Maciej Sypniewski

Linear and superlinear speedup in parallel FDTD processing

The nature of the FDTD method is that simulation of big and complicated passive microwave circuits and large antenna models requires a vast amount of computer operational memory and run-time. The relevance of fast parallel processing in the FDTD method proves to be significant and in most cases is performed on distributed memory (DM) machines using MPI library. We will focus on two main aspects: speedup efficiency optimization to achieve linear scaling with a number of CPUs (N), and importance of having proper parallel solution for both shared memory (SM) and distributed memory (DM) systems. Under some specific, but realistic circumstances, it is possible to achieve superlinear speedup - with efficiency far over 100%.

Parallel FDTD Calculation Efficiency on Computers with Shared Memory Architecture

Fast parallel processing of the FDTD method is becoming very important, and its efficiency on different multiprocessor computer architectures should be carefully examined. Is it possible to have the same versatile parallel FDTD simulator working efficiently on a set of very different computer architectures? In contrast to distributed memory systems, shared memory systems can be built with a variety of memory configuration types (and subtypes) as well as combined with distributed memory systems. The paper focuses on viability of different shared memory architectures for parallel FDTD processing. Our analysis concerns the processing method suitable for both shared and distributed memory systems. The results of experiments comparing parallel processing efficiency on different shared memory machines are presented.

New Applications of the FDTD Method with Enthalpy-Dependent Media Parameters

An original extension of the FDTD method to enthalpy-dependent media parameters is proposed. During one FDTD simulation, media parameters are repetitively modified in accordance with the extracted heating patterns. Generation of spurious modes is suppressed by converting E-, H- to D-, B- fields before media modification. The method correctly predicts run-away heating, diverging defrosting, and a variety of other effects, which until now could only be intuitively expected or experimentally observed. This opens new applications for the FDTD method in microwave power sectors including food industry, microwave oven design, and the newly emerging but highly competitive area of microwave chemistry and pharmacy.

Investigation of multithread FDTD schemes for faster analysis on multiprocessor PCs

Although the finite difference time domain (FDTD) method is a robust and efficient technique in the class of 3D electromagnetic solvers, many practical problems of microwave and antenna engineering still remain beyond reach because of long CPU times required. One idea to accelerate FDTD is though parallel implementations for multiprocessor computers. The target of our work is to develop FDTD schemes for such systems, making use of multithread programming techniques. We develop and investigate multithread FDTD schemes for PCs, Windows NT, based on the original concept of field vector decomposition.

Space-Selective Extraction of Q-Factors from FDTD Simulations

A direct technique of Q-factor extraction based on space-selective integration of dissipated power and electromagnetic energy is proposed, implemented in the FDTD software, and validated against analytical solutions and Prony method. An advantage of this technique over Prony is that losses caused by various materials and / or in various parts of the circuit can be independently evaluated during one FDTD simulation. Its disadvantage is the necessity of a separate simulation for each eigenmode.

Automatic design of high frequency structures using a 3-D FD TD simulator in an optimisation loop

A system for automatic design of microwave structures based on an electromagnetic FD-TD simulator working in an optimisation loop is presented. The paper describes the main elements of the system, which are: the FDTD simulator, the optimiser based on the Powell method, the parametric shape editor and the goal function calculator. Specific requirements imposed on each of the elements are discussed and the functioning of the entire system based on the QuickWave-3D Simulator by QWED is presented on practical examples.

Large 3D object FDTD simulations

It is possible currently to simulate FDTD circuits, containing a huge number of cells (order 109), on contemporary PCs. In this paper, the results of simulations of an electrically large 3D object are presented. Problems with big data scale were resolved during each step of the simulations - from preprocessing, through FDTD simulation, to postprocessing.

Improvements to parameter extraction techniques for FDTD simulations of handset antennas

Improved techniques for the extraction of return loss and antenna efficiency from FDTD simulations of hand-held transceivers have been proposed. The key point for both techniques resides in using a new scheme of antenna excitation via a 3D model of a thin TEM line, terminated with a template source. This permits one to extract the return loss within the line by the differential method previously developed and well tested for S-parameter analysis in microwave circuits, and now enhanced with TEM template filtering. Another advantage of applying the template source is that it provides accurate control of power delivered to the antenna. The radiation efficiency can then be calculated directly, without the need to integrate dissipated power over the whole computational domain. Depending on the problem complexity, this reduces the computing time by up to a factor of two, and at the same time ensures the 0.8 % accuracy even for the discretization of 10 cells per wavelength.

On the influence of arithmetic underflow rounding standard on the speed of FDTD modeling

This paper presents the influence of arithmetic underflow rounding operations on the speed of FDTD analysis. It is shown that the underflow treatment according to the IEEE standard 754 (commonly accepted and implemented in modern arithmetic processors) may sometimes result in drastic slowdown of the speed of computing. The effect is much more pronounced in some of the most modern and most used processors. The ways to circumvent the effect by specific software operations are discussed.

Faster analysis of microwave engineering problems with multithread FDTD multiprocessor PCs

Two multithread FDTD algorithms have been developed, and their applicability to faster analysis of microwave engineering problems has been demonstrated. MS-FDTD is based on low-order spatial decomposition and subcircuit calculations on separate processors. MF-FDTD performs parallel updating of the three orthogonal field components. The algorithms have been tested on 2- and 4-processor PC, where speed-up factors of 1.5 and 2.1 have been obtained for MF-FDTD and 1.5-1.7 for MS-FDTD respectively.