An earlier post made me ask...

In a DAW, I can't actually use a signal that goes above '0' because it
clips. But getting as hot a level as is practical when recording is more
important from the standpoint of resolution, no? I hear the term headroom
used and that seems incorrect (though my understanding may be all screwed
up). I want to record at as hot a level as I can without clipping because
then I represent the signal with as fine a RESOLUTION as possible. To my
ears, 24-bit vs 16-bit is a matter of eliminating a level of graininess on
the quieter passages (like fades and reverb tails).

When I record a signal that's using as much of the available 24 bits, I
don't have any more headroom - just a more detailed representation of the
signal and that means that summing and processing can work more accurately
on the backside.

Is that even CLOSE to correct?

TIA

dik

Dik Ledoux wrote:
I want to record at as hot a level as I can without clipping because
then I represent the signal with as fine a RESOLUTION as possible. To my
ears, 24-bit vs 16-bit is a matter of eliminating a level of graininess on
the quieter passages (like fades and reverb tails).

When I record a signal that's using as much of the available 24 bits, I
don't have any more headroom - just a more detailed representation of the
signal and that means that summing and processing can work more accurately
on the backside.

Correct in theory, but for most practical purposes a mic and premp
feeding a 24 bit converter at full digital scale means that the preamp
noise is being digitised with many bits of useless accuracy.

In other words, it's useful to think of the "graininness" of the
resolution as a noise signal that's added to the theoretically ideal
signal. In A-D engineering its proper name is quantization noise. If the
quantization noise is way below the mic/preamp noise, it's irrelevant.

16 bits represent a dynamic range of 96dB.
24 bits is 144dB.
A good mic preamp might have a s/n ratio of 120dB.
If you record with peaks 12 dB below full scale you've still got 22 bits
of resolution which is 124dB - you won't hear any graininess!

--
Anahata
-+- http://www.treewind.co.uk
Home: 01638 720444 Mob: 07976 263827

On Mon, 29 Aug 2005 15:31:53 -0500, "Dik Ledoux"
wrote:

An earlier post made me ask...

In a DAW, I can't actually use a signal that goes above '0' because it
clips. But getting as hot a level as is practical when recording is more
important from the standpoint of resolution, no? I hear the term headroom
used and that seems incorrect (though my understanding may be all screwed
up). I want to record at as hot a level as I can without clipping because
then I represent the signal with as fine a RESOLUTION as possible. To my
ears, 24-bit vs 16-bit is a matter of eliminating a level of graininess on
the quieter passages (like fades and reverb tails).

Surprisingly enough, it's really not. This is a very very common
misconception, and it's one that *seems* to be obvious, but
really isn't true.

When tracking, a noise is added to your desired signal. That
noise, called dither, is adjusted to be just large enough to
completely randomize all of the conversion's graininess.

What's really surprising is that the noise required is
smaller than the smallest possible signal.

Yeah, i know, wacky but true.

What this means is that a (theoretical, idealized) conversion
is *perfect*. Plus a tiny amount of noise.

Only true for perfect conversions, not valid on
alternate Thursdays, etc.

Good fortune,

Chris Hornbeck
"What I love about Jean-Luc Godard is that he is honest, smart,
and has no humility." -butterfinger, reviewing _Pierrot le fou_, 1965

On 8/29/05 4:31 PM, in article , "Dik
Ledoux" wrote:

An earlier post made me ask...

In a DAW, I can't actually use a signal that goes above '0' because it
clips. But getting as hot a level as is practical when recording is more
important from the standpoint of resolution, no?

Yes... And no

IN DIGITAL:
Your RESOLUTION is pretty much set by both your
SAMPLING RATE (frequency resolution) and
BIT RATE (dynamic resolution)

QUESTION YOU ASK: "Do I Have Enough Resolution?"
ANSWER: Anything past 16 bits CAN NOT BE REPRODUCED FOR LISTENING HUMANS in
any real listening situation outside of a severely isolated chamber and
allowing for hearing damage.
So,What Are You Recording?
- a super-crunched electric guitar track...
PROBABLE Max Dynamic Range: 15dB with a crest factor of.. What... 2dB?
(not counting damped-string humbuzzgrackle****)
WHO CARES where the record level is as long as the lights light up somewhere
around -12 dBfs max. you gain NOTHING recording hotter.

- a close-mic snare drum
PROBABLE Max Dynamic range: 90dB with a crest factor of ... What.. 80dB?
As long as it doesn't hit 0dBfs you;re good at 16bits or better... And yeah
you;re gonna compress it later if you haven;t already analog on tracking.

In analog (which is where you start before converting) the same issues
prevail...
you need good flat broad frequency response
(30-15kHz is adequate... More is better... But not MUCH better)

You need a good low noise floor... And since a lo analog noise floor when
using mics takes Some Considerable attention, you worry THERE, not digital.



I hear the term headroom
used and that seems incorrect (though my understanding may be all screwed
up). I want to record at as hot a level as I can without clipping because
then I represent the signal with as fine a RESOLUTION as possible. To my
ears, 24-bit vs 16-bit is a matter of eliminating a level of graininess on
the quieter passages (like fades and reverb tails).

WHAT QUIETER PASSAGES???
If you;re set up to listen at real listening levels,
(Katz refence here... Go read Digital Domain's Bob Katz Monitoring
Levels and claibration essay)
Say 85dB at your ears is -15dBfs... (and that's pretty dang LOUD for
anythign other than classical stuff that spends most of it's time around
-18) You CAN'T HEAR THE TAILS in a normal listening room. (
If you have a THX isolation rated type of mastering room or tracking room or
control room, sure you can tell everybody to hold their breath so you all
can try and analyse the LSB sonics)
If indeed you REALLY CAN hear this stuff, I'm really curious what your
quiet-room noise-floor figures are there! Sounds impressive! It cost
Hollywood more than several handfulls of kilobucks to get that kind of
quiet.


