Score function based totally blind equalizers

In this paper we propose two new totally blind block equalizers for linear modulations using approximated maximum likelihood criteria. Totally means here that we assume we don´t know the emitted symbols or the constellation (the modulation type). The coefficient update equations turn out to be functions of the equalizer output score function (SF). Because of our assumptions we have no closed form expression for the probability density function (PDF) and we propose to estimate the SF using the equalizer output samples. The performance of the equalizers is compared with the constant modulus algorithm (CMA). It shows that for a small number of symbols the proposed equalizers can remove all the inter symbol interference (ISI) while the CMA performs poorly. For severe channels the CMA appears to be more robust (our equalizers may not converge) but in this case our equalizers can be used after the CMA with very good results. Lastly when the algorithms converge, performance of our equalizers is similar to the decision directed one but in a totally blind way and without its drawbacks