CN108494488A - The SVM equalization methods based on DFE for short distance optical communication system - Google Patents
The SVM equalization methods based on DFE for short distance optical communication system Download PDFInfo
- Publication number
- CN108494488A CN108494488A CN201810145616.6A CN201810145616A CN108494488A CN 108494488 A CN108494488 A CN 108494488A CN 201810145616 A CN201810145616 A CN 201810145616A CN 108494488 A CN108494488 A CN 108494488A
- Authority
- CN
- China
- Prior art keywords
- dfe
- svm
- training
- sequence
- symbol
- 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
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/25—Arrangements specific to fibre transmission
- H04B10/2507—Arrangements specific to fibre transmission for the reduction or elimination of distortion or dispersion
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L25/03178—Arrangements involving sequence estimation techniques
- H04L25/03248—Arrangements for operating in conjunction with other apparatus
- H04L25/03254—Operation with other circuitry for removing intersymbol interference
- H04L25/03267—Operation with other circuitry for removing intersymbol interference with decision feedback equalisers
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L25/03178—Arrangements involving sequence estimation techniques
- H04L25/03305—Joint sequence estimation and interference removal
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L25/03178—Arrangements involving sequence estimation techniques
- H04L25/03312—Arrangements specific to the provision of output signals
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/03—Shaping networks in transmitter or receiver, e.g. adaptive shaping networks
- H04L25/03006—Arrangements for removing intersymbol interference
- H04L2025/03433—Arrangements for removing intersymbol interference characterised by equaliser structure
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Power Engineering (AREA)
- Physics & Mathematics (AREA)
- Electromagnetism (AREA)
- Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)
- Optical Communication System (AREA)
Abstract
The present invention is used for the SVM equalization methods based on DFE of short distance optical communication system, including：Step 1：Digital signal with training sequence is converted to optical signal by Electro-optical Modulation and sends, and training sequence is extracted through over-sampling after receiving terminal is converted to electric signal；Step 2：The feature vector that training symbol is built according to DFE structures, the feature value vector based on training sequence calculate optimal hyperlane using SVM；Step 3：Using hyperplane, equilibrium is realized to the information sequence judgement of input based on DFE structures, original signal is recovered using demodulation；Step 4：The symbol that length is set after preservation is balanced becomes characteristic value of next symbol to be detected by leading interference by feed back input；Step 5：Information sequence returns to step 2, re -training optimal hyperlane every certain length.The present invention solve the problems, such as signal in a fiber high-speed transfer when caused dispersion and the random distribution Gaussian noise etc. brought of system device cause system reception sensitivity to decline.
Description
Technical field
The present invention relates to optical transport technology fields, and in particular to it is a kind of for short distance optical communication system based on
SVM (support vector machines) equalization methods of Decision feedback equalization (decision feedback equalization).
Background technology
With the fast development of the new business such as cloud computing, HD video, virtual reality, demand of the terminal user for bandwidth
Increasing, long-distance optical fiber transmission technology is increasingly mature at present, can realize data high capacity transmission and performance is stablized relatively, but
It is that the bandwidth of its access net still reaches to the demand less than user, so the problem of here it is short distance optical transports to be solved.Due to
The upgrading of short distance optical transport technology is fast, huge, and input is high, so high-performance and low cost are always to determine that short distance light passes
The key factor of transferring technology evolution.The single channel data capacity of optic communication at present has reached 100Gb/s, but high-speed optical devices
Industrial chain has not yet been formed, expensive, is also not suitable for being widely applied.If the optical device using low bandwidth is modulated
Transmission, then can because of dispersion, interference between the problems such as bandwidth limits caused symbol (Inter Symbol Interference,
ISI).So in order to meet the requirement of optical transport high-performance and low cost, can be caused to offset channel fading with balancing technique
Intersymbol interference.
