in training sequence produces again for receiving end training sequence generator
Raw training sequence can repeat to generate since it is pseudo-random sequence.The training sequence feature that will be constructed in step 1
Vector and its corresponding label, constitute test set together.
Preferably, in step 3: the used Feature Engineering of valid data is consistent with the Feature Engineering in step 2, to guarantee phase
With the feature vector of dimension and structure, the distance in feature space can be calculated.By training set (including feature vector and label)
With the feature vectors of valid data as the input of KNN classifier.
Preferably, in step 4: being calculated with valid data k apart from nearest training data, then select k a most
Classification of the most classification of frequency of occurrence as valid data to be equalized in nearly data.The distance is Euclidean distance；K is
Odd number is to ensure the most and identical classification there is no multiple frequency of occurrence.
Equalization algorithm of the invention has fully considered the characteristics of high-speed band limits system, has devised and the system performance pair
The original low-dimensional data received is changed into suitable for the high dimensional data under the system model, Ke Yi great by the Feature Engineering answered
The performance of width promotion KNN algorithm.
Compared with prior art, the present invention have it is following the utility model has the advantages that
1, the present invention compared with traditional DFE, FFE balanced device based on DSP, can under less pumping coefficient condition,
It can larger reduction computation complexity.
2, algorithm performance of the invention obtains higher power compared to DFE and FFE, available better portfolio effect
Budget.
3, algorithm of the invention compare with other machine learning algorithms such as neural network, the principles such as support vector machines are simpler
It is single, it is easy to accomplish.
Detailed description of the invention
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon:
Fig. 1 is the KNN balancing principle schematic diagram based on Feature Engineering.
Fig. 2 is short distance optical communication applications system schematic.
Fig. 3 is the system based on 10G optical transceiving device after the transmission of 20km standard single-mode fiber, in different channels equilibrium
BER performance comparison figure under algorithm, FE-KNN are an example of inventive algorithm.
Specific embodiment
The present invention is described in detail combined with specific embodiments below.Following embodiment will be helpful to the technology of this field
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill of this field
For personnel, without departing from the inventive concept of the premise, several changes and improvements can also be made.These belong to the present invention
Protection scope.
The present invention provides a kind of KNN equalization algorithms based on Feature Engineering for short distance optic communication, such as Fig. 1 institute
Show.Based on Feature Engineering, respectively obtains and pre-process the feature vector by training sequence and valid data.In feature space,
The Euclidean distance for calculating each data in the feature vector and training set of each valid data, according to k nearest instruction
The classification for practicing data determines the classification of valid data, and the result of KNN classification is the output of balanced device.Specifically, including it is following
Step:
Step 1: the data comprising training sequence after receiving end is sampled are input to Feature Engineering module, Feature Engineering mould
Block is data each in training sequence by Feature Engineering, and construction feature vector simultaneously carries out characteristic processing；
Step 2: training sequence generator regenerates training sequence, as corresponding with the feature vector of training sequence
Label, composing training collection；
Step 3: its feature vector being obtained using identical Feature Engineering to valid data, by training set and valid data
Input of the feature vector as KNN classifier；
Step 4: according in feature space, the classification of k nearest training data of each valid data is determined effective
The classification of data, the result of KNN classification are balanced output.
Further, to hereafter need using term be first introduced:
1), Feature Engineering refers to the process for initial data being changed into the feature vector of training data, more preferable to obtain
Training data feature so that machine learning model approaches its upper limit.
2), the feature value vector X of n-th of symbol based on Feature EngineeringnIt can be described as:
Xn=[xn-(L-1)/2,…,xn-1,c×xn,xn+1,…,xn+(L-1)/2]
Wherein, xnIndicate current symbol, xn-(L-1)/2,…,xn-1Indicate prior symbol, xn+1,…,xn+(L-1)/2Indicate subsequent
Symbol, the weight of c current symbol can take the fixed value between 1~1.5.
3), training set is made of feature vector and label, can be described as:
{(X1,l1),(X2,l2),…,(XN,lN)}
4), distance calculated refers to the Euclidean distance in N-dimensional feature space, can be described as；
In step 1: the Feature Engineering that training sequence uses includes 1) feature construction method are as follows: it is obtained by tap delayer
Characteristic value to the sampled value of current data and its several front and back data as each data, construction feature vector, the length
Degree is determined by the length of channel intersymbol interference.2) characteristic processing method are as follows: assign weight in centre tapped characteristic value, increase it
Calculate apart from when generation influence, obtain final feature vector.The weight parameter can take consolidating between 1~1.5
Definite value.
In step 2: the label of each data in training sequence, the training regenerated for receiving end training sequence generator
Sequence can repeat to generate since it is pseudo-random sequence.By the training sequence feature vector constructed in step 1 and its
Corresponding label, constitutes test set together.