When I record a signal that's using as much of the available 24 bits, I
don't have any more headroom - just a more detailed representation of the
signal and that means that summing and processing can work more accurately
on the backside.

Is that even CLOSE to correct?

Dik Ledoux wrote:
In a DAW, I can't actually use a signal that goes above '0' because it
clips. But getting as hot a level as is practical when recording is more
important from the standpoint of resolution, no? I hear the term headroom
used and that seems incorrect (though my understanding may be all screwed
up). I want to record at as hot a level as I can without clipping because
then I represent the signal with as fine a RESOLUTION as possible. To my
ears, 24-bit vs 16-bit is a matter of eliminating a level of graininess on
the quieter passages (like fades and reverb tails).

No. This is not the 1980s. We have dither. Everything is pretty much
linear across the range, and when it's not linear it's usually so in a
way that cranking the levels up won't help.

When I record a signal that's using as much of the available 24 bits, I
don't have any more headroom - just a more detailed representation of the
signal and that means that summing and processing can work more accurately
on the backside.

No, it's not any more detailed, it just has a lower noise floor. That
is what dither does for you. Without proper converter dither, your
resolution would be affected by operating level. Thank God the days of
having to deal with that crap are over.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

In article ,
"Dik Ledoux" wrote:

When I record a signal that's using as much of the available 24 bits, I
don't have any more headroom

It's not headroom. It's better to think of it as foot room. The top of
the scale, 0dBFS, is the same; full scale is full scale. The extra
"room" is gained at the bottom of the range in the form of a lower noise
floor which gives you a wider available dynamic range.

From a practical standpoint, don't push your luck trying to hit full
scale. Leave some headroom. Today's systems don't need to be run so hot
to stay in the linear range. The old "16 bit" converters were often
lucky to give you 13-14 bits of useable resolution, and dither was not
well-used. In modern systems, leaving a few dB of headroom will not
result in grainy reverb tails etc. This doesn't mean that you should (in
the typical music application) leave 12 or more dB of headroom above
peaks, as the noise floor will indeed become an issue if you print
unnecessarily conservatively. However, a couple dB short of full scale
is far better than several instances of clipping.

--
Jay Frigoletto
Mastersuite
www.promastering.com

But how much resolution do I lose when you have to mix in digital domain 24
tracks recorded at 0db (24bit)?
If I understand it right 0db in a sample is one big number....
When I mix 24 0db tracks I have to make 24 0db tracks become 1 0db track
(the master) (I hope this is clear).
Since I cannot go over 0db in the mix I assume that every track has 1/24 of
resolution or something like that.
This sounds odd to me, where am I wrong in this thought?
In analog I can go over 0db in master buss... (at least +20db) while in
digital I cannot. Is that the point?
F.

"Federico" wrote in message


But how much resolution do I lose when you have to mix in
digital domain 24 tracks recorded at 0db (24bit)?

To review, your individual channels have no more than say 85
dB resolution, and 65 dB is more like it.

24 bit fixed point has 144 dB dynamic range.

32 bit floating point has about 1000 dB dynamic range.

If I understand it right 0db in a sample is one big
number....

which means that there is quite a bit of resolution.

When I mix 24 0db tracks I have to make 24 0db tracks
become 1 0db track (the master) (I hope this is clear).

Which implies attenuating each of the inbound tracks.

Since I cannot go over 0db in the mix I assume that every
track has 1/24 of resolution or something like that.

Actually, the amplitude of uncorrelated signals increases
geometrically, so the resolution is more than that,
depending on the tracks.

Hoever, lets go with 1/24. That is about 22 dB loss of
resolution. Even if you were mixing 16 bit tracks on a 16
bit bus, the available resolution (78 dB dynamic range)
would exceed the probable resolution of each track (65 dB
dynamic range).

This sounds odd to me, where am I wrong in this thought?
In analog I can go over 0db in master buss... (at least
+20db) while in digital I cannot. Is that the point?

Your analog mixer has headroom. Your software mixer may or
may not. But, headroom is just a mind game. It's all about
setting levels to avoid clipping.

"Jay-atldigi" wrote in message
...
In article ,
"Dik Ledoux" wrote:

When I record a signal that's using as much of the available 24 bits, I
don't have any more headroom

It's not headroom. It's better to think of it as foot room. The top of
the scale, 0dBFS, is the same; full scale is full scale. The extra
"room" is gained at the bottom of the range in the form of a lower noise
floor which gives you a wider available dynamic range.

From a practical standpoint, don't push your luck trying to hit full
scale. Leave some headroom. Today's systems don't need to be run so hot
to stay in the linear range. The old "16 bit" converters were often
lucky to give you 13-14 bits of useable resolution, and dither was not
well-used. In modern systems, leaving a few dB of headroom will not
result in grainy reverb tails etc. This doesn't mean that you should (in
the typical music application) leave 12 or more dB of headroom above
peaks, as the noise floor will indeed become an issue if you print
unnecessarily conservatively. However, a couple dB short of full scale
is far better than several instances of clipping.


And then there's the gain structure problem - analog electronics in the
front end struggling with high gains and levels necessary to reach digital
full scale.

There's an additional potential for signal degradation when an analog
console and outboard processing is used to mix such hot signals.

Predrag