Parallel SAR imaging on the KD-50 heterogeneous computer system

With the increased requirements on the precision of image products, the parallel imaging of high resolution SAR data has been receiving much more concerns nowadays. However, many existed parallel algorithms haven't given adequate attentions to both the architecture of a certain parallel computer and the computational features (e.g. matrix transposition) of a SAR imaging algorithm, which couldn't reach higher parallel efficiency. In this paper, strategies for parallel SAR imaging on the KD-50 heterogeneous computer system were discussed, and a CSA based parallel SAR imaging algorithm to harness the features of KD-50 hardware architecture was presented. The algorithm was implemented in a message passing interface (MPI) plus OpenMP specifications based hybrid programming environment, and tested by a block size of 65536 × 32768 simulative SAR data. Both the theoretical analysis and test result show that the algorithm had a good performance, which can cut down the time-consuming from 7080(s) to nearly 536(s).