CN109905170A - A kind of nonlinear distortion compensation algorithm and visible light communication device based on K-DNN - Google Patents
A kind of nonlinear distortion compensation algorithm and visible light communication device based on K-DNN Download PDFInfo
- Publication number
- CN109905170A CN109905170A CN201910041824.6A CN201910041824A CN109905170A CN 109905170 A CN109905170 A CN 109905170A CN 201910041824 A CN201910041824 A CN 201910041824A CN 109905170 A CN109905170 A CN 109905170A
- Authority
- CN
- China
- Prior art keywords
- signal
- dnn
- layer
- node
- input
- 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
Landscapes
- Optical Communication System (AREA)
Abstract
The invention belongs to technical field of visible light communication, specially a kind of nonlinear distortion compensation algorithm and visible light communication device based on K-DNN.It provides nonlinear distortion compensation algorithm, and step includes: that will be converted to parallel signal by the serial signal of self-adaptive linear equalisation device, the DNN balanced device as feature vector input kernel function auxiliary；Through K-DNN training, feature vector is ranged the classification of maximum probability, obtains original signal；The present apparatus uses above-mentioned nonlinear distortion compensation algorithm, and structure includes: transmitting terminal and receiving end, and receiving end carries out processed offline to signal by processing chip, comprising: compensates linear distortion using the adaptive post-equalizer of S-MCMMA；Nonlinear distortion compensation is carried out using K-DNN equilibrium and demapping, obtains original signal.Channel circumstance non-linear distortion caused by signal that the present invention can be convenient, complicated in the existing underwater wireless optic communication of effectively compensating has important Practical significance.
Description
Technical field
The invention belongs to technical field of visible light communication, and in particular to a kind of nonlinear distortion compensation calculation based on K-DNN
Method and visible light communication device.
Background technique
In recent years, due to the fast development of white light LEDs, the research of LED based visible light communication (VLC) technology attracts
The concern of more and more scientists.VLC has the advantages that high speed, electromagnetic-radiation-free, eye-safe and secure communication reliability.
In particular, existing underwater wireless communication means (acoustic communication, radio communication) is not able to satisfy the need of high speed and long distance transmission
It asks.In LED based VLC, since the wavelength of blue green light is located at the transmission window (absorption coefficient is small) of water, it is based on blue green led
VLC can realize the subsurface communication of relatively large distance and high speed.The same time.Therefore, LED based underwater wireless communication may
It is the promising solution of the following Underwater High Speed Communication network.
However, complicated underwater environment includes turbulent flow, scattering and diffusion, this leads to the distortion channel of nonlinearity.Cause
This, existing linear equalisation techniques, such as recurrence least square (RLS), minimum mean (LMS) and the modified cascade multimode of scalar
Algorithm (S-MCMMA) cannot effectively restore non-linear distortion signal.Therefore, eliminate the effective schemes of non-linear penalties for
Actual underwater VLC system is essential.Fortunately, it has been found that a large amount of machine learning algorithm is that reply is non-linear
The intelligence tool of process.One of hot topic as machine learning, deep learning (DNN) algorithm are widely used to face knowledge
Not, the fields such as speech recognition and image recognition.Since DNN has with the ability of arbitrary accuracy approximation arbitrary model, it can make
For the compensator to non-linear distortion signal.That is DNN can be automatically found by training process input feature vector and tag along sort it
Between relationship.However the training process of DNN balanced device needs up to thousands of times training the number of iterations, this occupies system very much
Computing resource.Therefore propose that a kind of DNN algorithm with less trained the number of iterations is significantly.
Summary of the invention
Nonlinear distortion compensation algorithm and the visible light communication dress that the purpose of the present invention is to provide a kind of based on K-DNN
It sets, to compensate channel circumstance non-linear distortion caused by signal complicated in existing underwater wireless optic communication.
Present invention firstly provides a kind of nonlinear distortion compensation algorithm for optic communication, this method is a kind of based on core letter
The pre-weighting method of the DNN balanced device (being denoted as K-DNN) of number auxiliary；The steps include: that self-adaptive linear equalisation device will be passed through first
Serial signal after carrying out linear equalization is converted to parallel signal, inputs K-DNN as feature vector；It is instructed by K-DNN model
Practice, feature vector is ranged that classification of maximum probability, obtains original signal；Wherein, K-DNN model training, using friendship
Entropy function is pitched as cost function, optimization is iterated to the parameter in K-DNN by back-propagation algorithm.
