A fast Gaussian maximum-likelihood method for blind multichannel estimation

Abstract

We propose a blind maximum-likelihood method for FIR multichannel estimation, denoted GML. The GML criterion is derived assuming the input symbols as Gaussian random variables. The performance of GML (computed based on the true symbol distribution) is compared through numerical evaluations to the optimally weighted covariance matching method: both methods are equivalent in a certain asymptotic sense. A fast implementation of the scoring algorithm is proposed to solve GML.