大规模光学相控阵分级SPGD算法及优化研究  

Study on Multi-Stage SPGD Algorithm of Large Scale Optical Phased Array

作　　者：韩荣磊 孙建锋[1,3] 侯培培[1] 董芳 许玲玲 张正伟 姜玉鑫 任伟杰 从海胜 张龙坤 李超洋 Han Ronglei;Sun Jianfeng;Hou Peipei;Dong Fang;Xu Lingling;Zhang Zhengwei;Jiang Yuxin;Ren Weijie;Cong Haisheng;Zhang Longkun;Li Chaoyang(Key Laboratory of Space Laser Communication and Detection Technology,Shanghai Institute of Optics and Fine Mechanics,Chinese Academy of Sciences,Shanghai 201800,China;Center of Materials Science and Optoelectronics Engineering,University of Chinese Academy of Sciences,Beijing 100049,China;Spaceborne Laser Engineering Department,Shanghai Institute of Optics and Fine Mechanics,Chinese Academy of Sciences,Shanghai 201800,China)

机构地区：[1]中国科学院上海光学精密机械研究所空间激光传输与探测技术重点实验室,上海201800 [2]中国科学院大学材料与光电研究中心,北京100049 [3]中国科学院上海光学精密机械研究所航天激光工程部,上海201800

出　　处：《中国激光》2023年第2期72-83,共12页Chinese Journal of Lasers

基　　金：国家重点研发计划(2020YFB0408302);国家自然科学基金(91938302);中国科学院战略性先导科技专项(XDB43030400);中国科学院重点部署项目(ZDRW-KT-2019-1-01-0302)。

摘　　要：随机并行梯度下降(SPGD)算法是一种应用广泛的最优化算法,在光学相控阵的多光束相位调控中具有重要作用。常规的SPGD算法在光学相控阵单元数目较大的应用场景下存在着迭代步数多、收敛速度慢等缺陷。为此提出了一种分级SPGD算法,将多光束进行分级,通过在多级使用SPGD算法对光束的调控来实现快速收敛和稳定维持。介绍了该算法的理论模型和流程,运用数值仿真的方式与传统算法进行了性能比较,结果表明,分级SPGD算法能够在大规模光学相控阵中显著提高收敛速度。搭建了光学相控阵实验系统进行分级SPGD算法的原理和可行性验证,在不同光束数目条件下实现了分级SPGD算法的闭环锁相,验证了算法的优越性,结果显示分级算法在32路光束时可以将迭代次数降低到常规算法的59.6%。Objective In the process of space laser communication networking,optical phased array(OPA) can replace the traditional mechanical turntable to realize the lightweight and miniaturization of laser terminals,and can quickly switch links among different terminals.The space laser communication network based on optical phased array will be the inevitable trend of the future development.In OPA technology,it is necessary to ensure the same phase of all the beams split from the same laser source.Therefore,the monitoring part is set to obtain the phase of beams to do the phase locking.Considering a large number of beams,it is not practical to calculate the phase directly,so it is suitable to use iterative algorithm to calculate the compensation.At present,there are many algorithms in the field of multivariable control,and stochastic parallel gradient descent(SPGD) algorithm has more advantages in convergence speed and effect,but with the increase of the number of OPA elements,the convergence speed of SPGD becomes significantly slower.There exist some optimized SPGD algorithms now,such as AdmSPGD,AdaDelSPGD and other schemes,which have optimized the step size,the gain parameter and other coefficients of SPGD,and improved the speed to a certain extent.In the experiments using SPGD algorithm,the largest number of beams in the array reported is written by the 107-channel fiber laser coherent synthesis based on SPGD algorithm.As the number of beams increases,the structure becomes more complex and the performance of the algorithm must be higher.Nowadays,OPA has been widely used in laser communication and lidar where there is always a large optical aperture and a large number of array elements are required.When the number of array elements reaches thousands,only optimizing the parameters of SPGD algorithm does not change the essence of control of all array elements based on one single evaluation parameter,which has limited performance improvement.Therefore,it is necessary to optimize the SPGD algorithm in a deeper level for the

关 键 词：激光光学 光学相控阵 分级随机并行梯度下降算法 相位调控 

分 类 号：TN929.1[电子电信—通信与信息系统]