You are here

Techniques to Improve the Accuracy of System Identification in Non-Gaussian and Time Varying Environments

Title: Techniques to Improve the Accuracy of System Identification in Non-Gaussian and Time Varying Environments.
49 views
10 downloads
Name(s): Ta, Minh Quang, 1977-, author
DeBrunner, Victor, professor directing dissertation
Chicken, Eric, outside committee member
DeBrunner, Linda, committee member
Roberts, Rodney, committee member
Department of Electrical and Computer Engineering, degree granting department
Florida State University, degree granting institution
Type of Resource: text
Genre: Text
Issuance: monographic
Date Issued: 2008
Publisher: Florida State University
Place of Publication: Tallahassee, Florida
Physical Form: computer
online resource
Extent: 1 online resource
Language(s): English
Abstract/Description: Estimation of a dynamical system under unknown influences is always subjected to uncertainty. Thus, reducing the estimation variance under external influences is absolutely desired and becomes the motivation for the field of System Identification. In this dissertation, the author proposes new techniques for system identification under two major general situations: offline estimation of fixed systems under the unknown non-Gaussian distributed measurement noise, and online-estimation of time-varying systems undergoing systematic (long term correlated) changes. For the first situation of offline estimating of fixed systems under the unknown non-Gaussian distributed measurement noise, a technique called Minimum Entropy Estimation is employed, which promises to be better than the traditional Least Square (LS) estimation method due to the ability to simultaneously estimate the system and the statistical property of the unknown measurement noise sequence. This method gives rise to two novel classes of generalized offline estimation algorithms being proposed in this dissertation: a method of estimating a Multiple-Input-Multiple-Output (MIMO) systems under unknown, independent and identically distributed (iid) non-Gaussian measurement noise, and a more general method of estimating a feedback structure under unknown, possibly colored, non-Gaussian distributed measurement noise. For the second situation of online estimation of time-varying systems undergoing systematic changes, a new method of Parameter-Filtering Adaptation (PFA) algorithm is proposed for the first time as an attempt to solve this problem and improve the estimation quality. Instead of updating the parameter based on the prediction error and an estimated value of the parameter at single time iteration (before the current one) as in the traditional adaptive algorithms, the new method improves the estimation quality of the system parameter by incorporating its prediction from all previous estimated values. The parameter prediction transfer function itself is also updated adaptively. The PFA algorithm is firstly considered in the context of IIR filter estimation to show the benefit of better local quadratic approximation for the time-varying, non-quadratic error surface. Its application in the spectral estimation of time-varying chirps utilizing Adaptive Notch Filters has shown an enormously better estimation of the instantaneous frequency. In the context of estimating time-varying systems using FIR filters, it is discovered that the PFA has a filtering effect on the sequence of the (estimated) parameters. Consequently it is shown in the dissertation that the sparser in the frequency domain (less frequency bandwidth) the parameter variations are, the better their estimation quality. Simulation on tracking of sinusoidal time-varying systems, as well as periodically switching systems shows that the PFA has a superior estimation quality with virtually no lag comparing to the traditional tracking methods.
Identifier: FSU_migr_etd-0311 (IID)
Submitted Note: A Dissertation submitted to the Department of Electrical and Computer Engineering in partial fulfillment of the requirements for the degree of Doctor of Philosophy.
Degree Awarded: December Semester, 2008.
Date of Defense: August 11, 2008.
Keywords: Non Gaussian Noise, System Identification, Sparse Representation, Adaptive Filter, Time Varying System
Bibliography Note: Includes bibliographical references.
Advisory Committee: Victor DeBrunner, Professor Directing Dissertation; Eric Chicken, Outside Committee Member; Linda DeBrunner, Committee Member; Rodney Roberts, Committee Member.
Subject(s): Electrical engineering
Computer engineering
Persistent Link to This Record: http://purl.flvc.org/fsu/fd/FSU_migr_etd-0311
Owner Institution: FSU

Choose the citation style.
Ta, M. Q. (2008). Techniques to Improve the Accuracy of System Identification in Non-Gaussian and Time Varying Environments. Retrieved from http://purl.flvc.org/fsu/fd/FSU_migr_etd-0311