"Jay - atldigi" wrote in message
In article , "Arny Krueger"
wrote:
"Tommi" wrote in message
"Arny Krueger" wrote in message
The idea that adding bits does not increase resolution is yet
another popular urban myth about digital. It's similar to the
urban myth that analog has resolution below the noise floor.
So, if you're recording, say, someone's vocals at both 16 and 24
bits, and the peaks are at -6dB to 0dB FS, does the 24 bit recording
represent more accurately the signal in that region than the 16-bit
version?
The 24 bit recording has the capability to represent the signal much
more accurately in *any* range from zero to max, than the 16 bit
recording.
I think you're suffering the myth, Arny. Let me quote from another
thread where Scott Dorsey is trying to explain the same thing that I
am, and I'll and try to explain it yet another way:
In article , (Scott
Dorsey) wrote:
A 16 bit number is significantly
smaller and therefore less precise than a 24 bit number.
Right.
So, in a nutshell. Moving from 16 bit to 24 bit, we have 8 extra
bits per sample to represent the analog wave which is a massive
gain.
Not really. It gives you more dynamic range, which is often wasted
anyway. 96 dB is an awful lot.
Scott was clearly addressing the "massive gain" part of the comment, not the
"8 extra bits per sample". That there are 8 extra bits per sample is an
inarguable fact that we can all agree on.
The 24 bit number is more precise than the 16 bit. True enough. What
that means in audio, however, is that the 24 bit word can describe
smaller values than the 16 bit word, thus signals that are lower in
level.
It means that, but it also means that a 24 bit word describes 255 additional
levels between every pair of levels that can be described by a 16 bit word.
It means both things. Again, this is inarguable and readily observable in
the real world.
The 16 bit number is already describing 96 dB of dynamic range
just fine.
Psychoacoustically 16 bits does do a fine job, however technically the 24
bit word codes 255 additional levels between each pair of levels described
by a 16 bit word.
If you want to carry the precision further and capture
signals that are lower, say to -144 dB, then 24 bits is your ticket.
24 bits also adds resolution in any region between -144 dB and full scale.
The myth is the dynamic equivalent to the argument that 4 samples on a
20kHz sine wave will render it more accurately than 2, and 8 samples
even more so. That's not true either.
That an accrete 24 bit representation of a signal has more resolution at any
level than a 16 bit representation of signal is readily observable in the
real world as soon as you have converters that are sufficiently accurate,
which we now have quite commonly.
Look at how this works with DC levels. In the following examples some
numbers may be off by 1 which is obviously practically irrelevant.
If you have a 16 bit converter with a 1 volt range, there are 65,535 levels
that can be uniquely described between 0 and 1 volt. 1.0000 volts is
represented by 65,535. 0.0000 volts is represented by 0. 0.50000 volts is
represented as 32,767 and there are 32,767 unique levels between 0 and
0.5000 volts and 32,767 more unique levels between 0.5000 volts and 1.000
volts. The smallest voltage 0 volts that can be coded is
3.0518509475997192297128208258309e-5 volts.
If you have a 24 bit converter with a 1 volt range, there are 16,776,960
levels that can be uniquely described between 0 and 1 volt. 1.0000 volts is
represented by 16,776,960. 0.0000 volts is represented by 0. 0.50000 volts
is represented as 8,388,480 and there are 8,388,480 unique levels between 0
and 0.5000 volts and 8388480 more unique levels between 0.5000 volts and
1.000 volts. The smallest voltage 0 volts that can be coded is
5.9605554283970397497520408941787e-8 volts.
Thus the 24 bit representation of voltages between 0 and 1 volt has both
more dynamic range and also more resolution than the 16 bit representation
of the same voltages. The same concept relates to audio signals.