[alsa-devel] wrong decibel data?

Mon Jun 7 11:03:11 CEST 2010

'Twas brillig, and Raymond Yau at 06/06/10 01:12 did gyre and gimble:
> 2010/5/28 Colin Guthrie <gmane at colin.guthr.ie>
> 
>> 'Twas brillig, and Colin Guthrie at 27/05/10 15:43 did gyre and gimble:
>>> PA should play nice with the softvol plugin so I don't think this is the
>>> bit that is at fault.
>>>
>>> I strongly suspect that the reason has already been correctly identified
>>> a while ago, which is that this card considers -48dB silent where as PA
>>> assumes this level is -200dB. I believe it was Raymond who pointed out
>>> the -48dB level in the HDA spec before on this list.
>>>
>>
> 
> if  floating point 0.0 is -inf dB , and 1.0 is 0dB ,
> 
> 0.5 is -6dB , 0.25 is -12 dB , 0.125 is -24dB and 0.0625 is -48dB

This is just a pure mapping from dB->linear, but as this linear mapping
is generally not "natural" there are several different approaches to
presenting this to users. In PA, a cubic mapping is used on top of this
basic conversion, to map to the percentage scale (0.0 to 1.0 if you like).

So I'm not sure what point you're making by providing these numbers. Can
you explain?

> how can PA master volume control at 10~15% equivalent to HDA 's -48dB ?

Not sure what you mean here, but I suspect it's the cubic mapping that
is confusing you.

Here is the function in PA's pulse/volume.c:

 pa_volume_t pa_sw_volume_from_linear(double v) {

    if (v <= 0.0)
        return PA_VOLUME_MUTED;

    /*
     * We use a cubic mapping here, as suggested and discussed here:
     *
     * http://www.robotplanet.dk/audio/audio_gui_design/
     *
http://lists.linuxaudio.org/pipermail/linux-audio-dev/2009-May/thread.html#23151
     *
     * We make sure that the conversion to linear and back yields the
     * same volume value! That's why we need the lround() below!
     */

    return (pa_volume_t) lround(cbrt(v) * PA_VOLUME_NORM);
 }

> Can you provide the pulseaudio log when you change the volume from 100% to
> 0% ?

I can provide the actual log output if you like but here is the output
from the following command:

(for i in 100 95 90 85 80 75 70 65 60 55 50 45 40 35 30 25 20 19 18 17
16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0; do echo
"============================== $i% =============="; amixer set Master
$i%>/dev/null; amixer -c0 get Master; amixer -c0 get PCM; echo
"PulseAudio:"; pacmd list-sinks| grep "index: 11" -A 13; done) 2>&1
>volume-test.txt

This basically sets the volume (via alsa->pulse plugin, but that's just
for convenience!) all the way down to 0. I get more fine grained below
20 as that is the interesting zone.

The output shows the scaling from 100% where both Master and PCM are
0dB, up to the point at around 16% where the Master channel is maxed out
at -46.5dB (0%) and PA starts instead manipulating the PCM control to
get more of it's range (up until this point PCM was only used to gain
more accuracy - i.e. when Master alone did not provide as fine grained a
setting as was needed; in this scenario, PA will use the PCM control to
get more accuracy and, if needed, it will also use software scaling on
top of that: for more info about this works see:
http://pulseaudio.org/wiki/PulseAudioStoleMyVolumes).

As you can see from the output, by the time we reach 2%, both Master and
PCM are fully maxed out at -46.5dB and -51dB respectively. At this point
PA wants a volume of  -101.93dB, so that means that -101.93 -
(-51+-46.5=-96.5) = -5.43 dB is performed in software.

By 1%, the software component of the reduction increases to -23.47dB to
give a total of -119.97dB. By 0% we reach -inf dB

Obviously the fact that the chip basically cuts off any audio when the
Master slider hits 0 (or perhaps when the combined volume reaches -48dB
- it's hard to tell) doesn't really play nicely with the real value of
-inf which we attempt to reach.

I'm not sure where this problem needs to be fixed. Oviously having a
Master and PCM slider whose range is far greater than the value of -48dB
is pointless. This configuration means that there are numerous "zero
point" configurations of the two sliders beyond which any further change
in value is useless. So disregarding PA completely, this setup is not ideal.

When PA is used, the value for -inf is actually configured by the system
and we attempt to scale to -inf (albeit via a cubic mapping from
percentage). If the volume literally cuts out at -48dB when dealing with
the h/w mixers, then there is a problem, but by the same token if the
-48dB level really is more like the silence we want to represent, then
perhaps trying to scale to -inf is pointless in itself and really the
range of the scale used in PA should be adjusted.

I don't know enough about this side of things to comment more accurately
than this, so when LinuxTag is over, hopefully Lennart can comment a bit
on this thread to add his opinions to the mix.

Col

Col

-- 

Colin Guthrie
gmane(at)colin.guthr.ie
http://colin.guthr.ie/

Day Job:
  Tribalogic Limited [http://www.tribalogic.net/]
Open Source:
  Mandriva Linux Contributor [http://www.mandriva.com/]
  PulseAudio Hacker [http://www.pulseaudio.org/]
  Trac Hacker [http://trac.edgewall.org/]
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: volume-test.txt
Url: http://mailman.alsa-project.org/pipermail/alsa-devel/attachments/20100607/60becfa1/attachment-0001.txt 