Natural resonance extraction from: multiple data sets using a genetic algorithm

Identification of radar targets based on their late-time, natural-resonance signature requires the accurate knowledge of target natural frequencies. Realistically, these frequencies must be determined from measurements of actual or scale model targets. Many approaches have been studied to extract the natural frequencies from measurement, including the E-pulse technique, several variants of Prony´s method, and the pencil-of-functions method. Perhaps the most straight-forward are those based on nonlinear least squares algorithms. Difficulty arises with least-squares techniques, however, because of their dependence on initial guesses for the natural frequencies and their tendency to get trapped in local minima. Also, constraints must often be placed on the ranges of the parameters as dictated by the measurement process and the principle of energy conservation. A genetic algorithm provides a natural minimization procedure for implementing constraints and avoiding initial guesses. Early studies have shown that a genetic algorithm provides results equivalent to the constrained E-pulse technique, with much less computer effort.<>