SBIR-STTR Award

The problem of estimating radar target parameters such as shape, size and material properties from monostatic scattering measurements is formulated within the context of statistical estimation theory and inverse scattering. A bayesian approach to this problem is proposed where the liklihood ratios between multiple target hypotheses are used to make an optimum target identification from sparse scattering data. A detailed analysis of this approach is presented using the polarization corrected physical optics approximation for cylindrically symmetric, conducting targets. It is shown that for such targets the statistical estimation problem becomes mathematically equivalent to the limited view problem to computed tomography when cast within a statistical framework. Extensions of the approach using alternative formulations of the inverse scattering problem are discussed.

The Phase I project titled Probabilistic Radar Inverse Scattering (PRIS) is reviewed and the results of the study are shown to indicate that the PRIS concept is applicable to a host of government and commercial applications. These applications include the use of PRIS computer codes to aid in the design and evaluation of radar systems and low observable vehicles and the use of PRIS codes in operational situations to detect and identify radar targets from limited and noisy radar returns. It is proposed to extend and generalize the PRIS algorithms developed in Phase I of this project and to implement these algorithms in a microprocessor workstation. The generalized algorithms will employ a number of scattering models which include the physical optics approximation, the born and distorted wave born approximations and a newly derived polarization corrected physical optics approximation that was developed in Phase I of the program. The workstation will have the capability of generating synthetic radar signatures for a large class of targets and for user selected radar system parameters. Performance bounds for target detection and identification in the presence of zero mean colored Gaussian noise will be generated as a function of target and radar parameters. Detection and identification of the target is performed using maximum likelihood estimation and matched filtering.