It is found through being retrieved to existing literature, short distance optical transport mainly disappears in area of light or in the electrical domain of receiving terminal at present
Except interference, such as Lei Xue, waits and deliver《Symmetric 100-Gb/s TWDM-PON in O-band based
on10G-Class Optical Devices Enabled by Dispersion-Supported Equalization》, profit
The equilibrium in area of light realized to light spectrum reshaping, dispersion compensation with delay interferometer, but since this method is high to requirement on devices and
Performance is unstable, is not suitable for short-range inexpensive high-transmission principle.For another example Junqi xia etc. are delivered《Investigation
on adaptive equalization techniques for 10G-glass optics based 100G-PON
system》, pass through DFE (decision feedback equalization), FFE (feed forward equalization) in the electrical domain of short distance high speed transmission system receiving terminal
Etc. balancing techniques carry out thermal compensation signal, improve the sensitivity of receiving terminal to a certain extent, and demonstrate the feasibility of the program,
And cost is relatively low, meets the principle of short distance low cost high-speed transfer.But the training sequence of tradition FFE and DFE is longer, centainly
Channel spending is improved in degree, and in the weak relatively serious situation of channel, effectively signal can not be compensated.
Invention content
For the defects in the prior art, the present invention provides a kind of for short distance optical communication system based on DFE's
SVM equalization methods.
The present invention adopts the following technical scheme that：
For the SVM equalization methods based on DFE of short distance optical communication system, include the following steps：
Step 1：Digital signal with training sequence is converted to optical signal by Electro-optical Modulation and sends, and is converted in receiving terminal
To extract training sequence through over-sampling after electric signal.
Step 2：The feature vector of training symbol is built according to DFE structures, the feature value vector based on training sequence utilizes
SVM calculates optimal hyperlane.
Step 3：Using hyperplane, equilibrium is realized to the information sequence judgement of input based on DFE structures, it is then demodulated
Recover original signal.
Step 4：The symbol of certain length after preservation is balanced becomes next symbol to be detected by preceding by feed back input
Lead the characteristic value of interference.
Step 5：Information sequence returns to step 2, re -training optimal hyperlane every certain length.
Preferably, the step 1, training sequence is pseudo-random sequence, and information sequence is inserted into one every certain length
Go here and there training sequence, among it every length set according to the influence degree of channel time-varying characteristics.
Preferably, in the step 2, the structure of feature vector：DFE structures are divided into feedforward and feedback two parts, then can use
Several preceding symbols of training symbol to be detected take the correct judgement of several rear symbols of training symbol to be detected as feedforward part
As feedback fraction, feedforward and feedback fraction are then combined into the feature value vector as current training symbol, then protected
Depositing the feature value vector of n-k training symbol, (wherein n is the number of whole training symbols, and k is that the tap of feedback filter is delayed
Device number).
Preferably, in the step 2, SVM calculates optimal hyperlane：By the training sequence feature value vector of structure, and connect
The training sequence (the latter is the former correct court verdict) that receiving end regenerates is trained into SVM together, is calculated most
Excellent hyperplane.
SVM (support vector machines) calculates optimal hyperlane and is as follows：
1), initialize hyperplane, ask each feature vector point to the distance of hyperplane, by the feature nearest apart from hyperplane to
Amount point is used as supporting vector.
2), in order to keep the robustness of hyperplane best, need to adjust hyperplane make supporting vector to the interval of hyperplane most
Bigization.
3), margin maximization equivalence in step 2 is converted to and seeks minimum, to meet the solution of convex optimization problem.
4) conditional extremum that lagrange multiplier approach seeks convex optimization problem, is introduced, hyperplane method vector, intercept and drawing are obtained
The correspondence of Ge Lang multipliers.
5) it, by the normal vector and intercept substitution former formula representated by Lagrange multiplier, is asked according to dual problem, SMO algorithms
Go out Lagrange multiplier, then obtains optimal hyperplane.
Preferably, the step 3 is based on DFE structures, regards the normal vector of optimal hyperlane as feedforward and feedback filter
In all tap coefficient set, information sequence, which is multiplied by tap delayer with the normal vector of optimal hyperlane, to be feedovered
The weighted sum of part and feedback fraction, the signal after judgement is equalized are then demodulated to recover original signal.
