自适应光学系统随机并行梯度下降算法  

Stochastic parallel gradient descent algorithm for adaptive optics system

作　　者：马慧敏[1] 张鹏飞[1] 张京会[1] 范承玉[1] 王英俭[1] 

机构地区：[1]中国科学院安徽光学精密机械研究所大气成分与光学重点实验室,合肥230031

出　　处：《强激光与粒子束》2010年第6期1206-1210,共5页High Power Laser and Particle Beams

基　　金：国家自然科学基金项目(10574127);中国科学院合肥物质科学物质研究院计算中心资助课题

摘　　要：随机并行梯度下降（SPGD）算法可以对系统性能指标直接优化来校正畸变波前。对基于SPGD算法的61单元自适应光学系统进行仿真模拟,分析了对不同初始静态畸变波前的校正能力,并比较了不同性能指标情况下的算法增益系数、扰动幅度值的选取及校正情况。仿真结果表明：算法收敛速度很大程度上依赖于增益系数和扰动幅度值,对畸变较大的波前,随机扰动幅度在0.50~0.85范围内,性能指标采用焦斑平均半径比采用斯特列尔比取得的校正效果好。The stochastic parallel gradient descent（SPGD） algorithm can optimize the system performance indexes directly to correct wavefront aberration.A 61-element adaptive optics system model based on SPGD algorithm was simulated.For different initial static aberrations,the algorithm＇s correction capabilities were analyzed.The selections of algorithm gain coefficient and perturbation amplitude were compared in the conditions of adopting different performance indexes,and so was the correction effects.Simulation results demonstrate that the algorithm＇s convergence rate depends on gain coefficient and perturbation amplitude to a great extent.For relatively severe aberrations with perturbation amplitude ranging from 0.50 to 0.85,the correction effects of using mean radius as the system performance index is better than that of using Strehl ratio.

关 键 词：自适应光学 随机并行梯度下降算法 数值仿真 波前畸变 

分 类 号：TP273.2[自动化与计算机技术—检测技术与自动化装置]