Estimador competitivo de bajo coste computacional para deconvolución dispersa

Abstract

This paper presents an efficient algorithm for computing the maximum a posteriori (MAP) estimate of a Bernouilli- Gaussian sequence distorted by a linear filter and corrupted by noise. The computational effort for the MAP estimator increases exponentially with the number of samples and is generally unfeasible. The approach used by Kormylo and Mendel in their pioneering work on the Single Most-Likely Replacement algorithm (SMLR) has been the sub-optimal reference to evaluate the efficiency of this new estimation method in computational effort and detection probability. This algorithm reduces drastically the SMLR computational load and allows real-time processing, because only few samples of the observation data are required by employing a windowing strategy. An extensive Monte Carlo analysis, using synthetic data, has been performed to compare the behaviour of the proposed estimators.