It is specifically further described as follows:
The DNN balanced device (being denoted as K-DNN) based on kernel function auxiliary, as shown in Figure 1, K-DNN includes one defeated
Enter layer, a kernel function layer, multiple hidden layers, an output layer and a Softmax classification layer；The input layer is to core letter
Node between several layers is correspondingly that i.e. in input layer a input node only transmits data in kernel function layer only
One corresponding core node；And the output of kernel function layer to each node between output layer can be multiplied by a weightIt is sent in all nodes of next layer, and is finally reached Softmax layers and classifies；WeightMatrix be expressed as Wl:
Wherein, i and j respectively represents i-th of node of current layer and next layer of j-th of node, l then represent first it is hidden
Layer (when l is equal to L+1, l represents output layer) is hidden, L is the total number of hidden layer, and m is the number of nodes of current layer, under n is
Primary number of nodes.
Serial signal after self-adaptive linear equalisation device carries out linear equalization is converted into after parallel signal as spy
Levy vectorIt is input to the input layer of K-DNN；It is distorted caused by non-linear and intersymbol interference (ISI)
Lead to center signalIt will receive the interference of its adjacent several signal in timing.Therefore, withAdjacent number
A signal can be input to the input layer of K-DNN by the elements in parallel as feature vector.
Kernel function layer is the second layer of K-DNN.In this layer, accelerate the training process of DNN using kernel function.By preparatory
Higher weight is assigned to the signal for closing on center signal, reaches pre- convergent effect, to reduce required for DNN training repeatedly
Generation number.The kernel function includes: linear kernel (LK), polynomial kernel (PK), cosine kernel (CK), expression are as follows:
Linear kernel (LK):
Polynomial kernel (PK):
Cosine kernel (CK):
Wherein, σ is the parameter for adjusting kernel function, t0At the time of for current demand signal, at the time of t is closing signal.
For the network inputted with n feature, kernel function K can be expressed as an amplitude correction vector
Hidden layer is clipped between kernel function layer and output layer.Wherein the node in the first hidden layer is to all nodes of preceding layer
Output summation, and the input as node, then by this input plus a biasing
Here, kiIt is the element of the amplitude correction vector of kernel function K.
In K-DNN, use Sigmoid as activation primitive, formula can be expressed as f (x)=max (0, x), wherein x
It is the output for the node being connected with ReLU.Accordingly, it is considered to activation primitive is arrived, the node expression in all hidden layers are as follows:
In output layer, number of nodes is suitable with the quantity of level classification of signal is sent, (for example, for PAM8, N0
=8).Then, by Softmax layers, classification results are obtained, are expressed are as follows:
For example, Lj=-7, -5, -3, -1,1,3,5,7
Wherein, OiIt is the output for exporting node layer, its expression formula are as follows:
Then, according to trained model, feature vector x is ranged that classification L of maximum probabilityp。
Finally, using entropy function is intersected as cost function, by back-propagation algorithm to the hyper parameter (W in K-DNNl
And bl) it is iterated optimization [1]；Obtain original signal.
The visible light communication device of the present invention also provides a kind of nonlinear distortion compensation algorithm based on K-DNN, the device
Using above-mentioned nonlinear distortion compensation algorithm；Its structure is as shown in Fig. 2, be divided into transmitting terminal and receiving end.It specifically includes:
Data prediction chip；Its function includes: that original bit sequence is mapped to real symbol to form PAM8 signal；
Using PS-Manchester coding for alleviating common-mode noise, and improve system performance；Carry out up-sampling and Nyquist filter
Wave；
Arbitrary waveform generator (AWG)；For generating electric signal according to input signal；
Hardware pre equalizer based on Bridged-T；For compensating the decaying of high fdrequency component to input electrical signal；
With 2 microcircuit amplifiers (EA), a bias-tee combination, transmitter, transmitting collimation convex lens；Signal warp
After EA amplification, electric signal and DC offset voltage are applied in LED light by the merging of bias-tee group, and the emitted device of LED light is logical
Collimation convex lens is crossed to be collimated；
Collectiong focusing convex lens, PIN photodiode；Under water and after free space transmission, PAM8 signal line focus
It is received after convex lens by PIN photodiode；
Electric amplifier EA, digital storage oscilloscope (OSA)；Signal is amplified by EA, and is connect by digital storage oscilloscope (OSA)
It receives, to carry out further signal processed offline；
Processed offline chip；Its function be processed offline is carried out to signal, including, it is synchronous to execute signal in order, power
Normalization and down-sampling, to obtain standardized PAM8 signal；Using differential decoding, for mitigating the influence of common-mode noise；It adopts
Linear distortion is compensated with the adaptive post-equalizer of S-MCMMA；Nonlinear distortion compensation is carried out using K-DNN equilibrium and demapping
And classification, to obtain original signal.
