On variability in high resolution range profiles

Radar target classification often uses feature sets derived from high range resolution profiles as an input to a classifier. However, rage profiles typically exhibit variability that is subsequently passed on to the feature vectors used in classification. This results in degraded classification performance as the variance in the feature vectors is too large and the mean distance between feature vector too small. In this paper we examine the fundamental causes of such variability. Specifically we demonstrate via modeling, simulation and experimentation that (i) the output of the matched filter is a sensitive function of the relative position of scatters distributed along the length of a target and (ii) the apparent location of a scatterers phase center is a sensitive function of viewing angle causing significant variability even over orientation changes of a fraction of a degree.