Gain correction in optimal filtering using floating point arithmetic

Abstract

The effects of the finite word length in a floating point implementation of the least squares filter is discussed. Optimal precomputable gains are given, and a computationally more attractive approximation is given.