Never normalize to 0dB?

[Tobiah]

Quite learned in the field of computer music, a guy
from the csound (synthesis program) list warns to normalize
to -3dB for fear of clipping in the DAC. The example he gives
is of the classic extra peaks that are seen when a square wave
is lowpass filtered. Here is some of his explanation:

The simplest example is a full-range digital square wave. The Gibbs-
effect spikes at the edges of the square wave (famously seen when you
make them additively) will exceed the peak value by an appreciable
margin, which a dac cannot be expected to cope with accurately - digital
peak may be very close to the width of the rail voltage. You may find
there is a certain amount of compression going on up there, which you may
or may not want.

My thinking is that this would have crossed the minds of the DAC makers,
and they would simply allow for the largest analog signal possible to
pass through unscathed. I just think I would have heard of this by now.
I fully expect that the largest sample on most CD's is going to be
32767. I didn't check though.

What say you?

Toby

[Don Pearce[_3_]]

On Wed, 11 Nov 2009 19:53:29 GMT, Tobiah > wrote:

>Quite learned in the field of computer music, a guy
>from the csound (synthesis program) list warns to normalize
>to -3dB for fear of clipping in the DAC. The example he gives
>is of the classic extra peaks that are seen when a square wave
>is lowpass filtered. Here is some of his explanation:
>
>The simplest example is a full-range digital square wave. The Gibbs-
>effect spikes at the edges of the square wave (famously seen when you
>make them additively) will exceed the peak value by an appreciable
>margin, which a dac cannot be expected to cope with accurately - digital
>peak may be very close to the width of the rail voltage. You may find
>there is a certain amount of compression going on up there, which you may
>or may not want.
>
>My thinking is that this would have crossed the minds of the DAC makers,
>and they would simply allow for the largest analog signal possible to
>pass through unscathed. I just think I would have heard of this by now.
>I fully expect that the largest sample on most CD's is going to be
>32767. I didn't check though.
>
>What say you?
>
>Toby

This has just been discussed in uk.rec.audio and Jim Lesurf (retired
scientist) produced a normal 16/44.1 audio file which overshot its
peaks genuinely by 5dB between samples.

The limiting caused by this would occur not in the audio stages, but
in the oversampling digital filter which would be required to create
that trajectory.

I'm sure he won't mind me posting a link to the files he generated,
one at FS and the other at FS - 10dB.

http://jcgl.orpheusweb.co.uk/temp/WaveFromHell.zip

And my analysis in Audition which confirmed his calculations

http://www.soundthoughts.co.uk/look/WFH0dBMono.html

So for full safety, -5dB is the highest level at which one should
normalize.

d

[Scott Dorsey]

Tobiah > wrote:
>
>My thinking is that this would have crossed the minds of the DAC makers,
>and they would simply allow for the largest analog signal possible to
>pass through unscathed. I just think I would have heard of this by now.
>I fully expect that the largest sample on most CD's is going to be
>32767. I didn't check though.
>
>What say you?

There are some very badly designed converters out there. You don't need
every bit of loudness. Leave 1dB or so at least.

Yes, most CDs are normalized to full scale, and they sound bad on some
earlier oversampling players too. You have no control over what kind of
player the end user will have, but you do have some control over headroom.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Geoff]

Tobiah wrote:
> Quite learned in the field of computer music, a guy
> from the csound (synthesis program) list warns to normalize
> to -3dB for fear of clipping in the DAC. The example he gives
> is of the classic extra peaks that are seen when a square wave
> is lowpass filtered. Here is some of his explanation:

Well he's not quite as 'learned' as he makes out. While an essence of truth
in what he says, he is over-compensationg by 1000%. Normalising to -0.3dB
(or so) is an accepted manner of avoiding nasties from all but the worst DA
( and filters). -3dB is extreme and unnecessary.

geoff

[Dan Mills]

On Thu, 12 Nov 2009 09:48:28 +1300, geoff wrote:

> Tobiah wrote:

> Well he's not quite as 'learned' as he makes out. While an essence of
> truth in what he says, he is over-compensationg by 1000%. Normalising
> to -0.3dB (or so) is an accepted manner of avoiding nasties from all but
> the worst DA ( and filters). -3dB is extreme and unnecessary.

Depends on how smart the normaliser is, the samples are NOT the signal
(except at the exact point they are taken) and NOT every set of samples
represents a valid audio signal, some will represent a signal outside the
valid range. Thus a dumb normalise may well need to leave a lot of
headroom where something smarter about level measurement could safely get
closer.

This is not a case of a problem with the DA, it is a case of a problem
with the assumption that any given set of samples that lie below 0dBFs
must necessarily represent a signal that can be converted.

This is particularly insidious in things like broadcast stereo generators
where there is an absolute, hard limit of 75Khz deviation in the regs, to
ensure compliance there are limiters out there that oversample to 0.5Mhz
to minimise overshoot.

Regards, Dan.

[Tobiah]

> Depends on how smart the normaliser is, the samples are NOT the signal
> (except at the exact point they are taken) and NOT every set of samples
> represents a valid audio signal, some will represent a signal outside
> the valid range. Thus a dumb normalise may well need to leave a lot of
> headroom where something smarter about level measurement could safely
> get closer.

I've heard a DAC described as a set of voltage gates, one for each
bit. The LSB has some voltage and each successive bit doubles. It
it worked like that, then at the highest digital peak, the gates
would all be open, and some maximum voltage would go out. So now,
all you have to do is be able to handle that level of signal in
all the rest of the electronics in the device.

Otherwise, you will tell me that the DAC is smart enough to overshoot
by examining the curve along the digital points. Even so, how stupid
would it be to allow clipping or compression when it's known how high
the curve can get?

[Dan Mills]

On Thu, 12 Nov 2009 19:40:54 +0000, Tobiah wrote:

> I've heard a DAC described as a set of voltage gates, one for each bit.
> The LSB has some voltage and each successive bit doubles. It it worked
> like that, then at the highest digital peak, the gates would all be
> open, and some maximum voltage would go out. So now, all you have to do
> is be able to handle that level of signal in all the rest of the
> electronics in the device.

And therein lies the problem...

You model is a reasonable black box representation of a DAC without any
filtering, but it is the filtering that turns your stepwise approximation
back into a pretty exact replica of the original signal, and it it in the
filtering that you can overshoot.

> Otherwise, you will tell me that the DAC is smart enough to overshoot by
> examining the curve along the digital points.

Just so for an oversampling part, and if it does not oversample then the
analogue reconstruction filter will have exactly this effect.

Even so, how stupid would
> it be to allow clipping or compression when it's known how high the
> curve can get?

If you take a suitably band limited signal where the peak level is below
the level corresponding to 0dBFs and run it into an ADC, then both the
set of sample values and the reconstructed signal from the DAC will also
peak below 0dBFs.
This is NOT equivalent to saying that any set of samples that fit below
0dBFs must represent a convertible signal.

Consider that the actual worst case is a set of samples having a
magnitude of 0dBFs and a sign equal to the sign of the corresponding
sample of the filters impulse response.... This could get absolutely
bloody huge, but given that the filters stop band must start below Fs/2,
this set of samples also probably represents something that could never
be recorded as it will contain energy above the stop band of the ADC
filter.

Regards, Dan.

[Don Pearce[_3_]]

On 12 Nov 2009 20:17:27 GMT, Dan Mills >
wrote:

>On Thu, 12 Nov 2009 19:40:54 +0000, Tobiah wrote:
>
>> I've heard a DAC described as a set of voltage gates, one for each bit.
>> The LSB has some voltage and each successive bit doubles. It it worked
>> like that, then at the highest digital peak, the gates would all be
>> open, and some maximum voltage would go out. So now, all you have to do
>> is be able to handle that level of signal in all the rest of the
>> electronics in the device.
>
>And therein lies the problem...
>
>You model is a reasonable black box representation of a DAC without any
>filtering, but it is the filtering that turns your stepwise approximation
>back into a pretty exact replica of the original signal, and it it in the
>filtering that you can overshoot.
>
>> Otherwise, you will tell me that the DAC is smart enough to overshoot by
>> examining the curve along the digital points.
>
>Just so for an oversampling part, and if it does not oversample then the
>analogue reconstruction filter will have exactly this effect.
>
>Even so, how stupid would
>> it be to allow clipping or compression when it's known how high the
>> curve can get?
>
>If you take a suitably band limited signal where the peak level is below
>the level corresponding to 0dBFs and run it into an ADC, then both the
>set of sample values and the reconstructed signal from the DAC will also
>peak below 0dBFs.
>This is NOT equivalent to saying that any set of samples that fit below
>0dBFs must represent a convertible signal.
>
>Consider that the actual worst case is a set of samples having a
>magnitude of 0dBFs and a sign equal to the sign of the corresponding
>sample of the filters impulse response.... This could get absolutely
>bloody huge, but given that the filters stop band must start below Fs/2,
>this set of samples also probably represents something that could never
>be recorded as it will contain energy above the stop band of the ADC
>filter.
>
>Regards, Dan.

I suggest you read my post earlier in this thread. The whole thing is
presented and actually carried out there. You can see everything you
have spoken of here.

d

[Les Cargill[_2_]]

Tobiah wrote:
> Quite learned in the field of computer music, a guy
> from the csound (synthesis program) list warns to normalize
> to -3dB for fear of clipping in the DAC. The example he gives
> is of the classic extra peaks that are seen when a square wave
> is lowpass filtered. Here is some of his explanation:
>
> The simplest example is a full-range digital square wave. The Gibbs-
> effect spikes at the edges of the square wave (famously seen when you
> make them additively) will exceed the peak value by an appreciable
> margin, which a dac cannot be expected to cope with accurately - digital
> peak may be very close to the width of the rail voltage. You may find
> there is a certain amount of compression going on up there, which you may
> or may not want.
>
> My thinking is that this would have crossed the minds of the DAC makers,
> and they would simply allow for the largest analog signal possible to
> pass through unscathed. I just think I would have heard of this by now.
> I fully expect that the largest sample on most CD's is going to be
> 32767. I didn't check though.
>
> What say you?
>
> Toby


I say -0.5dB or -1dB. There was actually quite a thread on the
subject in the last century on alt.music.4-track.

--
Les Cargill

[Jay Ts]

Les Cargill wrote:
> Tobiah wrote:
>> Quite learned in the field of computer music, a guy from the csound
>> (synthesis program) list warns to normalize to -3dB for fear of
>> clipping in the DAC. The example he gives is of the classic extra
>> peaks that are seen when a square wave is lowpass filtered.
>
> I say -0.5dB or -1dB. There was actually quite a thread on the subject
> in the last century on alt.music.4-track.

I got curious about this subject, and decided to try some real
world observations. I made a stereo .wav file (44 KHz, 16-bit)
using one track from WFH0dBMono.wav, and created a matching 315 Hz
sinewave in the other channel. This was played through my M-Audio
Audiophile 2496 with VLC, with the volume set at 100%. I recorded
this with a Tektronix 2232 DSO, and here's a picture of the scope's
screen:

http://jayts.com/images/WaveFromHell_vs_Sinewave.jpg

I placed the cursors on the maximum and minimum points in the waveforms,
and you can see in the upper left that there is very little difference
in amplitudes between the sinewave and "wave from hell". At the top
right, you can see the 1 over delta T (frequency) as 2x the wave's
frequency, because the cursors are measuring half of the wave.
The large grid squares are 2 volts, vertical resolution, and
keep in mind that this is a scope, not a meter, and hasn't
been calibrated lately, so vertical accuracy is about 5%, more-or-less.

To get a better idea of the wave from hell's peak, I expanded the
time scale, and got this:

http://jayts.com/images/WaveFromHell_Expanded.jpg

IMO,the 2496 handled the "wave from hell" neatly. No problem.
(In this shot, the waveform was shifted down by about 1 small
tick. Sorry. :)

Of course, this is a pro sound card, and there may be very cheap
converters with horrible filters that don't behave this way.

