Nicholas B. Pulsone

Adaptive Signal Detection in Non-Gaussian Interference

May 21, 1997
9:00 AM
306 Egan

Abstract

Coherent signal detection in non-Gaussian interference is presently of interest for adaptive array applications. Typically in this context, no prior knowledge of the correlation properties of the interference is available. As such, it is desirable that the probability of false alarm (i.e. the size of the test) of candidate array detection algorithms be independent of the unknown interference covariance matrix. For example, in airborne radar surveillance applications the constant false alarm rate (CFAR) feature enables the design of practical receivers that control the number of false alarms while providing reasonably good detection performance. Over the years, a number of receiver implementations for array detection that provide the desired CFAR feature have been developed assuming an unknown multivariate Gaussian interference model. However, in some non-Gaussian interference environments the performance of these receivers is unsatisfactory. The main purpose of this thesis is to develop and characterize the performance of a few generalized receivers for array detection in a class of non-Gaussian interference. As an example, this thesis considers an airborne radar application using reduced-dimension space-time adaptive processing (STAP). Clutter measurements made under the Mountain Top Program are analyzed to show that a zero-mean spherically invariant random vector (SIRV) model provides a good description of the observed space-time interference vectors. The observed interference vectors are shown to be characterized either by a 'dependent SIRV model' or by an 'independent SIRV model.' For both SIRV models, we develop and characterize the performance of several CFAR array detectors using a generalized likelihood ratio formulation. It is shown that CFAR array detectors under the 'dependent model' can be implemented without prior knowledge of the characteristic probability density function of the SIRV model. However, array detectors under the 'independent model' require prior knowledge of some parameters of the characteristic probability density function. Results showing the detection performance of several generalized receivers in SIRV interference are presented.

Thesis Committee:
Prof. R.S. Raghavan (advisor)
Prof. D.J. McLaughlin
Prof. E.S. Manolakos