[glen herrmannsfeldt]

In comp.dsp davew wrote:
(snip)

The VU meter is basically a bridge rectifier followed by a low pass
filter. So it's mean rectified, not mean squared. So a 1dB
difference for pure tone.

Well, you can just change the numbers on the scale to represent
the squared value. It is logarithmic (dB) in either case.

We don't tend to use rms or mean whatever
when talking about audio levels though, we just say "level" and that
seems to be good enough. It's understood that when you reach 0dBFS
you're in trouble shortly thereafter.

In EE, RMS can mean different things. In some cases, it is
the value that a sine would give, with the specified RMS value.
That is, for peak-to-peak reading VTVM or mean absolute value
reading analog meters, the calibration is for a sine.

The only reliable means to know when the limit has been reached is the
peak indicator (and then only if it's true peak which was the subject
of a recent thread or two). As far as a reasonable measure of
loudness, neither VU or peak or PPM are good enough, just a guide.

But even with true peak, for live recording it is too late by
the time you find the peak.

This has all changed no with loudness metering, but that's another
subject.

-- glen

[Steve Pope]

Al Clark wrote:

0 dBFS is a digital specification that represents the maximum level that a
data converter can convert. For example 0x7FFFFF... or 0x800000.. assuming
twos complement.

It follows that the level of all signals will be = 0 dBFS

It has nothing to do with the rms level at all.

I woulda said that a 0 dBFS signal has the RMS level of a sine wave
that just barely doesn't clip a converter (or, a hardlimited channel;
it does not need to be a converter).

(This is an important concept, of sorts, in that is shows that
an N bit converter has a full-scale-signal to quantization noise
ratio of 6*N + 2 dB, not the 6*N + 5 dB that some texts claim.)

One can debate these things. Most outcomes of such debates are
equivalent within a factor of two.

Steve

[Steve Pope]

In article , Randy Yates wrote:

Why do you say the VU measurements aren't RMS? Because of the meter
ballistics?

That's what I would traditionally say. A steady-state input into
a VU measurment gives a value that should correspond to RMS,
or at least average absolute value if it's a cheap VU meter.
But a dynamic input into a VU meter would have peaks attenuated
from RMS.

When I designed consoles, we chose a compromise --peakier than VU,
not as peaky as instant RMS. Got no complaints.

Don't make such meter decisions without consulting marketing, and
figuring out what the customers want. There is no one true way.

Steve

[glen herrmannsfeldt]

In comp.dsp Scott Dorsey wrote:
(snip)

It does seem that the FS applies to measuring devices, either
analog or digital meters.

NO. FS applies ONLY to digital system. When all the bits are
set to 1, the meter goes to FS.

Well, in at least one place FS is used for analog meters, and
that is the sensitivity of an analog voltmeter, commonly
specified in ohms/volt FS.

-- glen

[Don Pearce[_3_]]

On Sat, 20 Nov 2010 08:03:14 +0000 (UTC),
(Steve Pope) wrote:

Al Clark wrote:

0 dBFS is a digital specification that represents the maximum level that a
data converter can convert. For example 0x7FFFFF... or 0x800000.. assuming
twos complement.

It follows that the level of all signals will be = 0 dBFS

It has nothing to do with the rms level at all.

I woulda said that a 0 dBFS signal has the RMS level of a sine wave
that just barely doesn't clip a converter (or, a hardlimited channel;
it does not need to be a converter).

(This is an important concept, of sorts, in that is shows that
an N bit converter has a full-scale-signal to quantization noise
ratio of 6*N + 2 dB, not the 6*N + 5 dB that some texts claim.)

One can debate these things. Most outcomes of such debates are
equivalent within a factor of two.

Steve

Here's how the maths works.

Lets call the clipping point 0dB.
The biggest sine wave it can hold is -3dB RMS. (peaks just clip).
The lowest bit level is - 16 * 20log(2), or -96.3dB

Because the converter is perfect, the noise is TPD, which has an RMS
value 4.8dB below the 1 bit peak. So the noise level is -101.1dB

So the dynamic range is 101.1dB -3dB +4.8 or 98.1dB. That is the range
from a just-clipping sine wave, to a signal equal in RMS amplitude to
the noise.

d

[Mike Rivers]

In comp.dsp wrote:

Now we know when calibrating with a tone, 0 VU needs to be something
like -20dBFS to avoid clipping when audio at 0 VU replaces the tone.

On 11/20/2010 2:54 AM, glen herrmannsfeldt wrote:

I thought for CDs the number is supposed to be -12dB, but I am
not sure what is supposed to be at -12dB. Some CD recorders
have an arrow at -12dB. (And the meter isn't likely to
have VU ballistics.)

