Abstract
In this paper, a new trial on the parameter estimation for stochastic environmental systems has been proposed. The measure of statistical independency between an input signal and an intruded observation noise is newly adopted as the error evaluation measure for this parameter estimation. This measure can evaluate not only the neighbourhood of mean value (such as, related to the well-known least-square-error criterion), but also the end of level distribution form of random phenomenon (such as, used very often as one of actual environmental standards in the field of environmental noise evaluation). Concretely, the joint probability density function, P(x, v), for an input signal, x, and an observation noise, v, is positively expressed in some series expansion form to grasp the statistical independency hierarchically. Then, after naturally defining the deviation from independency by ε(x, v)_??_P(x, v)-P(x)P(v), several types of new estimation methods have been derived according to how to evaluate and then decrease this deviation. The validity and effectiveness of these proposed methods have been experimentally confirmed by applying them to the actual estimation problem for the sound insulation parameters in an acoustic field.