I suppose a better test would be to play the same file on one of
my Palm pocket computers. I might do that later, if I have any
more interest in this.

I think before I would allow -5 dB of headroom when normalizing,
I would want to be made aware of definite, REAL examples of devices
that need it. If the only examples were cheap toys, the owners are
probably already used to them sounding bad.

But if we're talking about iPods, and/or other common "hi-fi"
devices (that is, products that are considered high-quality by
their owners), this issue would require more serious consideration.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 13 Nov 2009 07:01:08 GMT, Jay Ts >
wrote:

>Les Cargill wrote:
>> Tobiah wrote:
>>> Quite learned in the field of computer music, a guy from the csound
>>> (synthesis program) list warns to normalize to -3dB for fear of
>>> clipping in the DAC. The example he gives is of the classic extra
>>> peaks that are seen when a square wave is lowpass filtered.
>>
>> I say -0.5dB or -1dB. There was actually quite a thread on the subject
>> in the last century on alt.music.4-track.
>
>I got curious about this subject, and decided to try some real
>world observations. I made a stereo .wav file (44 KHz, 16-bit)
>using one track from WFH0dBMono.wav, and created a matching 315 Hz
>sinewave in the other channel. This was played through my M-Audio
>Audiophile 2496 with VLC, with the volume set at 100%. I recorded
>this with a Tektronix 2232 DSO, and here's a picture of the scope's
>screen:
>
>http://jayts.com/images/WaveFromHell_vs_Sinewave.jpg
>
>I placed the cursors on the maximum and minimum points in the waveforms,
>and you can see in the upper left that there is very little difference
>in amplitudes between the sinewave and "wave from hell". At the top
>right, you can see the 1 over delta T (frequency) as 2x the wave's
>frequency, because the cursors are measuring half of the wave.
>The large grid squares are 2 volts, vertical resolution, and
>keep in mind that this is a scope, not a meter, and hasn't
>been calibrated lately, so vertical accuracy is about 5%, more-or-less.
>
>To get a better idea of the wave from hell's peak, I expanded the
>time scale, and got this:
>
>http://jayts.com/images/WaveFromHell_Expanded.jpg
>
>IMO,the 2496 handled the "wave from hell" neatly. No problem.
>(In this shot, the waveform was shifted down by about 1 small
>tick. Sorry. :)
>

The WFH is heavily clipped in this picture. That flat bit across the
bottom shows where the missing 5dB of headroom should be.

>Of course, this is a pro sound card, and there may be very cheap
>converters with horrible filters that don't behave this way.
>
>I suppose a better test would be to play the same file on one of
>my Palm pocket computers. I might do that later, if I have any
>more interest in this.
>
>I think before I would allow -5 dB of headroom when normalizing,
>I would want to be made aware of definite, REAL examples of devices
>that need it. If the only examples were cheap toys, the owners are
>probably already used to them sounding bad.
>

This isn't a question of devices. Anything that uses oversampling (and
I imagine that means everything) will suffer this 5dB of clipping with
this waveform.

What this really means is this. When you normalize, have a quick look
at all the peaks that approach mac. Expand them all the way and see
where the trajectory of the waveform goes *between* the samples. If it
goes over FS, back off the normalizing a bit so it doesn't. That will
guarantee proper normalization.

This is actually a fault in DAW software. Part of the normalization
process should be internal oversampling to create the true waveform,
and normalization based on those levels. That would guarantee
non-clipped output from any DAC.

d

[Jay Ts]

I, Jay Ts wrote:
> the 2496 handled the "wave from hell" neatly. No problem.

Ok, this is me, correcting myself. One of the reasons I don't
post more is that all too often, within 5 minutes of posting,
I realize I made a dumb error. ;-)

Here are two more screens from my scope, showing how
the Audiophile 2496 handles the "wave from hell" at -10 dB.
(This was included in the .zip file, and I carelessly ignored it.)

http://jayts.com/images/WaveFromHell-10db.jpg
http://jayts.com/images/WaveFromHell_Expanded.jpg

If you compare to the images I posted earlier, you can
see that in the file normalized to 0 dB, the card is
definitely clipping that odd peak.

Compare these two (hint: open each in a separate browser
tab), and you can see it easily:

http://jayts.com/images/WaveFromHell_Expanded.jpg
http://jayts.com/images/WaveFromHell-10db_Expanded.jpg

I'm not sure exactly what causes the clipping, but I
suspect it has to do with the spec for the AP 2496:

Peak Analog Output Signal: +2dBV (Consumer setting)

(Quoted from the user manual for the Audiophile 2496,
and yes, I have it set to "Consumer".) It looks like
the card's analog output circuitry is clipping the
waveform, because it has very little headroom. Or
is this being done in the digital part of the circuitry?

BTW, the top (sine wave) and bottom ("wave from hell")
waveforms are shifted by +1 and -1 large grid division,
respectively. For the -10 dB signals, the vertical
scale is now 0.5 volts/division in both images.

Also, if anyone wants these images for later, please
copy them off my website, because I will delete them
from the server at some time in the future.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 13 Nov 2009 09:06:46 GMT, Jay Ts >
wrote:

>I'm not sure exactly what causes the clipping, but I
>suspect it has to do with the spec for the AP 2496:

No it isn't. The clipping is inherent in the system. The only way that
this clipping could be avoided is by making the oversampling section
17 bits instead of 16 to accommodate the inter-sample peak.

d

[Jay Ts]

Don Pearce wrote:
> Jay Ts wrote:
>>I suppose a better test would be to play the same file on one of my Palm
>>pocket computers. I might do that later, if I have any more interest in
>>this.
>>
> This isn't a question of devices. Anything that uses oversampling (and I
> imagine that means everything) will suffer this 5dB of clipping with
> this waveform.

Well, I found my old Palm Tungsten E in the closet and tried it.
It was able to reproduce the waveform correctly, even at 0 dB.
So I suppose because it's a simple, low-power device, it
is actually better at this test than my pro audio card. :(

On the other hand, it's signal-to-noise ratio is horrible.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 13 Nov 2009 10:30:44 GMT, Jay Ts >
wrote:

>Don Pearce wrote:
>> Jay Ts wrote:
>>>I suppose a better test would be to play the same file on one of my Palm
>>>pocket computers. I might do that later, if I have any more interest in
>>>this.
>>>
>> This isn't a question of devices. Anything that uses oversampling (and I
>> imagine that means everything) will suffer this 5dB of clipping with
>> this waveform.
>
>Well, I found my old Palm Tungsten E in the closet and tried it.
>It was able to reproduce the waveform correctly, even at 0 dB.
>So I suppose because it's a simple, low-power device, it
>is actually better at this test than my pro audio card. :(
>
>On the other hand, it's signal-to-noise ratio is horrible.
>
>Jay Ts

Can we see the scope pic of that one?

d

[Jay Ts]

Don Pearce wrote:
> Jay Ts wrote:
>
>>I'm not sure exactly what causes the clipping, but I suspect it has to
>>do with the spec for the AP 2496:
>
> No it isn't. The clipping is inherent in the system. The only way that
> this clipping could be avoided is by making the oversampling section 17
> bits instead of 16 to accommodate the inter-sample peak.

You may be right, but something that I noticed earlier with the
AP 2496 is that if I play a 0 dBFS sinewave, and measure the output,
it comes out to a bit less than 3.6 volts peak-to-peak at the
analog outputs, which is equal to the +2 dbV that M-Audio claims
in its specs for "Peak Analog Output Signal".

I think what they are doing is to design the output circuitry so
that 0 dBFS of the digital signal is at +2 dBV for the analog
signal, which is clipping level for the analog circuitry. There's
no headroom, but by doing this, they get the dynamic range at the
outputs to their published spec of 104 dB.

Maybe if they did things the way you describe, they would have
to allow 5 dB more headroom in the analog output circuitry, and
then they could only get a dynamic range spec of 99 dB. And that
doesn't look as good to customers as 104 dB. So perhaps
they deliberately ignored this issue, and simply allowed the digital
processing to clip/limit the waveform. If so, I now wonder if this
is a common practice.

I don't know how the Audiophile 2496 works, so as far as I
know right now, this is just a speculative theory.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Jay Ts]

Don Pearce wrote:
> Jay Ts wrote:
>>Don Pearce wrote:
>>> Jay Ts wrote:
>>>>I suppose a better test would be to play the same file on one of my
>>>>Palm pocket computers. I might do that later, if I have any more
>>>>interest in this.
>>>>
>>> This isn't a question of devices. Anything that uses oversampling (and
>>> I imagine that means everything) will suffer this 5dB of clipping with
>>> this waveform.
>>
>>Well, I found my old Palm Tungsten E in the closet and tried it. It was
>>able to reproduce the waveform correctly, even at 0 dB. So I suppose
>>because it's a simple, low-power device, it is actually better at this
>>test than my pro audio card. :(
>>
>>On the other hand, it's signal-to-noise ratio is horrible.
>
> Can we see the scope pic of that one?

Ok:

http://jayts.com/images/WaveFromHell_PalmTE_0dB.jpg
http://jayts.com/images/WaveFromHell_PalmTE_0dB_Expanded.jpg

These were both made with the "wave from hell" normalized to 0 dBFS.
The output is about equal to the M-Audio Audiophile's, when playing
the -10 dBFS normalized "wave from hell". This is because the Palm's
analog output level is about 10 dB lower than that of the M-Audio card.

Compare these two:

http://jayts.com/images/WaveFromHell_PalmTE_0dB.jpg
http://jayts.com/images/WaveFromHell-10db.jpg

And look at the "Delta V" numbers at the upper left.
If you calculate the ratios (dV2/dV1) in both images,
you will see that they are both 1.63. (I only did this
once, so I don't know if they come out so nearly equal
every time. ;) This shows that the Palm is not clipping
the waveform. I regret that due to limitations of my scope,
I was not able to show the Delta V measurements in the "Expanded"
images. (It's a 1990's model, with limited digital waveform storage,
so at this horizontal scale, it could not hold a full cycle of
the waveforms.)

A couple of notes regarding these images:

1) Just to make sure I wasn't playing the wrong file, I played the
-10 dBFS normalized "wave from hell" after I did this, to make sure.
It displayed and measured 10 dB lower in amplitude.

2) Because there's so much ultrasonic garbage in the Palm's audio
outputs, I had to use custom-made scope probes to filter
off ultrasonic frequencies. If I didn't use those, you wouldn't
be able to recognize the waveforms! It was that bad.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 13 Nov 2009 12:19:07 GMT, Jay Ts >
wrote:

>Don Pearce wrote:
>> Jay Ts wrote:
>>>Don Pearce wrote:
>>>> Jay Ts wrote:
>>>>>I suppose a better test would be to play the same file on one of my
>>>>>Palm pocket computers. I might do that later, if I have any more
>>>>>interest in this.
>>>>>
>>>> This isn't a question of devices. Anything that uses oversampling (and
>>>> I imagine that means everything) will suffer this 5dB of clipping with
>>>> this waveform.
>>>
>>>Well, I found my old Palm Tungsten E in the closet and tried it. It was
>>>able to reproduce the waveform correctly, even at 0 dB. So I suppose
>>>because it's a simple, low-power device, it is actually better at this
>>>test than my pro audio card. :(
>>>
>>>On the other hand, it's signal-to-noise ratio is horrible.
>>
>> Can we see the scope pic of that one?
>
>Ok:
>
>http://jayts.com/images/WaveFromHell_PalmTE_0dB.jpg
>http://jayts.com/images/WaveFromHell_PalmTE_0dB_Expanded.jpg
>
>These were both made with the "wave from hell" normalized to 0 dBFS.
>The output is about equal to the M-Audio Audiophile's, when playing
>the -10 dBFS normalized "wave from hell". This is because the Palm's
>analog output level is about 10 dB lower than that of the M-Audio card.
>
>Compare these two:
>
>http://jayts.com/images/WaveFromHell_PalmTE_0dB.jpg
>http://jayts.com/images/WaveFromHell-10db.jpg
>
>And look at the "Delta V" numbers at the upper left.
>If you calculate the ratios (dV2/dV1) in both images,
>you will see that they are both 1.63. (I only did this
>once, so I don't know if they come out so nearly equal
>every time. ;) This shows that the Palm is not clipping
>the waveform. I regret that due to limitations of my scope,
>I was not able to show the Delta V measurements in the "Expanded"
>images. (It's a 1990's model, with limited digital waveform storage,
>so at this horizontal scale, it could not hold a full cycle of
>the waveforms.)
>
>A couple of notes regarding these images:
>
>1) Just to make sure I wasn't playing the wrong file, I played the
>-10 dBFS normalized "wave from hell" after I did this, to make sure.
>It displayed and measured 10 dB lower in amplitude.
>
>2) Because there's so much ultrasonic garbage in the Palm's audio
>outputs, I had to use custom-made scope probes to filter
>off ultrasonic frequencies. If I didn't use those, you wouldn't
>be able to recognize the waveforms! It was that bad.
>
>Jay Ts

What is the sine wave on each of these? Is it the other channel?

d

[Scott Dorsey]

Don Pearce > wrote:
>
>This isn't a question of devices. Anything that uses oversampling (and
>I imagine that means everything) will suffer this 5dB of clipping with
>this waveform.

