基于GPU的自适应光学控制系统研究

Research on Real-timeadaptive Optics Control System Based on GPU

为了提高自适应光学系统闭环校正的工作频率,满足天文望远镜系统对频率的要求,解决通常基于CPU的控制系统实时性差的问题,提出了一种基于GPU的自适应光学快速控制方法,即CUDA架构下的波前斜率计算、控制电压计算及波前重构计算方法。仿真表明,对于1172单元变形镜,系统计算延时可以满足100Hz自适应光学系统要求。对GPU控制方法进行实验验证,使用梯度下降算法,搭建了一套61单元变形镜的自适应光学系统。对于1024*1024像素图像(29*29子孔径数),61单元变形镜的自适应光学系统,整个系统延时小于5ms,系统的计算能力提升10倍。

This paper proposes wavefront slope calculation method,control voltage calculation and wavefront reconstruction method based on CUDA architecture.The simulation shows that for adaptive system with 1172-element deformation mirror,computing time delay can meet the requirement of 100Hz adaptive optics system.This paper sets up an adaptive optical system with 61-element deformation mirror,using the principle of gradient descent as control method,performing real-time wavefront aberration correction experiment with GPU.Experimental results demonstrate that for the adaptive optical system with 1024*1024 pixel image(29*29 subaperture)and 61-element deformation mirror,the calculation delay of the whole system can be controlled within 5ms,with a 10-fold increase in computing power.