Time-Frequency Brown Bag Seminar

Wednesday, March 24, 1999

12:30pm

EQuad E415

Speaker: Brani Vidakovic, ISDS, Duke University

Title: Bayesian Strategies for Wavelet Denoising

Abstract:

After a short introduction to Bayesian modeling in the wavelet domain I will discuss two applications, one in smoothing the periodograms of stationary time series, and the second in processing and denoising turbulence signals.

For the second application a short description is provided. Fully developed turbulence is a self-similar phenomena with universal and stable second order properties. That makes wavelets an appropriate tool in describing, summarizing, and modeling the turbulence measurements.

The following two tasks are discussed:

  1. Wavelet based ``Townsend's Denoising'' in which the measured atmospheric turbulence signal (wind velocity components, temperature, humidity, ozone and CO2 concentrations, etc.) is separated on the attached part (that preserves energy and fluxes) and the detached part (an energyless component).
  2. Denoising turbulence signals from the instrumentation noise [ozone measurements] is performed with help of Bayesian models in the wavelet domain. A novel shrinkage technique of ``supervised'' shrinkage is introduced with a motivation to preserve fluxes.
The research is done jointly with Gabriel Katul from NSOE, Duke University. More information can be found in Discussion Paper 98-30 at: http://www.stat.duke.edu/papers/working-papers-98.html .
back to TFBB page