It's just a number. It isn't "supposed" to be anything. I
suppose that 12 dB of headroom would be reasonable if you
were compressing everything that you recorded, as is the
case with a lot of pop music today ("the loudness war").
When you reduce the dynamic range, you reduce the need for a
lot of headroom since you have less uncertainty about the
peak level.

I sometimes record live high-school orchestra concerts.
Because it is hard to know the level, I record 24 bit, then
find the peak and RMS of each track. Then I figure out how many
bits to scale each track by so that peaks stay below FS, and
they should sound about right together.

I suppose that's one way of mixing when you aren't listening
and don't have controls handy. From a computer standpoint,
it makes a certain degree of sense, but from a musical and
artistic standpoint, it's absurd.


--
"Today's production equipment is IT based and cannot be
operated without a passing knowledge of computing, although
it seems that it can be operated without a passing knowledge
of audio." - John Watkinson

http://mikeriversaudio.wordpress.com - useful and
interesting audio stuff

[Mike Rivers]

On 11/19/2010 9:07 PM, Scott Dorsey wrote:

I don't know what a Fs/4 sine wave is.

I assumed he was talking about the frequency of the test
waveform, one fourth of the sample rate. Perhaps he's
looking at something tricky here (remember, this thread has
drifted over to comp.dsp where they know nothing about music
or audio - note my tag line), sampling a single frequency at
an integral ratio of the sample rate so that every sample
occurs at the same point on the waveform. That could lead
someone who doesn't believe in the sampling theorem to the
conclusion that the reconstructed waveform would be at the
wrong level.


--
"Today's production equipment is IT based and cannot be
operated without a passing knowledge of computing, although
it seems that it can be operated without a passing knowledge
of audio." - John Watkinson

http://mikeriversaudio.wordpress.com - useful and
interesting audio stuff

[Mike Rivers]

On 11/19/2010 10:44 PM, Randy Yates wrote:

What units would a typical professional digital audio system use to
measure RMS values of digital signals?

They wouldn't, because they don't care. a professional (or
even amateur) digital audio system doesn't care what the
level is until it reaches 0 dBFS.

You couldn't really be sure you were calculating the RMS
value correctly based on looking at individual samples since
there's a good chance that none of the samples occurred at
the peak of the waveform. You'd have to convert the digital
samples back to analog in order to accurately reconstruct
the waveform. I suppose there's an arithmetic way to do
that, but I'll leave it to the computer guys to figure that
out.

If I wanted to know the RMS value of a digital signal, I'd
play it back through a D/A converter with a known
relationship between volts and bits, measure the voltage
with an RMS voltmeter, and then convert that back to bits.

But I still don't understand your real question. I can read
the words you're writing, but I can't get the significance
of either the question or the answer.

--
"Today's production equipment is IT based and cannot be
operated without a passing knowledge of computing, although
it seems that it can be operated without a passing knowledge
of audio." - John Watkinson

http://mikeriversaudio.wordpress.com - useful and
interesting audio stuff

[Don Pearce[_3_]]

On Sat, 20 Nov 2010 06:50:53 -0500, Mike Rivers
wrote:

On 11/19/2010 9:07 PM, Scott Dorsey wrote:

I don't know what a Fs/4 sine wave is.

I assumed he was talking about the frequency of the test
waveform, one fourth of the sample rate. Perhaps he's
looking at something tricky here (remember, this thread has
drifted over to comp.dsp where they know nothing about music
or audio - note my tag line), sampling a single frequency at
an integral ratio of the sample rate so that every sample
occurs at the same point on the waveform. That could lead
someone who doesn't believe in the sampling theorem to the
conclusion that the reconstructed waveform would be at the
wrong level.

Sampling at an integer multiple of the frequency just makes the
process clean. There is no leakage between frequency bins, hence no
need for a windowing function to get clean sidebands. As you say, it
certainly doesn't result in wrong levels. In fact I would say that
where you have the option, it is the right choice to make.

I can see that to somebody unfamiliar with sampling theory, they could
assume that because - say - the peaks of the waveform never get
sampled, they will be missed. Not so, of course.

d

[Mike Rivers]

On 11/20/2010 12:00 AM, Randy Yates wrote:

What units would a typical professional digital audio system use to
measure RMS values of digital signals?
OK, well here are three examples:

http://www.digitalsignallabs.com/m-1.png

You'll note that all of the meters in your examples have
their scale calibrated in dBFS. Hence my answer several
messages back that you dismissed as "no information" is
exactly correct.

Where you have a digital meter that indicates both peak and
"RMS" they're both displayed on the same scale. The program
that calculates the position of the RMS meter bar has plenty
of time to look at samples and can make a reasonable
estimate of the RMS value regardless of the wave shape.

