Ablindadaptiveequalizerbasedona Lattice/all-pass Connguration

Networks for the blind channel equalization can be designed as a combination of a prewhitening lter (adjusted by means of 2nd-order statistics) and an inverse all-pass system (adjusted by 4th-order statistics). In this paper an appropriate FIR approximation of inverse all-pass systems is presented. On the other hand the inverse all-pass modelling can be avoided if the minimum phase prewhitening lter is replaced by a lattice connguration where the maximum phase (backward prediction error) output is used for decorrelation. In that case conventional recursive all-pass networks can be applied for phase correction. It will be shown that the convergence speed can be considerably increased due to the reduced number of parameters to be identiied. Under additive noise conditions a solution near the optimum MSE-solution is obtained.