Anything that uses oversampling and doesn't scale the signal down internally
will suffer this. But a lot of modern oversampling converters are
specifically designed to fix this issue.
--scott

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

[Scott Dorsey]

Don Pearce > wrote:
>On 13 Nov 2009 09:06:46 GMT, Jay Ts >
>wrote:
>
>>I'm not sure exactly what causes the clipping, but I
>>suspect it has to do with the spec for the AP 2496:
>
>No it isn't. The clipping is inherent in the system. The only way that
>this clipping could be avoided is by making the oversampling section
>17 bits instead of 16 to accommodate the inter-sample peak.

Which is, in fact, what the better AKM and Cirrus converters do.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Don Pearce[_3_]]

On 13 Nov 2009 08:38:40 -0500, (Scott Dorsey) wrote:

>Don Pearce > wrote:
>>On 13 Nov 2009 09:06:46 GMT, Jay Ts >
>>wrote:
>>
>>>I'm not sure exactly what causes the clipping, but I
>>>suspect it has to do with the spec for the AP 2496:
>>
>>No it isn't. The clipping is inherent in the system. The only way that
>>this clipping could be avoided is by making the oversampling section
>>17 bits instead of 16 to accommodate the inter-sample peak.
>
>Which is, in fact, what the better AKM and Cirrus converters do.
>--scott

I never knew that. Are you positive?

d

[Jay Ts]

Don Pearce wrote:
> Jay Ts wrote:
>>Don Pearce wrote:
>>> Jay Ts wrote:
>>>>Don Pearce wrote:
>>>>> Jay Ts wrote:
>>>>>>I suppose a better test would be to play the same file on one of my
>>>>>>Palm pocket computers. I might do that later, if I have any more
>>>>>>interest in this.
>>>>>>
>>>>> This isn't a question of devices. Anything that uses oversampling
>>>>> (and I imagine that means everything) will suffer this 5dB of
>>>>> clipping with this waveform.
>>>>
>>>>Well, I found my old Palm Tungsten E in the closet and tried it. It
>>>>was able to reproduce the waveform correctly, even at 0 dB. So I
>>>>suppose because it's a simple, low-power device, it is actually better
>>>>at this test than my pro audio card. :(
>>>>
>>>>On the other hand, it's signal-to-noise ratio is horrible.
>>>
>>> Can we see the scope pic of that one?
>>
>>Ok:
>>
>>http://jayts.com/images/WaveFromHell_PalmTE_0dB.jpg
>>http://jayts.com/images/WaveFromHell_PalmTE_0dB_Expanded.jpg
>>
> What is the sine wave on each of these? Is it the other channel?

Yes. I documented this earlier. Check my previous posts.
It's a 315 Hz sinewave, with the amplitude normalized to
0 dBFS or -10 dBFS, to match that of "the wave from hell".

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 13 Nov 2009 13:44:53 GMT, Jay Ts >
wrote:

>Don Pearce wrote:
>> Jay Ts wrote:
>>>Don Pearce wrote:
>>>> Jay Ts wrote:
>>>>>Don Pearce wrote:
>>>>>> Jay Ts wrote:
>>>>>>>I suppose a better test would be to play the same file on one of my
>>>>>>>Palm pocket computers. I might do that later, if I have any more
>>>>>>>interest in this.
>>>>>>>
>>>>>> This isn't a question of devices. Anything that uses oversampling
>>>>>> (and I imagine that means everything) will suffer this 5dB of
>>>>>> clipping with this waveform.
>>>>>
>>>>>Well, I found my old Palm Tungsten E in the closet and tried it. It
>>>>>was able to reproduce the waveform correctly, even at 0 dB. So I
>>>>>suppose because it's a simple, low-power device, it is actually better
>>>>>at this test than my pro audio card. :(
>>>>>
>>>>>On the other hand, it's signal-to-noise ratio is horrible.
>>>>
>>>> Can we see the scope pic of that one?
>>>
>>>Ok:
>>>
>>>http://jayts.com/images/WaveFromHell_PalmTE_0dB.jpg
>>>http://jayts.com/images/WaveFromHell_PalmTE_0dB_Expanded.jpg
>>>
>> What is the sine wave on each of these? Is it the other channel?
>
>Yes. I documented this earlier. Check my previous posts.
>It's a 315 Hz sinewave, with the amplitude normalized to
>0 dBFS or -10 dBFS, to match that of "the wave from hell".
>
>Jay Ts

OK, but that raises a problem. The WFH waveform is now too small,
because the main square wave part, ignoring the spike, should have the
same peak to peak amplitude as the sine wave. The spike should be
standing out above that (or missing if it is clipped).

d

[Scott Dorsey]

Don Pearce > wrote:
>On 13 Nov 2009 08:38:40 -0500, (Scott Dorsey) wrote:
>
>>Don Pearce > wrote:
>>>On 13 Nov 2009 09:06:46 GMT, Jay Ts >
>>>wrote:
>>>
>>>>I'm not sure exactly what causes the clipping, but I
>>>>suspect it has to do with the spec for the AP 2496:
>>>
>>>No it isn't. The clipping is inherent in the system. The only way that
>>>this clipping could be avoided is by making the oversampling section
>>>17 bits instead of 16 to accommodate the inter-sample peak.
>>
>>Which is, in fact, what the better AKM and Cirrus converters do.
>
>I never knew that. Are you positive?

Well, these days they are all sigma-delta converters... but yes... they
make the coefficients extra-long to deal with this stuff.

Steven Harris wrote a paper about it back when he was with Crystal Semi
(which later became Cirrus). I can find a cite if you're interested.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

[Don Pearce[_3_]]

On 13 Nov 2009 09:14:12 -0500, (Scott Dorsey) wrote:

>Don Pearce > wrote:
>>On 13 Nov 2009 08:38:40 -0500, (Scott Dorsey) wrote:
>>
>>>Don Pearce > wrote:
>>>>On 13 Nov 2009 09:06:46 GMT, Jay Ts >
>>>>wrote:
>>>>
>>>>>I'm not sure exactly what causes the clipping, but I
>>>>>suspect it has to do with the spec for the AP 2496:
>>>>
>>>>No it isn't. The clipping is inherent in the system. The only way that
>>>>this clipping could be avoided is by making the oversampling section
>>>>17 bits instead of 16 to accommodate the inter-sample peak.
>>>
>>>Which is, in fact, what the better AKM and Cirrus converters do.
>>
>>I never knew that. Are you positive?
>
>Well, these days they are all sigma-delta converters... but yes... they
>make the coefficients extra-long to deal with this stuff.
>
>Steven Harris wrote a paper about it back when he was with Crystal Semi
>(which later became Cirrus). I can find a cite if you're interested.
>--scott

That would great if you could, Scott. Thanks

d

[Jay Ts]

Don Pearce wrote:
>>> What is the sine wave on each of these? Is it the other channel?
>>
>>Yes. I documented this earlier. Check my previous posts. It's a 315 Hz
>>sinewave, with the amplitude normalized to 0 dBFS or -10 dBFS, to match
>>that of "the wave from hell".
>>
>>Jay Ts
>
> OK, but that raises a problem. The WFH waveform is now too small,
> because the main square wave part, ignoring the spike, should have the
> same peak to peak amplitude as the sine wave. The spike should be
> standing out above that (or missing if it is clipped).

Not sure what you're saying here. The "main square wave part",
as I interpret your expression, is at -6 dB (half the amplitude of
the sine wave). But it's not a square wave at all, really.
It's a tricky waveform to think about, so for reference, here
is a piece of the .wav file I used, as viewed in Audacity:

http://jayts.com/images/WFH-Audacity.png

The scope images I posted earlier look correct to me.

Ok? If not, please clarify.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 14 Nov 2009 05:47:09 GMT, Jay Ts >
wrote:

>Don Pearce wrote:
>>>> What is the sine wave on each of these? Is it the other channel?
>>>
>>>Yes. I documented this earlier. Check my previous posts. It's a 315 Hz
>>>sinewave, with the amplitude normalized to 0 dBFS or -10 dBFS, to match
>>>that of "the wave from hell".
>>>
>>>Jay Ts
>>
>> OK, but that raises a problem. The WFH waveform is now too small,
>> because the main square wave part, ignoring the spike, should have the
>> same peak to peak amplitude as the sine wave. The spike should be
>> standing out above that (or missing if it is clipped).
>
>Not sure what you're saying here. The "main square wave part",
>as I interpret your expression, is at -6 dB (half the amplitude of
>the sine wave). But it's not a square wave at all, really.
>It's a tricky waveform to think about, so for reference, here
>is a piece of the .wav file I used, as viewed in Audacity:
>
>http://jayts.com/images/WFH-Audacity.png
>
>The scope images I posted earlier look correct to me.
>
>Ok? If not, please clarify.
>
>Jay Ts

My mistake - that is right. That leaves us two alternatives for your
Palm. Either it has very competent and complex oversampling or it
doesn't oversample at all. Either will have that effect. As you say
there is an immense amount of stuff to get rid of ultrasonically, so
I'm guessing the latter is the right option.

d

[Peter Larsen[_3_]]

Don Pearce wrote:

>> http://jayts.com/images/WFH-Audacity.png

> My mistake - that is right. That leaves us two alternatives for your
> Palm. Either it has very competent and complex oversampling or it
> doesn't oversample at all. Either will have that effect. As you say
> there is an immense amount of stuff to get rid of ultrasonically, so
> I'm guessing the latter is the right option.

For a real world context the bright thing to do seems to be to take the
actual mix and mp3 encode-decode and see what happens. As much as 2.2 dB of
initial headroom may in my experience be required.

> d

Kind regards

Peter Larsen

[Don Pearce[_3_]]

On Sat, 14 Nov 2009 14:43:45 +0100, "Peter Larsen"
> wrote:

>Don Pearce wrote:
>
>>> http://jayts.com/images/WFH-Audacity.png
>
>> My mistake - that is right. That leaves us two alternatives for your
>> Palm. Either it has very competent and complex oversampling or it
>> doesn't oversample at all. Either will have that effect. As you say
>> there is an immense amount of stuff to get rid of ultrasonically, so
>> I'm guessing the latter is the right option.
>
>For a real world context the bright thing to do seems to be to take the
>actual mix and mp3 encode-decode and see what happens. As much as 2.2 dB of
>initial headroom may in my experience be required.
>
I think maybe you clicked the wrong message when you replied. None of
what you write has anything to do with the topic.

d

[Peter Larsen[_3_]]

Don Pearce wrote:

> On Sat, 14 Nov 2009 14:43:45 +0100, "Peter Larsen"
> > wrote:

>> Don Pearce wrote:

>>>> http://jayts.com/images/WFH-Audacity.png

>>> My mistake - that is right. That leaves us two alternatives for your
>>> Palm. Either it has very competent and complex oversampling or it
>>> doesn't oversample at all. Either will have that effect. As you say
>>> there is an immense amount of stuff to get rid of ultrasonically, so
>>> I'm guessing the latter is the right option.