Preferably, in the step 4, the symbol of certain length after equilibrium is preserved：Intersymbol interference is essentially from preceding symbol sequence
The leading interference of the hangover interference and rear sequence of symhols of row, so taking front and back sequence of symhols length will be with the length phase of intersymbol interference
Closely to reach preferable portfolio effect.Wherein, DFE structures are based on, the signal after equilibrium is exactly rear sequence of symhols, so according to dry
The rear sequence of symhols length disturbed preserves the symbol of balanced output signal corresponding length.
Preferably, the step 4, is based on DFE structures, and the symbol of equilibrium output enters feedback filter and is delayed by tap
Device preserves, using the characteristic value as next leading interference of symbol to be detected.
Preferably, the step 5 is long every certain information sequence because most of channels all have time-varying characteristics
Degree needs re -training hyperplane to ensure portfolio effect.
SVM is a kind of Machine Method based on Statistical Learning Theory, is equivalent to a largest interval grader.Pass through input
The characteristic value of each sample finds out in feature space maximum spaced linear grader (optimal hyperlane) to realize to data
Classification.And SVM has very strong robustness, it is only necessary to which small sample, which can train, meets the optimal super of sample characteristics distribution
Plane is to realize classification.And equilibrium is constantly to adjust the tap coefficient of balanced device by adaptive algorithm to do to offset intersymbol
It disturbs, makes balanced output constantly close to initial signal, to realize correctly judgement classification.And wherein SVM's is optimal super flat
Surface model is similar with equalizer model, it is possible to which consideration, which applies to the classification of SVM, improves performance in equilibrium.
The equalization methods of the present invention are follow-on SVM based on DFE.Wherein, DFE is eliminated by feedforward filter
Preceding symbol interferes the hangover of current symbol, is eliminated with feedback filter by being previously detected leading interference caused by symbol,
In this way after detecting and judging an information code element, so that it may just eliminate current symbol before continued code member after sensing to subsequent symbol
Interference.For the high efficiency calculated using SMO (sequence minimum) algorithm in SVM, the adaptive algorithm of DFE is replaced with
SMO.So the SVM based on DFE, takes preceding symbol and the symbol (rear symbol) of balanced output as current symbol to be detected by intersymbol
The characteristic value of interference, to construction feature vector, then whole feature vectors are input in SMO algorithms find out it is optimal super flat
Face makes decisions realization equilibrium with hyperplane to the information sequence of input again later, finally demodulated to recover original signal.It examines
The time-varying characteristics for considering channel are inserted into one section of training sequence every certain length in information sequence, are obtained by re -training
Optimal hyperlane ensures balanced effect.And SVM takes full advantage of the high efficiency of SMO algorithms, and it is adaptive compared to traditional DFE
Answering algorithm only needs shorter training sequence that can complete the equilibrium to channel, so more calculating optimal hyperlanes of SVM are still
Equilibrium can soon be completed.
Compared with prior art, the present invention has following advantageous effect：
1, the present invention is compared with traditional balancing technique, under the training sequence of equal length, balancing technique of the invention
The bit error rate is reduced, precision is improved.
2, the balancing technique bit error rate of the invention is smaller by the effect length of training sequence, and the shorter training sequence of use also can
Ensure low error rate, improves band efficiency.
3, balancing technique of the invention is higher to the adaptivity of channel, is adapted to the channel of different characteristics.
The present invention is used for the SVM equalization methods based on DFE of short distance optical communication system, extracts letter in receiving terminal first
Training sequence in number is then based on DFE structures to build symbol by leading and hangover interference feature vector, by SVM
Training calculate optimal hyperlane, finally the information sequence inputted based on DFE structures is made decisions using optimal hyperlane
It realizes balanced.The present invention solve signal in a fiber high-speed transfer when the caused dispersion and random distribution brought of system device
The problem of Gaussian noise etc. causes system reception sensitivity to decline.
