CN-121984595-A - Optical receiver construction method, optical receiver construction device, optical receiver construction apparatus, optical receiver construction communication system, and storage medium

CN121984595ACN 121984595 ACN121984595 ACN 121984595ACN-121984595-A

Abstract

The application discloses an optical receiver building method, a device, equipment, a communication system and a storage medium, relating to the field of network communication, comprising a first neural network equalizer for building a first communication link based on a neural network; the method comprises the steps of configuring model parameters of a second neural network equalizer to be initial model parameters of a first neural network equalizer, performing fine tuning training on the first neural network equalizer by using training data of a first communication link based on the initial model parameters to obtain a target neural network equalizer, wherein the target neural network equalizer is applicable to an optical receiver of the first communication link, and the first communication link is different from the second communication link. Through the migration multiplexing of knowledge, the construction process of the optical receiver is changed from the low offer one's services dynamic of repeatability to the high-efficiency adaptive learning, so that the deployment efficiency and the expandability of the optical fiber communication system are greatly improved.

Inventors

XU ZHAOPENG
HE ZHIXUE
CHEN HUI
ZHANG XU
LI XINGFENG
JI HONGLIN
SUN ZHONGLIANG
FAN LINSHENG
LIANG JUNPENG
WEI JINLONG

Assignees

鹏城实验室

Dates

Publication Date: 20260505
Application Date: 20260126

Claims (10)

1. A light receiving mechanism constructing method, characterized in that the light receiving mechanism constructing method comprises: constructing a first neural network equalizer of a first communication link based on the neural network; Configuring model parameters of a second neural network equalizer as initial model parameters of the first neural network equalizer, wherein the second neural network equalizer is obtained by training by taking transmission signals in a second communication link as training data; And performing fine tuning training on the first neural network equalizer by using training data of the first communication link based on the initial model parameters to obtain a target neural network equalizer, wherein the target neural network equalizer is applicable to an optical receiver of the first communication link, and the first communication link is different from the second communication link.
2. The method of constructing an optical receiver unit according to claim 1, wherein the step of performing fine-tuning training on the first neural network equalizer based on the initial model parameters using training data of the first communication link to obtain a target neural network equalizer comprises: determining a common characteristic of the first communication link and the second communication link according to the similarity of channel characteristics of the first communication link and the second communication link; Determining common parameters related to the common features in the initial model parameters; And under the condition of freezing the common parameters, performing fine tuning training on the first neural network equalizer by using the training data of the first communication link to fine tune differential parameters to obtain a target neural network equalizer, wherein the differential parameters are parameters except the common parameters in the initial model parameters.
3. The method of constructing an optical receiver according to claim 2, wherein the step of determining the common characteristic of the first communication link and the second communication link based on the similarity of the channel characteristics of the first communication link and the second communication link further comprises: Determining the channel difference degree of the first communication link and the second communication link according to the common characteristics; and determining a parameter value of the super parameter according to the channel difference, wherein the parameter value of the super parameter is used for guiding the fine tuning training of the first neural network equalizer.
4. The optical receiver architecture method according to any one of claims 1 to 3, further comprising, before the step of configuring the model parameters of the second neural network equalizer to the initial model parameters of the first neural network equalizer: Selecting a source system based on the data quantity of the transmission signal to extract the signal, and obtaining pre-training data of the source system; establishing a reference communication link of the source system based on a neural network, the reference communication link comprising a second neural network equalizer; and performing iterative training on the second neural network equalizer by using the pre-training data to obtain model parameters of the second neural network equalizer, wherein the model parameters can reflect a common distortion mode of the source system.
5. A method of constructing an optical receiver unit according to any one of claims 1 to 3, wherein the step of performing fine-tuning training on the first neural network equalizer based on the initial model parameters using training data of the first communication link to obtain a target neural network equalizer further comprises: calculating the error rate of the first communication link based on the target neural network equalizer; evaluating the performance of the optical receiver according to the error rate; and optimizing model parameters of the target neural network equalizer according to the performance of the optical receiver.
6. A light-receiving mechanism construction method according to any one of claims 1 to 3, wherein the initial model parameters include at least one of a convolution kernel, a weight matrix, an offset vector, and a residual structure.
7. An optical fiber communication system, comprising a transmitting end, a fanin device, a space division multiplexing optical fiber, a fanout device, and a receiving end, wherein the transmitting end comprises a plurality of optical transmitters, the receiving end comprises a plurality of optical receivers, and the optical receivers are constructed by adopting the optical receiver construction method according to any one of claims 1 to 6, wherein: the optical transmitter is used for transmitting a modulation signal; the fan-in device is used for coupling the modulation signal into the space division multiplexing optical fiber; the space division multiplexing optical fiber is used for transmitting the modulation signal to the fan-out equipment; the fan-out device is used for decoupling the modulation signals to obtain multichannel signals and processing signals of all channels of the multichannel signals; the optical receiver is used for receiving each path of signal after signal processing.
8. An optical receiver construction apparatus, characterized in that the optical receiver construction apparatus comprises: a building module for building a first neural network equalizer of a first communication link based on the neural network; The migration module is used for configuring model parameters of a second neural network equalizer as initial model parameters of the first neural network equalizer, wherein the second neural network equalizer is obtained by training by taking transmission signals in a second communication link as training data; And the fine tuning training module is used for carrying out fine tuning training on the first neural network equalizer by utilizing training data of the first communication link based on the initial model parameters to obtain a target neural network equalizer, wherein the target neural network equalizer is suitable for an optical receiver of the first communication link, and the first communication link is different from the second communication link.
9. An optical receiver construction apparatus comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program being configured to implement the steps of the optical receiver construction method according to any one of claims 1 to 6.
10. A storage medium, characterized in that the storage medium is a computer-readable storage medium, on which a computer program is stored, which computer program, when being executed by a processor, realizes the steps of the light receiving unit construction method according to any one of claims 1 to 6.

