| Paper: | SPTM-P6.8 | ||
| Session: | Non-stationary Signals and Time-Frequency Analysis | ||
| Time: | Wednesday, May 17, 16:30 - 18:30 | ||
| Presentation: | Poster | ||
| Topic: | Signal Processing Theory and Methods: Non-stationary Signals and Time-Frequency Analysis | ||
| Title: | PARAMETER OPTIMIZATION METHODS FOR THE EDS MODEL | ||
| Authors: | Stijn Lamens, Wim D'haes, University of Antwerp, Belgium | ||
| Abstract: | Various methods have been proposed to estimate the parameters of the exponentially damped sinusoidal (EDS) model, which is useful for modeling audio signals. As most estimation methods do not yield optimal parameter values, it is often useful to apply further optimization. In this work, various optimization algorithms which can be used for both frequency and damping factor parameters are developed and compared. The optimization methods include Newton and Gauss-Newton methods, which were improved by applying a regularization factor resulting in the Levenberg-Marquardt method. Estimation and optimization algorithms were implemented and tested in two different methodologies: a method where optimization is performed on all partials simultaneously, and a pseudo-simultaneous method where optimization is performed on individual partials in an iterative manner. | ||
©2018 Conference Management Services, Inc. -||- email: webmaster@icassp2006.org -||- Last updated Friday, August 17, 2012