Fast Electromagnetic Computing On Graphics Hardware

The analysis of the electromagnetic scattering characteristic of the target, especially the prediction of Radar Cross Section (RCS), is one of the important research areas in computational electromagnetics. It has wide application in many different fields, ranging from the scattering and radiation analysis of the antenna to the design of stealth and anti-stealth. Thus, it is vital important in theory and practice to study the electromagnetic scattering characteristic of the target.This thesis focuses on the frequency-domain numerical methods, which can be classified into high-frequency asymptotic methods and low-frequency numerical methods. Graphical electromagnetic computing (GRECO) and shooting and bounce ray method (SBR) are two widely used high-frequency asymptotic methods in nowadays. However, GRECO still fails to predict RCS in real time and to identify all visible wedges exactly, and SBR has the problem of time-consuming ray tube tracing and the split ray tube. These problems significantly affect the computational efficiency of high-frequency asymptotic methods. The method of moments (MoM), which is one of the most popular low-frequency numerical methods, solves the scattering problem by discretizing Maxwell's integral equation into the dense impedance matrix. However, due to the limited memory and low computing power of the computer, only the RCS of electrically small targets can be predicted by MoM.Aiming at the above-mentioned issues, we compare and analyze the similarity of the frequency-domain numerical methods and rendering algorithms in computer graphics, and try to improve the frequency-domain numerical methods by adopting the idea of real-time rendering algorithms and employing graphics hardware as the compute platform. In order to accelerate GRECO, we present a new architecture unifying the visibility computing and electromagnetic computing on GPUs. This architecture can detect visible wedges more exactly and predict the first-order scattered field in real-time.The proposed CUDA-based SBR fully implements the ray tube tracing and electromagnetic computing in CUDA. The ray tube tracing is based on the stackless kd-tree traversal algorithm and this implementation greatly accelerates the RCS prediction. We also introduce the beam-tracing based SBR, which inverses the conventional order of the ray tube generation and the ray tube tracing. During the ray tube tracing, the ray tube is dynamically divided into several ray tubes according to the geometry of the target, which results in avoiding the problem of the split ray tube and improving the computational accuracy of the scattering field. Moreover, beam tracing is also able to identify visible wedges, and the diffracted field of these visible wedges can be evaluated using TW-ILDC. In fact, SBR and TW-ILDC together can obtain a high-fidelity RCS result for most high-frequency scattering problems. Besides the PEC targets, the equivalent reflection coefficients of a layered medium is also presented and it extends the application of SBR to the coated target.For low-frequency numerical methods, we propose a CUDA-based MoM, which exploits the formidable of computing power on GPUs to enhance the computational efficiency. Additionally, memory management techniques, i.e., block-based impedance matrix and out-of-core, enlarge the electrical size of the target that can be solved on a single computer.Finally, the electromagnetic modeling and simulation software emX is introduced. emX not only integrates the frequency-domain numerical methods above, but also offers various features, such as geometric processing, target imaging, visualization, etc.The research of this thesis effectively integrates the frequency-domain numerical methods with rendering algorithms in computer graphics and graphics hardware, and provides a new direction to solve scattering problems of electrically large and complex targets. A large number of numerical results demonstrate that the combination of computational electromagnetics and computer graphics could improve the computational efficiency, accuracy, and scale of scattering problems.