home · mobile · calendar · defenses · 2000-2001 · 

Thesis Defense - Wilson

Compression of Turbulence Data using Wavelet-based Lossy Coding
John Wilson
Computer Science PhD Candidate

This research examines the use of wavelet-based lossy coding for the compression of turbulence data. The primary target is the compression of data produced by numerical simulations. The computer systems used for the simulations are becoming sufficiently fast that the output of intermediate results can significantly impact the performance of the simulation. Sufficiently fast data compression will reduce the time required for data output.

Before lossy coding can be used to compress data from turbulence simulations, it is necessary to determine the degree of compression achievable with acceptable error. Turbulence data and the error measures of interest differ sufficiently from typical coding situations that numerous issues arise. New cost functions/decision heuristics which better match the error measures need to be developed. The statistical characteristics of turbulence data are sufficiently non-Gaussian that experiments show that non-uniform quantization is potentially useful at high data rates. The nature of some of the error measures and concern with values derived from the data lead to situations where rate-distortion theory is lacking.

Committee: Oliver McBryan, Professor (Chair)
Elizabeth Bradley, Associate Professor
Harold (Hal) Gabow, Professor
Aimé Fournier, NCAR
Nicholas Brummell, APS/JILA
Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:20)