Designing of Two- Channel Causal-stable IIR Filter Banks by Spectral Factorization

摘要

A new method is proposed for designing of two-channel filter banks (FBs) with causal-stable IIR filters. By using IIR filters with a cosine-rolloff transition band, the flatness condition required for two-channel NPR FB is automatically satisfied. Instead of designing the frequency magnitude responses of the analysis filters, the power spectrums of the desired filters are designed by solving a quasi-convex problem. When the solution is found, the analysis filters desired can be obtained by spectral factorization. The polyphase components of the analysis filters are assumed to have an identical denominator to simplify the PR condition. Two-channel NPR IIR FB so obtained has a reasonably low reconstruction error and it can be employed as the initial guess to constrained nonlinear optimization software for designing the PR HR FB.