#### Date of Award

Spring 1995

#### Document Type

Dissertation - Restricted

#### Degree Name

Doctor of Philosophy (PhD)

#### Department

Electrical and Computer Engineering

#### First Advisor

Josse, Fabien

#### Second Advisor

Heinen, James A.

#### Third Advisor

Brown, Ronald H.

#### Abstract

A number of radar detection problems may be formulated as one where the presence of a signal is sought in a background of clutter, interference and receiver noise. The background noise environment is typically unknown or often varies with time and as a consequence, the noise statistics need to be estimated. For coherent detection algorithms, the parameters that need to be estimated are the elements of the noise covariance matrix. The covariance matrix of the noise/interference is typically estimated by taking a snapshot to obtain the received data from one or more range gates adjacent to the resolution cell under test, forming an analysis window, and calculating the noise statistics within this window. The spatial statistics of the noise typically varies with time, but for a finite size analysis window, the background may be assumed to be homogeneous in some scenarios. Several variants of coherent detection processors exist, but decision theoretic approaches to coherent adaptive detection primarily relies (either implicitly or explicitly) on estimating the noise covariance matrix, RN. In adaptive array detection problems the interference may sometimes be modeled as the sum of uncorrelated plane waves with different directions-of-arrival. The covariance matrix of the interference in such a case inherently has a Toeplitz structure. However, the sample covariance matrix obtained from the observed data is not Toeplitz. Vector processors designed to incorporate the a priori known structural information into the detection procedure are thus capable of providing improved performance...