In data prediction chip, original bit sequence is mapped to real symbol to form PAM8 signal, and uses PS-
Manchester coding alleviates common-mode noise and improves system performance.Then data prediction chip can up-sample signal
And nyquist filtering.Treated, and signal is input in the channel of arbitrary waveform generator (AWG) generates electric signal.AWG is raw
At PAM8 signal signal is filtered by pre equalizer, so that its frequency spectrum is met visible light channel.It is put by microcircuit
After big device (EA) amplification, electric signal and DC offset voltage merge the blue for being applied to silicon substrate LED light by Bias tee group
On chip.
After 1.2 meters of underwater and free space transmission, in receiving end, PAM8 signal is at receiver using commercialization
PIN photodiode receives.The signal received is amplified by EA, and is received by digital storage oscilloscope, further to carry out
Off-lined signal processing chip carries out recovery demodulation to signal.
Processed offline chip executes signal in order and synchronizes, and power normalization, down-sampling and differential decoding are (for mitigating
The influence of common-mode noise) to obtain standardized reception signal.Next, processed offline chip is calculated using self-adaptive linear equalisation
The linear distortion that method (MCMMA) compensation channel generates signal.However there are also remaining non-linear distortions for this signal.Then,
Processed offline chip carries out nonlinear distortion compensation and classification using the K-DNN equilibrium of proposition and de-mapping algorithm to be exported
Signal.K-DNN while balanced device, decision device as underwater VLC system in this process.
Fig. 3 describes the performance of traditional DNN balanced device and K-DNN balanced device in two kinds of nonlinear channels of height.By Fig. 3
(a) as it can be seen that in the system compared with high non-linearity, traditional DNN balanced device of β → 0 needs 1300 repetitive exercises to reach hard decision
Thresholding (HD-FEC) 3.8 × 10-3, and then 600 repetitive exercises of minimum only needs are the K-DNN balanced device with kernel function auxiliary
It can reach HD-FEC, the occupancy of computing resource is reduced to 53.85%.Equally in Fig. 3 (b), non-linear lesser system center
Function still can effectively accelerate the training rate of DNN balanced device.It can be seen that in Fig. 3 (a) of either high non-linearity or low
K-DNN in nonlinear Fig. 3 (b) can reach hard decision thresholding relative to traditional DNN balanced device faster.
The channel circumstance that the present invention can be convenient, complicated in the existing underwater wireless optic communication of effectively compensating is caused by signal
Non-linear distortion has important Practical significance.
[1]Rumelhart,David E.,Geoffrey E.Hinton,and Ronald J.Williams."
Learning representations by back-propagating errors."nature 323.6088(1986):
533.。
Detailed description of the invention
Fig. 1 is the K-DNN network structure of the embodiment of the present invention.
Fig. 2 is the system frame for realizing K-DNN equilibrium and classification in the embodiment of the present invention in PAM8 undersea optical communications system
Figure.
Fig. 3 is the comparison diagram of the embodiment of the present invention and other schemes.
Specific embodiment
To keep the technical problem to be solved in the present invention, technical solution and advantage clearer, below in conjunction with specific implementation
Example and attached drawing are described in detail.
The visible light communication device for the compensating non-linear distortion algorithms based on K-DNN that the embodiment of the invention provides a kind of,
The visible light communication device for solving the existing compensating non-linear distortion algorithms based on DNN needs a large amount of repetitive exercise numbers to occupy
The problem of a large amount of computing resources.