The idea behind having such a meter is not as a guide for
recording level, available headroom, or dynamic range, but
rather to give an indication of the perceived loudness at a
given time. It's somewhat useful to mastering engineers
because it gives them a sense of how heavily the program
material has been compressed in an attempt to get the
average/RMS/perceived volume level closer to the peak volume
level.

--
"Today's production equipment is IT based and cannot be
operated without a passing knowledge of computing, although
it seems that it can be operated without a passing knowledge
of audio." - John Watkinson

http://mikeriversaudio.wordpress.com - useful and
interesting audio stuff

[Scott Dorsey]

In article , Randy Yates wrote:

So to infer from your response, one definition of dBFS would be
something like this:

dBFS = 20 * log_10(XPEAK / FSPEAK),

Yes, I have already said that three times now in three different ways.

What units would a typical professional digital audio system use to
measure RMS values of digital signals?

I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

I have seen dozens of different time-averaged levels displayed in the
digital domain but none of them were really RMS.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Scott Dorsey]

glen herrmannsfeldt wrote:

Now we know when calibrating with a tone, 0 VU needs to be something
like -20dBFS to avoid clipping when audio at 0 VU replaces the tone.

I thought for CDs the number is supposed to be -12dB, but I am
not sure what is supposed to be at -12dB. Some CD recorders
have an arrow at -12dB. (And the meter isn't likely to
have VU ballistics.)

That's something different. The CD recorder meter, like almost all digital
meters, is peak-reading. If you set your average levels so that they peak
at -12dBFS, that gives you some headroom for an unexpected crescendo. It
is a good rule of thumb. It is, however, only a rule of thumb.
--scott


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

[Scott Dorsey]

Mike Rivers wrote:
On 11/20/2010 12:00 AM, Randy Yates wrote:

What units would a typical professional digital audio system use to
measure RMS values of digital signals?
OK, well here are three examples:

http://www.digitalsignallabs.com/m-1.png

You'll note that all of the meters in your examples have
their scale calibrated in dBFS. Hence my answer several
messages back that you dismissed as "no information" is
exactly correct.

Where you have a digital meter that indicates both peak and
"RMS" they're both displayed on the same scale. The program
that calculates the position of the RMS meter bar has plenty
of time to look at samples and can make a reasonable
estimate of the RMS value regardless of the wave shape.

Note that those "RMS" measurements aren't RMS at all but are
instantaneous weighted averages of various sorts.

And you'll note that they won't match one another at all on
different software or devices (UNLESS they match the ITU
standard, which lots of systems including Pro Tools do not).

If two devices give different values, which is correct?
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[KRAS]

Why does this have different than analog just because it is digital?

dBFS defines the reference for calculating the dB's just like dBV or
dBu or dBSPL or dBHL or any other dB measure. It does not define the
measure - only the reference.

0dBV is defined as the reference being 1V.

any other dBV is 20*log10(abs(xV/1V))

-3dBVrms, 0dBVpk, and +6dBVpkpk all describe the same sine wave - the
dBV describes the reference value while the qualifier rms, pk, and
pkpk describes the type of measure.

The confusion may because dBFS defines the reference level as the
maximum magnitude that can be digitally represented in a system, the
point of digital clipping. But you would still have to use the
qualifier to describe your measurement as pk, rms, or pkpk. The
reference only tells you how you are dividing to get the dB scale - it
does not tell you what your measurement type was.

dBFS = 20*log10(abs(x#/max#))

-3dBFSrms, 0dBFSpk, and +6dBFSpkpk describe the same maximum digital
sine wave - and if 0dBFS = 0dBV in the D2A/A2D calibration it has
equivalent meaning in the analog world of sine wave at analog
clipping Though not all D2A/A2D have that convenient of calibration,
nor is it likely the analog trim matches the analog clip so you can
certainly have more peak analog signal than 0dBV. And if you want
to describe that signal you would give the measure relative to 0V and
the qualifier for the type of measure.

[Don Pearce[_3_]]

On Sat, 20 Nov 2010 11:28:45 -0500, Dick Pierce
wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

But if you have two samples you can have an RMS value with a meaning,
albeit a pretty meaningless meaning. Within the terms of the act, one
would assume a complete cycle of waveform has been measured to reach a
conclusion about its RMS level.

d

[Steve King]

