Re: Suboctave filterbank?

Vladimir Vassilevsky wrote:
Andor wrote:
Is there an efficient method to implement a filterbank with suboctave
spacing of the bands, i.e. log(frequency)? The perfect reconstruction is
not required; low memory requirement is the primary goal.
A bunch of IIR bandpass filters?

Yes, filters, or Goertzels, or doing something like a slow Fourier
transform. The more efficient approach could be cascading several stages
of FFT and decimation.

Yup, this is sometimes known as "Bounded Q Transform" (in contrast to
"Constant Q").


Also, check out Judy Brown's "Calculation of a Constant Q Spectral
Transform" and "An Efficient Algorithm for the Calculation of a
Constant Q Transform" from her webpage:

http://www.wellesley.edu/Physics/brown/jbrown.html

Her algorithm is FFT-based, which isn't exactly low-memory.

The second article presents a very elegant idea!
Thank you, Andor.

I learned about this algorithm here in comp.dsp :-).

Regards,
Andor
.