An adapted filter bank for frequency estimation

In this paper, a parametric spectral estimation method using an adapted filter bank is presented. The subband decomposition is performed classically through filtering and decimation stages. The decision about stopping or carrying on the decomposition of a given node is taken according to a new stopping rule. The latter uses a measure of whiteness of the residuals resulting from the node considered. Using Monte Carlo simulations, the results achieved with the proposed method are compared to those obtained with other methods performing fullband and subband estimations, in the case of noisy exponentials. The analysis points out the advantages of the proposed method.