Optimization of the Perfectly Matched Layer for the FDTD Method

碩士 === 淡江大學 === 電機工程學系 === 91 === In is thesis, the characteristics of several useful sources to excite a FDTD code are investigated thoroughly. The includes hard source, simple transparent source and improved transparent source. The steady-state genetic algorithm ( SSGA ) is applied...

Full description

Bibliographic Details
Main Authors: Chien-Wei Liu, 劉建緯
Other Authors: Ching-Lieh Li
Format: Others
Language:zh-TW
Published: 2003
Online Access:http://ndltd.ncl.edu.tw/handle/65123119442102075161
Description
Summary:碩士 === 淡江大學 === 電機工程學系 === 91 === In is thesis, the characteristics of several useful sources to excite a FDTD code are investigated thoroughly. The includes hard source, simple transparent source and improved transparent source. The steady-state genetic algorithm ( SSGA ) is applied to optimize the conductivity profile of an anisotropic perfectly matched layer ( APML ). We will separate optimization method into two parts, traditional conductivity profile optimization and random conductivity profile optimization. Traditional conductivity profile optimization:The conductivity profile of a perfectly matched layer (PML) absorber is optimized to increase the absorption performance. The key parameters describing a polynomial-type conductivity profile for an anisotropic PML absorber are optimized by employing a steady-state genetic algorithm (SSGA). A Gaussian line source is used to excite a free space and the local error is minimized. The optimum polynomial parameters are firstly reported for PML absorbers of various layers. Numerical results show that PML absorption can be improved at least by 7~15 dB independent of the space domain size and the grid size. Random conductivity profile optimization:The general parameters defining the sublayers of a PML medium are optimized to minimize the global error of a 2D FDTD code excited by a Gaussian pulse source. The optimum parameters are first reported for PML absorbers of various layers. Numerical results show that the global error can be improved at least by 8~10 dB as compared to the usual simple polynomial conductivity profile.