Blind and semi-blind maximum likelihood methods for FIR multichannel identification

Abstract

We investigate maximum likelihood (ML) methods for blind and semi-blind estimation of multiple FIR channels. Two blind deterministic ML (DML) strategies are presented. In the first one, we propose to modify the iterative quadratic ML (IQML) algorithm in order to "denoise" it and hence obtain consistent channel estimates. The second strategy, called pseudo-quadratic ML (PQML), is naturally asymptotically denoised. Links between these two approaches are established and their global convergence is proved. Furthermore, we propose semi-blind ML techniques combining PQML with two different training sequence estimation methods and compare their performance. These semi-blind techniques, exploiting the presence of known symbols, outperform their blind version. They also allow channel estimation in situations where blind and training sequence methods fail separately. Simulations are presented to demonstrate the performance of all the proposed algorithms, and comparisons between them are discussed in a blind and/or semi-blind context.