In astronomical data processing, imaging of radio interferometry is one of the critical and key computationally intensive parts, and as the array size of radio interferometry grows up, the data to be processed scales up as well.In order to evaluate the wide field imaging of radio interferometry in detail, discuss the characteristics of the imaging and parallel computation implementation, and analyze the parallel computation effects of wide field imaging algorithm, this paper investigates the algorithms of wide field imaging of radio interferometry, analyzes the origination of the wide field imaging problem, simulates the observation experiment, and investigates the parallel computation of wide field imaging. All the works of this paper are summarized as follows.(1)The basic principles of radio interferometry data processing are studied. The relationship between the visibility received by radio interferometry and intensity of the observed source is analyzed, imaging method of radio interferometry is involved as well, which is also extended to the problem of wide field imaging.(2)The performance of wide field imaging is discussed in detail. The principle of the three-dimensional Fourier transform imaging method is analyzed and its computational complexity in imaging is discussed. The implementation procedures of Faceting are analyzed, and computational costs during imaging are presented too. The main characteristics of W-Projection and W-Stacking algorithms are analyzed. With the configuration of SKA1-low core array, thirty-six observation point sources at different locations are simulated. Imaging experiments are performed for Faceting, W-Projection and W-Stacking. Computational costs of the three imaging algorithms are compared based on experimental data. Point sources fits are performed on the dirty images obtained by the three imaging methods. Additionally the differences between three imaging methods in flux ratio and position offset of the point sources were analyzed.(3)This paper investigates parallel computing implementation of W-Projection and W-Stacking based on RASCIL. The gridding process which consumes most computing resources in imaging are diagnosed. According the basic principles investigation, the performance of gridding kernels which are mainly used in the data processing of radio interferometry at present is analyzed. Parallel computing implementation of W-Projection and W-Stacking based on RASCIL is achieved. Additionally, based on analysis of two imaging algorithms frameworks, Python code of RASCIL are optimized in several ways. Parallel computing on two imaging algorithms are performed by the distributed computing framework DASK, which employ the observation data of VLA-D, then parallel implementation strategies of the two imaging algorithms are obtained under experimental conditions. Results of parallel computing are analyzed and discussed, including execution time statistics, as well as the comparison of the two imaging algorithms under two parallel resource scaling methods, Strong Scaling and Weak Scaling. (4)The non-coplanar baseline effects of MUSER are investigated. The 𝛿𝑤 value distribution of MUSER during daily observations is analyzed. According to 𝛿𝑤 value distribution combined with the corresponding theory in wide field imaging, it is concluded that MUSER is affected by non-coplanar baseline effects during majority of the observations. The effects of non-coplanar baseline effects on MUSER imaging are quantified by simulated observation and imaging.The work in this paper makes a detailed and experimental analysis for the wide field imaging of radio interferometry. The parallel computing results of W-Projection and W-Stacking have a reference value for both data processing and wide field imaging work in SKA.
修改评论