Re: [hatari-devel] Improved YM2149 cycle accuracy

[Nicolas Pomarède <npomarede@xxxxxxxxxxxx>]

[cc-ing hatari-devel, please respond to the ML not to me directly :) ]

Le 28/07/2021 à 18:05, David Savinkoff a écrit :
> On Wed, Jul 28, 2021 at 6:26 AM Nicolas Pomarède wrote:

> > even if IIR convolves many samples, it does so only every 5.66 input
> > samples in the case of YM2149_Next_Resample_Nearest(), so you loose some
> > information between those 5.66 samples that can be meaningful when
> > period is <= 6.
> Hopefully when period is <= 6, the filter has attenuated some of the
> high frequency
> so that it aliases less without filtering to much of the desirable frequencies.

I don't agree because YM2149_Next_Resample_Nearest() will only process 
one input sample every 5.66 samples. So these samples might be 
attenuated by the filter, but the skipped part (the rest of the 5.66 
samples) will not be taken into account at all while it might contain 
valuable information.
For sound period <= 6 I really think that YM2149_Next_Resample_Nearest() 
is not good enough, especially for per=0 or per=1, as used in several 
digi replay routines.

> > > > This might be OK when emulating YM2149 for sound having period>=6, but
> > > > with very high freq sound, this gives bad results.
> > > >
> > > > IIRC this was hearable in the game Lethal Xcess in STF mode where they
> > > > play digi sound by setting period=0 on one voice and modulating volume
> > > > for this voice. In that case YM2149_Next_Resample_Weighted_Average_N()
> > > > gives much better result as it uses all samples from the 250kHz input
> > > > signal.
> > > >
> > > > Nicolas
> > >
> > >   From the snippet of code from sound.c below:
> > > - LowPassFilter(sample); is supposed to sound like a ST or STF
> > > - PWMaliasFilter (sample); is supposed to sound like a STE
> > >
> > >           /* Apply low pass filter ? */
> > >           if ( YM2149_LPF_Filter == YM2149_LPF_FILTER_LPF_STF )
> > >               sample = LowPassFilter ( sample );
> > >           else if ( YM2149_LPF_Filter == YM2149_LPF_FILTER_PWM )
> > >               sample = PWMaliasFilter ( sample );
> > >
> > >   From what you say... I suppose that PWMaliasFilter(sample); is ineffective
> > > and can be removed.
> >
> > I'm not sure it should be removed. There's this part in audio.c
> OK
> >                   if ( Config_IsMachineST() && nAudioFrequency >= 40000 )
> >                           YM2149_LPF_Filter = YM2149_LPF_FILTER_LPF_STF;
> >                   else
> >                           YM2149_LPF_Filter = YM2149_LPF_FILTER_PWM;
> >
> > so if PWMaliasFilter() is supposed to sound more like an STE, it should
> > be kept.
> By all means (sometimes I have opposing ideas in quantum entangled states).

OK, keep the PW filter for STE.

> > > However, LowPassFilter(sample); is supposed to give the charactistic ST
> > > sound (less sharp). This is the same ST sound that everyone thought was OK
> > > a few years ago. I'll calculate some sharper sounding coefficients soon.
> >
> > I don't disagree, it's just that the change is hearable between hatari
> > 2.3 and current devel version when I apply the patch you sent on 26/7/2021.
> > Maybe sound was too "sharp" before, I can't tell, some YM2149 musicians
> > could give their opinion on this compared to a real STF.
> Maybe the existing antialiasing algorithms are filtering some of the audio
> frequencies, and making the sound less sharp. The STE/STF filters could
> be adjusted to compensate to some extent.

We could have some kind of bass/trebble filter inside Hatari's UI so 
everyone one could adapt the filter as he like. But that would not be on 
top of my TODO list   for now ;)

Nicolas