>> For a real world context the bright thing to do seems to be to take
>> the actual mix and mp3 encode-decode and see what happens. As much
>> as 2.2 dB of initial headroom may in my experience be required.

> I think maybe you clicked the wrong message when you replied. None of
> what you write has anything to do with the topic.

The topic here is whether to normalize to 0 dB full scale, that at least is
what the header says. The newsgroup is rec.audio.pro(duction). As long as we
stay in the digital domain it works fine to normalize to 0 dB full scale, it
at conversion time there are possible issues.

In my universe it is very difficult to perceive less than 2 ms of clipping,
you may have more acute hearing than I have. My understanding is that the
clippings produced by the artificial waveform illustrated by jay will be of
short duration.

Coming up with waveform examples that catch poor DA converters is all very
fine, but in the context of audio production it matters a lot more whether
it plain sounds distorted on playback and mp3 encode-decode is guaranteed to
occur for just about any audio produced on this planet in this millenium.

Which is to say that a mp3 encode-decode of actual audio will indicate how
much headroom to leave in the actual audio file.

> d

Kind regards

Peter Larsen

[Don Pearce[_3_]]

On Sat, 14 Nov 2009 16:25:31 +0100, "Peter Larsen"
> wrote:

>Don Pearce wrote:
>
>> On Sat, 14 Nov 2009 14:43:45 +0100, "Peter Larsen"
>> > wrote:
>
>>> Don Pearce wrote:
>
>>>>> http://jayts.com/images/WFH-Audacity.png
>
>>>> My mistake - that is right. That leaves us two alternatives for your
>>>> Palm. Either it has very competent and complex oversampling or it
>>>> doesn't oversample at all. Either will have that effect. As you say
>>>> there is an immense amount of stuff to get rid of ultrasonically, so
>>>> I'm guessing the latter is the right option.
>
>>> For a real world context the bright thing to do seems to be to take
>>> the actual mix and mp3 encode-decode and see what happens. As much
>>> as 2.2 dB of initial headroom may in my experience be required.
>
>> I think maybe you clicked the wrong message when you replied. None of
>> what you write has anything to do with the topic.
>
>The topic here is whether to normalize to 0 dB full scale, that at least is
>what the header says. The newsgroup is rec.audio.pro(duction). As long as we
>stay in the digital domain it works fine to normalize to 0 dB full scale, it
>at conversion time there are possible issues.
>
>In my universe it is very difficult to perceive less than 2 ms of clipping,
>you may have more acute hearing than I have. My understanding is that the
>clippings produced by the artificial waveform illustrated by jay will be of
>short duration.
>
>Coming up with waveform examples that catch poor DA converters is all very
>fine, but in the context of audio production it matters a lot more whether
>it plain sounds distorted on playback and mp3 encode-decode is guaranteed to
>occur for just about any audio produced on this planet in this millenium.
>
>Which is to say that a mp3 encode-decode of actual audio will indicate how
>much headroom to leave in the actual audio file.
>
>> d
>
> Kind regards
>
> Peter Larsen
>
>

No. No better than the last.

MP3 conversion has no relevance here, but the behaviour of the typical
DAC is very much the issue, since it determines whether or not it is
safe to normalize to zero dB, considering that between samples there
will be peaks greater than 0dB that the DAC must handle. The purpose
of this thread has been to determine how big those peaks can be, and
what the resulting audio will be like. This is nothing to do with
catching out poor DACs.

d

[Peter Larsen[_3_]]

Don Pearce wrote:

> MP3 conversion has no relevance here

They ARE going to happen, so they are relevant.

> but the behaviour of the typical
> DAC is very much the issue, since it determines whether or not it is
> safe to normalize to zero dB, considering that between samples there
> will be peaks greater than 0dB that the DAC must handle. The purpose
> of this thread has been to determine how big those peaks can be, and
> what the resulting audio will be like. This is nothing to do with
> catching out poor DACs.

This thread started with a question about whether it was sane practice to
normalize to 0 dB. It seems to be an excellent and simple point that it is
not because there will be overshoot in mp3 encode-decode. This will happen
to the signal in the digital domain at the end user or in web-based
distribution or in radio station harddisk storage systems and will result in
overmodulation before it ever reaches the AD converter.

The point you can make Don, but don't, is that I could have followed up
earlier or to another post, however inter-sample overshoots are not likely
to be longer than 2 ms and thus, if cleanly clipped, not a cause for
concern. Constant clipping because of a mp3 encode-deocde cycle is much more
likely to be an actually audible issue.

> d

Kind regards

Peter Larsen

[Don Pearce[_3_]]

On Sat, 14 Nov 2009 18:00:19 +0100, "Peter Larsen"
> wrote:

>Don Pearce wrote:
>
>> MP3 conversion has no relevance here
>
>They ARE going to happen, so they are relevant.
>
>> but the behaviour of the typical
>> DAC is very much the issue, since it determines whether or not it is
>> safe to normalize to zero dB, considering that between samples there
>> will be peaks greater than 0dB that the DAC must handle. The purpose
>> of this thread has been to determine how big those peaks can be, and
>> what the resulting audio will be like. This is nothing to do with
>> catching out poor DACs.
>
>This thread started with a question about whether it was sane practice to
>normalize to 0 dB. It seems to be an excellent and simple point that it is
>not because there will be overshoot in mp3 encode-decode. This will happen
>to the signal in the digital domain at the end user or in web-based
>distribution or in radio station harddisk storage systems and will result in
>overmodulation before it ever reaches the AD converter.
>
>The point you can make Don, but don't, is that I could have followed up
>earlier or to another post, however inter-sample overshoots are not likely
>to be longer than 2 ms and thus, if cleanly clipped, not a cause for
>concern. Constant clipping because of a mp3 encode-deocde cycle is much more
>likely to be an actually audible issue.
>
Why are you so hung up on MP3 encode - decode? It is unrelated to what
we are talking about here. It may be interesting, OK, it is
interesting, but irrelevant.

d

[Peter Larsen[_3_]]

Don Pearce wrote:

> Why are you so hung up on MP3 encode - decode? It is unrelated to what
> we are talking about here. It may be interesting, OK, it is
> interesting, but irrelevant.

Because it is a very good real world reason not to release stuff that is
normalized to 0.0000 dB and because it hasn't been mentioned by other
participants.

In this context it is the 5 of spades. That does not mean that the
electonics design issues are not interesting - they are fascinating - but
this issue matters before the DA conversion for playback takes place.

> d

Kind regards

Peter Larsen

[Don Pearce[_3_]]

On Sat, 14 Nov 2009 19:48:56 +0100, "Peter Larsen"
> wrote:

>Don Pearce wrote:
>
>> Why are you so hung up on MP3 encode - decode? It is unrelated to what
>> we are talking about here. It may be interesting, OK, it is
>> interesting, but irrelevant.
>
>Because it is a very good real world reason not to release stuff that is
>normalized to 0.0000 dB and because it hasn't been mentioned by other
>participants.
>
>In this context it is the 5 of spades. That does not mean that the
>electonics design issues are not interesting - they are fascinating - but
>this issue matters before the DA conversion for playback takes place.
>

Then I would suggest you start a thread to discuss it.

d

[Les Cargill[_2_]]

Jay Ts wrote:
> Les Cargill wrote:
>> Tobiah wrote:
<snip>
>
> I think before I would allow -5 dB of headroom when normalizing,
> I would want to be made aware of definite, REAL examples of devices
> that need it. If the only examples were cheap toys, the owners are
> probably already used to them sounding bad.
>

That is my philosophy on the subject. -5dB would be too much
guardband, unless you know all listeners would be willing to
grab the volume control.

> But if we're talking about iPods, and/or other common "hi-fi"
> devices (that is, products that are considered high-quality by
> their owners),

*snort*. :) Now, I'd agree that identifying such devices
publicly would be a good thing. FWIW, if I can swing the
logistics of it, it'd be interesting to try this on
my daughter's iPhone. Don't have a scope, tho....

> this issue would require more serious consideration.
>
> Jay Ts


--
Les Cargill

[Les Cargill[_2_]]

Don Pearce wrote:
> On Sat, 14 Nov 2009 16:25:31 +0100, "Peter Larsen"
<snip>
>
> No. No better than the last.
>
> MP3 conversion has no relevance here, but the behaviour of the typical
> DAC is very much the issue, since it determines whether or not it is
> safe to normalize to zero dB, considering that between samples there
> will be peaks greater than 0dB that the DAC must handle. The purpose
> of this thread has been to determine how big those peaks can be, and
> what the resulting audio will be like. This is nothing to do with
> catching out poor DACs.
>
> d

Hang on. We have a cost associated with how people perceive our work.
This cost has two components - perceived level ( which we will be
penalized "unfairly" for if it's too low ) or having a D/A converter
mangle the signal.

This puts us in a position of having to compromise between those two
costs. It's a classic trade space, and any choice is *going* to
be arbitrary. I can't see normalizing to less than -1.0 dB, but that's
my personal arbitrary choice. And given JayTs input, that choice
may change - but I don't think real world signals have the kind of...
"artificial" spectrum his Wave from Hell has. Do an FFT on that thing -
it's brutal. Which is the point.

What really would serve our best interest is to have the performance of
all possible such devices known. That's not likely, but that would
be what we'd need. Normalizing to -5dB seems like it's too much,
although if it became a standard, that'd be just fine - 16 bits
has plenty of bandwidth for all peaks being < 5dB. But then there's
the Loudness Wars.....

--
Les Cargill

[Don Pearce[_3_]]

On Sat, 14 Nov 2009 16:39:59 -0500, Les Cargill
> wrote:

>Don Pearce wrote:
>> On Sat, 14 Nov 2009 16:25:31 +0100, "Peter Larsen"
><snip>
>>
>> No. No better than the last.
>>
>> MP3 conversion has no relevance here, but the behaviour of the typical
>> DAC is very much the issue, since it determines whether or not it is
>> safe to normalize to zero dB, considering that between samples there
>> will be peaks greater than 0dB that the DAC must handle. The purpose
>> of this thread has been to determine how big those peaks can be, and
>> what the resulting audio will be like. This is nothing to do with
>> catching out poor DACs.
>>
>> d
>
>Hang on. We have a cost associated with how people perceive our work.
>This cost has two components - perceived level ( which we will be
>penalized "unfairly" for if it's too low ) or having a D/A converter
>mangle the signal.
>
>This puts us in a position of having to compromise between those two
>costs. It's a classic trade space, and any choice is *going* to
>be arbitrary. I can't see normalizing to less than -1.0 dB, but that's
>my personal arbitrary choice. And given JayTs input, that choice
>may change - but I don't think real world signals have the kind of...
>"artificial" spectrum his Wave from Hell has. Do an FFT on that thing -
>it's brutal. Which is the point.
>
>What really would serve our best interest is to have the performance of
>all possible such devices known. That's not likely, but that would
>be what we'd need. Normalizing to -5dB seems like it's too much,
>although if it became a standard, that'd be just fine - 16 bits
>has plenty of bandwidth for all peaks being < 5dB. But then there's
>the Loudness Wars.....

