multiple channel filter bank

On Oct 8, 4:06&#4294967295;pm, emeb <ebromba...@gmail.com> wrote:
> On Oct 7, 6:16&#4294967295;pm, John <sampson...@gmail.com> wrote:
>
>
>
>
>
> > On Oct 7, 6:30&#4294967295;pm, "cinetron" <cinetron@n_o_s_p_a_m.passport.ca>
> > wrote:
>
> > > >cinetron wrote:
>
> > > >> I have a newbie question. I am about to embark on a project for which I
> > > am
> > > >> designing a multiple channel filter bank. This device will have an
> > > input
> > > >> signal from 20 to 10K Hz and and 8 outputs. The outputs will split the
> > > >> signal into 8 frequency bands approx.
> > > >> 20 to 100
> > > >> 100 to 200
> > > >> 200 to 300
> > > >> 300 to 400
> > > >> 400 to 500
> > > >> 500 to 600
> > > >> 600 to 700
> > > >> 700 to 10k
>
> > > >> Since I have never tackled a DSP project I thought I would see how
> > > others
> > > >> might tackle this. I am quite familiar to using PIC mucrocontrollers so
> > > am
> > > >> leaning towards using there dsPICs. I am considering using a dsPIC for
> > > the
> > > >> filters and a CS42518 2 in 8 out CODEC.
> > > >>http://www.cirrus.com/en/products/cs42518-28.html. I want at least 16
> > > bit
> > > >> audio. Another possibility I was thinking was to use 2 ADAU1701 Analog
> > > >> Devices processors as they have 4 built in DACs and their Sigma Studio
> > > >> Software looks quite slick.
>
> > >http://www.analog.com/en/audiovideo-products/audio-signal-processors/...
>
> > > >> My questions with respect to these two solutions:
> > > >> Can the Sigma Studio software handle 4 different filters with 4
> > > outputs.
> > > >> ( I still have to determine my filter characteristics so this may be
> > > hard
> > > >> to answer)
> > > >> Also can the dsPIC do 8 filters simultaneously and speak with the CODEC
> > > in
> > > >> real time.
> > > >> Thanks, these may not be the best solution to the problem so I'm open
> > > to
> > > >> better solutions.
>
> > > >1. For 8 channels of processing at audio sample rate, I would take a
> > > >real DSP like Blackfin-531 or TMS55xx.
>
> > > >2. Codec interfacing to the DSP is no problem.
>
> > > >3. Each band is likely to require two cascaded biquads at least.
>
> > > >4. The processing should be implemented with ~31 bit precision.
>
> > > >5. Do you need to update the filter settings during the run time?
>
> > > >6. dsPICs and Sigma Studios isn't an answer.
>
> > > >Vladimir Vassilevsky
> > > >DSP and Mixed Signal Design Consultant
> > > >http://www.abvolt.com
>
> > > Thanks Vladimir for the suggestion.
> > > Presently I don't have to update the filters at run time but eventually it
> > > would be nice to have that flexibility. I was thinking the processing
> > > overhead of 8 channels wouldn't be huge doing an FIR based filter for
> > > example since the input value doesn't change for each channel. All 8
> > > channels are processing the same input just spitting out different results
> > > based on different coefficients. I'm also considering 2 of these chips from
> > > quickfilter to do the filter work.http://www.quickfiltertech.com/html/content_page.php?content_id=7&
> > > Jim
>
> > Curious, how does a common filter input value save anything when the
> > filters are all different?
>
> > John
>
> if it's a bunch of FIR filters in direct form you can use the same x(n-
> m) pipeline for all of them. Not much of a savings though since you
> still have to do all the multiplies separately for each output SOP.
>
> Eric- Hide quoted text -
>
> - Show quoted text -

Right. Another way to do the filterbank is with FFTs and inverse FFTs
as described here:

www.3db-labs.com/01598092_MultibandFilterbank.pdf

A key thing to figure out is how many FIR taps your bandpass filter
requires and whether the
resulting FFT size can run on your platform.

John