中国光学, 2016, 9 (4): 432, 网络出版: 2016-07-26
随机并行梯度下降算法性能与变形镜排布规律的关系研究
Relationship between performance of stochastic parallel gradient descent algorithm and distribution rule of deformable mirror
波前整形 随机并行梯度下降算法 变形镜 wave-front shaping stochastic parallel gradient descent algorithm deformable mirror
摘要
对随机并行梯度下降算法(SPGD)性能与不同变形镜排布规律的关系进行了研究。以采用Roddier方法生成的由52项Zernike像差构成的畸变波前为整形对象, 对SPGD算法的收敛速率和整形效果与变形镜排布规律(单元数分别为19、21、32、37、45、60、61、77、91)之间的关系进行了仿真研究。结果表明: 从整体分析, 随着变形镜单元数逐渐增多, SPGD算法的收敛速率和整形效果均逐渐变差; 从局部分析, 由于变形镜元胞类型变化和边缘占空比的影响, 在渐变规律中产生了局部差异。
Abstract
The relationship between the performance of stochastic parallel gradient descent(SPGD) algorithm and distribution rule of deformable mirror is studied in this paper. The study object is distortion wave-front including 52 order Zernike aberrations created with Roddier method. The simulation research mainly focuses on relationship between convergence rate, shaping effect of SPGD algorithm and distribution rule of deformable mirrors with 19, 21, 32, 37, 45, 60, 61, 77, and 91 units. The results show that from the general analysis, the convergence and shaping effects of SPGD algorithm gradually become worse with actuator number of deformable mirror increases, but from the partial analysis, this trend create local difference due to different cell type and void ratio in the edge of deformable mirror.
陈惠颖, 王卫兵, 王挺峰, 郭劲. 随机并行梯度下降算法性能与变形镜排布规律的关系研究[J]. 中国光学, 2016, 9(4): 432. CHEN Hui-ying, WANG Wei-bing, WANG Ting-feng, GUO Jin. Relationship between performance of stochastic parallel gradient descent algorithm and distribution rule of deformable mirror[J]. Chinese Optics, 2016, 9(4): 432.