It is of the present invention to utilize K-DNN nonlinear distortion compensation algorithm in the underwater VLC system of PAM, comprising the following steps:
Step 1: the serial signal after self-adaptive linear equalisation device carries out linear equalization is converted into parallel signal,
Then using the parallel signal as feature vectorIt is input toK-DNNInput layer；Non-linear and symbol
Between distortion caused by crosstalk (ISI) lead to center signalIt will receive the dry of its adjacent several signal in timing
It disturbs.Therefore, withAdjacent several signals can be input to the input layer of K-DNN by the elements in parallel as feature vector；
Step 2: accelerating the training process of DNN using kernel function in kernel function layer.Center letter is closed on by giving in advance
Number signal assign higher weight, reach pre- convergent effect, to reduce the number of iterations required for DNN training.Core letter
Number can be expressed as:
Linear kernel (LK):
Polynomial kernel (PK):
Cosine kernel (CK):
Step 3: inputting a signal into full connection hidden layer.Node in hidden layer will be to the defeated of all nodes of preceding layer
Summation and the input as node out；It then will be plus a biasing by this input
In K-DNN, use ReLU as activation primitive, formula can be expressed as f (x)=max (0, x), wherein x is
The output for the node being connected with ReLU.Accordingly, it is considered to activation primitive is arrived, the node expression in all hidden layers are as follows:
Softmax layers are arrived step 3: signal is entered to enter and leave, classification results is obtained, expresses are as follows:
For example, Lj=-7, -5, -3, -1,1,3,5,7
Wherein, OiIt is the output for exporting node layer, its expression formula are as follows:
Step 4: feature vector x is ranged that classification L of maximum probability according to trained modelp；
Step 5: using entropy function is intersected as cost function, by back-propagation algorithm to the hyper parameter in K-DNN
(WlAnd bl) it is iterated optimization.
Implementation of the invention additionally provides a kind of visible light communication dress of compensating non-linear distortion algorithms based on K-DNN
It sets, specific implementation flow is as follows:
Step 1: original bit sequence is mapped to real symbol to form PAM8 signal；
Step 2: being encoded using PS-Manchester for alleviating common-mode noise and improving system performance；
Step 3: carrying out up-sampling and nyquist filtering；
Step 4: signal input arbitrary waveform generator generates electric signal；
Step 5: electric signal is input to the pre equalizer based on Bridged-T, to compensate the decaying of high fdrequency component；
Step 6: electric signal and DC offset voltage pass through biasing after by having the amplification of 2 microcircuit amplifiers (EA)
The merging of threeway group is applied in LED light, and LED emits light and collimated by collimating convex lens；
Step 7: PAM8 signal is at receiver using commercial PIN photoelectricity two under water and after free space transmission
Pole pipe receives；
Step 8: the signal received is amplified by EA, and by digital storage oscilloscope (OSA), with carry out further from
Line signal processing；
Step 9: execute signal in order in processed offline and synchronize, power normalization and down-sampling, to obtain standard
The PAM8 signal of change；
Step 10: using differential decoding, for mitigating the influence of common-mode noise；
Step 11: the adaptive post-equalizer using S-MCMMA compensates linear distortion；
Step 12: nonlinear distortion compensation and classification are carried out using K-DNN equilibrium and demapping, to obtain original letter
Number.
Device device sequence of movement is as follows:
First device: data prediction chip (first to third step)
First device: arbitrary waveform generator (the 4th step)
Second device: the hardware pre equalizer (the 5th step) based on Bridged-T
Third device: EA (the 6th step)
4th device: a bias-tee combines (the 6th step)
5th device: transmitter (the 6th step)
6th device: transmitting collimation convex lens (the 6th step)
7th device: collectiong focusing convex lens (the 7th step)
8th device: PIN (the 7th step)
9th device: electric amplifier (the 8th step)
Tenth device: OSA (the 8th step)
11st device: processed offline chip (the 9th step to the 12nd step).
Fig. 3 describes the performance of traditional DNN balanced device and K-DNN balanced device in two kinds of nonlinear channels of height.By Fig. 3
(a) as it can be seen that in the system compared with high non-linearity, traditional DNN balanced device of β → 0 needs 1300 repetitive exercises to reach hard decision
Thresholding (HD-FEC) 3.8 × 10-3, and then 600 repetitive exercises of minimum only needs are the K-DNN balanced device with kernel function auxiliary
It can reach HD-FEC, the occupancy of computing resource is reduced to 53.85%.Equally in Fig. 3 (b), non-linear lesser system center
Function still can effectively accelerate the training rate of DNN balanced device.It can be seen that in Fig. 3 (a) of either high non-linearity or low
K-DNN in nonlinear Fig. 3 (b) can reach hard decision thresholding relative to traditional DNN balanced device faster.
