Roundoff noise properties of IIR polynomial predictive filters

Many signals occurring in measurements and instrumentation technology can be well approximated by a polynomial of a low degree. Such signals can be effectively predicted using Polynomial Predictive Filters (PPFs). Although polynomial prediction has traditionally been done using FIR filters, the recently introduced IIR PPF structure appears promising due to its adaptable magnitude response function. When implementing a digital filter with finite precision arithmetic, if is necessary to understand the roundoff noise properties and quantization sensitivity of the filter structure. In this paper, we consider the roundoff noise properties of IIR PPFs. We concentrate on the most commonly occurring length K=2. Both direct form and “augmented FIR” structures are considered. We find that IIR PPFs generally generate high levels of roundoff noise with both filter structures, with the augmented FIR structure generally performing slightly better in most cases. In the light of our study, we recommend IIR PPFs only to be realized using high-precision arithmetic