Enhanced method of moments performance through efficient implementation and error estimation

Date
2023-03
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: The emergence of increasingly ambitious engineering projects has pushed the limits of traditional computational electromagnetics (CEM) simulation software. This has necessitated the development and implementation of new methods that can effectively utilise the available hardware resources of modern computing systems. Due to the increasing availability of large cluster computing systems, research into increasing the capabilities of electromagnetic (EM) field solvers in terms of performance and scale of the problems they can handle has mainly focused on distributed memory parallelisation schemes. The single-core performance of modern processors has however also continued to improve. New processors are also continually being released with increasingly large core counts. Thus designing implementations of solvers that can effectively utilise the available hardware on shared memory systems is crucial for application to both small and very large problems. To this end, this document presents the development and implementation of a numerical field solver that is designed to push the limits of single central processing unit (CPU) performance for the Method of Moments (MoM). This solver forms the core of a software library, the electromagnetic kernel library (EMK), that aims to provide utilities for performing numerical field simulations as well as for developing methods for large-scale antenna analysis. This solver is compared to the solver in a commercial CEM software suite, to assess its performance and accuracy. As well as efficiently computing numerical solutions, it is also important to verify the validity of such solutions and improve them by locating and eliminating sources of error. This is done through a process of error estimation, which seeks to measure the extent to which a numerical solution deviates from the true solution. To this end, a goal-oriented a posteriori error estimator is formulated for MoM-based EM analysis. Goal-oriented error estimators allow for the solution error to be determined with regard to a particular quantity of interest (QOI). This estimator is used in the computation of multi-port Zparameters by driving a simple mesh refinement algorithm, to assess its effectiveness in generating efficient meshes in terms of degrees of freedom
AFRIKAANS OPSOMMING: Toenemend ambisieuse en komplekse ingenieursprojekte bied nuwe uitdagings aan tradisionele, numeriese elektromagnetika (NEM) simulasie sagteware. Dit noodsaak die ontwerp en implementering van nuwe metodes om die beskikbare hardeware hulpbronne in moderne rekenaar stelsels optimaal te benut. Weens die toenemende beskikbaarheid van groot, komplekse groep-rekenaar stelsels, het navorsing om die vermoëns en effektiwiteit van NEM veldoplossers in die parallelle-verwerking konteks te verbeter, tot dusver meestal gefokus op verspreide-geheue stelsels. Die vermoëns van enkel-verwerker prosesseerders het egter ook intussen verbeter. Nuwe prosesseerders met toenemende aantalle verwerkingskerne word ook deurgaans bekendgestel. Die ontwerp en implementering van veldoplossers om doeltreffend gebruik te maak van al die beskikbare hardeware hulpbronne in gedeelde-geheue stelsels, is noodsaaklik vir die oplossing van beide klein en groot probleme. Hierdie verhandeling dokumenteer dus die ontwikkeling en implementering van ’n NEM veldoplosser wat ontwerp is om ’n enkele, multi-kern sentrale verwerkingseenheid maksimaal te benut vir moment-metode (MM) berekeninge. Hierdie veldoplosser vorm die basis van ‘n programmatuur-biblioteek, genaamd die elektromagnetiese kern biblioteek (EMK), met die doelwit om doeltreffende gereedskap te verskaf vir NEM simulasies in die algemeen, sowel as spesifiek vir die ontwikkeling van pasgemaakte metodes vir doeltreffende, grootskaalse antenna analise. Die veldoplosser se akkuraatheid en werkverrigting word geassesseer deur dit te vergelyk met kommersiële sagteware. Sowel as om numeriese oplossings doeltreffend te bereken, is dit ook belangrik om die geldigheid van die oplossings te toets en te verbeter deur bronne van foute op te spoor en uit die weg te ruim. Dit word gedoen deur middel van ’n foutberamingsproses, wat bepaal tot watter mate ’n gegewe numeriese oplossing afwyk van die ware oplossing. ’n Doelgerigte a posteriori foutberamer word dus geformuleer vir MM-gebaseerde elektromagnetiese analise. Die doelgerigte foutberamer is daartoe in staat om oplossingsfoute te bepaal, met betrekking tot ’n spesifieke kwantiteit van belang. Hierdie foutberamer word aangewend saam met ’n eenvoudige maas-verfyningsalgoritme, vir berekening van multipoort Z-parameters. Hierdeur word die vermoë van die foutberamer geassesseer, om mase op te stel wat doeltreffend is in terme van aantal vryheidsgrade teenoor akkuraatheid.
Description
Thesis (PhD)--Stellenbosch University, 2023.
Keywords
Citation