"Randy Yates" wrote in message
...
| Al Clark writes:
|
| 0 dBFS is a digital specification that represents the maximum level that
a
| data converter can convert. For example 0x7FFFFF... or 0x800000..
assuming
| twos complement.
|
| It follows that the level of all signals will be = 0 dBFS
|
| It has nothing to do with the rms level at all.
|
| The relationship between nominal rms levels and dBFS is loose.
| The more bits you assign for headroom, the less bits you have for low
| levels.
|
| A common professional audio tradeoff is 4dBu = -18dBFS. This would mean
| that a +22dBu sine wave would just fit into the converter range without
| clipping.
|
| It is also common that 0dBu = -18dBFS. This means the maximum input
level
| is +18dBu.
|
| Hey Al,
|
| I'm trying hard to see an answer to my question in what you wrote and
| failing.
|
| Let me respond to you with this question: If you had a meter that
| -24 dBFS with a Fs/4 sine wave, what would the peak value of the
| sine wave be?
|
| --Randy
|
|
|
|
| Al Clark
| www.danvillesignal.com
|
|
|
|
| Rick
|
|
| On Nov 19, 5:09 pm, Randy Yates wrote:
| Cross-posting to comp.dsp.
|
| --RY
|
|
|
| Randy Yates writes:
| Also, what reference level does an analog peak-reading meter
| use?
|
| --Randy
|
| Randy Yates writes:
|
| Hi,
|
| Some had responded here to my recent inquiry on levels that dBFS is
a
| peak measurement.
|
| If an RMS measurement needs to be made for a digital signal (i.e.,
on
| a
| digital mixing console or a ProTools plugin), what units are
| utilized?
| I
| thought they were dBFS, i.e., that dBFS was an RMS measurement.
| Apparently I am incorrect. Somebody please set me straight.
|
| --
| Randy Yates % "Ticket to the m
| oon, flight leaves here today
| Digital Signal Labs % from Satellite 2"
| % 'Ticket To The Moon'http://w
| ww.digitalsignallabs.com% *Time*, Electric Light Orchestra
|
|
|
| --
| Randy Yates % "She has an IQ of 1001, she has a
jumpsuit
| Digital Signal Labs % on, and she's also a
telephone."
| %
| http://www.digitalsignallabs.com % 'Yours Truly, 2095', *Time*, ELO

Randy, I think you and Bill Graham ought to get together. There's nothing
worse than a troll who doesn't think he's a troll.

Steve King

[Scott Dorsey]

Dick Pierce wrote:
Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! Mr. Pierce wins the kewpie doll!
--scott

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

[Mark]

But, regardless, there aren't any "different versions" of
RMS, just the one:

* * sqrt(sum(X1^2 + X2^2 ... +Xn^2) / n)



but there are different values of n...

and for a sine tone, the value of n doesn't matter because a tone is
the same all the time....., but for real audio the value of n does
matter and that is the crux of the discussion..

Mark

[rickman]

On Nov 20, 12:02*pm, (Scott Dorsey) wrote:
Dick Pierce wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! *Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. What is the instantaneous RMS value of -1...
+1. RMS doesn't have to be an integral or a sum. b

Rick

[Don Pearce[_3_]]

On Sat, 20 Nov 2010 11:43:03 -0800 (PST), rickman
wrote:

On Nov 20, 12:02*pm, (Scott Dorsey) wrote:
Dick Pierce wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! *Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. What is the instantaneous RMS value of -1...
+1. RMS doesn't have to be an integral or a sum. b

Rick

It isn't. It is the square root of the mean of the squares. In the
case of -1, +1 it is 1.

The inclusion of the term "mean" says that there must be at least two
measurements - any fewer and you can't have a mean.

d

[Randy Yates]

(Scott Dorsey) writes:

In article , Randy Yates wrote:

If an RMS measurement needs to be made for a digital signal (i.e., on a
digital mixing console or a ProTools plugin), what units are utilized? I
thought they were dBFS, i.e., that dBFS was an RMS measurement.
Apparently I am incorrect. Somebody please set me straight.

That depends entirely on which averaging standard you decide to use.
Most common is LKFS according to ITU BS.1771 loudness standard. You
will never, never see this in the US, but RTW standalone meters can
display it.

Scott, sorry but I didn't see this until just today. Thanks. In
searching for info on BS.1771 I also found this paper from Grim Audio,
which, at a cursory glance, looks like it touches on many of the same
issues I've been asking about here.

http://www.grimmaudio.com/whitepaper...%20journal.pdf

--
Randy Yates % "With time with what you've learned,
Digital Signal Labs % they'll kiss the ground you walk
% upon."
http://www.digitalsignallabs.com % '21st Century Man', *Time*, ELO

[rickman]

On Nov 20, 2:52*pm, (Don Pearce) wrote:
On Sat, 20 Nov 2010 11:43:03 -0800 (PST), rickman
wrote:



On Nov 20, 12:02 pm, (Scott Dorsey) wrote:
Dick Pierce wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. *What is the instantaneous RMS value of -1...
+1. *RMS doesn't have to be an integral or a sum. b

Rick

It isn't. It is the square root of the mean of the squares. In the
case of -1, +1 it is 1.

