JASMIN frame structure based on parallel multilevel fast multipole algorithm and its application
|School||University of Electronic Science and Technology|
|Course||Electromagnetic Field and Microwave Technology|
|Keywords||Multilevel fast multipole method Large target Pre- conditions Parallelization JASMIN framework|
Three-dimensional object rapid analysis of electromagnetic problems has been a hot research in computational electromagnetics. This paper focuses on multilevel fast multipole algorithm (MLFMA) carried out a series of optimization studies and parallel processing, and analysis using the optimized parallel algorithm for solving complex three-dimensional electromagnetic scattering and radiation TVU problem. This paper describes the electromagnetic scattering problems for metal surface analysis method for solving integral equation method and its - the method of moments, and elaborates on the method of moments of fast solution methods - fast multipole method and the multilevel fast multipole method of the basic principles and steps. Secondly, the paper studies the multilevel fast multipole method optimization strategy. Multipole mode by modifying the number, the direction of reducing the number of the angular spectrum, thus ensuring the accuracy of the case, reducing the memory storage and computation time. Aggregation configuration using the angular spectrum symmetry factor, making the electric field integral equation (EFIE) aggregation configuration memory storage capacity reduced by a factor of a quarter, while the mixed field integral equation (CFIE) is half of the original. This paper also made analysis of interpolation techniques to achieve a Gaussian interpolation and Lagrange interpolation technology optimization, and both before and after the interpolation interpolation techniques to improve the analysis done on the target radar error analysis. Again, this study used multilevel fast multipole methods preconditioning techniques. In this paper, the generalized minimal residual (GMRES) iterative method to add a block diagonal (BD) and sparse approximate inverse (SAI) preconditioner to accelerate the convergence of iterative integral equation and compare these two pre-conditions for consideration of specific construction techniques and acceleration effect. Then, we study parallel multilevel fast multipole algorithm for electromagnetic analysis of large complex target. First, based on the realization of shared memory parallel OpenMP standard multilevel fast multipole, and analyze its parallel efficiency. Second, will belong to the unstructured grid multilevel fast multipole algorithm transplanted into the structure of the grid JASMIN framework that leverages the framework of high portability and scalability, load balancing, good benefits, to achieve a parallel distributed memory architecture, and used for analysis of electromagnetic characteristics, while doing a parallel efficiency analysis. Parallel programs using C / Fortran mixed programming, data structures stored in double-precision, high-performance computing. Finally, the framework will be based on JASMIN parallel MLFMA for electromagnetic scattering and radiation analysis of the problem, and the numerical results with the analytical solution, commercial software simulation results, and based on the results OpenMP parallel MLFMA compares results demonstrate the reliability of parallel algorithms sex.