CN111988249B - Receiving end equalization method based on adaptive neural network and receiving end - Google Patents

Receiving end equalization method based on adaptive neural network and receiving end Download PDF

Info

Publication number
CN111988249B
CN111988249B CN202010686240.7A CN202010686240A CN111988249B CN 111988249 B CN111988249 B CN 111988249B CN 202010686240 A CN202010686240 A CN 202010686240A CN 111988249 B CN111988249 B CN 111988249B
Authority
CN
China
Prior art keywords
neural network
symbol
receiving end
adaptive neural
adaptive
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.)
Active
Application number
CN202010686240.7A
Other languages
Chinese (zh)
Other versions
CN111988249A (en
Inventor
张帆
明浩
阮小可
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Peking University
Original Assignee
Peking University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Peking University filed Critical Peking University
Priority to CN202010686240.7A priority Critical patent/CN111988249B/en
Publication of CN111988249A publication Critical patent/CN111988249A/en
Application granted granted Critical
Publication of CN111988249B publication Critical patent/CN111988249B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • H04L25/03165—Arrangements for removing intersymbol interference using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04J—MULTIPLEX COMMUNICATION
    • H04J14/00—Optical multiplex systems
    • H04J14/02—Wavelength-division multiplex systems
    • H04J14/0227—Operation, administration, maintenance or provisioning [OAMP] of WDM networks, e.g. media access, routing or wavelength allocation

Landscapes

  • Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Power Engineering (AREA)
  • Cable Transmission Systems, Equalization Of Radio And Reduction Of Echo (AREA)

Abstract

The invention discloses a receiving end equalization method based on a self-adaptive neural network and a receiving end. The method comprises the following steps: 1) Embedding a self-adaptive neural network in a digital signal processing flow of a receiving end, wherein the self-adaptive neural network comprises a neural network, a judgment unit and a loss calculation unit; 2) Training a neural network by using the received symbol generation characteristic vector as training data and using a corresponding transmitting symbol as a label; 3) Taking the parameters of the trained neural network as initialization parameters of the adaptive neural network; 4) Inputting the characteristic vector corresponding to the received symbol to be balanced into the adaptive neural network to obtain corresponding output which is recorded as y and is used as the symbol output after the balance; 5) y is judged by a judgment unit to obtain a pseudo label
Figure DDA0002587656290000011
6) The loss calculation unit calculates y and
Figure DDA0002587656290000012
the error between L and the gradient of L to the neural network parameters; 7) And calculating the average gradient to update the neural network parameters, and outputting the received symbols to be equalized subsequently after equalization processing.