Description of the drawings
By reading with reference to the following drawings by comparing other balancing techniques, other features, objects, and advantages of the present invention
It will become more apparent upon：
Fig. 1 is the structural schematic diagram of short distance high-speed optical transmission system.
Fig. 2 is traditional decision feedback equalization principle schematic.
Fig. 3 is the decision feedback equalization principle schematic based on SVM.
Fig. 4 be signal after short distance optical transmission system, receiving terminal using different equalization methods to signal progress
BER performance comparison figures after compensation, in figure：Horizontal axis is the luminous power of receiving terminal, and the longitudinal axis is BER expression bit error rate sizes, wherein
The BER performance comparisons of receiving terminal have respectively：By it is balanced, using RLS as the DFE of adaptive algorithm, with RLS be adaptively to calculate
The FFE of method, the decision feedback equalization based on SVM.
Specific implementation mode
With reference to specific embodiment, the present invention is described in detail.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 domain.For this purpose, the present invention is tested to verify balanced effect using 0,1 signal.
The present invention is used for the SVM equalization methods based on DFE of short distance optical communication system, and process is as shown in Figure 1, band first
There is the digital signal of training sequence to enter standard single-mode fiber after excessive rate Electro-optical Modulation to transmit, be visited by light in receiving terminal
It surveys device and converts optical signals to electric signal, then carrying out sampling to electric signal extracts corresponding training sequence, according to DFE structures
Feature vector of the training symbol by intersymbol interference is built, then the feature vector of training sequence is trained with SVM and is calculated
Optimal hyperlane is based on DFE structures, the signal after being equalized is made decisions to information sequence using hyperplane later, passes through
It crosses demodulation and recovers original signal, while the tap delayer that balanced output signal enters in feedback filter is preserved as under
The characteristic value of one leading interference of symbol to be detected.Specifically include following steps：
Step 1：0,1 signal with training sequence is converted to optical signal through excessive rate Electro-optical Modulation.
Step 2：Optical signal is transmitted by single mode optical fiber, is up-sampled after receiving terminal is converted to electric signal, is periodically extensive
Multiple, down-sampling, extracts training sequence.
Step 3：The feature vector that training symbol is built according to DFE structures, the feature value vector based on training sequence, profit
Optimal hyperlane is calculated with SVM.
Step 4：Using hyperplane, equilibrium is realized to the information sequence judgement of input based on DFE structures, it is then demodulated
Recover original signal.
Step 5：The symbol of certain length after preservation is balanced becomes next symbol to be detected by preceding by feed back input
Lead the characteristic value of interference.
Step 6：Information sequence returns to step 3, re -training optimal hyperlane every certain length.
Further the term in step is explained：
1, tradition DFE is made of two parts, feedforward part and feedback fraction.As shown in Fig. 2, output is as follows：
Wherein, x is input signal,It is exported for judgement, n, w are respectively the tap number and coefficient of feedforward filter, m, b
The respectively tap number and coefficient of feedback filter.Traditional decision feedback equalization is calculated by the training sequence of input with adaptive
Method constantly to adjust the tap coefficient of balanced device, to estimate the frequency characteristic and thermal compensation signal of channel.Wherein signal passes through
The hangover interference for crossing symbol before feed forward equalization is eliminated, the leading interference of symbol after feedback fraction is used for eliminating.And the base of the present invention
In the SVM of DFE be the weight for adjusting each characteristic value instead of adaptive algorithm using SMO algorithms.
2, the feature value vector of symbol is built with DFE structures
Known to 1, the input of decision feedback equalization includes two parts, and a part is stimulus part, and another part is
The feedback fraction of output signal.Then the feature vector of i-th of training symbol based on decision feedback equalization structure can be expressed as：
WhereinIt is xiOutput valve after judgement.
3, SVM (support vector machines)
SVM is a kind of largest interval grader, is to find a n dimension hyperplane in sample data to be divided into data
Two classifications keep the spacing of two classifications maximum.
SVM calculates optimal hyperlane and is as follows：
1), initialize hyperplane, ask each feature vector point to the distance of hyperplane, by the feature nearest apart from hyperplane to
Amount point is used as supporting vector.
