Full wave analysis of millimeter wave quasi-periodical structure for antenna applications by method of moments and its conjugate gradient solution on GPU/CPU platform

Abstract

In this paper, an accurate and efficient full wave analysis for millimeter wave quasi-periodical structure for antenna applications is realized. By extracting all the surface wave poles of the spectral domain multilayered Green's functions, the discrete complex image method (DCIM) can be adopted to fast evaluate their counter parts in spatial domain, and the mixed potential integral equation (MPIE) for method of moments (MoM) can be constructed. With the Jerusalem cross as a patch element on a dielectric layer, a quasi-periodical structure can be organized as the geometries gradually varied. By illuminate this kind of structure with an incident plane wave, the distribution of electric current on the metal surface can be obtained by solving the MoM equation and the forward scattering cross section can be fast calculated by the conjugate gradient (CG) algorithm written in CUDA and realized in NVIDIA graphic process unit (GPU). A layered structure with a 17×17 Jerusalem crosses at 30GHz was calculated and very good agreements have been found between the proposed method and commercial EM simulator (CST), while an improvement on efficiency is realized.Index Terms -Quasi-periodical structure, layered structure, MoM, multilayered Green's functions, CUDA.

High Performance Computing of Complex Electromagnetic Algorithms Based on GPU/CPU Heterogeneous Platform and Its Applications to EM Scattering and Multilayered Medium Structure

“…(1) A novel realization of OpenMP-CUDA-MLFMM method for EM scattering problem, in which the nearfield matrix filling and the sparse matrix-vector production (MVP) are optimized, together with a warplevel parallel scheme for aggregation/disaggregation and the use of texture memory for 2D local interpolation/anterpolation (2) A developed ACA method for EM scattering problem, in which the near-field matrix filling realizes a 100% efficiency enhancement and a thread-block-level parallel and a register-reusable schemes are applied to matrix compression and far-field MVP, respectively; also, a double-buffer technique for double precision is proposed to further enhance the efficiency of MVP computation [25] (3) A deeply optimized CG-FFT method for planar multilayered structure, in which the hardware instruction "__shufl_down" instead of shared memory was firstly used in the summation of MVP [26] and the current distribution of a large-scale flat lens was obtained efficiently.…”