光学学报, 2015, 35 (11): 1111003, 网络出版: 2015-11-03
基于混合优化算法的分块镜共相位误差校正
Co-Phasing Error Correction for Segmented Primary Mirror Based on Hybrid Optimization Algorithm
成像系统 自适应光学 分块镜 共相位误差 混合优化算法 随机并行梯度下降 遗传算法 imaging systems adaptive optics segmented mirror co-phasing error hybrid optimization algorithm stochastic parallel gradient descent genetic algorithm
摘要
对采用分块拼接式主镜的大口径空间望远镜,有必要研究适合在轨应用的共相位误差校正方法。以像清晰度函数作为评价函数,通过仿真分析了单色光和白光照明2种情况下分块镜共相位误差与评价函数的变化关系。仿真结果表明:采用单色光照明时,分块镜平移误差校正的动态范围小于0.5 λ;采用白光照明时,评价函数存在局部极值,必须采用全局最优化算法。仿真结果表明,当平移误差大于0.5 λ时,采用随机并行梯度下降(SPGD)算法易陷入局部极值,而采用遗传算法(GA)可以搜索到全局最优解,但迭代速度较慢。提出了采用基于SPGD算法和遗传算法的混合优化算法校正分块镜共相位误差。当平移误差在[-λ, +λ]范围内时,混合优化算法的校正精度可优于10 nm。
Abstract
For large aperture space telescope with segmented primary mirror, it is necessary to study in-orbit cophasing error correction method. The image sharpness function is taken as metric function. The relationships between co- phasing errors of segmented mirror and the metric function are analyzed by simulation under monochromatic illumination or white- light illumination. Simulation results show that under monochromatic illumination, the dynamic range of piston error correction is less than 0.5 λ. Under white-light illumination, global optimization algorithm should be adopted because of the local maximums of metric function. When piston error is greater than 0.5 λ, stochastic parallel gradient descent (SPGD) algorithm is prone to falling into local maximum and genetic algorithm is capable of finding out the global optimum but with relatively slow iteration speed. A hybrid optimization algorithm based on SPGD algorithm and genetic algorithm is proposed for co-phasing error correction of segmented mirror. When piston error is between - λto λ, the correction accuracy of the hybrid optimization algorithm can be better than 10 nm.
孙雪, 董冰. 基于混合优化算法的分块镜共相位误差校正[J]. 光学学报, 2015, 35(11): 1111003. Sun Xue, Dong Bing. Co-Phasing Error Correction for Segmented Primary Mirror Based on Hybrid Optimization Algorithm[J]. Acta Optica Sinica, 2015, 35(11): 1111003.