Description

Receiving end equalization method based on adaptive neural network and receiving end
Technical Field
The invention belongs to the field of optical communication transmission, relates to a receiving end balancing method, and particularly relates to a receiving end balancing method based on a self-adaptive neural network.
Background
The optical fiber communication transmission system has the advantages of large capacity, low cost and the like. Overcoming the linear and nonlinear damage suffered by the optical signal in the transmission process has become a first problem facing further increasing the capacity of the optical fiber communication transmission system. Conventional solutions to this problem have two categories, one being Time Domain Equalization (TDE) and the other being Frequency Domain Equalization (FDE).
In recent years, neural networks have been successfully applied to optical fiber communication transmission systems due to their strong fitting capability, so as to compensate for linear and nonlinear damages suffered by optical signals during transmission. As a data-driven algorithm, the neural network can compensate signals in both the time domain and the frequency domain.
The method for compensating signal damage by adopting the neural network is mainly divided into two stages of training and balancing. In the training stage, the neural network optimizes the network parameters thereof according to the training sequence; and in the equalization stage, compensating the signals by adopting a neural network subjected to parameter optimization. Since neural networks have a strong fitting ability, fully trained neural networks are generally able to compensate signals well. However, since the parameters of the neural network are not updated once the training is completed, the compensation performance of the neural network may be reduced sharply if the signal is subjected to different impairment characteristics in the training and equalization stages or if the transmission channel is changed over time.
Disclosure of Invention
Aiming at the technical problems in the prior art, the invention aims to provide a receiving end equalization method based on an adaptive neural network.
In order to realize the purpose, the invention adopts the following technical scheme:
1. the signal itself may be processed by one or more equalization algorithms at the receiving end, and the adaptive neural network may be embedded at any position in the digital signal processing flow on the premise of ensuring that the adaptive neural network input is symbol information. After the embedded position of the adaptive neural network is determined, the receiving symbols of the optical fiber communication transmission system are collected and stored at the embedded position.
2. And generating a feature vector by using the stored received symbol according to a specific mode as training data, and using the symbol sent by a corresponding sending end as a label to train the neural network. The eigenvector corresponding to the kth received symbol is marked as x k 。
3. And taking the parameters of the trained neural network as initialization parameters of the adaptive neural network. The self-adaptive neural network provided by the invention comprises a neural network, a judgment unit and a loss calculation unit; namely, an adaptive structure capable of realizing parameter updating is added on the basis of a neural network; wherein the neural network may be an Artificial Neural Network (ANN) or other neural network.
4. And sequentially inputting the characteristic vectors x corresponding to the received symbols to be balanced into the adaptive neural network, wherein the output corresponding to the network is the balanced symbols and is marked as y. If the sending symbol corresponding to the receiving symbol is a real number, y is a real number; if the transmitted symbol is complex, y is a vector containing two elements corresponding to the real part and the imaginary part of the transmitted symbol, respectively.
5. The output y of the self-adaptive neural network is judged to obtain a pseudo label
Figure BDA0002587656270000021
The judgment mode is
Figure BDA0002587656270000022
Wherein y is i Indicating the ith symbol in the transmit symbol set S.
6. Calculating the error between the network output and the pseudo tag using a squared error loss function
Figure BDA0002587656270000023
And gradient of error L to network parameter θ
Figure BDA0002587656270000024
θ represents all parameters of the neural network.
7. Calculating the average gradient corresponding to the B feature vectors
Figure BDA0002587656270000025
Wherein B is not less than 1,g k Representing the gradient corresponding to the k-th feature vector.
Figure BDA0002587656270000026
y k And
Figure BDA0002587656270000027
respectively representing the kth output of the adaptive neural network and its corresponding pseudo label.
8. Updating the network parameters:
Figure BDA0002587656270000028
where η is the learning rate of the adaptive neural network.
The invention also discloses a receiving end based on the self-adaptive neural network, which is characterized by comprising the self-adaptive neural network, a receiving end and a receiving end, wherein the self-adaptive neural network is used for balancing the received symbols in the digital signal processing flow; the self-adaptive neural network comprises a neural network, a judgment unit and a loss calculation unit; training the neural network by using the characteristic vector generated by the received symbol as training data and using a transmitting symbol corresponding to the received symbol as a label; taking the parameters of the trained neural network as the initialization parameters of the adaptive neural network;
