Re: [Alsa-user] multichannel multiband FFT-based equalizer - README

[Sergei S. <ste...@li...>]

Fons,

we are making rounds

I want JUST to increase amplitude of 8.5 * Fs / N, and
not to touch anything else.

Your suggestion to "also *decrease* the gain in 7 and 10"
will change amplitudes of 7 and 10, and that's exactly what
I want to avoid.

I am trying to say that if one uses ONLY (k * Fs / (2 * N)) frequencies
(0 <= k <= N/2) discrete Fourier transform behaves the same way as
classical continuous Fourier transform.

Whenever other frequencies are used, all kinds of artifacts
come.

So, if you are going to implement the code using
"also *decrease* the gain in 7 and 10" approach, you've
already failed the contest, because I asked

"
please implement in the given 8 points example a filter an equalizer,
changing 1.5 * Fs / 8, but not affecting other frequencies
"
- please pay attention to "but not affecting other frequencies"
part.

Even more, "boost bins 8 and 9" is also against the request
to change gain of only 8.5.

Regards,
  Sergei.

On Tue, 3 Jan 2006 03:01:17 +0100
fons adriaensen <fon...@sk...> wrote:

> Sergei,
> 
> > Whatever implementation of yours should be limited to simple
> > FFT-based equalizer similar to the one I published. That is,
> > data from no more than two adjacent FFT buffers can be used.
> 
> I will take up this challenge, but not immediately. Right now,
> I'm working until almost 3 hours after midnight to finish my
> papers for LAC2006. After that it will be the same to finish
> the code...
> 
> Hint: if you want to boost e.g. 8.5 * Fs / N, you should not only
> boost bins 8 and 9, but also *decrease* the gain in 7 and 10.
> Look at the values of sin(x)/x for x = (i + 0.5) * pi.
> 
> -- 
> FA (going to sleep now)
> 
> 
> 
>