基于随机并行梯度下降算法的自适应光学实时并行处理机  

Real-Time Parallel Wavefront Processor for Adaptive Optics Based on Stochastic Parallel-Gradient-Descent Algorithm

作　　者：王彩霞[1,2,3] 李梅[1,2] 李新阳[1,2] 叶红卫[1,2] 

机构地区：[1]中国科学院自适应光学重点实验室,四川成都610209 [2]中国科学院光电技术研究所,四川成都610209 [3]中国科学院研究生院,北京100049

出　　处：《光学学报》2010年第11期3076-3081,共6页Acta Optica Sinica

基　　金：国家自然科学基金(60978050);四川省青年科技基金(09ZQ026-014)资助课题

摘　　要：基于随机并行梯度下降(SPGD)算法的自适应光学系统通过直接优化系统的性能评价函数来控制波前校正器以补偿光束中存在的波前畸变。但由于算法收敛速度的影响,在一定程度上限制了SPGD在自适应光学系统中的应用。在对SPGD控制算法分析的基础上,充分提取和发掘算法内在的并发性,采用流水线和并行处理技术,设计并实现了基于现场可编程门阵列(FPGA)加数字信号处理器(DSP)的单指令流多数据流(SIMD)结构实时并行处理机,实现了SPGD控制算法由表达层到结构层的优化映射。该处理机应用在激光光束净化自适应光学系统中,同时实现了对变形镜和倾斜镜的控制。实验结果表明,采用基于SPGD算法自适应光学实时并行波前处理机具有很快的收敛速度,可以有效地校正激光出光过程中的光束波前相差和光束漂移误差。The adaptive optical system based on stochastic parallel gradient descent algorithm（SPGD） compensates wavefront aberration by direct optimization of a system performance metric.The convergence rate is the key factor that limits the SPGD algorithm application in practical adaptive optics system.On the basis of analysis of SPGD algorithm in which the concurrency is embedded,a pipeline single instruction and multipe data（SIMD） architecture parallel real-time wavefront processor is deslgned and implementd in which field-programmable gate array（FPGA） and digital signal processing（DSP） are introduced.The optimal mapping of the control algorithm from representative layer to the architecture is realized.The wavefront processor is applied to a laser beam cleanup adaptive optical system to control the deformable mirror and fast steering mirror.The experimental results show that real-time parallel wavefront processor based on SPGD algorithm can compensate the wavefront aberration and tilt excursion of the laser beam involved effectively and that the convergence rate is enhanced enormously.

关 键 词：自适应光学 随机并行梯度下降算法 波前处理机 并行处理 单指令流多数据流结构 

分 类 号：TP273.2[自动化与计算机技术—检测技术与自动化装置]