Dissertation > Excellent graduate degree dissertation topics show
Electromagnetic scattering integral equation method GPUbased parallel numerical solution
Author: XuJunYing
Tutor: NieZaiPing
School: University of Electronic Science and Technology
Course: Electromagnetic Field and Microwave Technology
Keywords: Graphics processing unit CUDA programming model Parallel method of moments Parallel multilevel fast multipole method
CLC: O441.4
Type: Master's thesis
Year: 2010
Downloads: 118
Quote: 1
Read: Download Dissertation
Abstract
The large target electromagnetic scattering analysis is an important direction in computational electromagnetics , its main feature is the large number of unknowns , calculated for a long time , hardware resource requirements . Order to quickly solve the target scattering characteristics , usually in largescale computer cluster system , highperformance multicore CPU server platform application of MPI , OpenMP programming techniques to achieve a parallel fast solution . Recent years , parallel computing technology on a single platform  based on the advent of the graphics processing unit GPU CUDA programming model promotion was successful application , the paper based GPU CUDA programming model is implemented in a standalone platform for electromagnetic scattering integral equation method for parallel computing . This inheritance , the specific content to achieve parallel processing on the GPU platform based on the method of moments of the integral equation method and the multilevel fast multipole method . Method of moments has completed about 140 times speedup on GPU platform multilevel fast multipole method on GPU platform about 7 times speedup . The downside is that the current platform on the method of moments can calculate the number of unknowns is only less than one million , multilevel fast multipole method speedup is very low . In this context , read carefully and fully understand the program , test each part of the code running efficiency improvement program for the deficiencies : the method of moments , from the original impedance matrix , all the data stored in memory instead stored in the host memory , then the data needed in the calculation batches sequentially read into the memory , small workstation the computable unknown significantly improved , but speedup declined slightly ; for multilevel fast multipole method , in a multipolar polymerization step , from the original individual threads required data is read from the memory changed to read the required data from the shared memory , reducing the thread directly to the number of times data is read from the memory , and improve the speed of the thread reads data at the expense of a certain percentage of redundant data in shared memory , in a multipolar configuration steps , from the angular spectrum of point serial expand instead angular spectrum point packet so that the angular spectrum of points within each group, expand the spectrum does not overlap , and the angular spectrum point of each set of parallel expand on the current platform , optimized , multilevel fast multipole algorithm from 7 to double the speed ratio is now 14 to double the speed ratio .

Related Dissertations
 GPUaccelerated simulation of highspeed particle collisions,TP332
 GPUBased Parallel Artificial Fish Swarm Algorithm and Its Applications,TP391.41
 Realtime Rendering Based on GPU for Largescale Virtual Environment,TP391.41
 GPUbased parallel speedup of OpenFOAM,O35
 Motion Analysis of Mitral Valve in Children Based on Three Dimensional Echocardiography,R445.1
 GPGPUbased optimization SILK Speech Codec,TN912.3
 GPU based raycasting all terrain rendering algorithms to achieve,TP391.41
 Design of Embedded Graphics Processing Unit Based on ARM and FPGA,TP391.41
 GPU Based Motion Path Matching and Crowd Simulation,TP391.9
 Investigation on the Key Technology of Virtual Environment for Distributed Simulation System,TP391.9
 Design and Implementation of Realistic Graphics Processing System for Mobile Devices,TP391.41
 Research on the Technology of the Threedimensional Surface Reconstruction from Scattered Point Cloud,TP391.41
 GPUBased Accelerated FiniteDifference TimeDomain for Electromagnetic Radiation and Scattering Simulations,O441.4
 Imaging Algorithm of Synthetic Aperture Radar Based on Graphics Processing Unit,TN957.52
 Study on Accurate Segmentation Algorithms 3d Medical Images,TP391.41
 Research on Several Problems of Elliptic Curves Related to Cryptography,TN918.1
 Study of Power System Operating State Visualization,TP391.41
 Design of Graphics Processing Unit for Embedded Systems,TP368.1
 The Research and Implementation on Large Data Sets Volume Rendering Based on GPU,TP391.41
 Realtime Fluid Simulation Based on Spectral Methods,TP391.9
CLC: > Mathematical sciences and chemical > Physics > Electromagnetics,electrodynamics > Electricity and Magnetism > Electromagnetic waves and electromagnetic fields
© 2012 www.DissertationTopic.Net Mobile
