Two kinds of parameter estimation methods based on the information theory and an application to the acoustical measurement

Abstract

In this paper, two kinds of parameter estimation methods for overcoming the obstruction of the background noise co-existing in many actual acoustical or vibrational measurements are studied. As it is very important in an estimation problem for the measurement with the background noise of non-Gaussian type to grasp first the statistical properties of noise from all viewpoints as precise as possible, two kinds of information quantities which enable us to evaluate the distance between two different types of distribution forms or the uncertainty between two random physical quantities (e. g., the input and the output of the acoustical system) are newly introduced as the estimation criteria, and two kinds of sequential estimation algorithms which can be applied to the estimation for the general non-linear acoustical system with the background noise of arbitrary distribution type are explicitly derived. Furthermore, the legitimacy and the effectiveness of these algorithms are confirmed by applying to the actual parameter estimation problem of the reverberation room under the background noise from an acoustical viewpoint.