We present an efficient method of designing polynomial predictive IIR
filters via feedback extension of FIR predictive filters. The advantag
e of the W structures over their FIR counterparts is a more desirable
frequency response obtained with considerably smaller memory and compu
tational overhead. In this paper, we show that the IIR structure is ab
le to predict complex-valued signals as well. This is demonstrated by
the prediction of complex Rayleigh fading signal, where the W structur
e is found to be clearly superior to the FIR structure in performance.