Parameter estimation of decaying exponentials by projection on the parameter space

Abstract We propose in this paper a new estimation method for the parameters of a linear combination of exponentials flcorrupted by some noise. It is shown that this method does not require any a priori information about the nature of the noise probability density. The algorithm is used to estimate parameters of synthetic signals and to estimate effective pulmonary ncapillary pressure. This algorithm is compared to the method of Kumaresan and Tufts which is based on linear prediction and singular value decomposition. Our experiments demonstrate increased accuracy over the method of Kumaresan and Tufts.