Research On Wide-field Imaging Parallel Algorithm Based On Low-frequency Radio Telescope Array

The Square Kilometre Array(SKA)is the world's largest integrated aperture radio telescope array under construction in the international astronomy community.Its receiving area reaches one square kilometer,and it provides a great opportunity for humans to understand the universe.But in the first phase of SKA,the observed astronomical data has reached 10 Pbit per second.It is clear that such a huge amount of data is a severe test for astronomers.Due to the high image quality of large low-frequency interference arrays(such as SKA1-LOW),the non-coplanar baseline effects or the influence of the w-term must be considered.We called it as the wide-field imaging or the large-field imaging.Astronomers have proposed several wide-field imaging methods: three-dimensional(3D)Fourier transform imaging,faceting,w-projection,w-stacking,and warped snapshots etc.For these methods,although the distortion of the image can be corrected and the detailed information of the edge of the image can be clearly observed,the complexity of algorithm increases the imaging time in the imaging processing.For example,the faceting algorithm performs projection projection processing on a non-coplanar image.When more segmented cutting planes are used,more imaging running time is required.And the w-projection algorithm processes the convolution kernel function so that the three-dimensional data is mapped to the two-dimensional plane,but this algorithm requires a large amount of time to calculate and store the convolution kernel function.All of these algorithms are 100 to 1000 times slower than the simple two-dimensional Fourier imaging algorithm,so it is necessary to accelerate the optimization of the algorithm in order to cope with the imaging of massive astronomical data in the future.The main contents and results of this paper are:1?In the paper,we study the acceleration of the uv-faceting algorithm.Firstly,analyze the principle of the faceting algorithm in detail,than propose two parallel acceleration methods,namely a uv-faceting imaging method based on MPI+Open MP and a uv-facteing imaging method based on MPI+CUDA.At the same time,we optimize the two time-consuming steps,data reading and gridding,in the uv-faceting parallel algorithm.We use real observation data for image verification experiments,single-node experiments and multi-node experiments,and get final image.As a result,the scalability of the two parallel optimization methods is verified,and they can reduce 13-14 times than un-parallel method.2?At the same time,the acceleration of w-projection algorithm is studied.By analyzing the implementation principle of w-projection,we propose two parallel acceleration methods,namely a w-projection parallel method based on MPI + Open MP and a w-projection parallel method based on MPI + CUDA.They are suitable for fast imaging of large-scale low-frequency interference arrays of massive data.By comparing these two methods with the original imaging methods,it shows that the w-projection parallel methods based on MPI + Open MP and MPI + CUDA can obviously improve the imaging efficiency and reduce 13-19 times imaging time than un-paralle method,increase the resolution of the image,and have good scalability and robustness.3?Finally,we study a hybrid algorithm in this paper,which is based on faceting algorithm and w-projection algorithm,called hybrid "w-facets" algorithm.It provides a new technical supporting for wide-field imaging methods.We analyze the proposed background and research significance of the hybrid w-facets algorithm,derive the specific algorithm implementation formula,and use actual observation data for image verification experiments and performance test analysis.They show that the imaging quality of hybrid w-facets algorithm is better than single faceting algorithm or w-projection algorithm,and it has the advantages of good robustness and small dynamic range near bright sources,but it will inevitably increase the loss of imaging time.