Description

Optical receiver construction method, optical receiver construction device, optical receiver construction apparatus, optical receiver construction communication system, and storage medium Technical Field The present application relates to the field of network communications, and in particular, to a method, an apparatus, a device, a communication system, and a storage medium for constructing an optical receiver. Background To meet the explosive growth of data center traffic demands for 800G/1.6T and even higher optical interface capacities, high-speed, high-density parallel optical interconnects have become a necessary choice. The transmission architecture of multichannel parallel transmission such as space division multiplexing (Space Division Multiplexing, SDM) technology, especially Multi-core Fiber (MCF), effectively reduces the number of lasers by creating multiple parallel spatial channels in the same Fiber, remarkably reduces the cost, size and power consumption of the transmitting end, and exhibits remarkable advantages. On the receiving end, since different channel impairments, such as specific noise, crosstalk and nonlinear effects, are introduced in the transmission process of each parallel channel, a proprietary receiver needs to be configured for each channel and precise nonlinear equalization needs to be performed. The conventional scheme generally adopts an independent training strategy from zero for the equalizer configuration of the receiver of each channel, and can realize the performance optimization of each channel, but essentially superposes the construction process of the receiver of a single channel, so that the construction complexity of the receiver of the whole communication system is linearly increased along with the number of channels, the training time is long, and the deployment efficiency and the expandability of the communication system are seriously restricted. The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present application and is not intended to represent an admission that the foregoing is prior art. Disclosure of Invention The application mainly aims to provide an optical receiver construction method, an optical receiver construction device, an optical receiver construction equipment, a communication system and a storage medium, and aims to solve the technical problems that an existing receiver construction mode is high in construction complexity, long in training time and capable of affecting deployment efficiency and expandability of the communication system. In order to achieve the above object, the present application provides a light receiving mechanism construction method, comprising: constructing a first neural network equalizer of a first communication link based on the neural network; Configuring model parameters of a second neural network equalizer as initial model parameters of the first neural network equalizer, wherein the second neural network equalizer is obtained by training by taking transmission signals in a second communication link as training data; And performing fine tuning training on the first neural network equalizer by using training data of the first communication link based on the initial model parameters to obtain a target neural network equalizer, wherein the target neural network equalizer is applicable to an optical receiver of the first communication link, and the first communication link is different from the second communication link. In an embodiment, the step of performing fine tuning training on the first neural network equalizer based on the initial model parameters by using training data of the first communication link to obtain a target neural network equalizer includes: determining a common characteristic of the first communication link and the second communication link according to the similarity of channel characteristics of the first communication link and the second communication link; Determining common parameters related to the common features in the initial model parameters; And under the condition of freezing the common parameters, performing fine tuning training on the first neural network equalizer by using the training data of the first communication link to fine tune differential parameters to obtain a target neural network equalizer, wherein the differential parameters are parameters except the common parameters in the initial model parameters. In an embodiment, after the step of determining the common characteristic of the first communication link and the second communication link according to the similarity of the channel characteristics of the first communication link and the second communication link, the method further includes: Determining the channel difference degree of the first communication link and the second communication link according to the common characteristics; and determining a parameter value of the super parameter according to the