Subject: Re: [linux-audio-dev] Reference amplitudes and LADSPA (again)
From: Paul Davis (pbd_AT_Op.Net)
Date: Fri Feb 01 2002 - 20:09:19 EET
>>In a nutshell I think that there should be a reference 0dB (LADSPA) level,
>>defined as 1.0f.
for whatever reason, its been my assumption all along ...
>A/D converter should output values in range (-1,1); that is
>what I have always assumed.
i don't want to be too pedantic, but "A/D converter" is not a very
useful phrase. nothing that we do in software does A/D or D/A
conversion. i think what you mean is that the source that moves data
to and from an audio interface should clip values to the range -1..+1,
and map them to the relevant (probably integral) values for the h/w
(e.g. -32768.. +32767 for 16 bit h/w). [ even this is not necessarily
where A/D or D/A happens either, since some audio interfaces are 100%
digital).
however, this clipping+mapping not necessarily a good idea. some h/w
might come with an assumption that a 16 bit signal value less than
32767 was 0dB, and that values above this represent +dB values. i
don't know of any audio interfaces that do this, but most HDR systems
and digital mixers either work this way by default, or can be set to
do so.
its therefore unwise to try to map the -1..+1 range of a LADSPA audio
value to anything except to pin 1.0f as 0dB, and anything that follows
directly from this. having said this, all of my code assumes that 0dB
represents "full range" for the relevant bit size, and i can't imagine
people doing anything else for now. notice, however, that if you were
reading sample values from a Mackie D8B or an Alesis M20 ADAT, its
meters might tell you the signal level reached 0dB, but the sample
values never reached 1.0f (or the full range integral value, whichever
you prefer).
obviously, LADSPA hosts and clients are free to deliver any values
they want to, as long as they understand the consequences of those
values falling outside the -1..+1 range.
--p
This archive was generated by hypermail 2b28 : Fri Feb 01 2002 - 20:06:14 EET