Dear good people,

In a nutshell: modulated filter banks[lapped transforms] are discrete time-frequency analysis tools where the sinusoidal waveforms are localized with a prototype filter[window function]. Based on the sheer count of published papers, one would also expect to find some pieces of useful code from the web. If such collections exist, they seem to be hidden from the search engines.

But in case your research/study interests coincide with critically sampled perfect reconstruction systems, the following link might be useful:

http://www.students.tut.fi/~jalhava/lt/intro.html

You can download there a collection of MFB algorithms and some design routines. The www-pages also serve as a tutorial for the LT-toolbox. The m-file package is still under developement so I have not verified the functionality with other MATLAB versions than r2007a (windows).

A major rewrite is under consideration but there still some unsolved problems:

* Is it possible to factorize any 2-channel PR system into lifting steps?
Namely, I can generate such low-delay biorthogonal low/highpass system where MATLAB filt2ls-function fails. Is this a wavelet toolbox "feature" or what?

* Non-linear optimization problem solver for OCTAVE that is comparable with the MATLAB's fmincon.

* 2-D filter banks: What is the fastest "direction" to apply the filtering? Block-wise, column-wise, does it make sense to divide the actual filtering into modulation and polyphase parts?

* Handling the signal edges without allocating new space for the extended data matrix

But anyway, any contribution for this non-profit project is welcome. Improvements, faster code and new features? Please, share your code.

Best subband processing wishes,
Juuso Alhava Start a New Thread