Research On High Performance Algorithms Based On Feature Basis Functions And Parallel Techniques

Fast methods for solving equations of method of moments are active topics of computational electromagnetic society,and characteristic basis function method is a fast and effective direct solver for large scale problems.In this thesis,ultra-band characteristic basis function methods,and its multi-core-CPU-based and GPU-based parallel versions are studied.The main contributions of this thesis are as follows:1.One kind of ultra-band CBFM based on interpolating scheme is proposed.The main profiles of the proposed method are its interpolation of impedance matrices and ultra-band characteristic basis functions.By virtual of interpolations of impedance matrix and UCBF,the global time-consuming and required memory for analyzing planar circuits can be dramatic reduced.2.One multi-core-CPU-based parallel realization of the above method is designed by virtual of OpenMP.In that realization,serial parts of the multi-frequency scanning are adopted in a paralleled manner,which can fast the interpolation-based ultra-band CBFM on workstations and PCs.3.One CUDA-based parallel realization of the interpolation-based UCBF is designed to fast the matrix-reduction operations.In that realization,the two-level parallel scheme is applied to fast the matrix-matrix multiplying operations in the matrix-reduction step,which makes UCBFM can be further improved by the GPU platform.