[Al Clark]

rickman wrote in news:27189e84-f6c3-4f34-8d1d-
:

On Nov 19, 10:42*pm, (Eric Jacobsen) wrote:
On Fri, 19 Nov 2010 20:32:08 -0500, Randy Yates
wrote:



Mike Rivers writes:

On 11/19/2010 7:52 PM, Randy Yates wrote:

My question is this: What is the definition of dBFS?

Decibels relative to full scale.

Nonsense. All you've given is the meaning of the acronym, not an
engineering definition of the unit. This is similar to stating the
definition of RMS is "root mean square."

[...]
But it's not defined that way.

I'm not asking how it's not defined. I'm asking how it is defined (in a
sensible way).

I think you're asking what color the sky is, and people are telling
you "blue", but you're expecting a wavelength or something, so you're
not accepting the answer.

As you know, dB measurements are always relative to some reference
level. * With dBFS the reference level is Full Scale of the converter
or number system or whatever. * The ratio of the level measured to the
Full Scale level provides the argument for the logarithm, and the
scaled result is dBFS.

If you can do dBm, or dBW, or dBC, you should be able to do dBFS.

The odd part is that you're not seeing this after being told correctly
what it is several times.

I think you are using an inappropriate metaphor. It is more like
Randy is asking what is the elephant like and the blind men are all
telling him something different in these two threads. One person says
0 dBFS is a sample of all 1's and all 0's is -96 dBFS (I won't even go
into what is wrong with that one)! Another describes how a VU meter
works. Yet another tells him 0 dBFS is the peak clipping point (that
one alone actually says somethng).

I'm not sure I want to jump in again at all, but here are a couple of
points:

1. I think I concur with Erik Jacobsen on definitions.

2. 0 dBFS does not mean the level of all 1s as someone suggested. It is the
value of the full scale range of the converter which is virtually always
expressed as twos complement in the audio world. This is either
0x7FFFFF...etc depending or word length for positive peaks or 0x80000...
for negative peaks. If we really wanted to nitpik, I suppose it should be
the positive value which is 1 bit less than the absolue value of full scale
negative. This distinction is meaningless in dB when the bit depth is
large.

3. dBFS does not by itself refer to rms levels at all. In a practical
system, there will be a relationship to the rms level of a sine wave to
dBFS. This is because the crest factor of a sine wave is fixed at 3dB. A
square wave has a crest factor of 0dB, music and voice has a crest factor
3dB in almost all cases. We can relate a the rms level of a +4dBu sinusoid
to an equivalent dBFS value only when we know the conversion. This may be
18 or 20 dB (or something else) and simply establishes the balance between
headroom and low level noise.

4. RMS measurements will also vary. In most cases, we will be using
exponential averaging with some arbitrary time constant. It really doesn't
matter whether we a considering signals in either the digital domain or
analog domain. With exponential averaging, the most recent signals
(samples) have more weight than earlier signals (samples). A long time
constant will yield a measurement that reflects the overall long term
level. A shorter time constant will accent more current events. A sinusoid
would measure the same assuming that the averaging filter has settled to a
steady state value. VU meters are similar where averaging is at least
partially the result of meter balistics. Like a typical low cost
multimeter, they may not be TRMS either.

Al Clark
www.danvillesignal.com







None of the blind men are really right and none are wrong. In the
meantime no coherent picture of the dBFS elephant has emerged and more
disjointed statements are made on the topic.

Another metaphor is that this is a can of worms!

Rick