The inclusion of the term "mean" says that there must be at least two
measurements - any fewer and you can't have a mean.

d

I don't recall any such restriction on N. Certainly the measurement
has just as much meaning with N = 1 as any greater N. Think of it as
a limit as N approaches 1 then. The point is that it is as valid a
measurement for a single point as it is for many points. It
represents the equivalent voltage that would produce the same power as
DC of the same voltage.

Rick

[Don Pearce[_3_]]

On Sat, 20 Nov 2010 12:53:16 -0800 (PST), rickman
wrote:

On Nov 20, 2:52*pm, (Don Pearce) wrote:
On Sat, 20 Nov 2010 11:43:03 -0800 (PST), rickman
wrote:



On Nov 20, 12:02 pm, (Scott Dorsey) wrote:
Dick Pierce wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. *What is the instantaneous RMS value of -1...
+1. *RMS doesn't have to be an integral or a sum. b

Rick

It isn't. It is the square root of the mean of the squares. In the
case of -1, +1 it is 1.

The inclusion of the term "mean" says that there must be at least two
measurements - any fewer and you can't have a mean.

d

I don't recall any such restriction on N. Certainly the measurement
has just as much meaning with N = 1 as any greater N. Think of it as
a limit as N approaches 1 then. The point is that it is as valid a
measurement for a single point as it is for many points. It
represents the equivalent voltage that would produce the same power as
DC of the same voltage.


When you do it for a single point, the term RMS ceases to have
meaning. For a single point it is just the voltage. For two points and
above, RMS volts times RMS current give average power.

d

[Les Cargill]

rickman wrote:
On Nov 20, 2:52 pm, (Don Pearce) wrote:
On Sat, 20 Nov 2010 11:43:03 -0800 (PST),
wrote:



On Nov 20, 12:02 pm, (Scott Dorsey) wrote:
Dick wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. What is the instantaneous RMS value of -1...
+1. RMS doesn't have to be an integral or a sum. b

Rick

It isn't. It is the square root of the mean of the squares. In the
case of -1, +1 it is 1.

The inclusion of the term "mean" says that there must be at least two
measurements - any fewer and you can't have a mean.

d

I don't recall any such restriction on N. Certainly the measurement
has just as much meaning with N = 1 as any greater N. Think of it as
a limit as N approaches 1 then.

N is discrete, not continuous. That's the problem. You can't have half
a sample... there is no application of limits. N=1 is a degenerate case.

The point is that it is as valid a
measurement for a single point as it is for many points. It
represents the equivalent voltage that would produce the same power as
DC of the same voltage.

Rick

Right. It's proportional to the instantaneous voltage measure. But it's
degenerate as a weighted vector measure.

--
Les Cargill

[Randy Yates]

(Don Pearce) writes:
[...]
When you do it for a single point, the term RMS ceases to have
meaning. For a single point it is just the voltage.

Correction: it is the magnitude of the voltage.

I won't even get into whether this is a deterministic or random signal
and, if the latter, how the mean is analytically defined independent of
time...
--
Randy Yates % "The dreamer, the unwoken fool -
Digital Signal Labs % in dreams, no pain will kiss the brow..."
%
http://www.digitalsignallabs.com % 'Eldorado Overture', *Eldorado*, ELO

[Al Clark]

(Don Pearce) wrote in
:

On Sat, 20 Nov 2010 12:53:16 -0800 (PST), rickman
wrote:

On Nov 20, 2:52*pm, (Don Pearce) wrote:
On Sat, 20 Nov 2010 11:43:03 -0800 (PST), rickman
wrote:



On Nov 20, 12:02 pm, (Scott Dorsey) wrote:
Dick Pierce wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS
values ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. *What is the instantaneous RMS value of
-1... +1. *RMS doesn't have to be an integral or a sum. b

Rick

It isn't. It is the square root of the mean of the squares. In the
case of -1, +1 it is 1.

The inclusion of the term "mean" says that there must be at least two
measurements - any fewer and you can't have a mean.

d

I don't recall any such restriction on N. Certainly the measurement
has just as much meaning with N = 1 as any greater N. Think of it as
a limit as N approaches 1 then. The point is that it is as valid a
measurement for a single point as it is for many points. It
represents the equivalent voltage that would produce the same power as
DC of the same voltage.


When you do it for a single point, the term RMS ceases to have
meaning. For a single point it is just the voltage. For two points and
above, RMS volts times RMS current give average power.

d

A mean of 1 sample is valid. RMS of a one sample measurement is technically
valid as well, but perhaps not particularly useful.

OTOH, If you knew that the signal was DC, a single sample might not be
meaningless at all. With an AC signal, even a small number of samples may
not yield a particularly good rms value.

This issue doesn't begin to tradeoff exponential versus linear averaging,
which changes the result as well. A very fast exponential averaging time
approaches the 1 sample case.