It should be noted that the device is device corresponding with above method embodiment, own in above method embodiment
Implementation can also reach identical technical effect suitable for the embodiment of the device.
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
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (2)
1. a kind of nonlinear distortion compensation algorithm based on K-DNN, which is characterized in that be balanced using the DNN of kernel function auxiliary
Device carries out pre-weighting, and the DNN balanced device based on kernel function auxiliary is denoted as K-DNN；The steps include: that adaptive line will be passed through first
Serial signal after balanced device progress linear equalization is converted to parallel signal, inputs K-DNN as feature vector；By K-DNN
Model training ranges feature vector that classification of maximum probability, obtains original signal；Wherein, K-DNN model training,
Using entropy function is intersected as cost function, optimization is iterated to the parameter in K-DNN by back-propagation algorithm；Wherein:
The K-DNN includes an input layer, a kernel function layer, multiple hidden layers, an output layer and one
Softmax classification layer；The input layer is one-to-one, i.e. in input layer a input to the node between kernel function layer
Node only transmits data to unique corresponding core node in kernel function layer；Kernel function layer is to each between output layer
The output of node is all multiplied by a weightIt is sent in all nodes of next layer, and is finally reached Softmax layers and is divided
Class；WeightMatrix be expressed as Wl:
Wherein, i and j respectively represents i-th of node of current layer and next layer of j-th of node, and l then represents first of hidden layer,
When l is equal to L+1, l represents output layer, and L is the total number of hidden layer, and m is the number of nodes of current layer, and n is next time
Number of nodes；
Serial signal after self-adaptive linear equalisation device carries out linear equalization be converted into after parallel signal as feature to
AmountIt is input to the input layer of K-DNN；Being distorted caused by non-linear and intersymbol interference (ISI) causes
Center signalThe interference of the several signals adjacent in timing by it；Therefore, withAdjacent several signals
The input layer of K-DNN is input to by the elements in parallel as feature vector；
The kernel function layer is the second layer of K-DNN；In this layer, accelerate the training process of DNN using kernel function；The core letter
Number includes: linear kernel (LK), polynomial kernel (PK), cosine kernel (CK), expression are as follows:
Linear kernel (LK):
Polynomial kernel (PK):
Cosine kernel (CK):
Wherein, σ is the parameter for adjusting kernel function, t0At the time of for current demand signal, at the time of t is closing signal；
For the network inputted with n feature, kernel function K can be expressed as an amplitude correction vector
The hidden layer is clipped between kernel function layer and output layer；Wherein the node in the first hidden layer is to all nodes of preceding layer
Output summation, and the input as node, then by this input plus a biasing
Here, kiIt is the element of the amplitude correction vector of kernel function K；
In K-DNN, use Sigmoid as activation primitive, be expressed as f (x)=max (0, x), wherein x is and ReLU phase
The output of node even；Node expression in view of activation primitive, in all hidden layers are as follows:
In the output layer, number of nodes is suitable with the quantity of level classification of signal is sent, and then, by Softmax layers, obtains
To classification results, expression are as follows:
Wherein, OiIt is the output for exporting node layer, its expression formula are as follows:
Then, according to trained model, feature vector x is ranged that classification L of maximum probabilityp；
Finally, using entropy function is intersected as cost function, by back-propagation algorithm to the hyper parameter (W in K-DNNlAnd bl)
It is iterated optimization；Obtain original signal.