I wouldn't suggest normalizing to -5dB. Normalize to zero, but make
sure it is actually zero, to ascertain which you do need to zoom into
the peaks and see where they really go. If they go above zero between
samples, drag it all down a little so they don't. Otherwise there will
be clipping. You may be ok with clipping, but that is another story.

d

[Scott Dorsey]

Don Pearce > wrote:
>
>I wouldn't suggest normalizing to -5dB. Normalize to zero, but make
>sure it is actually zero, to ascertain which you do need to zoom into
>the peaks and see where they really go. If they go above zero between
>samples, drag it all down a little so they don't. Otherwise there will
>be clipping. You may be ok with clipping, but that is another story.

This would be reasonable if you actually trusted the envelope display
to be a real waveform display. I don't.

There are some systems where this is the case, but don't assume it is
unless you have absolutely verified it. Because sadly, most aren't.
--scott

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

[Peter Larsen[_3_]]

Don Pearce wrote:

> On Sat, 14 Nov 2009 19:48:56 +0100, "Peter Larsen"
> > wrote:

>> Don Pearce wrote:

>>> Why are you so hung up on MP3 encode - decode? It is unrelated to
>>> what we are talking about here. It may be interesting, OK, it is
>>> interesting, but irrelevant.

Also btw. because so many people seem to be unaware of the built in
overshoot, one that is a real problem, I don't think the academically
interesting waveform from hell is. For the hardware designers it is a good
point to the effect that they should leave the upper bit of the hardware
unused and above 0dB, but for the sound engineer it matters more that mp3's
will clip stuff that goes to 0 dB.

>> Because it is a very good real world reason not to release stuff
>> that is normalized to 0.0000 dB and because it hasn't been mentioned
>> by other participants.
>>
>> In this context it is the 5 of spades. That does not mean that the
>> electonics design issues are not interesting - they are fascinating
>> - but this issue matters before the DA conversion for playback takes
>> place.

> Then I would suggest you start a thread to discuss it.

Please clarify this, are you suggesting that I should not write about not
normalizing to 0dB in a thread about not normalizing to 0dB?

> d

Kind regards

Peter Larsen

[Jay Ts]

Les Cargill wrote:
> Jay Ts wrote:
>> Les Cargill wrote:
>>> Tobiah wrote:
> <snip>
>>
>> I think before I would allow -5 dB of headroom when normalizing, I
>> would want to be made aware of definite, REAL examples of devices that
>> need it. If the only examples were cheap toys, the owners are probably
>> already used to them sounding bad.

Since I identified that my M-Audio card is one of those "cheap toys",
I have changed my opinion, and now I want to see examples of real
audio tracks (e.g., mixdowns, masters, simple recordings, or softsynth
renderings) in which the overshoot is significant and results in
audible clipping.

And I mean, not just playing the WFH with a sampler and claiming
it is a perfectly reasonable example of a synth sound. ;-)

> That is my philosophy on the subject. -5dB would be too much guardband,
> unless you know all listeners would be willing to grab the volume
> control.
>
>> But if we're talking about iPods, and/or other common "hi-fi" devices
>> (that is, products that are considered high-quality by their owners),
>
> *snort*. :) Now, I'd agree that identifying such devices publicly would
> be a good thing. FWIW, if I can swing the logistics of it, it'd be
> interesting to try this on my daughter's iPhone. Don't have a scope,
> tho....

You can try this:

http://www.zelscope.com/

I use it for some things, in addition to my Tektronix scope.
It's not as full-featured or high-bandwidth, but you can
plug your soundcard's inputs into the outputs, play the
WFH, and see it on Zelscope. I wish the deveopers would work
on the app some more so it would be more than a "crippled
oscilloscope emulator", but it does basically work.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[hank alrich]

Les Cargill > wrote:

> Don Pearce wrote:
> > On Sat, 14 Nov 2009 16:25:31 +0100, "Peter Larsen"
> <snip>
> >
> > No. No better than the last.
> >
> > MP3 conversion has no relevance here, but the behaviour of the typical
> > DAC is very much the issue, since it determines whether or not it is
> > safe to normalize to zero dB, considering that between samples there
> > will be peaks greater than 0dB that the DAC must handle. The purpose
> > of this thread has been to determine how big those peaks can be, and
> > what the resulting audio will be like. This is nothing to do with
> > catching out poor DACs.
> >
> > d
>
> Hang on. We have a cost associated with how people perceive our work.
> This cost has two components - perceived level ( which we will be
> penalized "unfairly" for if it's too low ) or having a D/A converter
> mangle the signal.
>
> This puts us in a position of having to compromise between those two
> costs. It's a classic trade space, and any choice is *going* to
> be arbitrary. I can't see normalizing to less than -1.0 dB, but that's
> my personal arbitrary choice. And given JayTs input, that choice
> may change - but I don't think real world signals have the kind of...
> "artificial" spectrum his Wave from Hell has. Do an FFT on that thing -
> it's brutal. Which is the point.
>
> What really would serve our best interest is to have the performance of
> all possible such devices known. That's not likely, but that would
> be what we'd need. Normalizing to -5dB seems like it's too much,
> although if it became a standard, that'd be just fine - 16 bits
> has plenty of bandwidth for all peaks being < 5dB. But then there's
> the Loudness Wars.....
>
> --
> Les Cargill

Y'all might find some enlightening discussion here:

http://recforums.prosoundweb.com/index.php/m/280407/0/

and here:

http://tinyurl.com/yha798o

--
ha
shut up and play your guitar

[hank alrich]

Les Cargill > wrote:

> Jay Ts wrote:
> > Les Cargill wrote:
> >> Tobiah wrote:
> <snip>
> >
> > I think before I would allow -5 dB of headroom when normalizing,
> > I would want to be made aware of definite, REAL examples of devices
> > that need it. If the only examples were cheap toys, the owners are
> > probably already used to them sounding bad.
> >
>
> That is my philosophy on the subject. -5dB would be too much
> guardband, unless you know all listeners would be willing to
> grab the volume control.
>
> > But if we're talking about iPods, and/or other common "hi-fi"
> > devices (that is, products that are considered high-quality by
> > their owners),
>
> *snort*. :) Now, I'd agree that identifying such devices
> publicly would be a good thing. FWIW, if I can swing the
> logistics of it, it'd be interesting to try this on
> my daughter's iPhone. Don't have a scope, tho....
>
> > this issue would require more serious consideration.
> >
> > Jay Ts
>
>
> --
> Les Cargill

As an aside (while absorbing beer after the flight from Austin and car
ride from Reno), iPods sond remarkably better playing back .wav files
that they do playing back low bitrate mp3's, for some strange reason.

Point is merely a reminder (to self, as much as anyone), of GIGO.

--
ha
shut up and play your guitar

[Don Pearce[_3_]]

On 14 Nov 2009 17:39:06 -0500, (Scott Dorsey) wrote:

>Don Pearce > wrote:
>>
>>I wouldn't suggest normalizing to -5dB. Normalize to zero, but make
>>sure it is actually zero, to ascertain which you do need to zoom into
>>the peaks and see where they really go. If they go above zero between
>>samples, drag it all down a little so they don't. Otherwise there will
>>be clipping. You may be ok with clipping, but that is another story.
>
>This would be reasonable if you actually trusted the envelope display
>to be a real waveform display. I don't.
>
>There are some systems where this is the case, but don't assume it is
>unless you have absolutely verified it. Because sadly, most aren't.
>--scott

Easy to verify with that waveform from hell file. If the peak rises
5dB above the samples, the display is being generated by the proper
sinc function.

d

[Don Pearce[_3_]]

On Sat, 14 Nov 2009 23:38:16 -0800, (hank alrich)
wrote:

>Les Cargill > wrote:
>
>> Don Pearce wrote:
>> > On Sat, 14 Nov 2009 16:25:31 +0100, "Peter Larsen"
>> <snip>
>> >
>> > No. No better than the last.
>> >
>> > MP3 conversion has no relevance here, but the behaviour of the typical
>> > DAC is very much the issue, since it determines whether or not it is
>> > safe to normalize to zero dB, considering that between samples there
>> > will be peaks greater than 0dB that the DAC must handle. The purpose
>> > of this thread has been to determine how big those peaks can be, and
>> > what the resulting audio will be like. This is nothing to do with
>> > catching out poor DACs.
>> >
>> > d
>>
>> Hang on. We have a cost associated with how people perceive our work.
>> This cost has two components - perceived level ( which we will be
>> penalized "unfairly" for if it's too low ) or having a D/A converter
>> mangle the signal.
>>
>> This puts us in a position of having to compromise between those two
>> costs. It's a classic trade space, and any choice is *going* to
>> be arbitrary. I can't see normalizing to less than -1.0 dB, but that's
>> my personal arbitrary choice. And given JayTs input, that choice
>> may change - but I don't think real world signals have the kind of...
>> "artificial" spectrum his Wave from Hell has. Do an FFT on that thing -
>> it's brutal. Which is the point.
>>
>> What really would serve our best interest is to have the performance of
>> all possible such devices known. That's not likely, but that would
>> be what we'd need. Normalizing to -5dB seems like it's too much,
>> although if it became a standard, that'd be just fine - 16 bits
>> has plenty of bandwidth for all peaks being < 5dB. But then there's
>> the Loudness Wars.....
>>
>> --
>> Les Cargill
>
>Y'all might find some enlightening discussion here:
>
>http://recforums.prosoundweb.com/index.php/m/280407/0/
>

Unfortunately the SSL link referenced in the first post is dead.

Some real enlightenment is to be found here

http://www.audiomisc.co.uk/HFN/OverTheTop/OTT.html

d

[Jay Ts]

Don Pearce wrote:
> Some real enlightenment is to be found here
>
> http://www.audiomisc.co.uk/HFN/OverTheTop/OTT.html

Thanks. That's more the kind of discussion I was hoping for.

After reading about the cases of the 1812 Overture and Queen
recordings, I decided to try another kind of artificial test
file. I put a 1 kHz, -6 dBFS sinewave in one channel, and a
1 kHz, -6 dBFS square wave in the other.

Here is a cycle of the waveforms from the .wav file, as
viewed in Audacity:

http://jayts.com/images/Sine_Square_1kHz-5dB-Audacity.png

And here is the waveform, played at full volume with VLC
and as viewed by Zelscope, recording the audio card's inputs:

http://jayts.com/images/Sine_Square_1kHz-5dB-Zelscope.png

When played into my Tektronix oscilloscope, I measured +1.9 dB
excursions in the square wave, beyond the amplitude of the sine
wave, in both the rising and falling edges of the square wave.
This matches what the author of the Over The Top webpage says
he found in the recordings: "The results show excursions reaching
peaks up to +2dBFS [sic.]."

I used a 1 kHz square wave because I was thinking imaginatively that
guitars reach up to around 1500 Hz at the 24th fret of the high E string.
Put that through some heavy metal distortion, and process it
carelessly, and you've got something very close to a square wave
(that is, with the tops and bottoms clipped to horizontal, flat lines
somewhere in the digital world). Toss that into the "loudness wars", and
you'll have the flat lines "normalized" to 0 dBFS.

There are two things on my mind now:

1. Have we been doing normalizing wrong? Waveform editors are
happy to show waveforms as "connect the dots", essentially
using a pencil and ruler technique. This is not really correct,
and leads the user into a false sense of confidence. Worse, they
also are happy to "normalize to 0 dBFS", when they are actually
normalizing to maybe +2 or at the extreme, +5 dB.

Is it time we bugged the programmers for a WYSIWYG waveform
view, and a "smart normalize" checkbox in the normalize
function's dialog box?

2. Even if the output device's D/A converter clips the waveform,
is that clipping ever actually audible? In most cases, it's
just 2 dB at most, and it's happening at a very high frequency.

In both the WFH and the waveform I showed earlier, the
oscillations that are sending the signal over 0 dB are at
about 20+ KHz. Harmonic distortion resulting from the clipping
produces a 2nd harmonic around 40+ KHz. It may be audible to
insects and bats, assuming it makes it through the player's
analog output circuit.

So again, is this really worth caring about?

