Design of FIR predictors as a tapped cascaded interconnection of subfilters

The use of computationally efficient substructures for predictive FIR filters is proposed. When recursive running-sum blocks are used as subfilters, only a small number of multipliers are needed for predictor implementation. A design algorithm is described, based on using time-domain constraints to define the composition of the impulse response for least-squares optimization. Polynomial and sinusoidal predictors are given as examples, and are compared with the direct form structures.