2), in order to keep the robustness of hyperplane best, need to adjust hyperplane make supporting vector to the interval of hyperplane with
It maximizes.
3), margin maximization equivalence in step 2 is converted to and seeks minimum, to meet the solution of convex optimization problem.
4), introduce lagrange multiplier approach and seek the conditional extremum of convex optimization problem, obtain hyperplane method vector sum intercept with
The correspondence of Lagrange multiplier.
5) it, by the normal vector and intercept substitution former formula representated by Lagrange multiplier, is asked according to dual problem, SMO algorithms
Go out Lagrange multiplier, then obtains optimal hyperplane.
Since traditional channel model is similar with the hyperplane equation that SVM is generated, it is possible to consider to use hyperplane equation
Carry out approximated channel model, the compensation of signal is then realized using the classification capacity of hyperplane in SVM, by judgement, judgement
As a result it is balanced output.
Channel model：
Hyperplane equation：WTX+b=0
Step 1：The inserted mode of training sequence is to be inserted into training sequence every certain length in data-signal, is used for
The synchronization of receiving terminal and channel estimation.
Training sequence described in step 1 is 0,1 sequence of pseudorandom generated by sequencer.
Light modulation described in step 1 uses external modulation mode, the external modulator that modulated signal control laser is followed by,
The intensity of its output light is set to become with signal using physical effects such as the electric light of modulator, acousto-optics.
Step 2：Optical signal is converted to electric signal by photodetector, and photodetector uses photodiode.
Step 2：Signal is sampled using high speed storing oscillograph, is carried out first with the sample rate of 32 haplotype data rates
Up-sampling, then Timed Recovery extraction clock signal, Synchronous Digital Hierarchy carry out down-sampling extraction instruction with data rate again later
Practice sequence.
The structure of feature vector in step 3：DFE structures are divided into feedforward and feedback two parts, then can use midamble code to be detected
Several preceding symbols of member take the correct judgement of several rear symbols of training symbol to be detected to be used as feedback section as feedforward part
Point, feedforward and feedback fraction are then combined into the feature value vector as current training symbol, then preserve n-k training
The feature value vector of symbol (wherein n is the number of whole training symbols, and k is the tap delayer number of feedback filter).
SVM calculates optimal hyperlane in step 3：By the training sequence feature value vector of structure, given birth to again with receiving terminal
At training sequence (the latter is the former correct court verdict) be trained together into SVM, calculate optimal hyperlane.
Step 4：Based on DFE structures as, the normal vector of optimal hyperlane is regarded to pumping all in feedforward and feedback filter
Head coefficient sets, information sequence, which is multiplied by tap delayer with the normal vector of hyperplane, obtains feedforward section point and feedback fraction
Weighted sum, make decisions and be judged to 0 if weighted sum is less than 0, otherwise be judged to 1, the result of judgement is balanced output, so
It is demodulated afterwards to recover original signal.
The symbol of certain length after preservation is balanced in step 5：Intersymbol interference is interfered essentially from the hangover of preceding sequence of symhols
With the leading interference of rear sequence of symhols, so taking front and back sequence of symhols length will be with the similar length of intersymbol interference to reach preferable
Portfolio effect.DFE structures are wherein based on, the signal after equilibrium is exactly rear sequence of symhols, so according to the rear symbol sequence of interference
Row length preserves the symbol of balanced output signal corresponding length.
Step 5：Based on DFE structures, the symbol of equilibrium output enters feedback filter and is preserved by tap delayer,
Using the characteristic value as next leading interference of symbol to be detected.
Step 6：Because most of channels all have time-varying characteristics, need to instruct again every certain information sequence length
Practice optimal hyperlane to ensure portfolio effect.