If you have an issue with exponential averaging, you can send me all your
multimeters

Al Clark
www.danvillsignal.com

[Peter Larsen[_3_]]

glen herrmannsfeldt wrote:

It seems to me that there is still some uncertainty in the
meaning of dBFS.

No. There are 10 kind of readers here, those that understand binary and
those that do not.

Well, consider that CDs are considered to have 96dB (or some
similar number) of dynamic range. That is comparing a full
scale signal (just about impossible in a live recording)

Oh no, what is difficult to some is to stay in the comfy -10 to -5 zone re.
FS instead of being at 0 dB FS for a number of consecutive samples.

-- glen'

Kind regards

Peter Larsen

[Ian Bell[_2_]]

Randy Yates wrote:
Hi,

Some had responded here to my recent inquiry on levels that dBFS is a
peak measurement.

If an RMS measurement needs to be made for a digital signal (i.e., on a
digital mixing console or a ProTools plugin), what units are utilized? I
thought they were dBFS, i.e., that dBFS was an RMS measurement.
Apparently I am incorrect. Somebody please set me straight.


dBFS is NOT a measurement method (peak or rms) but a specification for a signal
level. Unlike dBm, dBu and dBV is has NO SPECIFIC PHYSICAL VALUE - it is simply
the largest value that a digital system can represent.

Cheers

ian

[rickman]

On Nov 20, 3:56*pm, (Don Pearce) wrote:
On Sat, 20 Nov 2010 12:53:16 -0800 (PST), rickman
wrote:



On Nov 20, 2:52 pm, (Don Pearce) wrote:
On Sat, 20 Nov 2010 11:43:03 -0800 (PST), rickman
wrote:

On Nov 20, 12:02 pm, (Scott Dorsey) wrote:
Dick Pierce wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. What is the instantaneous RMS value of -1...
+1. RMS doesn't have to be an integral or a sum. b

Rick

It isn't. It is the square root of the mean of the squares. In the
case of -1, +1 it is 1.

The inclusion of the term "mean" says that there must be at least two
measurements - any fewer and you can't have a mean.

d

I don't recall any such restriction on N. *Certainly the measurement
has just as much meaning with N = 1 as any greater N. *Think of it as
a limit as N approaches 1 then. *The point is that it is as valid a
measurement for a single point as it is for many points. *It
represents the equivalent voltage that would produce the same power as
DC of the same voltage.

When you do it for a single point, the term RMS ceases to have
meaning. For a single point it is just the voltage. For two points and
above, RMS volts times RMS current give average power.

d

I disagree. In for a single point it is just a trivial case. Clearly
the formula still has meaning and it still works perfectly. Saying it
has no meaning, has no meaning... ;^)

Rick

[rickman]

On Nov 20, 3:57*pm, Les Cargill wrote:
rickman wrote:
On Nov 20, 2:52 pm, (Don Pearce) wrote:
On Sat, 20 Nov 2010 11:43:03 -0800 (PST),
wrote:

On Nov 20, 12:02 pm, (Scott Dorsey) wrote:
Dick *wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. *What is the instantaneous RMS value of -1....
+1. *RMS doesn't have to be an integral or a sum. b

Rick

It isn't. It is the square root of the mean of the squares. In the
case of -1, +1 it is 1.

The inclusion of the term "mean" says that there must be at least two
measurements - any fewer and you can't have a mean.

d

I don't recall any such restriction on N. *Certainly the measurement
has just as much meaning with N = 1 as any greater N. *Think of it as
a limit as N approaches 1 then.

N is discrete, not continuous. That's the problem. You can't have half
a sample... there is no application of limits. N=1 is a degenerate case..

*The point is that it is as valid a
measurement for a single point as it is for many points. *It
represents the equivalent voltage that would produce the same power as
DC of the same voltage.

Rick

Right. It's proportional to the instantaneous voltage measure. But it's
degenerate as a weighted vector measure.

--
Les Cargill

Who are you calling a degenerate??? :-(

Rick

[Scott Dorsey]

In article , Randy Yates wrote:
(Scott Dorsey) writes:

That depends entirely on which averaging standard you decide to use.
Most common is LKFS according to ITU BS.1771 loudness standard. You
will never, never see this in the US, but RTW standalone meters can
display it.

Scott, sorry but I didn't see this until just today. Thanks. In
searching for info on BS.1771 I also found this paper from Grim Audio,
which, at a cursory glance, looks like it touches on many of the same
issues I've been asking about here.

It seems like most of the messages I have sent, you haven't seen.

Let me reiterate he

If it says dBFS, it is a peak-reading meter that reads relative to the
highest digital value on the system.

If it is some kind of average reading meter, it is not reading dBFS, but
is reading something else. Because there are so many different standards
for average reading, precisely WHAT it is measuring can be hard to tell.

