Re: AW: Bessel-filter problems

[Sergei Steshenko]

--- On Wed, 4/27/11, Chipmuenk <address@hidden> wrote:

> From: Chipmuenk <address@hidden>
> Subject: Re: AW: Bessel-filter problems
> To: address@hidden
> Date: Wednesday, April 27, 2011, 4:12 AM
> Hello Bernd,
> 
> given a sampling rate of 5 GHz (or 1 GHz as well), you can
> only use the
> simplest of DSP in the fastest available technology, i.e.
> you would have to
> design your own ASIC (on which hardware do you plan to do
> your number
> crunching?)
> 
> Even with a custom-specific IC, you would have to use very
> short wordlengths
> to perform multiplications and additions in the given time
> frame.
> 
> Even with a cascaded filter implementation (much more
> robust to coefficient
> and data quantization than direct form filters), I would
> guess you probably
> need coefficients with at least 12 ... 14 bit wordlength
> and about the same
> for your data. 
> 
> If your coefficients are too short, your filter may easily
> become unstable
> (poles outside unit circle) or you may end up with massive
> errors in your
> filter characteristic. Oppenheim / Schafer, "Discrete-time
> signal
> processing" gives a good desription of these effects in
> fixpoint filters.
> 
> What about an analog solution? You're way up in the RF
> domain.
> 
> Good luck,
> 
> Christian
> 
> --
> View this message in context: 
> http://octave.1599824.n4.nabble.com/Bessel-filter-problems-tp3473247p3477805.html
> Sent from the Octave - General mailing list archive at
> Nabble.com.
> _______________________________________________

Maybe it's already been suggested here, but why not run overlapping
windowed FFT transforms on the sampled signal ?

Hundreds of thousands of samples can be easily dealt with using the
approach, and arbitrary filter can be applied.

Arbitrary in a sense amplitude vs frequency and phase vs frequency can
be whatever.

Regards,
  Sergei.