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 15 Nov 2009 13:01:01 GMT, Jay Ts >
wrote:

>Don Pearce wrote:
>> Some real enlightenment is to be found here
>>
>> http://www.audiomisc.co.uk/HFN/OverTheTop/OTT.html
>
>Thanks. That's more the kind of discussion I was hoping for.
>
>After reading about the cases of the 1812 Overture and Queen
>recordings, I decided to try another kind of artificial test
>file. I put a 1 kHz, -6 dBFS sinewave in one channel, and a
>1 kHz, -6 dBFS square wave in the other.
>
>Here is a cycle of the waveforms from the .wav file, as
>viewed in Audacity:
>
>http://jayts.com/images/Sine_Square_1kHz-5dB-Audacity.png
>
>And here is the waveform, played at full volume with VLC
>and as viewed by Zelscope, recording the audio card's inputs:
>
>http://jayts.com/images/Sine_Square_1kHz-5dB-Zelscope.png
>
>When played into my Tektronix oscilloscope, I measured +1.9 dB
>excursions in the square wave, beyond the amplitude of the sine
>wave, in both the rising and falling edges of the square wave.
>This matches what the author of the Over The Top webpage says
>he found in the recordings: "The results show excursions reaching
>peaks up to +2dBFS [sic.]."
>
Yes, but in the case of the square wave those excursions are artefacts
caused by the Gibbs phenomenon.

>I used a 1 kHz square wave because I was thinking imaginatively that
>guitars reach up to around 1500 Hz at the 24th fret of the high E string.
>Put that through some heavy metal distortion, and process it
>carelessly, and you've got something very close to a square wave
>(that is, with the tops and bottoms clipped to horizontal, flat lines
>somewhere in the digital world). Toss that into the "loudness wars", and
>you'll have the flat lines "normalized" to 0 dBFS.
>
>There are two things on my mind now:
>
>1. Have we been doing normalizing wrong? Waveform editors are
> happy to show waveforms as "connect the dots", essentially
> using a pencil and ruler technique. This is not really correct,
> and leads the user into a false sense of confidence. Worse, they
> also are happy to "normalize to 0 dBFS", when they are actually
> normalizing to maybe +2 or at the extreme, +5 dB.
>

Audacity uses a dot-to-dot method, but my editor of choice (Audition)
uses a proper sinc filter to generate true trajectories. But it only
does that when you are zoomed in close. When looking at a wide view,
it is dot-to-dot just like Audacity, so you miss any overs.

> Is it time we bugged the programmers for a WYSIWYG waveform
> view, and a "smart normalize" checkbox in the normalize
> function's dialog box?
>
Would be good, but it will be a slow process due to the need to
oversample and reconstruct the entire tune in software.

>2. Even if the output device's D/A converter clips the waveform,
> is that clipping ever actually audible? In most cases, it's
> just 2 dB at most, and it's happening at a very high frequency.
>

Whether it is audible or not depends. If every peak in a long note has
a couple of dB lopped off the top, definitely. But most peaks are a
drum hit, and no, this won't be audible on those.

> In both the WFH and the waveform I showed earlier, the
> oscillations that are sending the signal over 0 dB are at
> about 20+ KHz. Harmonic distortion resulting from the clipping
> produces a 2nd harmonic around 40+ KHz. It may be audible to
> insects and bats, assuming it makes it through the player's
> analog output circuit.
>
No, that is not how it is. Those peaks are the result of the sum of
all the harmonics, not just a high one. And of course because this
clipping is happening in the digital domain it will result in severe
aliasing right into the heart of the audio band.

> So again, is this really worth caring about?
>

Oh yes. Even SSL have designed a VST plugin PPM to respond to
intersample peaks.

d

[Scott Dorsey]

Jay Ts > wrote:
>
>1. Have we been doing normalizing wrong? Waveform editors are
> happy to show waveforms as "connect the dots", essentially
> using a pencil and ruler technique. This is not really correct,
> and leads the user into a false sense of confidence. Worse, they
> also are happy to "normalize to 0 dBFS", when they are actually
> normalizing to maybe +2 or at the extreme, +5 dB.

Normalizing sets the maximum digital value up to where the maximum digital
value should be. There's nothing wrong with this. It still _is_ normalizing
the digital value to 0dBFS and it still _is_ producing the maximum digital
level without clipping.

The issue is that some converters cannot handle some waveforms with peaks
at 0dBFS. That is an issue with bad converter design, not with the concept
of normalization. Unfortunately, there are a lot of bad converters out
there.

> Is it time we bugged the programmers for a WYSIWYG waveform
> view, and a "smart normalize" checkbox in the normalize
> function's dialog box?

WYSIWYG waveform view is essential. Sonic has had it for years, but it's
a different display than the envelope view (at least on my very old NuBus
system). The problem is that people look at envelopes and they think they
are looking at waveforms when they are not.

A smart normalize or a smart limiter could be a useful function if you
are trying to eke out the absolute highest signal level without clipping.

>2. Even if the output device's D/A converter clips the waveform,
> is that clipping ever actually audible? In most cases, it's
> just 2 dB at most, and it's happening at a very high frequency.

Like all clipping, I bet it depends on the waveform. In an era where
people are applying aggressive amounts of limiting even to folk and
classical music, I think the clipping from overshoot could become an
issue because there are just so many peaks so close to the line in so
many recordings.

If it was an occasional peak hitting the mark now and then, and there
was an overshoot for a sample or two, it wouldn't be an issue at all.
It's really a very minor problem, until you turn it into a big problem
by putting huge number of FS peaks in.


> In both the WFH and the waveform I showed earlier, the
> oscillations that are sending the signal over 0 dB are at
> about 20+ KHz. Harmonic distortion resulting from the clipping
> produces a 2nd harmonic around 40+ KHz. It may be audible to
> insects and bats, assuming it makes it through the player's
> analog output circuit.
>
> So again, is this really worth caring about?

To my mind, the solution for this is to stop doing abusive limiting and
to live with an occasional crunched peak on some older converters.
--scott

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

[Les Cargill[_2_]]

hank alrich wrote:
> Les Cargill > wrote:
>
>> Don Pearce wrote:
>>> On Sat, 14 Nov 2009 16:25:31 +0100, "Peter Larsen"
>> <snip>
>>> No. No better than the last.
>>>
>>> MP3 conversion has no relevance here, but the behaviour of the typical
>>> DAC is very much the issue, since it determines whether or not it is
>>> safe to normalize to zero dB, considering that between samples there
>>> will be peaks greater than 0dB that the DAC must handle. The purpose
>>> of this thread has been to determine how big those peaks can be, and
>>> what the resulting audio will be like. This is nothing to do with
>>> catching out poor DACs.
>>>
>>> d
>> Hang on. We have a cost associated with how people perceive our work.
>> This cost has two components - perceived level ( which we will be
>> penalized "unfairly" for if it's too low ) or having a D/A converter
>> mangle the signal.
>>
>> This puts us in a position of having to compromise between those two
>> costs. It's a classic trade space, and any choice is *going* to
>> be arbitrary. I can't see normalizing to less than -1.0 dB, but that's
>> my personal arbitrary choice. And given JayTs input, that choice
>> may change - but I don't think real world signals have the kind of...
>> "artificial" spectrum his Wave from Hell has. Do an FFT on that thing -
>> it's brutal. Which is the point.
>>
>> What really would serve our best interest is to have the performance of
>> all possible such devices known. That's not likely, but that would
>> be what we'd need. Normalizing to -5dB seems like it's too much,
>> although if it became a standard, that'd be just fine - 16 bits
>> has plenty of bandwidth for all peaks being < 5dB. But then there's
>> the Loudness Wars.....
>>
>> --
>> Les Cargill
>
> Y'all might find some enlightening discussion here:
>
> http://recforums.prosoundweb.com/index.php/m/280407/0/
>
> and here:
>
> http://tinyurl.com/yha798o
>


Zounds! Thank you.

--
Les Cargill

[Les Cargill[_2_]]

Jay Ts wrote:
> Don Pearce wrote:
<snip>
>
> So again, is this really worth caring about?
>
> Jay Ts

Probably not. The examples given were "found in nature",
but verge on ... malpractice.

Isn't this just a gain staging problem? IOW, a device
which purports to properly (re)produce PCM should handle this -
it should be able to output an untrashed waveform with
the +5db "overs". I would not be very prepared to
compromise on this. Anything that does not properly
manage *all* possible input data is, by definition,
broken.

If devices exist which don't handle it, well... bummer.
Doctor, doctor it hurts when I do that....

Now, if the receiving analog device doesn't have enough
headroom to *receive* the Wave From Hell, then you have
to pad it. Or live with it.

--
Les Cargill

[Geoff]

Peter Larsen wrote:
> Don Pearce wrote:
>
>> MP3 conversion has no relevance here
>
> They ARE going to happen, so they are relevant.
>
>> but the behaviour of the typical
>> DAC is very much the issue, since it determines whether or not it is
>> safe to normalize to zero dB, considering that between samples there
>> will be peaks greater than 0dB that the DAC must handle. The purpose
>> of this thread has been to determine how big those peaks can be, and
>> what the resulting audio will be like. This is nothing to do with
>> catching out poor DACs.
>
> This thread started with a question about whether it was sane
> practice to normalize to 0 dB. It seems to be an excellent and simple
> point that it is not because there will be overshoot in mp3
> encode-decode.


Why should I pander to those who wish to destroy the music into MP#.

I'll continue doing what I do, and MP3 encoder writers should cope with
THEIR PROBLEM with pre-attenuation in the input to their encoding
algorithms.

geoff

[Peter Larsen[_3_]]

geoff wrote:

> Why should I pander to those who wish to destroy the music into MP#.

Scott's law: "he who pays decides". The paying public is gonna do
encode-decode it, so we do need to be aware of what happens when they do it.

> I'll continue doing what I do, and MP3 encoder writers should cope
> with THEIR PROBLEM with pre-attenuation in the input to their encoding
> algorithms.

THAT is a very good point, thanks!

> geoff

Kind regards

Peter Larsen

[Eric B[_2_]]

I approach it differently. I never normalize. What i do instead is put
a limiter on the master and watch how much gain reduction has
occurred. If it is more than minimal I reduce the level going to it.
This is unscientific but it seems to work. Digital zero is a limit not
to be f==ked with. When I record (track) I always aim at -12 to -15
with acoustic music and -20 with rock. During acquisition headroom is
cheap. During mix and master you are playing loudness war games.
Normalizing removes, or may remove, dynamic differences between songs.
There may be subjective reasons to have one song be quieter than
another.
I guess my point is that there is no reason to push the meters like
we did with analog. There is nothing to be gained except broadcast
dial volume and IMHO this is a silly non musical goal.
This discussion is very interesting though.
Best,
Eric B

[Jay Ts]

Don Pearce wrote:
> Unfortunately the SSL link referenced in the first post is dead.

The new page for the free X-ISM plugin is here:

http://www.solidstatelogic.com/support/X-ISM/

If you click on the Download tab near the top, they want you to
register and login before you are allowed to download anything.

If anyone gets that to work, please let me know.
I created an account, but the login didn't work. :(

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 16 Nov 2009 12:46:53 GMT, Jay Ts >
wrote:

>Don Pearce wrote:
>> Unfortunately the SSL link referenced in the first post is dead.
>
>The new page for the free X-ISM plugin is here:
>
>http://www.solidstatelogic.com/support/X-ISM/
>
>If you click on the Download tab near the top, they want you to
>register and login before you are allowed to download anything.
>
>If anyone gets that to work, please let me know.
>I created an account, but the login didn't work. :(
>
>Jay Ts

The logon worked for me, and I downloaded the meter, but it wasn't
terribly useful. It only operates in real time on playback. Also it
crashed Audition every time I used it. I would rather stick with
looking at the envelope, finding the highest bits and zooming out to
check for overs.

d

[Peter Larsen[_3_]]

Eric B wrote:

> I approach it differently. I never normalize. What i do instead is put
> a limiter on the master and watch how much gain reduction has
> occurred.

