Research On Parallel Implementation Of Characteristic Basis Function Method

It is very important and practical to analyze the electromagnetic scattering characteristics of objects quickly and efficiently. However, as the electrical dimension of the target increases, the memory requirement and CPU time are barriers to the solution by a single PC. To overcome this problem, parallel computing was employed in computational electromagnetics.Characteristic basis function method (CBFM) is an effectively approach for the solution of electromagnetic scattering problems. In this method, the dimension of coefficient matrix can be controlled by selecting the size of subdomain, and the matrix size can be reduced dramatically. Because it is based on dividing the large structure into small blocks, it is very suitable to be used in parallel computing.In this dissertation, two parallel methods based on static load balance have been presented by combining MPI with Adaptively Modified Characteristic Basis Function Method (AMCBFM). The first one is based on parallel MoM, which uses parallel methods to generate MoM matrix and solve matrix equation of every block. The size of matrix is very small in CBFM, so that the lowest efficiency part of parallel MoM can be avoided in this method, and we can get higher efficiency. The second one sends the serial number of blocks to related nodes according to a certain rule. Every node only needs to calculate the local blocks. The calculation of every level of CBF only needs to communicate once. The whole task can be divided equally. We have tested this two parallel methods on PC clusters. The obtained results confirm that the accuracy of the proposed two methods is equal to AMCBFM, and efficiency of the second one is better.In order to compute in heterogenous parallel computing environment, a new kind of parallel method has been put forward by combining hybrid scheduling arithmetic with AMCBFM, which distribute the task dynamically depend on the computational performance of the local node. This method has perfect performance in heterogenous parallel computing environment.All the numerical results listed in this dissertation have demonstrated the validity and high efficiency of our research works. Parallel AMCBFM is an effectively approach for the efficient solution of electromagnetic scattering of electrically large objects.