Estimation Methods and
|Steven T. Smith
MIT Lincoln Laboratory
244 Wood Street, Room J-149M
Lexington, MA 02173-9108
tel: (781) 981-3106
Abstract Synthetic wideband processing as proposed by Cuomo, Piou, and Mayhan employs parametric estimation methods to construct wideband data from widely separated subbands. The goal is to obtain ultrawideband resolution from two or more sensors operating at different frequencies greater than that achievable using conventional methods. The key is accurate estimation of the target return's signal parameters, which in this talk are assumed to be those of an all-pole model. The estimation algorithm consists of an initialization step to compute rough estimates of the pole parameters, followed by a maximum likelihood estimation step to compute ML estimates of poles and their associated amplitudes. In this talk we describe the fundamental accuracy and resolution bounds associated with this estimation problem, compare the performance of different superresolution algorithms (ESPRIT versus root-MUSIC) for the initialization stage, and compare different optimization algorithms (Newton-Raphson versus various quasi-Newton methods) for ML estimation. One interesting question addressed is the true resolution achieved by widely separated subbands; it is shown, for example, that by separating two subbands each of width B Hz by a bandgap of MHz, one achieves a resolution improvement of about sqrt (M/B). Experimental data provided by Lincoln Laboratory's Divisions 3 and 9, as well as simulated data, is used to compare different algorithms.
Direct comments and questions to: email@example.com
© MIT Lincoln Laboratory. All rights reserved.