In step 3: the used Feature Engineering of valid data is consistent with the Feature Engineering in step 2, to guarantee identical dimensional
With the feature vector of structure, the distance in feature space can be calculated.By training set (including feature vector and label) and effectively
Input of the feature vector of data as KNN classifier.
In step 4: being calculated with valid data k apart from nearest training data, then select in k nearest data
Classification of the most classification of frequency of occurrence as valid data to be equalized.The distance is Euclidean distance；K is odd number with true
Protect the most and identical classification there is no multiple frequency of occurrence.
A kind of equal balance system of the KNN based on Feature Engineering for short distance optic communication is present embodiments provided, such as Fig. 2 institute
Show, comprising: optical transmitter module, optical receiver module and fiber channel, the optical transmitter module is by the high speed comprising training sequence
Optical signal is sent to receiving module by optical fiber, converts optical signal into corresponding electric signal by the optical receiver module, sampling
The KNN equalizer module based on Feature Engineering, the information data after the equilibrium that terminal decision obtains are inputted afterwards.
Electric signal generation module includes: digital signal module, training sequence generator, training sequence insertion module, high speed
Rate modulation module, low bandwidth optical modulator；The digital signal module connection training sequence is inserted into module, inserts on the head of data
Enter the training sequence as caused by training sequence generator, the output end and high rate modulation mould of the training sequence insertion module
Block is connected, and generates the high-speed electric signal for needing to transmit, and finally completes electro-optical signal conversion by low bandwidth light modulation.
Optical receiver module include: low bandwidth photodetector, real-time oscilloscope, Feature Engineering module, KNN balance module,
Demodulation module, data outputting unit；Wherein, photodetector by the electric signal received after real-time oscilloscope samples,
The output end of the real-time oscilloscope is connected with Feature Engineering module, obtains the feature vector of training sequence and valid data；Institute
The feature vector for stating balance module received training sequence and valid data judges that valid data are nearest with it in feature space
The classification of k training data determines its classification by classifications most in k closest approach, completes channel equalization；Signal after equilibrium is logical
It is exported after crossing demodulation module, data outputting unit, realizes the reception of user data.
The present embodiment verifies FE-KNN algorithm parameter are as follows: training sequence length is 3000, and filter tap number is 5.Signal
It is modulated using NRZ-OOK, valid data add training sequence overall length to be 250000.The adaptive algorithm that DFE and FFE are used is gradient
Least square (RLS), training length are 3000, and filter tap number is 11.
Fig. 3 is high speed low cost optical Transmission system respectively back-to-back and after the transmission of 20km optical fiber, based on it is different
BER performance comparison figure under weighing apparatus, in figure: horizontal axis is the optical power of receiving end, and unit is dBm, and the longitudinal axis is that BER indicates the bit error rate
Size takes the BER under 1 × 10^-3 as receiver sensitivity, and FE-KNN is an example of inventive algorithm, DFE and FFE
It is two kinds of classical equalization algorithms based on Digital Signal Processing.It can be seen that the condition of algorithm of the invention in less tap number
Under achieve performance more preferably than traditional DFE and FFE.
In conclusion using the present invention is based on the equalization algorithms of the KNN of Feature Engineering more effective device band can must be compensated
Tolerance system and optical fiber add up caused distorted signals.More compared to filter tap number required for DFE and FFE algorithm
It is few, and portfolio effect is more excellent.The principle of KNN is simple simultaneously, easy to accomplish.Therefore, algorithm of the invention, the relatively good application of energy
In the demand of high speed short distance photosystem.
The present invention is used for the KNN equalization algorithm based on Feature Engineering of short distance optic communication, comprising: it is based on Feature Engineering,
It respectively obtains and pre-processes the feature vector by training sequence and valid data.In feature space, each valid data according to
The classification of k nearest training data determines the classification of valid data, and the result of KNN classification is the output of balanced device.
The present invention solves when using inexpensive bandwidth device transmission two-forty signal due to bandwidth limitation and optical fiber cumulative dispersion, is drawn
The problem of distorted signals, intersymbol interference for entering etc. cause system transmission performance to decline.Meanwhile it is balanced compared to traditional DFE, FFE
The available more preferable portfolio effect of device, while the order of balanced device is reduced, thus it is suitable for the inexpensive short distance of next-generation high speed
From photosystem.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or modify within the scope of the claims, this not shadow
Ring substantive content of the invention.In the absence of conflict, the feature in embodiments herein and embodiment can any phase
Mutually combination.