For example, the average meters on Pro Tools don't seem to match anything
else or meet any known standard. The ballistics are faster than VU.

If you actually need to have consistent and accurate average metering on
digital systems, you use BS.1771 metering. Most people don't, though.

Calling something RMS when it produces a weighted average is not correct.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Les Cargill]

rickman wrote:
On Nov 20, 3:57 pm, Les wrote:
rickman wrote:
On Nov 20, 2:52 pm, (Don Pearce) wrote:
On Sat, 20 Nov 2010 11:43:03 -0800 (PST),
wrote:

On Nov 20, 12:02 pm, (Scott Dorsey) wrote:
Dick wrote:

Scott Dorsey wrote:
I don't know, because I have never seen actual instantaneous RMS values
ever displayed anywhere.

The term "instantaneous RMS value" is itself meaningless.

BINGO! Mr. Pierce wins the kewpie doll!
--scott

Except that it is wrong. What is the instantaneous RMS value of -1...
+1. RMS doesn't have to be an integral or a sum. b

Rick

It isn't. It is the square root of the mean of the squares. In the
case of -1, +1 it is 1.

The inclusion of the term "mean" says that there must be at least two
measurements - any fewer and you can't have a mean.

d

I don't recall any such restriction on N. Certainly the measurement
has just as much meaning with N = 1 as any greater N. Think of it as
a limit as N approaches 1 then.

N is discrete, not continuous. That's the problem. You can't have half
a sample... there is no application of limits. N=1 is a degenerate case.

The point is that it is as valid a
measurement for a single point as it is for many points. It
represents the equivalent voltage that would produce the same power as
DC of the same voltage.

Rick

Right. It's proportional to the instantaneous voltage measure. But it's
degenerate as a weighted vector measure.

--
Les Cargill

Who are you calling a degenerate???:-(

Rick

Easy babe!

--
Les Cargill

[Steve Pope]

Don Pearce wrote:

On Sat, 20 Nov 2010 08:03:14 +0000 (UTC),

I woulda said that a 0 dBFS signal has the RMS level of a sine wave
that just barely doesn't clip a converter (or, a hardlimited channel;
it does not need to be a converter).

(This is an important concept, of sorts, in that is shows that
an N bit converter has a full-scale-signal to quantization noise
ratio of 6*N + 2 dB, not the 6*N + 5 dB that some texts claim.)

One can debate these things. Most outcomes of such debates are
equivalent within a factor of two.

Here's how the maths works.

Lets call the clipping point 0dB.
The biggest sine wave it can hold is -3dB RMS. (peaks just clip).
The lowest bit level is - 16 * 20log(2), or -96.3dB

Because the converter is perfect, the noise is TPD, which has an RMS
value 4.8dB below the 1 bit peak. So the noise level is -101.1dB

So the dynamic range is 101.1dB -3dB +4.8 or 98.1dB. That is the range
from a just-clipping sine wave, to a signal equal in RMS amplitude to
the noise.

I agree with the result, but what is "TPD"? (I'm thinking "triangular"
something...)



S.

[Mark]

On Nov 20, 11:26*pm, (Steve Pope) wrote:
Don Pearce wrote:
On Sat, 20 Nov 2010 08:03:14 +0000 (UTC),
I woulda said that a 0 dBFS signal has the RMS level of a sine wave
that just barely doesn't clip *a converter (or, a hardlimited channel;
it does not need to be a converter).

(This is an important concept, of sorts, in that is shows that
an N bit converter has a full-scale-signal to quantization noise
ratio of 6*N + 2 dB, not the 6*N + 5 dB that some texts claim.)

One can debate these things. *Most outcomes of such debates are
equivalent within a factor of two.
Here's how the maths works.

Lets call the clipping point 0dB.
The biggest sine wave it can hold is -3dB RMS. (peaks just clip).
The lowest bit level is - 16 * 20log(2), or -96.3dB

Because the converter is perfect, the noise is TPD, which has an RMS
value 4.8dB below the 1 bit peak. So the noise level is -101.1dB
So the dynamic range is 101.1dB -3dB +4.8 or 98.1dB. That is the range
from a just-clipping sine wave, to a signal equal in RMS amplitude to
the noise.

I agree with the result, but what is "TPD"? *(I'm thinking "triangular"
something...)

S.

TPD= triangular probability density,,

typically referring to a description of the noise used for dither..

Mark

[glen herrmannsfeldt]

In comp.dsp Mike Rivers wrote:
(snip on dBFS and such)
(then I wrote)

I sometimes record live high-school orchestra concerts.
Because it is hard to know the level, I record 24 bit, then
find the peak and RMS of each track. Then I figure out how many
bits to scale each track by so that peaks stay below FS, and
they should sound about right together.

