Gpu-based Hardware Acceleration Method And Its Applications In The Numerical Solution Of Integral Equations

The analysis of the scattering characteristic from electrically-large object has always been attracting general research interest due to the requirement of practical engineering problems. As the electrical size of object increases too much, the limited resources of personal computer becomes the bottleneck of the whole performance. To solve this problem, the parallel computing technology has been introduced, such as the cluster. Recently, a newly-emerged massively parallel processor– the graphics processing units (GPU) have been applied into the field of computational electromagnetics.The parallel method of moment (MoM) is implemented on the GPU. The acceleration ratio has exceeded 100, which is much higher than the conventional parallel systems. In addition, the matrix splitting algorithm is devised to break the video memory limits. The impedance matrix is divided into multiple regions and each region is calculated separately. While one region is calculated, it is read back from the video memory and store in the system memory, where the whole impedance matrix is assembled together. In this way, the program is not limited by the capacity of video memory and thus owns the ability to solve practical engineering problem.In order to further strengthen the ability of solving electrically-large scattering problems, the GPU-based parallel MoM is combined with a newly-emerged phase-extracted basis. In this way, the electrically-large object could be solved accurately with much lesser unknowns. Thus the full-scale military object could be solved efficiently on a single personal computer.Since the memory requirement and computational complexity are both O ( N 2). When the number of unknowns becomes very big, the resources usage will exceed the limit of personal computer. This thesis analyzes the key process of parallel multi-level fast multipole algorithm and implemented it on GPU. All results show good agreement with accurate results, which proves the accuracy of our code.The research work in this thesis has introduced the leading-edge advancement of computer science into the field of computational electromagnetics. The electrically- large complex scattering problem can now be solved fast and accurately on a single personal computer, which provide a highly-efficient method of analysis. In the meanwhile, the work in this thesis also proves the possibility of computational electromagnetics to benefit from the GPU-based hardware speed-up method, which will boost more applications of GPU in this field.