A fractionally spaced DFE with subband decorrelation

We proposed a modification of the "classic" fractionally spaced decision feedback equaliser (FS-DFE) in order to increase the slow convergence rate of the oversampled feed-forward section for LMS type algorithms. This is performed by employing a subband structure for the latter part of the FS-DFE, which results in a decorrelation of the input. We motivate why an oversampled subband decomposition is beneficial and comment on the selection of the filter banks. As an additional benefit, computational savings arise if judicious implementation of the filter banks is combined with the internal decimation and expansion operations in the FS-DFE. Simulations for a severe multipath environment are presented.