Include mainly following steps as shown in figure 3, being the DFE based on SVM：
Based on DFE structures, training sequence is inputted, takes feedforward list entries x and feedback judgement output sequenceTo build training
The feature value vector of sequence recycles SVM to be trained the feature vector of training sequence and calculates optimal hyperlane.Using super
Plane adjudicates the information sequence of input based on DFE structures, is equalized output signal while the tap into feedback filter
Delayer preserves, the characteristic value as next detection leading interference of symbol.Compared to traditional DFE, its is simple in structure, no
Numerous tap coefficients is needed to carry out thermal compensation signal, it is only necessary to which training sequence, can be directly to input signal come after determining hyperplane
It makes decisions, reduces circuit complexity.
The training sequence length for verifying SVM-DFE performances is 500, is inserted between the length of training sequence and is divided into 40000, information
The total length of sequence is 125000, is inserted into training sequence three times in total.The equalization algorithm that the DFE and FFE wherein to compare is used
It is RLS (least square method), training sequence length is 2000.
Fig. 4 is that the modulation device of 10G bandwidth generates the signal of 25G rates, transmits, is connecing by 20km standard single-mode fibers
Receiving end signal is compensated using different equalization methods after BER performance comparison figures, in figure：Horizontal axis is the light work(of receiving terminal
Rate, the longitudinal axis are that BER indicates bit error rate size, take 1 × 10-3BER sensitivity as receiver.By comparing, it can be seen that base
Traditional DFE, FFE are substantially better than in the decision feedback equalization of SVM, it is lower in received optical power, it also can be preferably
Thermal compensation signal.
In conclusion the SVM using the present invention based on DFE can well solve low bandwidth device transmission high speed signal
Receiving sensitivity caused by generated symbol distortion, intersymbol interference etc. declines problem.Compared with traditional balancing technique, hence it is evident that carry
High receiving sensitivity still can effectively dock receipts signal compensation, and required instruction in the case that received optical power is lower
It is less to practice sequence, reduces the spending of channel to a certain extent.Thus the SVM based on DFE can preferably be applied to short distance
Optical transmission system, and meet low cost, high power capacity transmission requirement.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or change within the scope of the claims, this not shadow
Ring the substantive content of the present invention.In the absence of conflict, the feature in embodiments herein and embodiment can arbitrary phase
Mutually combination.
Claims (8)
1. the SVM equalization methods based on DFE for short distance optical communication system, which is characterized in that include the following steps：
Step 1：Digital signal with training sequence is converted to optical signal by Electro-optical Modulation and sends, and electricity is converted in receiving terminal
After signal training sequence is extracted through over-sampling；
Step 2：The feature vector that training symbol is built according to DFE structures, the feature value vector based on training sequence utilize
SVM calculates optimal hyperlane；
Step 3：Using hyperplane, equilibrium is realized to the information sequence judgement of input based on DFE structures, is recovered using demodulation
Original signal；
Step 4：The symbol that length is set after preservation is balanced becomes next symbol to be detected by leading dry by feed back input
The characteristic value disturbed；
Step 5：Information sequence returns to step 2, re -training optimal hyperlane every certain length.
2. the SVM equalization methods based on DFE according to claim 1 for short distance optical communication system, which is characterized in that step
In rapid 1, training sequence is pseudo-random sequence, and information sequence is inserted into a string of training sequences every certain length.
3. the SVM equalization methods based on DFE according to claim 1 for short distance optical communication system, feature exists
In, in step 2, the structure of feature vector：DFE structures are divided into feedforward and feedback two parts, then can use training symbol to be detected
Several preceding symbols take the correct judgement of several rear symbols of training symbol to be detected to be used as feedback fraction as feedforward part, after
Feedforward and feedback fraction are combined as to the feature vector of current training symbol, preserve the feature vector of n-k training symbol,
In, n is the number of whole training symbols, and k is the tap delayer number of feedback filter.
4. the SVM equalization methods based on DFE according to claim 1 or 3 for short distance optical communication system, feature
It is, in step 2, SVM calculates optimal hyperlane process：By the training sequence feature vector of structure, given birth to again with receiving terminal
At training sequence be trained together into SVM, calculate optimal hyperlane.
