Implementation And Optimization Of GPU-based CLEAN Algorithm

Radio astronomy has become an important research field in astronomy.The world's largest radio interference array square kilometer array(SKA),and "China's eye" the construction of the 500-meter aperture spherical radio telescope(FAST)etc,scientists from all over the world have been aware of that huge astronomical to astronomical data processing and computer field has brought both new opportunities and challenges.In recent years all the world saw that China's efforts in the field of radio astronomy research and equipment development of visible,China centimeter-decimeter frequdency spectrum Radioheliograph,like MUSER is completed,interference array imaging become a key research field of data processing.For imaging in MUSER based on GPU,real-time requirements.In this paper,we use the CLEAN algorithm to process the image by observation to get the scientific observation images,to eliminate the effects of sparse samping in the UV plane to get the CLEAN image,then the astronomers get the sun's activity rule to reduce the influence of the solar activity on human,our works is as follows:First,In view of the interference matrix fast imaging,because of the effects of sparse samping in the UV plane lead to artifacts,we use CLEAN algorithm to process the dirty image,by studying the CUDA parallel architecture characteristic and advantage of the characteristics of the GPU multiprocessor,discussed the concrete application mode and method of data processing in MUSER.Second,By analyzing the execution process of the Hogbom CLEAN,multi-scale CLEAN and Clark CLEAN algorithm,the CUDA kernel function running on the device(GPU)is designed for the parallel parts of the algorithm.Using the Python and PyCUDA extensions to implement the CLEAN processing function,based on the GPU and CUDA implement the Hogbom CLEAN,multi-scale CLEAN and Clark CLEAN algorithm,which improves the efficiency of the MUSER to the deconvolution algorithm.At the same time,the selection of the convolved kernel is discussed based on the GPU multi-scale CLEAN algorithm.Third,We analysis and compares the deconvolution operation efficiency and advantages of the CLEAN algorithm of MUSER.Then,we summarizes the applicable envieonment of CLEAN algorithm in MUSER.In this paper we had done some related research,the interference of array imaging,for MUSER on the image quality requirements,the traditional back convolution algorithm optimization research,the study of radio astronomy subject has certain reference value.