EM Algorithm for Sequence Estimation over Gauss-Markov ISI Channels¹

Richard Perry, W. Andrew Berger ${}^{\dagger}$ and Kevin Buckley
ECE Department, Villanova University, Villanova, PA 19085
${}^{\dagger}$Physics/EE Department, University of Scranton, Scranton, PA 18510
perry@ece.villanova.edu, (610)-519-4969, FAX: (610)-519-4436

Abstract:

This paper presents a new algorithm, based on an EM (Expectation-Maximization) formulation, for ML (maximum likelihood) sequence estimation over unknown ISI (inter-symbol interference) channels with random channel coefficients which have a Gauss-Markov fast time-varying distribution. By using the EM formulation to marginalize over the channel coefficient distribution, maximum-likelihood estimates of the transmitted sequence are obtained. This EM algorithm is shown to perform better, in terms of BER, than existing algorithms which perform jointly-optimal sequence and channel estimation, or which do not take into account fast time-varying channel effects.