Robust estimation of source frequencies for superimposed sinusoidal signals
Abstract
New estimation methods for frequencies of source signals are investigated. The received observations are described as a sum of sinusoids (or complex exponentials) with various noise assumptions. The noise processes are assumed to be white, colored (Autoregressive process), contaminated Gaussian, and unknown distributed random processes. In this thesis, the robustness of estimates is emphasized. First, a new Linear Prediction method is investigated. The estimate using the proposed LP method is shown to be statistically consistent (in the MS sense) unlike other LP estimates. Second, a new decentralized processing scheme is introduced to improve the accuracy of sensor estimates of source signals for geographically distributed array sensors. Robust methods are investigated for combining sensor estimates. Third, the influence function and asymptotic variance of robust frequency estimates are derived. Fourth, new theories for high resolution frequency estimates are developed. Definitions and theorems are introduced and provide the legitimate tools for any frequency estimate to determine whether or not the estimate is a high resolution frequency estimate. Finally, a new 2-D frequency estimation method is proposed when the noise process obeys an unknown distribution. The new method is numerically efficient and performs as well as two dimensional model-based estimates.
Degree
Ph.D.
Advisors
Kashyap, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.