Quantitative comparison between Matrix Pencil method and state-space-based methods for radar object identification

Abstract

Approximating a function by a sum of complex exponentials to identify the nature of electromagnetic scatterers is treated with three model-based parameter-estimation methods. In particular, the Matrix Pencil Method and two State-Space-Based Harmonic-Retrieval methods are compared quantitatively. It is known that these methods generate very similar results in the absence of noise in the data and have only minor differences between them when the data is contaminated by noise. Since a quantitative comparison of the three methods has not been reported in the literature, this paper compares them by determining how accurately and quickly they predict the poles of the transient impulse responses of five electromagnetic systems: thin wire; perfectly conducting sphere; finite closed cylinder; dielectric sphere; composite metallic-dielectric sphere. It is important to note that these techniques are applied directly to the data and not to the covariance matrix, as noise statistics require additional information that is not available and the noise in electromagnetic scenes is generated typically by undesired signals in the form of a base-line shift in the measurement hardware instead of thermal background noise.