On MLSE algorithms for unknown fast time-varying channels

We consider maximum-likelihood sequence estimation (MLSE) algorithms for unknown, time-varying intersymbol interference communication channels. We assume a statistical channel model, and marginalize over model parameters to derive expectation-maximization (EM) algorithms for both time-independent Gaussian and Gauss-Markov models, and we contrast these with direct MLSE and computationally efficient per-survivor processing implementations. We identify a general concern associated with the convergence of EM-based discrete parameter (e.g., symbol) estimators.