Date of Award

2019-01-01

Degree Name

Master of Science

Department

Electrical Engineering

Advisor(s)

Raymond C. Rumpf

Abstract

The use of a matrix-free, memory-efficient approach to generate large-scale spatially variant lattices (SVL) was explored. A matrix-free iterative SVL generation algorithm was formulated and then implemented with a tremendous memory reduction observed. The algorithm consists of solving first-order central finite-differences along the entirety of the problem space point-by-point to obtain the grating phase function Φ(𝑠⃗) to which all desired spatially variant lattice properties are applied to. The algorithm was studied to identify key areas of data and task parallelism to exploit in heterogeneous computing systems consisting of clusters of central processing units (CPU) and graphics processing units (GPU) combinations. A sequential version of the algorithm was implemented in MATLAB to study memory usage; the sequential implementation proved to cut memory usage in the generation of large-scale SVLs when compared to the traditional matrix approach. Some preliminary efforts were made to map the algorithm to a GPU with the use of the Compute Unified Device Architecture (CUDA); these efforts are presented as the basis to improve the scaling problems inherent with traditional matrix approaches.

Language

en

Provenance

Received from ProQuest

File Size

41 pages

File Format

application/pdf

Rights Holder

Manuel Fernando Martinez

Share

COinS