[[email protected]]

I've been reading this thread on prosoundweb:

http://recforums.prosoundweb.com/ind...sg/4918/0/96/0

What is the difference between "sample value" and "signal" that Paul
Frindle is talking about, with respect to digital metering? Aren't
these one and the same? What is the difference between a
"reconstructed" and "unreconstructed" signal? If running a test signal
through a high pass filter increases the sample values, why do some
meters not show this? This is highly non-intuitive stuff, ouch!

Sean B

[Jim Gilliland]

Apologies if this reply appears more than once. My news server appears
not to be responding.

wrote:
I've been reading this thread on prosoundweb:

http://recforums.prosoundweb.com/ind...sg/4918/0/96/0

What is the difference between "sample value" and "signal" that Paul
Frindle is talking about, with respect to digital metering? Aren't
these one and the same? What is the difference between a
"reconstructed" and "unreconstructed" signal? If running a test signal
through a high pass filter increases the sample values, why do some
meters not show this? This is highly non-intuitive stuff, ouch!

A digital signal is made up of mathematical samples that represent the
waveform. In some cases, those samples - even though they fall properly
within the proper range of the digital scale - may represent a signal
that peaks beyond the zero point. In effect, the actual signal peak
falls between two samples, and even though the sample points fall
somewhere under 0dbfs, the actual signal peak would not have done so.

There is nothing inherently wrong with this - the signal is still being
properly represented by the set of samples. But because the DA
converters in most consumer equipment may have headroom limitations that
prevent it from properly reconstructing that signal, it is problematic
for digital audio. The resulting audio may become distorted in playback.

But more important is what may happen if such a signal is created early
in the recording chain, so that it gets routed through various
processing plug-ins. The mathematics involved in EQ, dynamics
processing, and other such effects will take in the original set of
samples and create a new one that may well contain some samples that DO,
in fact, fall above 0dbfs. This doesn't mean that there was anything
"wrong" with the original signal, nor the resulting one, but still the
result may not fit within the range of a normal digital signal.

And, as the thread points out, this can occur even when you think you
are LOWERING the level of the signal.

The solution is to leave yourself plenty of headroom when recording your
original tracks, keep plenty of headroom when processing and mixing your
tracks, and don't increase your level anywhere near full scale until you
are actually producing the final result. And keep in mind that even in
that case, some (most?) playback equipment may produce distortion when
playing back certain sets of samples that properly fall below 0dbfs.

[Meindert Sprang]

"Mike Rivers" wrote in message
oups.com...
By keeping the maximum digital recorded level below -6 dBFS, you leave
some headroom in the D/A converter, sufficient to reproduce the actual
waveform peaks accurately, for most sources. -6 dBFS is a good,
practical working value, not a rule. The rub is that novices working
with a DAW will be disturbed about "tracks not being hot enough"
because -6 dB is half amplitude, so the squiggles will cover less than
half the area of the track's waveform view. It's a visual problem.

On some DAW software you can set the scale of the "squiggles" to
logarithmic, so they resemble VU meter readings. Then the "visual problem"
is gone.

Meindert

[Arny Krueger]

wrote in message
ups.com
I've been reading this thread on prosoundweb:

http://recforums.prosoundweb.com/ind...sg/4918/0/96/0

What is the difference between "sample value" and
"signal" that Paul Frindle is talking about, with respect
to digital metering? Aren't these one and the same?

No, for the reasons described by others. This effect has been discussed here
in detail in the past.

Consider two samples right at or just below FS, and a signal that approaches
the first sample going more or less vertically and up, and approaches the
second sample going more or less vertically, but down. The two samples
describe a signal that peaks at some point in-between the two samples, with
an amplitude that is well beyond FS.

This can happen in the real world. As others have pointed out, you avoid
this by having enough headroom.

What is the difference between a "reconstructed" and
"unreconstructed" signal?

An "unreconstructed" signal is in the digital domain.

A "reconstructed" signal is in the analog domain.

If running a test signal
through a high pass filter increases the sample values,
why do some meters not show this?

A peak-reading meter would show this problem more reliably than an
average-reading meter. A meter that worked on the basis of reporting sample
values would not show this problem. A peak-reading meter operating in the
analog domain might show this problem.

Some if not most DAW software shows this problem if it exists, when
representing waveforms on the screen. I've seen it many times with Audition,
for example.

This is highly non-intuitive stuff, ouch!

It is very rare that this problem causes audible problems, even when it
exists. Again, the general solution is allowing plenty of headroom.

[Scott Dorsey]

What is the difference between "sample value" and "signal" that Paul
Frindle is talking about, with respect to digital metering? Aren't
these one and the same? What is the difference between a
"reconstructed" and "unreconstructed" signal? If running a test signal
through a high pass filter increases the sample values, why do some
meters not show this? This is highly non-intuitive stuff, ouch!

Okay, imagine this. You are sampling a 1 KHz sine wave, at 40 KHz. Now,
the way things happen, you don't happen to be sampling exactly on the peaks
of the waveform. You have a sample before the peak and a sample right after
the peak, but none of the samples are right on the peaks.

Now, you take this file and your normalize it, so the highest values in
the file are at full scale.

When you play it back and it goes through the reconstruction filter, the
peaks of the waveform are now ABOVE the highest voltage coming of the
D/A ladder, because the reconstruction filter builds the peak of the
waveform up so that it matches the original waveform.

As a result, the highest output level on the analogue signal can be actually
higher than the highest sample on the digital side.

My personal feeling is that this should be a non-issue, and if it's a problem
for you, you're trying WAY TOO HARD to make things loud and you should consider
a little bit of headroom. There's nothing wrong with headroom.
--scott

--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Mike Rivers]

Meindert Sprang wrote:

On some DAW software you can set the scale of the "squiggles" to
logarithmic, so they resemble VU meter readings. Then the "visual problem"
is gone.

That's great news. I was wondering when someone was going to think of
that. But still, some people won't be satisfied unless they see the
peaks of the waveform touching the border.