Re: [foo-devel] sinc interpolation questions

[Gerhard E. <ger...@im...>]

dear martin,

On Aug 27, 2004, at 7:49 AM, martin rumori wrote:
> this means that you are looking tap/2 samples back, but only tap/2-1
> samples in advance.  i think, the latter should read
>
> for (si = 1; si <= tot; si++, ti2 += tinc)

Yes, I think you are right.

> double xMax = _taps / 2 - .5;
>
> if (x > xMax)
> {
>   return 0;
> }
>
> that would mean applying a rectangular window in addition to the hann
> window by zeroing the second half of the last tap region.  again, with
> many taps the difference is almost zero, but for small tapsizes it
> could make a difference.

Yes, I understand. I didn't realize when I wrote that.

> i encountered that when debugging the kaiser window (still a bug in
> the bessel approximation in CVS, but fixed locally).  i used a sweep
> 20-20000 Hz @ 48k, transposed down by 6 octaves.  tested with both the
> windows and different tapsizes, had a look at the sprectrum and got
> somehow strange results.

It is great that you are introducing a Kaiser window. I was not aware 
that this can improve performance with small number of taps. Your stay 
in Banff is very successful in any sense!

> when fixing the abovementioned issues, i got more meaningful results.
> now the kaiser window gives much better results than the hann when
> using smaller tapsizes, i. e. the aliasing reaches the -60dB later
> when getting towards nyquist.  for large tapsizes, however, the
> difference doesn't really matter.

That sounds like a great result!

> with the hann, when fixing either one of the bugs, i got distortion
> immediately, since 1) we are doing a symmetric lookup, but get a zero
> for one of the last taps, or 2) we are getting the correct value for
> the last tap in looking back, but zero for the corresponding one by
> ommiting the lookup for the other side.

you mean if you fix only one of the issues at a time, this is what you 
get?

> does this make sense anyhow or am i missing something?

Makes perfectly sense to me. Looking forward to using the Kaiser window 
and propagating the changes to the MSP-verion of the resampling code. 
Does the Kaiser window also have a down side?

Best regards from Krems,

Gerhard