A randomised algorithm for improving source and channel estimates by exploiting the finite alphabet property

Abstract

Many channel identification algorithms do not take into account the fact that the source symbols belong to a finite alphabet. This paper proposes an algorithm which exploits this finite alphabet property to refine the estimates of the source symbols and channel parameters obtained by any identification algorithm. The key idea is to reformulate the problem as minimising a quadratic cost function on a set of discrete points. Since it is known that this problem is NP-hard, a randomised algorithm is proposed, which guesses the direction in which to search for the minimum.