Claims (5)
1. being used for the KNN equalization algorithm based on Feature Engineering of short distance optic communication, which comprises the steps of:
Step 1: the data comprising training sequence after receiving end is sampled are input to Feature Engineering module, using Feature Engineering mould
Block is each data construction feature vector in training sequence, and carries out characteristic processing；
Step 2: training sequence generator regenerates training sequence, as mark corresponding with the feature vector of training sequence
Label, composing training collection；
Step 3: its feature vector being obtained using identical Feature Engineering to valid data, by the feature of training set and valid data
Input of the vector as KNN classifier；
Step 4: according in feature space, the classification of k nearest training data of each valid data determines valid data
Classification, the result of KNN classification is balanced output.
2. the KNN equalization algorithm based on Feature Engineering according to claim 1 for short distance optic communication, feature exist
In the step 1 is specific as follows: the Feature Engineering that training sequence uses includes 1) feature construction method: pass through tap delayer
Characteristic value of the sampled value of current data and its several front and back data as each data is obtained, construction feature vector should
Length is determined by the length of channel intersymbol interference；2) weight, the power characteristic processing method: are assigned in centre tapped characteristic value
Weight parameter takes the fixed value between 1~1.5.
3. the KNN equalization algorithm based on Feature Engineering according to claim 1 for short distance optic communication, feature exist
In the step 2 is specific as follows: the label of each data in training sequence regenerates for receiving end training sequence generator
Training sequence, the training set is made of the corresponding label of the training sequence feature vector constructed in step 1.
4. the KNN equalization algorithm based on Feature Engineering according to claim 3 for short distance optic communication, feature exist
In the step 3 is specific as follows: the used Feature Engineering of valid data is consistent with the Feature Engineering in step 2, by the training
Input of the feature vector of collection and valid data as KNN classifier.
5. the KNN equalization algorithm based on Feature Engineering according to claim 1 for short distance optic communication, feature exist
In the step 4 is specific as follows: being calculated with valid data k apart from nearest training data, select k nearest data
Classification of the most classification of middle frequency of occurrence as valid data to be equalized；Wherein, the distance is Euclidean distance, and k is
Odd number.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910023077.3A CN109740690A (en) | 2019-01-10 | 2019-01-10 | The KNN equalization algorithm based on Feature Engineering for short distance optic communication |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910023077.3A CN109740690A (en) | 2019-01-10 | 2019-01-10 | The KNN equalization algorithm based on Feature Engineering for short distance optic communication |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109740690A true CN109740690A (en) | 2019-05-10 |
Family
ID=66364344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910023077.3A Pending CN109740690A (en) | 2019-01-10 | 2019-01-10 | The KNN equalization algorithm based on Feature Engineering for short distance optic communication |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109740690A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110190909A (en) * | 2019-06-06 | 2019-08-30 | 北京邮电大学 | A kind of signal equalizing method and device for optic communication |
| CN111313971A (en) * | 2020-02-28 | 2020-06-19 | 杭州电子科技大学 | Lightgbm equalization system and method for improving IMDD short-distance optical communication system |
| CN112733917A (en) * | 2020-12-31 | 2021-04-30 | 杭州电子科技大学 | Channel equalization method based on SHAP feature optimization |
| CN114124223A (en) * | 2021-11-26 | 2022-03-01 | 北京邮电大学 | Method and system for generating convolutional neural network optical fiber equalizer |
| CN114866145A (en) * | 2021-01-20 | 2022-08-05 | 上海诺基亚贝尔股份有限公司 | Method, apparatus, device and computer readable medium for optical communication |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102790734A (en) * | 2011-05-18 | 2012-11-21 | 中国科学院声学研究所 | Linear adaptive equalizer based on channel estimation |
| CN104202288A (en) * | 2014-08-27 | 2014-12-10 | 江苏中兴微通信息科技有限公司 | Data receiving and sending method and device of mixed carrier modulation MIMO system |
| CN108123908A (en) * | 2017-12-14 | 2018-06-05 | 杭州电子科技大学 | A kind of improvement SVM equalization methods and system for NG-PON |
| CN108965178A (en) * | 2018-06-29 | 2018-12-07 | 苏州大学 | Intelligent adaptive balanced device and Equilized demodulation method based on machine learning |
-
2019
- 2019-01-10 CN CN201910023077.3A patent/CN109740690A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102790734A (en) * | 2011-05-18 | 2012-11-21 | 中国科学院声学研究所 | Linear adaptive equalizer based on channel estimation |
