Blind Identification Based on Expectation-Maximization Algorithm Coupled With Blocked Rhee–Glynn Smoothing Estimator

In this letter, we consider blind estimation of channel parameters over a frequency-selective channel. We use a blocked Rhee-Glynn smoothing estimator to derive E-step in the expectation-maximization (EM) algorithm. The proposed algorithm copes with the curse of dimensionality of a forward-backward algorithm; meanwhile, it is easy to parallelize, which is amenable to a modern computing hardware and speeds up the estimation of channel parameters. The experiment results show that the proposed algorithm is close to the Baum-Welch algorithm in terms of convergence of channel coefficients and outperforms the EM algorithm coupled with a joined two-filter smoothing algorithm in terms of convergence of channel coefficients and running time.