Novel System Inversion Algorithm With Application to Oversampled Perfect Reconstruction Filter Banks

We derive a novel algorithm for linear (discrete-time) system inversion with decision delay and frequency weighted norm criterion. Complexity of the algorithm grows only linearly with the decision delay. The algorithm also applies to certain singular cases where optimal inverses may be nonunique. In that case, the set of optimal inverses is parametrized. A Scilab implementation of the algorithm is provided. Applications in oversampled perfect reconstruction filter banks are given.