Semi-Blind Maximum-Likelihood Multichannel Estimation with Gaussian Prior for the Symbols using Soft Decisions

We present Maximum-Likelihood (ML) approaches to semi-blind estimation of multiple FIR channels. The first approach, DML, is based on a deterministic model. The second one, GML is based on a Gaussian model in which the input symbols are considered as Gaussian random variables: this model leads to better and more robust performance than DML. Algorithms are presented to solve DML and GML and the significant improvement of GML w.r.t. DML is demonstrated. A soft decision strategy is also presented to improve ML performance: the most reliable decisions taken at the output of an equalizer built from a semi-blind ML channel estimate are treated as known symbols and semi-blind ML is reiterated with an augmented number of known symbols.