Research On Key Technologies Of Dual Optical Comb Spectroscopy Based On Intelligent Algorithms

With the advent of the era of artificial intelligence（AI）,the interdisciplinary research related to AI is in the ascendant,and AI technology is widely used in industry and academia.In the field of photonics,AI has a profound impact on the development of optical communication,optical quantum and super-resolution spectroscopy.Optical frequency comb（OFC）,as a cutting-edge technology of precision measurement,has attracted extensive attention in photonics research.Free-running dual optical comb spectroscopy（DCS）is a vital research direction of OFC,which is mainly used in the fields of distance precision measurement and material detection.It can produce highly coherent dual optical combs with a single cavity laser and has the advantages of simple structure and no common-mode noise.However,for this technique,the environmental disturbance will affect the stability of the DCS,so that the detection process is difficult to achieve efficiently.The application of AI algorithm will provide a new idea for the stability research of the key technology of DCS.This paper introduces the basic theory and related technologies of OFC,summarizes the application of intelligent algorithm in OFC technology in the past,and applies the intelligent algorithm to the research of the DCS.The main contributions of this paper are to propose a new dual-comb spectral recovery algorithm and to design a stable mode-locked recovery algorithm for mode-locked laser,which provide technical support for high-precision measurement and detection of DCS.This paper mainly includes the following two tasks:1.The intelligent temporal alignment algorithm based on machine learning is designed to realize the alignment and recovery of the DCS spectral signals.A free-running single-cavity mode-locked dual-optical comb laser is constructed to detect the gas cell containing ¹²CO.The intelligent temporal alignment（ITA）algorithm is designed to establish the mapping function between the noised signal and the pure signal,so as to achieve the denoising of the detected signal and restore the absorption spectrum.Comparing the result with the HITRAN database,it is found that the gas absorption spectrum in the range of 1564-1570 nm is in good agreement with the frequency position,shape and intensity.This verifies the ability of the ITA algorithm to correct and restore the absorption spectrum.2.Based on the deep deterministic policy gradient in reinforcement learning,a low-latency deep-reinforcement（DELAY）algorithm is designed to realize the real-time control of polarization state of mode-locked laser to ensure the stable mode-locked state of laser.After combining the algorithm with a mode-locked fiber laser,the effectiveness of the DELAY algorithm is verified.The polarization state of laser is disturbed by simulating ambient vibration to make it lose mode-locked state.Then,the DELAY algorithm decision model trained in the actual laser environment is used to adjust the polarization controller based on the state of the laser to quickly restore the mode-locked state.Through thousands of experiments,it is found that the laser can return to stable mode-locked state in an average of 1.948 s,which is 62.8%of the fastest average recovery time in the previous published work.