Using noise-feedback in approximating ML sequence estimation for channels with infinite intersymbol interference

We present a novel scheme for approximating maximum-likelihood sequence estimation for channels that can be modelled recursively. In order to reduce the infinite intersymbol interference of such channels. We propose to prefilter the channel output. In this way, the number of required states in a subsequent Viterbi detector can be reduced. To compensate for the resulting noise correlation, a modified branch metric is proposed. Compared to a recently presented decision-feedback sequence estimator, which reduces the number of states by feeding back preliminary data decisions, our scheme feeds back preliminary noise estimates. Both schemes are shown to have the same error probability. However, if the channel can be modelled recursively, our new one is computationally less costly