An adaptive DFE for high data rate applications

A non-uniform number representation and filter transformation techniques are used to increase the throughput rate of an adaptive transversal DFE for applications such as HIPERLAN. The DFE input data is non-uniformly quantised and represented by a signed power-of-two (SPT) number. Using this number representation, multipliers can be replaced with barrel shifters and adders. The mean quantisation noise power using SPT input data is examined. The delayed least mean square algorithm (DLMS) is used for training the DFE. The delay in the filter coefficient update, together with transformation techniques, results in a DFE structure, realisable as the cascade of a series of modular sections