2. a kind of visible light communication device based on nonlinear distortion compensation algorithm described in claim 1, be divided into transmitting terminal and
Receiving end, which is characterized in that use above-mentioned nonlinear distortion compensation algorithm, specific structure includes:
Data prediction chip；Its function includes: that original bit sequence is mapped to real symbol to form PAM8 signal；Using
PS-Manchester coding improves system performance for alleviating common-mode noise；Carry out up-sampling and nyquist filtering；
Arbitrary waveform generator (AWG)；For generating electric signal according to input signal；
Hardware pre equalizer based on Bridged-T；For compensating the decaying of high fdrequency component to input electrical signal；
With 2 microcircuit amplifiers (EA), a bias-tee combination, transmitter, transmitting collimation convex lens；Signal is put through EA
After big, electric signal and DC offset voltage are applied in LED light by the merging of bias-tee group, and the emitted device of LED light passes through standard
Straight convex lens is collimated；
Collectiong focusing convex lens, PIN photodiode；Under water and after free space transmission, PAM8 signal line focus convex lens
It is received after mirror by PIN photodiode；
Electric amplifier EA, digital storage oscilloscope (OSA)；Signal is amplified by EA, and is received by digital storage oscilloscope (OSA),
To carry out further signal processed offline；
Processed offline chip；Its function be processed offline is carried out to signal, including, it is synchronous to execute signal in order, power normalizing
Change and down-sampling, to obtain standardized PAM8 signal；Using differential decoding, for mitigating the influence of common-mode noise；Using S-
The adaptive post-equalizer of MCMMA compensates linear distortion；Nonlinear distortion compensation is carried out using K-DNN equilibrium and demapping and is divided
Class, to obtain original signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910041824.6A CN109905170B (en) | 2019-01-17 | 2019-01-17 | K-DNN-based nonlinear distortion compensation algorithm and visible light communication device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910041824.6A CN109905170B (en) | 2019-01-17 | 2019-01-17 | K-DNN-based nonlinear distortion compensation algorithm and visible light communication device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109905170A true CN109905170A (en) | 2019-06-18 |
| CN109905170B CN109905170B (en) | 2021-09-17 |
Family
ID=66943857
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910041824.6A Active CN109905170B (en) | 2019-01-17 | 2019-01-17 | K-DNN-based nonlinear distortion compensation algorithm and visible light communication device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109905170B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110830112A (en) * | 2019-10-16 | 2020-02-21 | 青岛海信电器股份有限公司 | Visible light communication method and device |
| CN111565160A (en) * | 2020-05-18 | 2020-08-21 | 中山大学 | Combined channel classification, estimation and detection method for ocean communication system |
| CN112257795A (en) * | 2020-10-28 | 2021-01-22 | 重庆邮电大学 | Spatial light communication system nonlinear compensation method based on naive Bayes |
| CN112737682A (en) * | 2020-12-31 | 2021-04-30 | 华南师范大学 | LED nonlinear distortion compensation method and device and visible light communication system |
| CN112865866A (en) * | 2021-01-20 | 2021-05-28 | 重庆邮电大学 | Visible light PAM system nonlinear compensation method based on GSN |
| CN112910562A (en) * | 2021-01-15 | 2021-06-04 | 清华大学深圳国际研究生院 | Communication method based on probability shaping |
| CN115102616A (en) * | 2022-05-29 | 2022-09-23 | 复旦大学 | Underwater wireless green light communication transmission system based on plastic optical fiber extension communication link |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105007118A (en) * | 2015-06-10 | 2015-10-28 | 重庆邮电大学 | Neural network equalization method used for indoor visible light communication system |
| CN105207963A (en) * | 2015-08-13 | 2015-12-30 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Balancing method based on probability nerve network and applied to visible light communication system |
| CN207835454U (en) * | 2017-07-21 | 2018-09-07 | 华南理工大学 | A kind of neural net equalizer based on visible light communication |
| US20190007143A1 (en) * | 2015-12-30 | 2019-01-03 | Universidad De Santiago De Chile | System and Method for Communication by Means of Visible Light for Underground Tunnels |
-
2019
- 2019-01-17 CN CN201910041824.6A patent/CN109905170B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105007118A (en) * | 2015-06-10 | 2015-10-28 | 重庆邮电大学 | Neural network equalization method used for indoor visible light communication system |
| CN105207963A (en) * | 2015-08-13 | 2015-12-30 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Balancing method based on probability nerve network and applied to visible light communication system |
