Date of Award

5-2022

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Engineering and Sciences

First Advisor

William Arrasmith

Second Advisor

Anthony Smith

Third Advisor

Luis Daniel Otero

Fourth Advisor

Adrian M. Peter

Abstract

The Well-Optimized Linear Finder (WOLF) high-speed, phase-dominant, transfer function estimation method was developed by Professor William W. Arrasmith at Florida Institute of Technology. The novel, high-speed, transfer function estimation method is robust and can be applied to atmospheric turbulence compensation (ATC)/blind deconvolution optical imaging problems. The WOLF methodology uses a diversitybased approach with an adapted error metric to quickly remove effects of atmospheric turbulence and system noise effects present in an incoherent, optical imaging system. In this research, we improve performance of the WOLF algorithm by investigating the impact of applying parallel processing technology to pre-calculate an expanding set of constant complex exponential phase difference sums that lie at the core of the WOLF methodology. Depending on the number of entrance pupil plane sample points in the image, these complex exponential phase differences can range from an initial single complex exponential phase difference term to sums of millions of complex exponential phase difference terms. We use order analysis on the WOLF algorithm to evaluate theoretical implications of pre-calculating the constant exponential phase difference chain terms in parallel, before they are needed. We validate theoretical predictions by using computer simulations to isolate timing associated with determination of the constant complex exponential phase difference terms. A conservative estimate of approximately 88.7 percent faster performance can be achieved by implementing pre-calculation of the sums of constant complex exponential phase difference terms. A representative 256 x 256-pixel image is used in our analysis and computer simulation. The computer used in the study is an un-modified 2014 MacBook Pro computer with 2.8 GHz (Quad-Core, Intel Core i7), 16 GB of 1600 MHz DDR3 memory, and a NVIDIA GeForce GT 750M 2 GB video card running Matlab 2020b. Removing atmospheric turbulence from the 256 x 256 image takes approximately 8 seconds using the non-optimized WOLF algorithm without taking advantage of parallel processing or pre-calculation of the constant complex exponential phase difference terms.

Download

Share

COinS