And the limiter-max is how many dB below 0 dB full scale, the outcome is the
same.

> If it is more than minimal I reduce the level going to it.
> This is unscientific but it seems to work. Digital zero is a limit not
> to be f==ked with. When I record (track) I always aim at -12 to -15
> with acoustic music and -20 with rock. During acquisition headroom is
> cheap. During mix and master you are playing loudness war games.
> Normalizing removes, or may remove, dynamic differences between songs.
> There may be subjective reasons to have one song be quieter than
> another.

Yes. And in classical concert recording to have entire works quieter than
other works so as to allow for an intentionally noisy tutti. Mostly that is
simple: set gain right and forget, afterwards unclip or normalize (possibly
combined with hard limit) if it was wrong. One never really knows with
violins and sopranos, but most other sonic implements have foreseeable
output, so set and forget usually just implies to leave gain where it was at
last similar recording.

> I guess my point is that there is no reason to push the meters like
> we did with analog. There is nothing to be gained except broadcast
> dial volume and IMHO this is a silly non musical goal.

Leave that sin to the broadcasters, as Orban explained doing it twice is
worse. Not all broadcaster level optimizing is sonically bad even tho' it
sometimes misrepresents what is going on in a work. Misrepresenting is a
word that implies the possibility of a Droit Morale violation.

> This discussion is very interesting though.

Indeed!

> Eric B

Kind regards

Peter Larsen

[Jay Ts]

Don Pearce wrote:
> The logon worked for me, and I downloaded the meter, but it wasn't
> terribly useful. [...]

Thanks for the review. If I can get it to work without
crashing, I think it will be useful here. Most of what
I'm doing now is playing live, and I actually don't
have recordings to look at. If it works in a live setup,
then I can use it.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On 16 Nov 2009 13:11:55 GMT, Jay Ts >
wrote:

>Don Pearce wrote:
>> The logon worked for me, and I downloaded the meter, but it wasn't
>> terribly useful. [...]
>
>Thanks for the review. If I can get it to work without
>crashing, I think it will be useful here. Most of what
>I'm doing now is playing live, and I actually don't
>have recordings to look at. If it works in a live setup,
>then I can use it.
>
>Jay Ts

Not sure what you can do with it live. It is a VST plugin for a DAW,
not a stand-alone application.

d

[Frank]

On 16 Nov 2009 12:46:53 GMT, in 'rec.audio.pro',
in article <Re: Never normalize to 0dB?>,
Jay Ts > wrote:

>Don Pearce wrote:
>> Unfortunately the SSL link referenced in the first post is dead.
>
>The new page for the free X-ISM plugin is here:
>
>http://www.solidstatelogic.com/support/X-ISM/
>
>If you click on the Download tab near the top, they want you to
>register and login before you are allowed to download anything.
>
>If anyone gets that to work, please let me know.
>I created an account, but the login didn't work. :(
>
>Jay Ts

Jay, if you can't login at http://www.solidstatelogic.com/downloads/
then try http://solid-state-logic.locotalk.com/ instead. If you can
successfully login there, then you can download the file from there.

[lcargill]

On Nov 15, 2:53 pm, Les Cargill > wrote:
> hank alrich wrote:
> > Les Cargill > wrote:
>
> >> Don Pearce wrote:
> >>> On Sat, 14 Nov 2009 16:25:31 +0100, "Peter Larsen"
> >> <snip>
> >>> No. No better than the last.
>
> >>> MP3 conversion has no relevance here, but the behaviour of the typical
> >>> DAC is very much the issue, since it determines whether or not it is
> >>> safe to normalize to zero dB, considering that between samples there
> >>> will be peaks greater than 0dB that the DAC must handle. The purpose
> >>> of this thread has been to determine how big those peaks can be, and
> >>> what the resulting audio will be like. This is nothing to do with
> >>> catching out poor DACs.
>
> >>> d
> >> Hang on. We have a cost associated with how people perceive our work.
> >> This cost has two components - perceived level ( which we will be
> >> penalized "unfairly" for if it's too low ) or having a D/A converter
> >> mangle the signal.
>
> >> This puts us in a position of having to compromise between those two
> >> costs. It's a classic trade space, and any choice is *going* to
> >> be arbitrary. I can't see normalizing to less than -1.0 dB, but that's
> >> my personal arbitrary choice. And given JayTs input, that choice
> >> may change - but I don't think real world signals have the kind of...
> >> "artificial" spectrum his Wave from Hell has. Do an FFT on that thing -
> >> it's brutal. Which is the point.
>
> >> What really would serve our best interest is to have the performance of
> >> all possible such devices known. That's not likely, but that would
> >> be what we'd need. Normalizing to -5dB seems like it's too much,
> >> although if it became a standard, that'd be just fine - 16 bits
> >> has plenty of bandwidth for all peaks being < 5dB. But then there's
> >> the Loudness Wars.....
>
> >> --
> >> Les Cargill
>
> > Y'all might find some enlightening discussion here:
>
> >http://recforums.prosoundweb.com/index.php/m/280407/0/
>
> > and here:
>
> >http://tinyurl.com/yha798o
>
> Zounds! Thank you.
>
> --
> Les Cargill

Eternal September is apparently down, so this is coming in from
Google.

What I saw was that reducing gain by < 2dB, around 1.4 dB, would stop
the X-ISM
meter from lighting up. Not -5dB, but -1.4 something.


---
Les Cargill

[Tobiah]

>
> As an aside (while absorbing beer after the flight from Austin and car
> ride from Reno), iPods sond remarkably better playing back .wav files
> that they do playing back low bitrate mp3's, for some strange reason.

Um, because low bitrate mp3's sound worse than cd quality .wav files
on *any* device? Or maybe the beer switched the sense of your
statement :)

[hank alrich]

Tobiah > wrote:

> >
> > As an aside (while absorbing beer after the flight from Austin and car
> > ride from Reno), iPods sond remarkably better playing back .wav files
> > that they do playing back low bitrate mp3's, for some strange reason.
>
> Um, because low bitrate mp3's sound worse than cd quality .wav files
> on *any* device? Or maybe the beer switched the sense of your
> statement :)

I was being humorous in context. <g> People slag iPods, and while they
aren't the greatest sound sources, what you get out does depend on what
you put in.

--
ha
shut up and play your guitar

[Geoff]

hank alrich wrote:
> Tobiah > wrote:
>
>>>
>>> As an aside (while absorbing beer after the flight from Austin and
>>> car ride from Reno), iPods sond remarkably better playing back .wav
>>> files that they do playing back low bitrate mp3's, for some strange
>>> reason.
>>
>> Um, because low bitrate mp3's sound worse than cd quality .wav files
>> on *any* device? Or maybe the beer switched the sense of your
>> statement :)
>
> I was being humorous in context. <g> People slag iPods, and while they
> aren't the greatest sound sources, what you get out does depend on
> what you put in.

.... but it's better to have 3,000,000 songs that sound like crap than to
have only half a million that sound as good as a small battery-powered
pocket device can deliver, no ?

geoff

[Jay Ts]

Frank wrote:
> Jay Ts wrote:
>
>>Don Pearce wrote:
>>> Unfortunately the SSL link referenced in the first post is dead.
>>
>>The new page for the free X-ISM plugin is here:
>>
>>http://www.solidstatelogic.com/support/X-ISM/
>>
>>If you click on the Download tab near the top, they want you to register
>>and login before you are allowed to download anything.
>>
>>If anyone gets that to work, please let me know. I created an account,
>>but the login didn't work. :(
>
> Jay, if you can't login at http://www.solidstatelogic.com/downloads/
> then try http://solid-state-logic.locotalk.com/ instead. If you can
> successfully login there, then you can download the file from there.

That worked. Thank you.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[hank alrich]

geoff > wrote:

> hank alrich wrote:
> > Tobiah > wrote:
> >
> >>>
> >>> As an aside (while absorbing beer after the flight from Austin and
> >>> car ride from Reno), iPods sond remarkably better playing back .wav
> >>> files that they do playing back low bitrate mp3's, for some strange
> >>> reason.
> >>
> >> Um, because low bitrate mp3's sound worse than cd quality .wav files
> >> on *any* device? Or maybe the beer switched the sense of your
> >> statement :)
> >
> > I was being humorous in context. <g> People slag iPods, and while they
> > aren't the greatest sound sources, what you get out does depend on
> > what you put in.
>
> ... but it's better to have 3,000,000 songs that sound like crap than to
> have only half a million that sound as good as a small battery-powered
> pocket device can deliver, no ?
>
> geoff

No.

--
ha
shut up and play your guitar

[Frank]

On 17 Nov 2009 01:50:30 GMT, in 'rec.audio.pro',
in article <Re: Never normalize to 0dB?>,
Jay Ts > wrote:

>Frank wrote:

>> Jay, if you can't login at http://www.solidstatelogic.com/downloads/
>> then try http://solid-state-logic.locotalk.com/ instead. If you can
>> successfully login there, then you can download the file from there.
>
>That worked. Thank you.
>
>Jay Ts

You're very welcome!

--
Frank, Independent Consultant, New York, NY
[Please remove 'nojunkmail.' from address to reply via e-mail.]
Read Frank's thoughts on HDV at http://www.humanvalues.net/hdv/
(also covers AVCHD and XDCAM EX).

[Les Cargill[_2_]]

Jay Ts wrote:
> Don Pearce wrote:
>> Unfortunately the SSL link referenced in the first post is dead.
>
> The new page for the free X-ISM plugin is here:
>
> http://www.solidstatelogic.com/support/X-ISM/
>
> If you click on the Download tab near the top, they want you to
> register and login before you are allowed to download anything.
>
> If anyone gets that to work, please let me know.
> I created an account, but the login didn't work. :(
>
> Jay Ts

I got it to work. It shows that setting gain to -1.4dB ( or so )
on your Wave from Hell will obviate any between-sample overs.

--
Les Cargill

[Don Pearce[_3_]]

On Tue, 17 Nov 2009 21:00:33 -0500, Les Cargill
> wrote:

>I got it to work. It shows that setting gain to -1.4dB ( or so )
>on your Wave from Hell will obviate any between-sample overs.

So it doesn't work, then? -5dB is the correct level.

d

[Les Cargill[_2_]]

Don Pearce wrote:
> On Tue, 17 Nov 2009 21:00:33 -0500, Les Cargill
> > wrote:
>
>> I got it to work. It shows that setting gain to -1.4dB ( or so )
>> on your Wave from Hell will obviate any between-sample overs.
>
> So it doesn't work, then? -5dB is the correct level.
>
> d


I don't know. If -5dB is the right answer, then yes, it
doesn't work.

--
Les Cargill

[Les Cargill[_2_]]

Don Pearce wrote:
> On Wed, 18 Nov 2009 18:01:47 -0500, Les Cargill
> > wrote:
>
>> Don Pearce wrote:
>>> On Tue, 17 Nov 2009 21:00:33 -0500, Les Cargill
>>> > wrote:
>>>
>>>> I got it to work. It shows that setting gain to -1.4dB ( or so )
>>>> on your Wave from Hell will obviate any between-sample overs.
>>> So it doesn't work, then? -5dB is the correct level.
>>>
>>> d
>>
>> I don't know. If -5dB is the right answer, then yes, it
>> doesn't work.
>
> 5dB (or rather 5.1dB) is the right answer.

Sorry. I remain unconvinced. I'd really have to do it all
hands on to believe 5.1dB. The more I think about it, the
more this reminds me of Gibbs phenomenon, and I don't buy
that being mathematically inevitable, either ( which is to
say I don't have deep enough DSP chops to actually
understand the thing - maybe some day).