I suppose that's one way of mixing when you aren't listening
and don't have controls handy. From a computer standpoint,
it makes a certain degree of sense, but from a musical and
artistic standpoint, it's absurd.

Well, the idea is that they should sound about the same level,
such that one shouldn't want to run up and change the volume
control for each track. I don't think I could do that very
well just listening to them, trying to memorize the average
level over a 15 minute track. If the peaks are about the
same, and RMS about the same then I figure that they will sound
about right. If both peak and RMS are different by about the
same amount (I might have changed the record level), then I
adjust by about that amount. Usually that works.
For a recent recording, the four RMS values were -36, -39, -41,
and -35, all dbFS, peaks were -10, -11, -14, -11 dbFS, all
with the same record level.

-- glen

[glen herrmannsfeldt]

In comp.dsp Dick Pierce wrote:

(snip, I wrote)

In EE, RMS can mean different things.

Not in any "EE" I ever encountered. It means one thing:

sqrt(sum(X1^2 + X2^2 ... +Xn^2) / n)

It used to be, for DVMs, that was called True RMS. When
they figured out how to actually do that electrically.

Maybe they all do now, so it isn't a problem anymore,
but it didn't used by be that way. Also, it isn't
for the peak-to-peak reading VTVM, normally calibrated
with an RMS (assuming sine) scale.

Now, what's partly at the root of this discussion is that
if we were to scale our system such that we called

max(abs(X1, X1 ...Xn))

"full scale," what is the ratio of the RMS value of that
set to the full scale value of that set.

-- glen

[Peter Larsen[_3_]]

Randy Yates wrote:

Also, what reference level does an analog peak-reading meter
use?

http://www.klay.com/klay/world_audio_levels.jpg

link supplied by Hank Alrich in some other context. Roger Orban made a
"multi-standard loudness meter" program some time ago, perhaps someone can
remember the download link, it is very illustrative.

--Randy

Kind regards

Peter Larsen

[robert bristow-johnson]

On Nov 20, 1:43*pm, Mark wrote:
But, regardless, there aren't any "different versions" of
RMS, just the one:

* * sqrt(sum(X1^2 + X2^2 ... +Xn^2) / n)

but there are different values of n...

and for a sine tone, *the value of n doesn't matter because a tone is
the same all the time....., but for real audio the value of n does
matter *and that is the crux of the discussion..

and the mean might not be equally weighted.

RMS{ x[n] } = sqrt( SUM{ h[k]*(x[n-k])^2 } )
k

where the weighting coefficients are normalized so that

SUM{ h[k] } = 1
k

essentially, you square the signal, run it through a low-pass filter
with DC gain of 1, then square root the output of the LPF.

there are many different versions of RMS. an infinite number of ways
to do it.

r b-j

[Ken[_8_]]

On Sun, 21 Nov 2010 08:36:13 +0100, "Peter Larsen"
wrote:

Also, what reference level does an analog peak-reading meter
use?

http://www.klay.com/klay/world_audio_levels.jpg

link supplied by Hank Alrich in some other context. Roger Orban made
a "multi-standard loudness meter" program some time ago, perhaps
someone can remember the download link, it is very illustrative.


http://www.orban.com/meter/

[davew]

On Nov 21, 7:56*am, robert bristow-johnson
wrote:
On Nov 20, 1:43*pm, Mark wrote:

But, regardless, there aren't any "different versions" of
RMS, just the one:

* * sqrt(sum(X1^2 + X2^2 ... +Xn^2) / n)

but there are different values of n...

and for a sine tone, *the value of n doesn't matter because a tone is
the same all the time....., but for real audio the value of n does
matter *and that is the crux of the discussion..

and the mean might not be equally weighted.

* * RMS{ x[n] } = sqrt( SUM{ h[k]*(x[n-k])^2 } )
* * * * * * * * * * * * *k

where the weighting coefficients are normalized so that

* * SUM{ h[k] } = 1
* * *k

essentially, you square the signal, run it through a low-pass filter
with DC gain of 1, then square root the output of the LPF.

there are many different versions of RMS. *an infinite number of ways
to do it.

r b-j

This thread is about dBFS. Strictly speaking, it has nothing to do
with peak, VU, PPM, RMS etc. etc. but for audio, in the music and
broadcast industry, it refers to peak magnitude or VU, or PPM or RMS
in some cases and possibly other measures I haven't come across
referred to full scale of the metered range. Americans generally use
VU, the UK and Europe use PPM generally.

0dBFS has units of FS. 0dBFS = one FS. FS is full scale whatever,
but commonly refers to the value 2^(number_of_bits-1). We also use
dBFS24, dBFS32 when referring to 24 and 32 bit audio for example.