Spectral estimation and matched filter performance

Several spectral estimation procedures are compared, on the basis of the performance of moving average matched filters designed with the spectral estimates as a priori information. Many spectral estimation methods do not impose the constraint of nonnegativity on the spectral estimate which in the matched filter context may lead to a design that performs worse than if no filter were used at all. Classical unbiased correlation estimates are modified so as to impose nonnegative definiteness, and the corresponding matched filter performances are evaluated. The parametric Burg and Itakura methods are ordered in terms of positive definiteness, but simulations show that such ordering is not reflected directly in the corresponding matched filter performances.