MMSE equalizers for DMT systems with and without crosstalk

This paper deals with finding optimal equalizers in the presence of intersymbol interference (ISI) and colored noise for systems using DMT (discrete multitone) modulation. Previously proposed, frequency-domain training of the equalizer performs better than its time-domain counterpart. However, frequency-domain training does not include the effects of noise. Therefore the trained equalizer minimizes only the ISI and not the overall mean square error. So far, no non-iterative method for computing the optimal equalizer exists. We find a way of computing the optimal equalizer that maximizes the SNR in the presence of ISI and colored noise. We also show how the effect of the noise can be incorporated into the training algorithm. Finally, we show that significant performance gains can be achieved using these new techniques.