5. the SVM equalization methods based on DFE according to claim 1 for short distance optical communication system, feature exists
In, in step 3, be based on DFE structures, the normal vector of optimal hyperlane is considered as to tap system all in feedforward and feedback filter
Manifold close, information sequence by tap delayer is multiplied with the normal vector of hyperplane obtain feedforward section divide and feedback fraction add
Quan He, the signal after judgement is equalized are rear demodulated to recover original signal.
6. the SVM equalization methods based on DFE according to claim 1 for short distance optical communication system, feature exists
In, in step 4, the symbol of setting length after preservation is balanced：Intersymbol interference is interfered essentially from the hangover of preceding sequence of symhols with after
The leading interference of sequence of symhols takes front and back sequence of symhols length to be imitated with the similar length of intersymbol interference to reach preferable equilibrium
Fruit；Wherein, DFE structures are based on, the signal after equilibrium is exactly rear sequence of symhols, is preserved according to the rear sequence of symhols length of interference
The symbol of balanced output signal corresponding length.
7. the SVM equalization methods based on DFE for short distance optical communication system according to claim 1 or 6, feature
It is, in step 4, is based on DFE structures, the symbol of equilibrium output enters feedback filter and preserved by tap delayer, to make
For the characteristic value of next leading interference of symbol to be detected.
8. the SVM equalization methods based on DFE according to claim 1 for short distance optical communication system, feature exists
In in step 5, needing re -training hyperplane to ensure portfolio effect every set information sequence length.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810145616.6A CN108494488B (en) | 2018-02-12 | 2018-02-12 | DFE-based SVM (support vector machine) equalization method for short-distance optical communication system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810145616.6A CN108494488B (en) | 2018-02-12 | 2018-02-12 | DFE-based SVM (support vector machine) equalization method for short-distance optical communication system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108494488A true CN108494488A (en) | 2018-09-04 |
| CN108494488B CN108494488B (en) | 2021-04-30 |
Family
ID=63340387
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810145616.6A Active CN108494488B (en) | 2018-02-12 | 2018-02-12 | DFE-based SVM (support vector machine) equalization method for short-distance optical communication system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108494488B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109818889A (en) * | 2019-02-19 | 2019-05-28 | 杭州电子科技大学 | Equalization algorithm for the SVM classifier optimization in high-order PAM optical transmission system |
| CN110020695A (en) * | 2019-04-19 | 2019-07-16 | 杭州电子科技大学 | K-means non-uniform quantizing algorithm for filter bank multi-carrier modulation optical communication system |
| CN112187683A (en) * | 2019-07-01 | 2021-01-05 | 默升科技集团有限公司 | Parallel mixed signal equalization for high speed serial links |
| CN112598072A (en) * | 2020-12-28 | 2021-04-02 | 杭州电子科技大学 | Equalization method of improved Volterra filter based on weight coefficient migration of SVM training |
| WO2021254222A1 (en) * | 2020-06-19 | 2021-12-23 | 华为技术有限公司 | Method for determining training sequence, and related device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102340466A (en) * | 2011-10-25 | 2012-02-01 | 西安电子科技大学 | Method for designing adaptive decision feedback equalizer based on support vector machine |
| CN103957176A (en) * | 2014-04-30 | 2014-07-30 | 华南理工大学 | Self-adaptive RLS decision feedback equalizing system and implementing method thereof |
| US20150063828A1 (en) * | 2013-08-27 | 2015-03-05 | Broadcom Corporation | Generalized Transmit Pre-Coding for Optical and Backplane Channels |
-
2018
- 2018-02-12 CN CN201810145616.6A patent/CN108494488B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102340466A (en) * | 2011-10-25 | 2012-02-01 | 西安电子科技大学 | Method for designing adaptive decision feedback equalizer based on support vector machine |
| US20150063828A1 (en) * | 2013-08-27 | 2015-03-05 | Broadcom Corporation | Generalized Transmit Pre-Coding for Optical and Backplane Channels |
