[LAU] maximum input level, or normalization and dc offset correction?

Bill Gribble grib at billgribble.com
Fri Jul 12 18:59:03 UTC 2013



On Fri, 2013-07-12 at 14:17 -0400, Ricardus Vincente wrote:

>  Record fairly hot to tape, but don't clip. If you see red lights turn
> it down a bit. That said, recording at 24 bit leaves tons of resolution
> and you don't need to record quite as loudly as we did in the 16 bit
> ADAT days.

I think you are underselling the benefits of 24-bit recording here.
It's not necessary to record "hot" at all.  

AFAICT "best practice" with modern equipment is to basically work as if
you were on an analog console; that is, clipping at no less than +20 dB
with respect to your "0 dB" working level.  That means you record with
signals working at RMS around -20dBFS and then you just don't worry
about clipping unless your signal is truly hugely dynamic, and in that
case you make allowances as necessary with gain and/or limiting. 

I tend to believe anything Mike Rivers says, and his article on gain
staging
(http://mikeriversaudio.files.wordpress.com/2010/10/gainstructure.pdf)
indicates that you really do need 20 dB of headroom over RMS to deal
with recording real music (see the "Crest Factor" section in the piece
mentioned above, which is mandatory reading in any case).  

3dB corresponds to roughly 1 bit of amplitude, so working at -20dBFS
means you are giving up about 6-7 bits of resolution to headroom,
leaving something like 17 bits in your "working" range, plus whatever
headroom you use for transients.  In your final production step, during
mastering, you will likely be exporting to 16-bit PCM for CD, so you
aren't really giving up anything except the stress of constantly
watching for overs. 

Practically, working this way in a DAW takes some getting used to,
because the "waveform" display of your tracks at -20dBFS looks pretty
much like a flat line.  You don't get those pretty wave shapes until you
zoom in. 

I have seen arguments, which I can neither refute nor support, that
working within the last 1-2 db of your available dynamic range on input
may compromise signal quality as well, as you are (for cheaply designed
prosumer hardware) pushing against the voltage rails of the analog
circuitry in front of the A/D and possibly running into current
limitations.  This is "plausible" but could also be total bunk, and I
wouldn't take either side in a bar bet. 

Thanks,
Bill Gribble 






More information about the Linux-audio-user mailing list