Cramer-Rao bounds for semi-blind, blind and training sequence based channel estimation

Abstract

Two channel estimation techniques are often opposed: training sequence based estimation in which a sequence of symbols known by the receiver is used and blind equalization in which the channel and/or the symbols are determined from the received signal only. The purpose of semi-blind techniques would be to adapt blind techniques in order to profit from the existence of a training sequence. A first approach to assess the performance of semi-blind methods is proposed. We study Cramer-Rao Bounds for blind, semi-blind and training-sequence based channel estimates for a deterministic as well as a Gaussian symbol model, and compare them theoretically as well as in numerical evaluations. Furthermore, an example of comparison between the corresponding maximum likelihood estimation methods is given.