Decision feedback equalization using fractionally spaced feedback filtering and vector quantization

We propose a generalization of the conventional fractionally spaced decision feedback equalization (FS-DFE) structure which replaces the symbol spaced feedback filter with a fractionally spaced feedback filter and replaces the scalar decision quantizer with a vector decision quantizer. We draw two significant findings regarding the design and performance of the new structure. First, when optimized under the commonly used correct past decision assumption the new structure is functionally equivalent to an appropriately configured conventional FS-DFE. However, we show that this equivalence, like many other purportedly optimal designs in the DFE literature, is a misleading manifestation of the strong assumption of correct past decisions. By taking due regard for the likelihood of decision errors (and error propagation) are demonstrate that the new structure when optimized using a maximum a posteriori criterion outperforms the conventional FS-DFE