the neural network is used for balancing a characteristic vector x corresponding to an input receiving symbol to obtain a corresponding output which is recorded as y and is used as the balanced symbol output; if the sending symbol corresponding to the receiving symbol is a real number, y is a real number; if the transmitting symbol corresponding to the receiving symbol is a complex number, y is a vector including a real part and an imaginary part of the transmitting symbol;
the judgment unit is used for judging y to obtain a pseudo label
Figure BDA0002587656270000029
The loss calculation unit is used for calculating the network output y and the pseudo label
Figure BDA00025876562700000210
Error L between, and gradient of error L to the neural network parameter θ
Figure BDA00025876562700000211
Then calculating the average gradient corresponding to the B feature vectors
Figure BDA00025876562700000212
Then updating the neural network parameters
Figure BDA0002587656270000031
Equalizing and outputting subsequent received symbols to be equalized, wherein B is more than or equal to 1,g k Representing the gradient corresponding to the k-th feature vector; eta is the learning rate of the adaptive neural network.
Compared with the prior art, the invention has the following positive effects:
the adaptive neural network equalizer has higher equalization performance than the conventional neural network equalizer, and the specific experimental result is shown in fig. 4.
Drawings
Fig. 1 is a schematic structural diagram of a training phase of an optical fiber communication transmission system according to an embodiment of the present invention.
Fig. 2 is a schematic structural diagram of an adaptive neural network according to an embodiment of the present invention.
Fig. 3 is a schematic structural diagram of an equalization stage of an optical fiber communication transmission system according to an embodiment of the present invention.
FIG. 4 is a schematic diagram of the experimental results of the 64Gbaud, 16QAM signal 960km transmission in the WDM system according to the embodiment of the present invention.
Detailed Description
The present invention will be described in further detail below with reference to specific examples and the accompanying drawings.
1. The optical signal at the receiving end is processed by front-end digital processing (dispersion compensation, phase recovery, etc.) to obtain a received symbol, and a corresponding characteristic vector is generated according to the received symbol. The kth received symbol s k Feature vector x of k Composed of itself and 2L adjacent received symbols before and after it, i.e. x k ＝[Re(s k-L ),Im(s k-L ),...,Re(s k ),Im(s k ),...,Re(s k+L ),Im(s k+L )]。Re(s k ) Representing complex symbols s k Real part of (i), im(s) k ) Representing complex symbols s k The imaginary part of (c). The value of L needs to be optimized according to an actual system, and the higher the transmission rate of a general system is, the longer the transmission distance is, the larger the value of L is.
2. N to be generated tr The characteristic vectors are used as training data, corresponding sending symbols are used as labels, and a neural network is trained by adopting a square error loss function.The above process is illustrated by figure 1. If the transmitted symbol is a real number, the symbol itself is used as a label; if the transmitted symbol is complex, the real part and imaginary part of the complex signal form a one-dimensional vector as a label.
3. And taking the parameters of the trained neural network as initialization parameters of the adaptive neural network. The specific structure of the adaptive neural network is shown in fig. 2. The j th symbol s to be equalized j Generated feature vector x j As input to the adaptive neural network, its output is denoted as y j . Output y j Obtaining a pseudo label after judgment
Figure BDA0002587656270000032
The decision rule is as follows:
Figure BDA0002587656270000041
wherein y is i Which represents the ith symbol in the transmit symbol set S, which is composed of 16 complex symbols, taking a 16QAM signal as an example. Computing network output y using a squared error loss function k And a pseudo tag
Figure BDA0002587656270000042
Error between
Figure BDA0002587656270000043
And gradient of L to network parameter θ
Figure BDA0002587656270000044
4. Calculating the average gradient corresponding to the continuous B feature vectors:
Figure BDA0002587656270000045
wherein B is more than or equal to 1. Updating parameters of the adaptive neural network according to the calculated average gradient:
Figure BDA0002587656270000046
where η is the learning rate of the adaptive neural network. It is noted that when the adaptive neural network is used to equalize signal impairments, the network parameters are updated every B symbols of equalization.
5. And continuously balancing subsequent data by adopting the adaptive neural network after the parameters are updated. The above process is illustrated by fig. 3.
FIG. 4 is a schematic diagram of the 960km experimental results of WDM system transmission using 64Gbaud, 16QAM signals according to an embodiment of the present invention. The horizontal axis is the fiber input power and the vertical axis is the signal Q 2 The value is obtained. The equalization result of the adaptive neural network and the traditional neural network shows that the adaptive neural network has higher equalization performance, because the adaptive neural network can dynamically adjust the network parameters of the adaptive neural network according to the damage characteristics of the signal.
The above embodiments are only intended to illustrate the technical solution of the present invention and not to limit the same, and a person skilled in the art can make modifications or equivalent substitutions to the technical solution of the present invention without departing from the spirit and scope of the present invention, and the scope of protection of the present invention should be subject to the claims.