| US20190007143A1 (en) * | 2015-12-30 | 2019-01-03 | Universidad De Santiago De Chile | System and Method for Communication by Means of Visible Light for Underground Tunnels |
| CN207835454U (en) * | 2017-07-21 | 2018-09-07 | 华南理工大学 | A kind of neural net equalizer based on visible light communication |
Non-Patent Citations (1)
| Title |
|---|
| NAN CHI 等: "Gaussian kernel-aided deep neural network equalizer utilized in underwater PAM8 visible light communication system", 《OPTICS EXPRESS》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110830112A (en) * | 2019-10-16 | 2020-02-21 | 青岛海信电器股份有限公司 | Visible light communication method and device |
| CN111565160A (en) * | 2020-05-18 | 2020-08-21 | 中山大学 | Combined channel classification, estimation and detection method for ocean communication system |
| CN111565160B (en) * | 2020-05-18 | 2021-03-26 | 中山大学 | Combined channel classification, estimation and detection method for ocean communication system |
| CN112257795A (en) * | 2020-10-28 | 2021-01-22 | 重庆邮电大学 | Spatial light communication system nonlinear compensation method based on naive Bayes |
| CN112257795B (en) * | 2020-10-28 | 2022-10-18 | 重庆邮电大学 | Spatial light communication system nonlinear compensation method based on naive Bayes |
| CN112737682A (en) * | 2020-12-31 | 2021-04-30 | 华南师范大学 | LED nonlinear distortion compensation method and device and visible light communication system |
| CN112737682B (en) * | 2020-12-31 | 2022-02-18 | 华南师范大学 | LED nonlinear distortion compensation method and device and visible light communication system |
| CN112910562A (en) * | 2021-01-15 | 2021-06-04 | 清华大学深圳国际研究生院 | Communication method based on probability shaping |
| CN112910562B (en) * | 2021-01-15 | 2022-02-11 | 清华大学深圳国际研究生院 | Communication method based on probability shaping |
| CN112865866A (en) * | 2021-01-20 | 2021-05-28 | 重庆邮电大学 | Visible light PAM system nonlinear compensation method based on GSN |
| CN115102616A (en) * | 2022-05-29 | 2022-09-23 | 复旦大学 | Underwater wireless green light communication transmission system based on plastic optical fiber extension communication link |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109905170B (en) | 2021-09-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109905170A (en) | A kind of nonlinear distortion compensation algorithm and visible light communication device based on K-DNN | |
| Chi et al. | Gaussian kernel-aided deep neural network equalizer utilized in underwater PAM8 visible light communication system | |
| CN108123908B (en) | Improved SVM (support vector machine) equalization method and system for NG-PON (NG-Passive optical network) | |
| Fang et al. | Deep learning detection method for signal demodulation in short range multipath channel | |
| Li et al. | 100Gbps IM/DD transmission over 25km SSMF using 20G-class DML and PIN enabled by machine learning | |
| CN111565160B (en) | Combined channel classification, estimation and detection method for ocean communication system | |
| Li et al. | Deep learning based end-to-end visible light communication with an in-band channel modeling strategy | |
| Niu et al. | End-to-end deep learning for long-haul fiber transmission using differentiable surrogate channel | |
| CN112598072A (en) | Equalization method of improved Volterra filter based on weight coefficient migration of SVM training | |
| Fei et al. | Underwater wireless optical communication utilizing low-complexity sparse pruned-term-based nonlinear decision-feedback equalization | |
| Bansbach et al. | Spiking neural network decision feedback equalization | |
| CN114301529A (en) | Volterra equalization method and system based on multi-symbol processing | |
| Li et al. | Deep-learning-based multi-user framework for end-to-end fiber-MMW communications | |
| Bi et al. | Machine learning classifier based on FE-KNN enabled high-capacity PAM-4 and NRZ transmission with 10-G class optics | |
| Jones | Machine learning methods in coherent optical communication systems | |
| Saidutta et al. | VAE for joint source-channel coding of distributed Gaussian sources over AWGN MAC | |
| Mortada et al. | Optical wireless communication performance enhancement using Hamming coding and an efficient adaptive equalizer with a deep-learning-based quality assessment | |
| Xiao et al. | Optimizations and investigations for transfer learning of iteratively pruned neural network equalizers for data center networking | |
| Lu et al. | Signal recovery in optical wireless communication using photonic convolutional processor | |
| CN114124223B (en) | Convolutional neural network optical fiber equalizer generation method and system | |
| CN114520693A (en) | Semantic optical communication system and method for text transmission | |
| Ney et al. | Unsupervised ANN-based equalizer and its trainable FPGA implementation | |
| Liu et al. | Transfer-learning based convolutional neural network for short-reach optical interconnects | |
| Li et al. | Optical equalization using photonic reservoir computing with optical analog signal injection | |
| Karanov | End-to-end deep learning in optical fibre communication systems |
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 |