[hatari-devel] Hatari v1.6 features?

[David Savinkoff]

Hi All,

I've been secretly doing things in regard with digi-sound and have
come up with a few things, including:

- another way to expand the ym-table from 16 to 32 (done)
- ym-table with geometric mean interpolation (done)
- better dc level shifter (will do on a rainy day soon)

A quick 'n dirty improvement can be made simply by changing
#define DC_ADJUST_BUFFERLEN		512		/* must be a power of 2 */
to 8192 in sound.c

There is an emulation bug that makes noise in the XMAS 2004 demo
when Miss Claus shows up. I think that some ym register is not fully
emulated.

Hopefully the per=0 issue is trivial.

Digi-sound is going to get a lot better

David
__________________In response to:_____________________________
Oct 15, 2011 02:26:35 PM, hatari-devel at lists.berlios.de wrote:

Le 15/10/2011 22:02, Anders Eriksson a écrit :
> On Sat, 15 Oct 2011, Eero Tamminen wrote:
>
>> Hi,
>>
>> As Nicolas mentioned that he's planning to release Hatari v1.6
>> before end of the year, I updated the Hatari release notes
>> (FDC stuff needs more details I think).
>>
>>
>> What things are still to be completed before this?
>
>
> Hi,
>
> David Savinkoff sent a patch for much better YM sample sound than in
> Hatari 1.5. It seems to have gone unnoticed by most people (I havn't
> seen it discussed and I don't think it went upstream?).
>
> The audio quality is very close to a real ST, so stuff like sample demos
> and soundtrackers sound a lot better and noise-free.
>
> Here's Davids original mail, I also attach his patch.


Hello

it's true that David patch was not included so far ; I still have to 
ponder if it's better to use this patch or to use volume tables as 
measured on a real ST.

This patch improves linear mixing in the sense that instead of 
summing/averaging volumes it also applies a filter similar to the one in 
the ST. But it doesn't take into account the period of each voice in 
this operation.

Final output signal depends on the volume of the 3 voices, but also on 
the period of each of this voice, as all 3 of them are wired together 
and the feedback effect will be different when freq changes.

So, regarding quality, I think we have this choices (from lower to better) :

  - linear mixing
  - david's version with filter
  - table mixing measured on a real ST

But even the 3rd method is not perfect, because those tables are often 
built for 1 or 2 different frequencies (per=0 and per!=0)

Sound conversion from digital to "analog" (ie the ST's signal) is a 
complex matter, it's quite subjective and hard to build a good model 
that reproduces the wiring "errors" in the ST.


Nicolas