研究太阳磁场活动产生的机理，获得整个太阳活动区的大视场高分辨力图像对太阳物理学家而言至关重要。传统自适应光学（CAO）只能在较小的视场内改善图像质量，无法满足大视场高分辨力天文观测的需求。多层共轭自适应光学（MCAO）是近年来自适应光学领域的研究热点之一，其基本原理是将湍流的大气分成若干层，然后对每一层的波前畸变进行分层测量和校正，消除大气湍流对成像系统的影响，从而实现大视场范围内的高分辨力成像。

实时控制器是太阳MCAO系统的运算核心。相对于CAO系统而言，其哈特曼波前传感器的子孔径数以及变形镜的单元数更多；相对于夜天文MCAO系统而言，其使用的波前斜率探测算法更加复杂。因此其最大的难点在于大计算量的前提下，实时性的保证。为了动态补偿大气湍流的变化，帧频通常要求高达数百帧甚至几千帧每秒，而且要求实时控制器必须在一帧内完成相应的计算，因而实时性要求非常高。本文主要针对太阳MCAO系统实时控制技术开展了研究，并且基于FPGA+多核DSP的定制化硬件平台开展了相应算法的并行化实现。

首先，论文简要介绍了大气的时间和空间统计特性，太阳MCAO的基本理论以及国内外发展现状，对太阳MCAO系统的实时控制器的设计提出了更高的要求。在此基础上，根据MCAO实时处理算法的特点，进一步分析了计算量和实时性的需求。

其次，根据计算量和实时性的需求分析，比较了多种并行计算平台的优缺点。通用的多核CPU的优点是编程灵活，资源丰富，然而由于系统调度或者中断响应，会引起时间抖动，影响系统性能；多DSP阵列虽然具有很好的实时性，但是对于哈特曼波前传感器中大量的子孔径和子区域图像，需要使用很多片DSP，这通常会引起非常复杂的外围电路设计。FPGA使用硬件描述语言来定义硬件，可以灵活的并行化处理大量的子孔径和子区域图像，与此同时，多核DSP十分擅长浮点数运算，尤其是大矩阵的并行优化。最终提出了基于FPGA+多核DSP的异构计算平台，并且完成了相应算法的映射。

然后，针对波前探测，在FPGA中采用了多种并行加速相结合的方式实现了750倍的硬件加速；针对波前复原和波前控制，在多核DSP内完成了相应代码的并行优化，实现了8倍的加速。整体计算延时得到大大的降低，可以满足实时性需求。

最后，针对实际的两层共轭自适应光学系统，使用了1个37子孔径的多视线相关哈特曼波前传感器（每个子孔径中选择5个子区域），2块变形镜（地表层151单元和高层37单元）分别共轭在0 km和2-5 km。在室内针对5路激光点源以及在室外针对太阳均开展了相应的实验，实时控制器成功地得到了应用，并获得了室内MCAO的点源校正图像以及太阳地表层自适应光学（GLAO）的校正图像。实时控制器在800 Hz的相机采样频率下获得了116.2 μs的计算延时，系统的0 dB误差带宽达到30 Hz。

该套实时控制器是首套基于FPGA+多核DSP架构的太阳MCAO实时控制器，能够运行在CAO，GLAO，MCAO三种工作模式下，自由切换，并且具有较好的扩展性和兼容性，升级也十分方便。本课题的研究对于太阳MCAO系统的实现具有重要的研究价值和实际工程指导意义。

It is of vital importance for solar physicists to obtain the large field of view (FOV) and high-resolution images of the whole solar activity area and to study the mechanism of solar magnetic field activity. Conventional adaptive optics (CAO) can only improve the image quality in a small FOV, and cannot meet the needs of large FOV and high-resolution astronomical observation. Multi-conjugate adaptive optics (MCAO) is one of the research focus in recent years from adaptive optical field. Its basic principle is to divide the atmospheric turbulence into several layers, and then measure and correct the wavefront distortion of each layer, eliminate the influence of atmospheric turbulence on the imaging system, and finally achieve the large FOV and high-resolution images.

The real-time controller is the computing core of the solar MCAO system. Compared with the CAO system, the number of subapertures of the Shack-Hartmann wavefront sensor and the elements of the deformable mirror in the solar MCAO system are even more. Compared with the night astronomical MCAO system, the wavefront slope detection algorithm is more complicated. Therefore, the biggest difficulty lies in the guarantee of the real-time under the large amount of computing.

In order to dynamically compensate for the change of atmospheric turbulence, the frame rate is usually up to hundreds or even thousands of frames per second, and the corresponding calculation must be completed within a frame, so the real-time requirements are very high. In this paper, the real-time control technology of the solar MCAO system is studied, and the parallel algorithm is implemented based on the hardware platform of FPGA+ multi-core DSP.

First of all, the paper briefly introduces the temporal and spatial statistical characteristics of the atmosphere, the basic theory of the solar MCAO and the development of the domestic and international, and puts forward higher requirements for the design of the real-time controller of the solar MCAO system. On this basis, according to the characteristics of MCAO real time processing algorithm, the demand of computation and real-time is further analyzed.

Secondly, according to the analysis of the requirement of computation and real-time, we compare the advantages and disadvantages of several parallel computing platforms. General multi-core CPU has the advantages of flexible programming and rich in resources. However, due to the system scheduling or interrupt response, it will cause the time jitter and affect the performance of the system. DSP arrays have good real-time performance, but often need large number of DSPs for the Shack-Hartmann wavefront sensor with a large number of subaperture and subfield images, that will usually cause the complicated peripheral circuit design. FPGA often use the hardware description language to define the hardware, and can flexibly process a large number of subaperture and subfield images. At the same time, multi-core DSP is fit for floating point operations, especially the parallel optimization of large matrices. Finally, a heterogeneous platform based on FPGA and multi-core DSP is proposed, and the mapping of the corresponding algorithm is completed.

Then, according to the wavefront detection, a variety of parallel combinations are used to accelerate the computations in the FPGA to achieve 750 times hardware acceleration; for the wavefront reconstruction and wavefront control, the corresponding code are completed with the parallel optimization in the multi-core DSP and to achieve a speed up of 8 times. The overall computation delay was greatly reduced and can meet the real time requirement.

Finally, according to the two-conjugate adaptive optics system, a MD-WFS which contains 37 subapertures (5 subfields in each subaperture) is used to wavefront detection and two DMs (151 actuators and 37 actuators) are conjugated to 0 km and 2-5 km respectively. We developed an experiment in door with 5 laser point sources and outside the door with the Sun. The real-time controller was successfully applied and the images of the indoor point source correction MCAO and the solar ground layer adaptive optics (GLAO) were obtained. The real-time controller achieves a calculated delay of 116.2 μs at the sampling frequency of 800 Hz, and the 0dB error bandwidth of the system is up to 30 Hz.

The real-time controller is the first set of solar MCAO real-time controller based on FPGA and multi-core DSP, can run in three work modes including CAO, GLAO, and MCAO, free switching, and it has better scalability and compatibility, upgrade is also very convenient. The research of this subject has very important research value and practical engineering significance for the realization of solar MCAO system.