Claims (10)

1. A receiving end equalization method based on an adaptive neural network comprises the following steps:
1) Embedding a self-adaptive neural network in a digital signal processing flow of a receiving end, wherein the self-adaptive neural network is used for balancing a received symbol; the self-adaptive neural network comprises a neural network, a judgment unit and a loss calculation unit;
2) Training the neural network by using the characteristic vector generated by the received symbol as training data and using a transmitting symbol corresponding to the received symbol as a label; wherein the eigenvector corresponding to the kth received symbol is marked as x k ；
3) Taking the parameters of the trained neural network as initialization parameters of the adaptive neural network;
4) Sequentially inputting the characteristic vectors x corresponding to the received symbols to be equalized into the adaptive neural network to obtain corresponding outputs which are recorded as y and serve as equalized symbol outputs; if the sending symbol corresponding to the receiving symbol is a real number, y is a real number; if the transmitting symbol corresponding to the receiving symbol is a complex number, y is a vector including a real part and an imaginary part of the transmitting symbol;
5) y is processed by the judgment unit to obtain a pseudo label
Figure FDA0003914727800000011
6) Loss calculation unit calculates network output y and pseudo label
Figure FDA0003914727800000012
Error L between, and gradient of error L to the neural network parameter θ
Figure FDA0003914727800000013
7) Calculating the average gradient corresponding to the B feature vectors
Figure FDA0003914727800000014
Wherein B is not less than 1,g k Representing the gradient corresponding to the k-th feature vector; then updating the neural network parameters
Figure FDA0003914727800000015
And carrying out equalization processing on the subsequent receiving symbols to be equalized and then outputting the symbols, wherein eta is the learning rate of the adaptive neural network.
2. The method of claim 1, wherein the loss computation unit computes the net output y and the pseudo label using a squared error loss function
Figure FDA0003914727800000016
Error between
Figure FDA0003914727800000017
3. The method of claim 1, wherein the kth received symbol s k Corresponding feature vector x k Is composed of itself and 2L adjacent received symbols before and after it.
4. The method of claim 3, wherein the feature vector x k ＝[Re(s k-L ),Im(s k-L ),...,Re(s k ),Im(s k ),...,Re(s k+L ),Im(s k+L )](ii) a Wherein, re(s) k ) Representing a complex symbol s k Real part of, im(s) k ) Representing complex symbols s k The imaginary part of (c).
5. The method of claim 1, wherein the decision unit decides in such a way that
Figure FDA0003914727800000018
Wherein y is i Indicating the ith symbol in the transmit symbol set S.
6. A receiving end based on a self-adaptive neural network is characterized by comprising the self-adaptive neural network, a receiving end and a processing device, wherein the self-adaptive neural network is used for equalizing a received symbol in a digital signal processing flow; the self-adaptive neural network comprises a neural network, a judgment unit and a loss calculation unit; training the neural network by using the characteristic vector generated by the received symbol as training data and using a transmitting symbol corresponding to the received symbol as a label; taking the parameters of the trained neural network as the initialization parameters of the adaptive neural network;
the neural network is used for balancing a characteristic vector x corresponding to an input receiving symbol to obtain a corresponding output which is recorded as y and is used as a balanced symbol output; if the sending symbol corresponding to the receiving symbol is a real number, y is a real number; if the transmitting symbol corresponding to the receiving symbol is a complex number, y is a vector including a real part and an imaginary part of the transmitting symbol;
the judgment unit is used for judging y to obtain a pseudo label
Figure FDA0003914727800000021
The loss calculation unit is used for calculating the network output y and the pseudo label
Figure FDA0003914727800000022
Error L between, and gradient of error L to the neural network parameter θ
Figure FDA0003914727800000023
Then calculating the average gradient corresponding to the B feature vectors
Figure FDA0003914727800000024
Then updating the neural network parameters
Figure FDA0003914727800000025
Equalizing and outputting subsequent received symbols to be equalized, wherein B is more than or equal to 1,g k Representing the gradient corresponding to the k-th feature vector; eta is the learning rate of the adaptive neural network.
7. The receiving end according to claim 6, characterized in that the kth received symbol s k Corresponding feature vector x k Is composed of itself and 2L adjacent received symbols before and after it.
8. The receiving end of claim 7, wherein the eigenvector x is characterized by k ＝[Re(s k-L ),Im(s k-L ),...,Re(s k ),Im(s k ),...,Re(s k+L ),Im(s k+L )](ii) a Wherein, re(s) k ) Representing a complex symbol s k Real part of, im(s) k ) Representing complex symbols s k The imaginary part of (c).
9. The receiving end of claim 6, wherein the loss calculating unit calculates the network output y and the pseudo label using a squared error loss function
Figure FDA0003914727800000026
Error therebetween
Figure FDA0003914727800000027
10. The receiving end of claim 6, wherein the decision unit decides in a manner of
Figure FDA0003914727800000028
Wherein y is i Indicating the ith symbol in the transmit symbol set S.
CN202010686240.7A 2020-07-16 2020-07-16 Receiving end equalization method based on adaptive neural network and receiving end Active CN111988249B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010686240.7A CN111988249B (en) 2020-07-16 2020-07-16 Receiving end equalization method based on adaptive neural network and receiving end

