Research On FDTD And High-order SFDTD Parallel Computing Based On GPU

As a new multi-disciplinary science, computational electromagnetics (CEM) plays a key role in electromagnetic engineering. Electromagnetic field theory, numerical method, and computer technology are integrated in CEM,which can be operated on the modern computer platform. In recent years, algorithms of CEM have been rapidly developed. The main methods involve finite-element method, moment method, and finite-difference time-domain (FDTD) method, which are respectively used to discretize vector wave functions, integral equations, and Maxwell’s equations.However,these methods are not perfect.The consuming time-the storage and the precision are the drawbacks that are not overcomed yet.To solve aforemetioned problems, systematical researches have been done on parallel algorithms using compute unified device architecture(CUDA). It is predicted that parallelization is one of the major trends in numerical calculation.Parallel technology based on GPU needs less consuming time and fewer hardware configrations. It breaks many limitations of the GPU computing model with the advent of CUDA and has become a trend in parallel computing fields.The dissertation mainly focuses on the basic derivation and application of this method.The main studies are as follows:(1)The fundamental properties of FDTD method and its natural parallelism are analyzed. The possibility of FDTD parallel method based on GPU is discussed.(2)The high-order symplectic FDTD (SFDTD) scheme which is relatively new and more complex is studied. SFDTD has better numerical precision and stability than the traditional FDTD method.However,it costs a lot of time in simulating electromagnetic problems. SFDTD(4,4) is put forward and the possibility of SFDTD parallel method based on GPU is discussed.(3)Two-dimensional FDTD parallel method based on GPU is implemented. The algorithm is further optimized by using texture memory.Comparing with the traditional calculation, this algorithm we presented has enough precision and significant improvments in the calculation speed under different sizes of Yee cells.(4)Two-dimensional SFDTD parallel method based on GPU is implemented. The calculation speed of SFDTD based on Fermi is accelerated under different sizes of Yee cells. Comparing with the traditional calculation,the CUDA parallel SFDTD has enough precision and high speed.