Prediction of the EM-Algorithm Speed of Convergence with Cramer-Rao Bounds
Cédric Herzet, Luc Vandendorpe · 2007
This paper aims at characterising the (mean) speed of convergence of the EM algorithm. We derive, under some simplifying assumptions, a relation between the EM algorithm mean convergence rate (MCR) and Cramer-Rao bounds (CRBs) associated to the so-called incomplete and complete data sets defined within the EM algorithm framework. We illustrate our derivations in the ease of carrier-phase estimation based on the EM algorithm, As far as our simulation setups are concerned, we show that the (mean) EM-algorithm behavior may be well predicted by means of the proposed CRB-based impression.