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010686240.7A CN111988249B (en) 2020-07-16 2020-07-16 Receiving end equalization method based on adaptive neural network and receiving end

Publications (2)

Publication Number Publication Date
CN111988249A CN111988249A (en) 2020-11-24
CN111988249B true CN111988249B (en) 2023-01-17

Family

ID=73437765

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010686240.7A Active CN111988249B (en) 2020-07-16 2020-07-16 Receiving end equalization method based on adaptive neural network and receiving end

Country Status (1)

Country Link
CN (1) CN111988249B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112965384B (en) * 2021-02-08 2021-11-16 南京工业大学 Global nonsingular rapid terminal sliding mode model-free method for adaptive neural network
CN113283592B (en) * 2021-05-13 2022-08-05 北京大学 Receiving end equalization method and system based on low-complexity LSTM

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110636020A (en) * 2019-08-05 2019-12-31 北京大学 Neural network equalization method for adaptive communication system

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110636020A (en) * 2019-08-05 2019-12-31 北京大学 Neural network equalization method for adaptive communication system

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
AdaNN: Adaptive Neural Network-Based Equalizer via Online Semi-Supervised Learning;qingyi zhou,等;《Journal of Lightwave Technology》;20200408;全文 *
DeepUNet: A Deep Fully Convolutional Network for Pixel-Level Sea-Land Segmentation;Ruirui Li，等;《IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing》;20180531;全文 *

Also Published As

Publication number Publication date
CN111988249A (en) 2020-11-24

Similar Documents

Publication Publication Date Title
Kechriotis et al. Using recurrent neural networks for adaptive communication channel equalization
CN111988249B (en) Receiving end equalization method based on adaptive neural network and receiving end
CN112598072B (en) Equalization method of improved Volterra filter based on weight coefficient migration of SVM training
CN108123908B (en) Improved SVM (support vector machine) equalization method and system for NG-PON (NG-Passive optical network)
CN112036543B (en) Time domain equalizer combining neural network equalization and linear equalization and equalization method
CN108494488B (en) DFE-based SVM (support vector machine) equalization method for short-distance optical communication system
CN110267127A (en) Method, apparatus and computer-readable medium for Low-cost passive optical network
CN109818889B (en) Equalization algorithm for SVM classifier optimization in high-order PAM optical transmission system
CN111917474B (en) Implicit triple neural network and optical fiber nonlinear damage balancing method
CN112381122B (en) Adaptive machine learning equalization method and communication system
CN109740690A (en) The KNN equalization algorithm based on Feature Engineering for short distance optic communication
CN112865866A (en) Visible light PAM system nonlinear compensation method based on GSN
CN114301529B (en) Volterra equalization method and system based on multi-symbol processing
CN112887237B (en) Method for designing complex channel equalizer of optical fiber communication system
CN112511472B (en) Time-frequency second-order equalization method based on neural network and communication system
Qin et al. Recurrent neural network based joint equalization and decoding method for trellis coded modulated optical communication system
CN114204993B (en) Nonlinear equalization method and system based on polynomial mapping feature construction
CN115882959B (en) Nonlinear compensation method based on machine learning
Chen et al. Computationally efficient pre-distortion based on adaptive partitioning neural network in underwater visible light communication
Wang et al. Low-complexity nonlinear equalizer based on artificial neural network for 112 Gbit/s PAM-4 transmission using DML
Chen et al. Machine learning of SVM classification utilizing complete binary tree structure for PAM-4/8 optical interconnection
CN113283592B (en) Receiving end equalization method and system based on low-complexity LSTM
Reza et al. 30-Gb/s PAM-8 transmission with directly-modulated laser using machine learning equalizer
CN115378507B (en) Noise equalization method and device suitable for multi-eigenvalue nonlinear frequency division multiplexing system
CN115276818B (en) Deep learning-based optical-load wireless transmission link demodulation method

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