Gerhard Doblinger filter bank design

Design of perfect-reconstruction filterbanks

The MATLAB program prfib.m referenced in

Doblinger, G.; , “A Fast Design Method for Perfect-Reconstruction Uniform Cosine-Modulated Filter Banks,” Signal Processing, IEEE Transactions on , vol.60, no.12, pp.6693-6697, Dec. 2012
doi: 10.1109/TSP.2012.2217139
URL of PDF at IEEE Xplore: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6294460&isnumber=6357313

is available here. Note that this program can be used to design both critically sampled and oversampled cosine-modulated filter banks (see demo file test_prfib.m contained in prfib.zip).

Paper Abstract:

In this correspondence, we present a new and fast design algorithm for perfect-reconstruction (PR), maximally decimated, uniform, cosine-modulated filter banks. Perfect reconstruction is obtained within arithmetic machine precision. The new design does not need numerical optimization routines and is significantly faster than a competing method based on second-order cone programming (SOCP). The proposed design algorithm finds the optimum solution by iteratively solving a quadratic programming problem with linear equality constraints. By a special modification of the basic algorithm, we obtain PR filter banks with high stopband attenuations. In addition, fast convergence is verified by designing PR filter banks with up to 128 channels.