Research On High Precision Frequency Locking System Based On Deep Learning

In recent years,with the development of technologies such as the Internet of Things and 5G,the rapidly increasing information communication traffic has put forward higher requirements for the data transmission capacity and quality of optical communication systems.The rapidly developing coherent optical technology can effectively improve the quality of optical fiber communication.It has the advantages of high sensitivity and diverse modulation formats,and has broad application prospects in high-speed and large-capacity optical communication systems.However,the problem of frequency drift is one of the main problems affecting the performance of receivers in coherent optical transmission systems,and it also limits the quality of coherent optical communication.Usually,optical phase-locked loop,laser frequency stabilization and other technologies are used to solve the frequency drift problem in coherent optical transmission.Optical phase-locked loops can be divided into two types:analog optical phase-locked loops and digital optical phase-locked loops.Due to the limited locking bandwidth of the analog optical phaselocked loop,the cumbersome locking process and the high production cost,this paper uses the digital optical phase-locked loop technology to build a laser frequency locking system.The frequency value of the beat frequency signal between the laser output light can be regarded as time series data,and the frequency value of the next time step can be predicted,which can be used for feedback control calculation and advance compensation,which can improve the frequency locking accuracy.The deep learning method has achieved better performance than traditional mathematical methods in sequence data prediction.Therefore,in the digital optical phase-locked loop system in this paper,a time series prediction model based on deep learning theory is built to achieve high-precision feedback control.The main research contents of this paper include:1.The related principles of fiber lasers and common time series prediction models are studied.The factors affecting the output of singlefrequency lasers and the frequency drift in the fiber link are focused on.The commonly used laser frequency-locking techniques based on laser frequency stabilization and optical phase-locked loops are introduced,and the reasons for choosing digital optical phase-locked loops are analyzed in this paper.At the same time,time series prediction models based on statistics,machine learning and deep learning are studied,focusing on the advantages of deep learning models in time series prediction,and providing theoretical guidance for the implementation of feedback control algorithms in frequency-locked systems.2.The design scheme of an intelligent digital frequency locking system with portability based on deep learning is proposed and experimentally verified.The system uses the Raspberry Pi embedded development platform as the master controller,including two parts:the beat frequency detection system and the feedback control system.The control algorithm in the feedback control system,based on the proportional-integralderivative control algorithm,introduces a time series prediction model based on the long-short-term memory network to compensate for the adjustment error caused by the system delay,so as to achieve more refined feedback compensation.At the same time,the dual control of fast and slow loops is introduced into the feedback control system,which prolongs the locking time and improves the robustness.The frequency locking of the frequency difference signal between the two local lasers and the signal light after passing through the optical fiber link and the local oscillator light is respectively realized.The locking time is up to 12 hours and 35 hours respectively,which proves the effectiveness of the designed frequency locking system.And the experimental results show that the accuracy of the frequency locking system based on deep learning is significantly higher than that of the frequency locking system based on the proportionalintegral-derivative algorithm,which verifies the feasibility of the application of deep learning in the laser frequency locking technology.To sum up,the high-precision frequency locking system based on deep learning designed in this paper has the characteristics of simple structure,high locking accuracy and long locking time,which provides a feasible solution for solving the problem of frequency drift in coherent optical systems.