| CN104202288A (en) * | 2014-08-27 | 2014-12-10 | 江苏中兴微通信息科技有限公司 | Data receiving and sending method and device of mixed carrier modulation MIMO system |
| CN108123908A (en) * | 2017-12-14 | 2018-06-05 | 杭州电子科技大学 | A kind of improvement SVM equalization methods and system for NG-PON |
| CN108965178A (en) * | 2018-06-29 | 2018-12-07 | 苏州大学 | Intelligent adaptive balanced device and Equilized demodulation method based on machine learning |
Non-Patent Citations (2)
| Title |
|---|
| 李建平: "宽带数字阵列雷达通道均衡算法研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 王丹石: "弹性光网络中的信号处理关键技术与应用研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110190909A (en) * | 2019-06-06 | 2019-08-30 | 北京邮电大学 | A kind of signal equalizing method and device for optic communication |
| CN111313971A (en) * | 2020-02-28 | 2020-06-19 | 杭州电子科技大学 | Lightgbm equalization system and method for improving IMDD short-distance optical communication system |
| CN111313971B (en) * | 2020-02-28 | 2021-06-22 | 杭州电子科技大学 | Lightgbm equalization system and method for improving IMDD short-distance optical communication system |
| CN112733917A (en) * | 2020-12-31 | 2021-04-30 | 杭州电子科技大学 | Channel equalization method based on SHAP feature optimization |
| CN112733917B (en) * | 2020-12-31 | 2024-04-05 | 杭州电子科技大学 | Channel equalization method based on SHAP feature optimization |
| CN114866145A (en) * | 2021-01-20 | 2022-08-05 | 上海诺基亚贝尔股份有限公司 | Method, apparatus, device and computer readable medium for optical communication |
| CN114866145B (en) * | 2021-01-20 | 2024-02-09 | 上海诺基亚贝尔股份有限公司 | Method, apparatus, device and computer readable medium for optical communication |
| CN114124223A (en) * | 2021-11-26 | 2022-03-01 | 北京邮电大学 | Method and system for generating convolutional neural network optical fiber equalizer |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109740690A (en) | The KNN equalization algorithm based on Feature Engineering for short distance optic communication | |
| CN108123908B (en) | Improved SVM (support vector machine) equalization method and system for NG-PON (NG-Passive optical network) | |
| CN109818889B (en) | Equalization algorithm for SVM classifier optimization in high-order PAM optical transmission system | |
| Li et al. | Convolutional neural network-aided DP-64 QAM coherent optical communication systems | |
| CN108494488B (en) | DFE-based SVM (support vector machine) equalization method for short-distance optical communication system | |
| CN109039472A (en) | A kind of data center's optic communication dispersive estimates and management method based on deep learning | |
| CN112733917B (en) | Channel equalization method based on SHAP feature optimization | |
| CN113193920A (en) | Probability shaping PAM-4 signal transmission method and device | |
| CN112598072A (en) | Equalization method of improved Volterra filter based on weight coefficient migration of SVM training | |
| CN114039670B (en) | Nonlinear damage compensation method for mode division multiplexing communication system | |
| CN111313971B (en) | Lightgbm equalization system and method for improving IMDD short-distance optical communication system | |
| CN108521384B (en) | SVM (support vector machine) equalization method and system for maximizing variance data selection for low-bandwidth high-speed optical communication | |
| Bi et al. | Machine learning classifier based on FE-KNN enabled high-capacity PAM-4 and NRZ transmission with 10-G class optics | |
| Bi et al. | A powerful equalizer based on modified SVM classifier without nonlinear kernel enabled 100-Gb/s NG-EPON system with 10-G class | |
| Xi et al. | Simplified SVM equalization algorithm based on single hyperplane training enabled 50Gb/s PAM-4/8 with 10-G optics in NG-PON system | |
| Zhang et al. | Convolutional neural network equalizer for Short-reach optical communication systems | |
| CN114124223B (en) | Convolutional neural network optical fiber equalizer generation method and system | |
| Li et al. | Attention-assisted autoencoder neural network for end-to-end optimization of multi-access fiber-terahertz communication systems | |
| Li et al. | End-to-end learning for chromatic dispersion compensation in optical fiber communication | |
| Chen et al. | Machine learning of SVM classification utilizing complete binary tree structure for PAM-4/8 optical interconnection | |
| CN112737688A (en) | DSP (digital signal processor) equalization method and system based on channel characteristic migration | |
| Wang et al. | Low-complexity nonlinear equalizer based on artificial neural network for 112 Gbit/s PAM-4 transmission using DML | |
| Xi et al. | Applications of machine learning on nonlinear frequency division multiplexing optic-fiber communication systems | |
| CN113938198B (en) | Optical fiber transmission system, LDA-based method and module for simplifying nonlinear equalizer | |
| LI et al. | SVM for constellation shaped 8QAM PON system |
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 |
Application publication date: 20190510 |
|
| RJ01 | Rejection of invention patent application after publication |