随机并行梯度下降自适应光学系统中算法收敛速度的仿真研究  

Simulation Study of Convergence Rate of Stochastic Parallel Gradient Descent Technique for Adaptive Optics

作　　者：孙穗[1] 梁永辉[1] 王三宏[1] 

机构地区：[1]国防科学技术大学光电科学与工程学院,长沙410073

出　　处：《光电工程》2011年第12期6-12,共7页Opto-Electronic Engineering

摘　　要：在随机并行梯度下降(SPGD)自适应光学(AO)中,SPGD算法通过直接对系统的性能评价函数进行优化从而校正波前像差,有可能实现实时校正,具有很强的应用潜力。SPGD算法的收敛速度与控制单元的数目、性能评价函数的选取、增益系数及扰动幅度的取值等密切相关。通过仿真的方法,研究了控制单元数对SPGDAO系统在静态波前校正中收敛速度的影响,得出了收敛所需迭代步数与控制单元数成线性关系的结论;仿真了SPGDAO系统在不同迭代速率或不同风速下的动态波前校正,为系统控制单元数的选取提供了参考。Adaptive Optics （AO） based on Stochastic Parallel Gradient Descent （SPGD） corrects the wave-front aberration by directly optimizing the performance evaluation function of the system. It may realize real-time correction and has great application potential. The convergence rate of SPGD algorithm is closely related with the number of the control units, the selection of the performance evaluation function, the value of gain coefficient and disturbance amplitude Through the simulationmethod, the influence of the number of control units on the convergence rate was studied when SPGD AO system was used for static wave-front aberration correction. A conclusion was reached that there was a linear relationship between the number of control units and the required number of iterative for convergence. SPGD AO system used for dynamic wave-front aberration correction under different iterative rate and wind speed was simulated, and suggestions for the selection of number of control units were provided.

关 键 词：自适应光学 随机并行梯度下降算法 控制变量 

分 类 号：O43[机械工程—光学工程]