Study Of Real-time Wavefront Processing Algorithm For Adaptive Optics System Based On GPU

The adaptive optics is a combination of optics,electronics and automatic control theory and other disciplines as a new technology.The adaptive optics system has the ability in the correction of deformable mirror's surface in the optics system and the realtime correction of dynamic wavefront aberration and the ability in improving the imaging quality of the optics system.A typical adaptive optics system usually consists of three parts.The real-time wavefront processor is the core of the adaptive optics system,and the time delay caused by the related data transmission and operation is very important for the real-time performance of the adaptive optics system.Based on the hardware platform and the algorithm existing in the laboratory,this paper has selected the hardware and improved the algorithm for the real-time wavefront processor,and has successfully applied it in the 349-unit adaptive optics system,which plays a valuable role in the realization of the large aperture ground-based high resolution imaging telescope meeting the requirements for the adaptive optics system in terms of processing scale,operation delay,image quality and other aspects.First of all,this paper introduces the basic concept of adaptive optics system and the origin and development of adaptive optics technology,and the three parts of the closed-loop adaptive optics,and emphatically introduces the operation form and data flow of real-time wavefront processor.Then the performance index of adaptive optics system is introduced.Finally,the time delay of real-time wavefront processing is analyzed.It is pointed out that based on the parallel pipeline processing of every frame,the restoration part of algorithm is processed in parallel by using the decomposability of matrix multiplication.Secondly,according to the advantages of all kinds of hardware,this paper determines the real-time wavefront processor with FPGA+GPU+CPU as the main hardware,including FPGA wavefront slope processor,GPU matrix multiplication processor and the main control computer.In addition,the hardware structure of GPU,the heterogeneous processing of CPU+GPU,and the execution model of CUDA are introduced in detail.Finally,the performance indicators of high-performance computing independent graphics card NVIDIA Tesla K80 has been introduced,and the advantage of GPU as matrix multiplication processor is pointed.Thirdly,this paper introduces algorithm processing for the real-time wavefront processor in adaptive optics system.According to the working principle of SH wavefront sensor,wavefront slope algorithm has been determined.According to the requirements of the control system,the direct wavefront gradient method has been chosen as the restoration algorithm.According to control model,the ProportionIntegration control algorithm has been designed.Finally,the real-time wavefront processor designed in this paper is applied in the laboratory 349-unit deformable mirror adaptive optics system,and the experimental results are analyzed in detail.The experimental results show that the 349-unit adaptive optical real-time wavefront processor can effectively correct the atmospheric turbulence frequency components which is below 100 Hz.With NVIDIA Tesla K80 being used as the GPU matrix multiplication processor to measure the real-time wavefront processing,the time delay is less than 235?s.The experimental results successfully demonstrate the feasibility of the GPU-based real-time wavefront processor designed in this paper.