I'm doing some work right now (not *work* work, just Adult
Continuing Ed on my own ) with IIR filters ala Bessel,
Chebyshev, Butterworth so perhaps there's hope. I'm inching
slowly towards understanding sinc and Heaviside, which I
expect to turn the light on.

> I suspect they have tried
> to use an algorithm that is a bit faster than the proper sinc
> function.
>

No clue here. All I know is what I saw. Can you try it, to
get another set of eyes on it?

> d

--
Les Cargill

[Jay Ts]

Les Cargill wrote:
> Don Pearce wrote:
>> Les Cargill wrote:
>>> Don Pearce wrote:
>>>> Les Cargill wrote:
>>>>
>>>>> I got it to work. It shows that setting gain to -1.4dB ( or so ) on
>>>>> your Wave from Hell will obviate any between-sample overs.
>>>> So it doesn't work, then? -5dB is the correct level.
>>>
>>> I don't know. If -5dB is the right answer, then yes, it doesn't work.
>>
>> 5dB (or rather 5.1dB) is the right answer.
[...]
>> I suspect they have tried
>> to use an algorithm that is a bit faster than the proper sinc function.
>>
> No clue here. All I know is what I saw. Can you try it, to get another
> set of eyes on it?

I tried it, and it took just -1.938 dB of gain to keep
X-ISM's "DIG" warning light from coming on.

Jay Ts
--
To contact me, use this web page:
http://www.jayts.com/contact.php

[Don Pearce[_3_]]

On Thu, 19 Nov 2009 20:50:24 -0500, Les Cargill
> wrote:

>>> I don't know. If -5dB is the right answer, then yes, it
>>> doesn't work.
>>
>> 5dB (or rather 5.1dB) is the right answer.
>
>Sorry. I remain unconvinced. I'd really have to do it all
>hands on to believe 5.1dB. The more I think about it, the
>more this reminds me of Gibbs phenomenon, and I don't buy
>that being mathematically inevitable, either ( which is to
>say I don't have deep enough DSP chops to actually
>understand the thing - maybe some day).

No, it isn't Gibbs phenomenon, although that is also visible along the
flats of the square wave. This is a real peak made by local summing of
harmonics. It has been created mathematically and does indeed climb
5.1dB above the pair of samples either side.

d

[Arny Krueger]

"Les Cargill" > wrote in message

> Don Pearce wrote:
>> On Wed, 18 Nov 2009 18:01:47 -0500, Les Cargill
>> > wrote:
>>
>>> Don Pearce wrote:
>>>> On Tue, 17 Nov 2009 21:00:33 -0500, Les Cargill
>>>> > wrote:
>>>>
>>>>> I got it to work. It shows that setting gain to
>>>>> -1.4dB ( or so ) on your Wave from Hell will obviate
>>>>> any between-sample overs.
>>>> So it doesn't work, then? -5dB is the correct level.
>>>>
>>>> d
>>>
>>> I don't know. If -5dB is the right answer, then yes, it
>>> doesn't work.
>>
>> 5dB (or rather 5.1dB) is the right answer.
>
> Sorry. I remain unconvinced. I'd really have to do it all
> hands on to believe 5.1dB. The more I think about it, the
> more this reminds me of Gibbs phenomenon, and I don't buy
> that being mathematically inevitable, either ( which is to
> say I don't have deep enough DSP chops to actually
> understand the thing - maybe some day).

I'm with you. I've long been aware of the two most common reasons for
unexpected high peak levels:

(1) Samples that are <= FS, but are connected by a smooth curve that is >
FS.

(2) Uncompressed mixes where every blue moon every channel peaks at the same
sample time and you get a rogue peak > FS.

IME you can clip a sample or three very few minutes and nobody is the wiser
except maybe some meter heads. This is vastly different than having 1% or
more of the samples clipped which I am not recommending.

Tracks where an occastional peak is -5 dB and everything is even softer are
problematical for listeners because they will sound weak as compared to
other music they are being played with.

I normalize to -1 dB as a rule. If there are only a few samples that are at
the peak level, I go out and manually attenuate them with an envelope.

[Mark]

Re Gibbs,

it is actually interesting to note that if you start with a 1 Vp-p
square wave and then REMOVE all the harmonics, the resulting
fundamental sine wave left over will actually be LARGER then the
square wave you started with, I think it is 1.43 Vp-p but I could be
wrong about the exact value.

The harmonics acually serve to flatten and reduce the p-p value and if
you remove them, the p-p increases.

Anther interesting thing, if you take say a 1 kHz square wave and
remove the 5th harmonic so there is no 5 kHz energy present and look
at the result, it will look like it has 5 kHz in it. The square wave
needs the 5 kHz to be flat, and if you remove the 5 kHz the result
osciallates at 5 kHz.

Nature is interesting.

Mark

[Les Cargill[_2_]]

Mark wrote:
> Re Gibbs,
>
> it is actually interesting to note that if you start with a 1 Vp-p
> square wave and then REMOVE all the harmonics, the resulting
> fundamental sine wave left over will actually be LARGER then the
> square wave you started with,

Not "area under the curve" larger, but perhaps peakier. If you remove
harmonics that cancel other harmonics...

I think it is 1.43 Vp-p but I could be
> wrong about the exact value.
>
> The harmonics acually serve to flatten and reduce the p-p value and if
> you remove them, the p-p increases.
>
> Anther interesting thing, if you take say a 1 kHz square wave and
> remove the 5th harmonic so there is no 5 kHz energy present and look
> at the result, it will look like it has 5 kHz in it. The square wave
> needs the 5 kHz to be flat, and if you remove the 5 kHz the result
> osciallates at 5 kHz.
>
> Nature is interesting.
>
> Mark
>
>


--
Les Cargill

[Mark]

On Nov 20, 7:35*pm, Les Cargill > wrote:
> Mark wrote:
> > Re Gibbs,
>
> > it is actually interesting to note that if you start with a 1 Vp-p
> > square wave and then REMOVE all the harmonics, *the resulting
> > fundamental sine wave left over will actually be LARGER then the
> > square wave you started with,
>
> Not "area under the curve" larger, but perhaps peakier. If you remove
> harmonics that cancel other harmonics...
>
> * I think it is 1.43 Vp-p but I could be
>
>
>
> > wrong about the exact value.
>
> > The harmonics acually serve to flatten and reduce the p-p value and if
> > you remove them, the p-p increases.
>
> > Anther interesting thing, if you take say a 1 kHz square wave and
> > remove the 5th harmonic so there is no 5 kHz energy present and look
> > at the result, it will look like it has 5 kHz in it. *The square wave
> > needs the 5 kHz to be flat, and if you remove the 5 kHz the result
> > osciallates *at 5 kHz.
>
> > Nature is interesting.
>
> > Mark
>
> --
> Les Cargill

I had a chance to look it up..

a 1 Vp-p square wave, if all the harmonic are removed will be a
1.2732 Vp-p sine wave.

It's 4/pi actually.

Mark

[Don Pearce[_3_]]

On Fri, 20 Nov 2009 19:35:14 -0500, Les Cargill
> wrote:

>Not "area under the curve" larger, but perhaps peakier. If you remove
>harmonics that cancel other harmonics...

How would harmonics cancel other harmonics? They are at different
frequencies and can't cancel each other.

d

[Mark]

On Nov 21, 1:11*am, (Don Pearce) wrote:
> On Fri, 20 Nov 2009 19:35:14 -0500, Les Cargill
>
> > wrote:
> >Not "area under the curve" larger, but perhaps peakier. If you remove
> >harmonics that cancel other harmonics...
>
> How would harmonics cancel other harmonics? They are at different
> frequencies and can't cancel each other.
>
> d

there is an animation near the bottom of this page..

http://en.wikipedia.org/wiki/Trigonometric_functions

they don't cancel out, they flatten out the peak of the sine wave,
the animation makes it very clear..

Mark

[Don Pearce[_3_]]

On Sat, 21 Nov 2009 21:16:03 -0800 (PST), Mark >
wrote:

>On Nov 21, 1:11*am, (Don Pearce) wrote:
>> On Fri, 20 Nov 2009 19:35:14 -0500, Les Cargill
>>
>> > wrote:
>> >Not "area under the curve" larger, but perhaps peakier. If you remove
>> >harmonics that cancel other harmonics...
>>
>> How would harmonics cancel other harmonics? They are at different
>> frequencies and can't cancel each other.
>>
>> d
>
>there is an animation near the bottom of this page..
>
>http://en.wikipedia.org/wiki/Trigonometric_functions
>
>they don't cancel out, they flatten out the peak of the sine wave,
>the animation makes it very clear..
>
>Mark

That is just showing the construction of a square wave. They can
flatten or sum to a narrow peak, depending on their relative phase.

d

[Mark]

On Nov 22, 3:56*am, (Don Pearce) wrote:
> On Sat, 21 Nov 2009 21:16:03 -0800 (PST), Mark >
> wrote:
>
>
>
> >On Nov 21, 1:11*am, (Don Pearce) wrote:
> >> On Fri, 20 Nov 2009 19:35:14 -0500, Les Cargill
>
> >> > wrote:
> >> >Not "area under the curve" larger, but perhaps peakier. If you remove
> >> >harmonics that cancel other harmonics...
>
> >> How would harmonics cancel other harmonics? They are at different
> >> frequencies and can't cancel each other.
>
> >> d
>
> >there is an animation near the bottom of this page..
>
> >http://en.wikipedia.org/wiki/Trigonometric_functions
>
> >they don't cancel out, *they flatten out the peak of the sine wave,
> >the animation makes it very clear..
>
> >Mark
>
> That is just showing the construction of a square wave. They can
> flatten or sum to a narrow peak, depending on their relative phase.
>
> d

Yes the harmonics have to be the right amplitude and phase to make a
square wave.

and if you take away all the harmonics, the sine wave that you have
left is 4/pi bigger then the square wave which I found surprising the
first time i saw it.

Mark

[Les Cargill[_2_]]

Don Pearce wrote:
> On Fri, 20 Nov 2009 19:35:14 -0500, Les Cargill
> > wrote:
>
>> Not "area under the curve" larger, but perhaps peakier. If you remove
>> harmonics that cancel other harmonics...
>
> How would harmonics cancel other harmonics? They are at different
> frequencies and can't cancel each other.
>
> d


They're not the same frequencies. If this
wasn't true, the effect of filtering a set of
frequencies would result in less max. amplitude,
and it doesn't always.

For lack a better term, removing destructively
interferring related sub wave forms must result in
something "peakier".

--
Les Cargill

[Les Cargill[_2_]]

Mark wrote:
> On Nov 22, 3:56 am, (Don Pearce) wrote:
>> On Sat, 21 Nov 2009 21:16:03 -0800 (PST), Mark >
>> wrote:
>>
>>
>>
>>> On Nov 21, 1:11 am, (Don Pearce) wrote:
>>>> On Fri, 20 Nov 2009 19:35:14 -0500, Les Cargill
>>>> > wrote:
>>>>> Not "area under the curve" larger, but perhaps peakier. If you remove
>>>>> harmonics that cancel other harmonics...
>>>> How would harmonics cancel other harmonics? They are at different
>>>> frequencies and can't cancel each other.
>>>> d
>>> there is an animation near the bottom of this page..
>>> http://en.wikipedia.org/wiki/Trigonometric_functions
>>> they don't cancel out, they flatten out the peak of the sine wave,
>>> the animation makes it very clear..
>>> Mark
>> That is just showing the construction of a square wave. They can
>> flatten or sum to a narrow peak, depending on their relative phase.
>>
>> d
>
> Yes the harmonics have to be the right amplitude and phase to make a
> square wave.
>
> and if you take away all the harmonics, the sine wave that you have
> left is 4/pi bigger then the square wave which I found surprising the
> first time i saw it.
>
> Mark
>
>
>


So there has to be destructive interference removed with the harmonics
taken away? I *vaguely* recall this being the case, but that was all
some time ago.

--
Les Cargill