Development and analysis of parametric exponential modeling techniques with application to radar signal processing

In this dissertation we address several signal processing aspects of the radar target identification problem. The focus of this research is the development and analysis of parametric signal processing techniques for extracting scattering features of radar targets from frequency domain scattering measurements. Using these characteristics, or features, automatic target recognition can be accomplished. This work investigates parametric modeling techniques for extracting these scattering features.; We first present a model in which signals consist of damped exponential terms in noise and demonstrate how the parameters of this model relate to radar target scattering centers. We provide a complete Cramer-Rao bound (CRB) derivation for the parameters of this model. The CRB provides a performance bound for the estimation of the model parameters. Expressions for the CRBs of the parameters of the damped exponential model with one set of poles and multiple sets of amplitude coefficients are derived; this corresponds to multiple polarization measurements in the radar problem.; We then present the total least squares-Prony (TLS-Prony) method for estimating the model parameters and an analysis of the estimated parameter statistics. We develop an analytical expression for the covariance matrix of the estimated parameters for this method. We verify these analytical results using Monte-Carlo simulations studies. We also compare the TLS-Prony variance results to the corresponding CRBs for several cases.; Next, we introduce a modified TLS-Prony method which incorporates data decimation. The use of data decimation reduces the computational complexity in a variety of applications by allowing several low order estimations to be performed in place of one high order estimation. We perform an analysis of pole variance statistics for the modified TLS-Prony method. This analysis is used to explain and quantify the characteristics of decimation. We show that by using decimation, one can obtain performance results comparable to the nondecimated case, but at a fraction of the computational cost.; We conclude with a summary of the research and its impact. We also suggest areas for future research.