S.M. Jesus sjesus@ualg.pt
SiPLAB-FCT, Universidade do Algarve
Faro, Portugal
Comments: download pdf file
Ref.: in New Acoustics Selected Topics, Ed.
Ranz-Guerra and GAllego - Juarez, pp.139-152, CSIC, Madrid (España),
ISBN 84-00-08111-0, 2003.
Abstract : Model-based correlators are in nowdays widely used in
disciplines such ascommunications and underwater acoustics. These class of
methods generally make strong assumptions on the data underlying physical
model and develop dedicated processors that fully exploit the model structure.
When the assumed model is a good description of the data, these methods achieve
very high quality results since, in the mismatch free case, they are optimal
for the single signal in white noise scenario. However, in presence of model
mismatch this result is degraded. This paper presents a theoretical study
that addresses the optimality of model-based correlators when compared with
straight correlators in a model mismatch situation. It is shown that model-based
correlators can suffer severe degradation in relatively mild mismatch scenarios
and - more importantly - in that case model-based correlators are outperformed
by simple signal correlators. The theory is supported by a few realistic
examples drawn from underwater acoustics.