Robust joint channel and noise estimation in Bayesian blind equalizers

摘要

In this paper, we discuss the problem of reducing the complexity of Bayesian approaches to the joint channel and data detection issues in digital communications. Taking ideas from Bayesian Analysis and Evolutionary Computation, we propose and compare different equalization schemes whose complexity is kept moderate by applying pruning and merging operations over an otherwise growing population of channel estimates. Already existing deterministic methods for complexity limitation as well as new stochastic ones are analyzed and compared. Using the capability of Bayesian equalizers to estimate the symbol error probability during operation, a convergence control mechanism is introduced. The joint channel and noise Bayesian estimation is shown to offer more robust BER estimates.