[ardour-users] peak identifier

[Mark Knecht]

On 10/4/05, Paul Davis <paul at linuxaudiosystems.com> wrote:
> > But think of it this way: If 0db on a channel meter represents 100% of
> > what the system can do then:
> >
> > 1) Is there any value at all to going higher?
>
> when sending data to a physical output port, none.

Right. My mains never go over and in fact are typically a lot lower
than I think other folks mix at. After reading Bob Katz's book and
doing some serious listening I am very convined that a mix at 12db
down with a power amp 12db louder sounds better than letting the value
going into my final D/A get close to rail.

> for intermediate signal routing, sure. there is some loss of precision
> when going over the 0dB level, but very little as long as you don't go
> way over. for bus submixes, its an invaluable and by now standard way to
> avoid digital clipping.

OK, as I thought. Now, how much over?

>
> > 2) I can raise volumes on individual tracks so that they meter above
> > 0db. Am I distorting that audio, or am I just going above some
> > predefined level?
>
> a 32 bit float has 24 bits available for the mantissa, so we can store
> any 24 bit sample value without loss of any precision. if you mix two
> maxed out 24 bit values together, we have to start to use the exponent
> part (its no longer 1), and the precision starts to drop slightly.

OK, so if we go something like 3db over in the mixed signal (1 bit in
the exponent?) and the cost of that is to lose a bit at the back end
of 24 bits thien that isn't a problem at all. Since I'm eventually
going to dither down to 16 bits anyway I think that I Can afford ot
lose 3-5 bits on the bottom and it will likely never be audible at 16
bits, and from a noise POV it probably isn't audible in my studio
environment due to background noise, etc.

Agreed?

>
> > 3) If I'm not distorting that specific audio then how high can it go
> > before I do?
>
> louder than you can imagine before it actually clips. the theoretical
> headroom is enormous.

Right. If it's moved to float at that point and the value is 256 time
bigger, etc., then it's huge.

Assuming I haven't medded up my train of thought description too much
here. I'm not the least bit concerned about the actual numbers but
rather the basic concepts.

Note: My submix buses typically sum 3-8 channels so it can be quite a
bit more than two maxed out signals.

Thanks for responding.

Cheers,
Mark