Normal and uniform noise - violation of the assumption on noise distribution in model identification

Mathematical modelling under uncertainty together with the field of applied statistics represent tools useful in many practical domains. Widely accepted assumption of normal (Gaussian) noise has created the basis for theoretical and algorithmic solutions of respective tasks. However, many continuous variables are strictly bounded and their uncertainty may have origin in various physical processes which causes a non-normal distribution of their noise. Furthermore, adaptation of algorithms based on normal model for identification of models with bounded noise can distort the estimates due to inconsistent handling of uncertainty. This report describes a study to compare results of estimation algorithms based on assumption of normal and uniform noise. Data sequences processed by the algorithms have normal noise bounded by a low limit with respect to standard deviation. We illustrate disparity between noise assumption and a true noise distribution and its influence on the quality of the estimates. It is a part of an effort to develop theory and fast algorithms for estimation with bounded noise, applicable in practice