Jie Evaluation Algorithms For Different Antenna Array With Parallel Processing Research

Astronomical radio observations is an important tool in the study of celestial bodies, but because of radio telescope sampling points Incomplete, so incomplete UV coverage causes the observed image to have many false ingredients, so the need for acquisition of the image data is restored by cleaning algorithm,In order to improve the efficiency of the implementation of both algorithms discussed GPU multithreaded parallel and two clean algorithms implemented in parallel on a GPU, In addition, this paper discusses how to use the GPU for processing huge amounts of data to CSRH.Firstly, using40antenna simulate three formations of T-array、Y-array�'�spiral array. The UV coverage of the antenna array can be obtained by the relation of each antenna in the antenna array. Then we discuss the UV coverage distribution of three formations. The UV coverage of T-array and Y-array is uneven, and calculating redundancy of both formations is larger than spiral array, sampling points will become less. We give dirty beam of the three formations, and analysis dirty beam of the three formations sidelobe characteristics. The antenna of T-array and Y-array distribute in a specific direction, so their dirty beam exist prominent sidelobe in certain directions while the distribution of the spiral array sidelobe is irregular and smaller than the two formations, so it is suitable for image restoration. Also spiral array with a suitable image restore is confirmed by the experiments of the next clean algorithm.Second to discuss the Hogbom algorithm and Clark algorithm, two algorithms focus on image cleaning effect under the UV coverage of different antenna. The experimental results that spiral array is more suitable for the formation of the image restoration. Meanwhile, discussing the impact of image restoration for the relevant parameters of the two cleaning algorithm:hi Hogbom algorithm, the size of y (loop gain), number of loop iterations, and exit condition for size, the depth of cleaning; In Clark algorithm, the number of the main iterative loop and the minor iterative loop, and Clark algorithm limited parameters. Then to use the GPU parallel implement the Hogbom algorithm and Clark algorithm. Since the astronomical image data is relatively large, only dependent on the ability of the CPU to process data is difficult to efficient completion, so in this paper, we implement Hogbom algorithm and Clark algorithm by the use of the GPU multithreaded parallel. We analysis Hogbom algorithm and Clark algorithm execution process, two algorithms are mainly Fourier transform and the operation of the data matrix, according to multithreaded programming ideological basis CUDA, both algorithms suitable for data-parallel way, also the implementation uses the merge mode access matrix data and shared memory storage to improve the speed of data access, thus to further improve the efficiency of the implementation of the algorithm.The last, because CSRH produce large amount of data, it will be difficult to meet CSRH fast and real-time processing of massive data, in order to ensure the real-time of the CSRH massive data processing, and consider CSRH high resolution on time, space and spectrum, so CSRH require reliable, high-speed, parallel data processing algorithm which is implemented on CPU-GPU. According to CSRH characteristics, we expound the advantages of GPU processing data compared to other treatment, as well as the Fourier transform CSRH data processing depends on is implemented to achieve efficient and real-time on the GPU. At the same time, we discuss some of the difficulties about the use of GPU that processes CSRH massive data and algorithm parallel implementation difficult, and how to design efficient for CSRH system characteristics, accurate processing of massive data parallel algorithms.