Optimum Blind Multichannel Equalization Using the Linear Prediction Algorithm

The linear prediction (LP) algorithm estimates a zero-forcing (ZF) equalizer from the channel output´s second-order statistics (SOS). We show that it allows, in fact, to determine a whole subspace of (but not all) ZF equalizers. The one obtained by maximizing the signal-to-noise ratio (SNR) at the equalizer output not only outperforms the original LP equalizer, but also attains the lowest achievable (by any no-delay ZF equalizer) mean-square error (MSE). The increase in complexity is only marginal, but equalization performance can be enhanced significantly, especially in the (practical) case of a small leading channel coefficient