| CN103957176A (en) * | 2014-04-30 | 2014-07-30 | 华南理工大学 | Self-adaptive RLS decision feedback equalizing system and implementing method thereof |
Non-Patent Citations (1)
| Title |
|---|
| 宋恒 等: "基于非单点模糊支持向量机的判决反馈均衡器", 《电子与信息学报》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109818889A (en) * | 2019-02-19 | 2019-05-28 | 杭州电子科技大学 | Equalization algorithm for the SVM classifier optimization in high-order PAM optical transmission system |
| CN109818889B (en) * | 2019-02-19 | 2021-12-28 | 杭州电子科技大学 | Equalization algorithm for SVM classifier optimization in high-order PAM optical transmission system |
| CN110020695A (en) * | 2019-04-19 | 2019-07-16 | 杭州电子科技大学 | K-means non-uniform quantizing algorithm for filter bank multi-carrier modulation optical communication system |
| CN112187683A (en) * | 2019-07-01 | 2021-01-05 | 默升科技集团有限公司 | Parallel mixed signal equalization for high speed serial links |
| CN112187683B (en) * | 2019-07-01 | 2023-08-04 | 默升科技集团有限公司 | Apparatus and method for parallel mixed signal equalization for high speed serial links |
| WO2021254222A1 (en) * | 2020-06-19 | 2021-12-23 | 华为技术有限公司 | Method for determining training sequence, and related device |
| CN112598072A (en) * | 2020-12-28 | 2021-04-02 | 杭州电子科技大学 | Equalization method of improved Volterra filter based on weight coefficient migration of SVM training |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108494488B (en) | 2021-04-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108494488A (en) | The SVM equalization methods based on DFE for short distance optical communication system | |
| CN108880692B (en) | Modulation format recognition and optical signal-to-noise ratio monitoring method for coherent optical communication system | |
| CN108123908A (en) | A kind of improvement SVM equalization methods and system for NG-PON | |
| CN108650201A (en) | Channel equalization method, interpretation method based on neural network and corresponding equipment | |
| CN109039472A (en) | A kind of data center's optic communication dispersive estimates and management method based on deep learning | |
| CN109740690A (en) | The KNN equalization algorithm based on Feature Engineering for short distance optic communication | |
| CN112598072A (en) | Equalization method of improved Volterra filter based on weight coefficient migration of SVM training | |
| CN109818889B (en) | Equalization algorithm for SVM classifier optimization in high-order PAM optical transmission system | |
| CN114039670B (en) | Nonlinear damage compensation method for mode division multiplexing communication system | |
| CN103004098A (en) | Digital filter device, digital filtering method, and control program for digital filter device | |
| Li et al. | 100Gbps IM/DD transmission over 25km SSMF using 20G-class DML and PIN enabled by machine learning | |
| CN102437978B (en) | Method and device for balancing digital microwaves | |
| CN108521384B (en) | SVM (support vector machine) equalization method and system for maximizing variance data selection for low-bandwidth high-speed optical communication | |
| Yi et al. | Neural network-based equalization in high-speed PONs | |
| Maiti et al. | Modified nonlinear decision feedback equalizer for long-haul fiber-optic communications | |
| CN113542172B (en) | Elastic optical network modulation format identification method and system based on improved PSO clustering | |
| CN111313971A (en) | Lightgbm equalization system and method for improving IMDD short-distance optical communication system | |
| CN101651643B (en) | Blind equalization method for wavelet neural network based on space diversity | |
| 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 | |
| CN113452638B (en) | Orthogonal chirp multiplexing optical transmission method and system | |
| Zhang et al. | An improved end-to-end optical transmission system based on deep learning | |
| Chen et al. | Machine learning of SVM classification utilizing complete binary tree structure for PAM-4/8 optical interconnection | |
| Glentis et al. | Electronic dispersion compensation of fiber links using sparsity induced volterra equalizers | |
| CN107302506A (en) | A kind of double judgement blind balance methods of palarization multiplexing coherent optical communication system | |
| Wang et al. | Low-complexity nonlinear equalizer based on artificial neural network for 112 Gbit/s PAM-4 transmission using DML |
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 |