Date of Award

9-1995

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Department of Electrical and Computer Engineering

First Advisor

Robert Riggins, PhD

Abstract

The derivation of the power spectral density of the optimal input for system identification is addressed in this research. Optimality is defined in information theoretic terms, with entropy quantifying the parameter information content of the input and output measurement sequences pertaining to a discrete time plant. The maximization of entropy is performed in the context of three different scenarios. First, the case in which the average output power of the plant is constrained is considered. Second, input average power is constrained. Finally, the optimization is carried out unconstrained, but with penalties applied to both the input and output powers. Although the focus of this research is the enhancement of the parameter identification potential of general System Identification algorithms, a new and efficient System Identification algorithm that employs Iterated Weighted Least Squares is derived. Experimental evidence is presented which clearly illustrates the superiority of this algorithm. Furthermore, experiments are documented which corroborate and validate the maximum entropy based theory for optimal input design presented in this dissertation.

AFIT Designator

AFIT-DSG-ENG-95-S-04

DTIC Accession Number

ADA297483

Download

Share

COinS