Login / Register
 
Solutions for True Peak / Intersample Peak limiting that really work?
New Reply
Subscribe
#61
24th July 2011
Old 24th July 2011
  #61
Lives for experience
 
minister's Avatar
 
Joined: Nov 2004
Location: minneapolis, mn
Posts: 2,301
My Recordings/Credits

minister is offline
Quote:
Originally Posted by Alexey Lukin View Post
I
The summary: my DAC can successfully play inter-sample peaks up to 3 dB, but not all DACs are created the same. Some of them may clip on the first waveform. This brings the issue of limiting not only digital levels, but also analog levels (which has been described in many papers by TC Electronic).
Alexey, is this one of the papers to which you refer?

http://www.tcelectronic.com/media/ni...0_0dbfs_le.pdf
#62
24th July 2011
Old 24th July 2011
  #62
Lives for gear
 
Alexey Lukin's Avatar
 
Joined: Dec 2007
Posts: 970

Alexey Lukin is offline
Yes, it's definitely one of them. TC are big advocates of inter-sample peak awareness.
#63
24th July 2011
Old 24th July 2011
  #63
Lives for experience
 
minister's Avatar
 
Joined: Nov 2004
Location: minneapolis, mn
Posts: 2,301
My Recordings/Credits

minister is offline
Quote:
Originally Posted by Alexey Lukin View Post
Yes, it's definitely one of them. TC are big advocates of inter-sample peak awareness.
Thanks Alexey. Is this increase in peak values after PCM decoding (which is another way of saying what we are talking about with inter-sample peaks and reconstruction) related to the Gibbs effect or phenomenon? Or are they separate issues?
#64
24th July 2011
Old 24th July 2011
  #64
Lives for gear
 
Alexey Lukin's Avatar
 
Joined: Dec 2007
Posts: 970

Alexey Lukin is offline
They are related, but I'd rather say that it's the result of ringing in the impulse response of a reconstructing low-pass filter. Ringing is sometimes referred to as Gibbs phenomenon, but it's not strictly mathematically correct (and TC's explanation in page 3 is a bit confusing).
#65
24th July 2011
Old 24th July 2011
  #65
Lives for experience
 
minister's Avatar
 
Joined: Nov 2004
Location: minneapolis, mn
Posts: 2,301
My Recordings/Credits

minister is offline
Quote:
Originally Posted by Alexey Lukin View Post
They are related, but I'd rather say that it's the result of ringing in the impulse response of a reconstructing low-pass filter. Ringing is sometimes referred to as Gibbs phenomenon, but it's not strictly mathematically correct (and TC's explanation in page 3 is a bit confusing).
Ok, thanks Alexey! It was a slightly afield curiosity I had since I came across it a few years back and don't understand it all fully. And thanks for sparing me the Mathematical explanation which I know you understand. High School calculus, Trig, and Algebra was a long time ago for me and you would quickly lead me into the deep weeds..........

AFAIK, Dan Lavry believes that 6dB is not a real world figure, but a 3dB increase can happen in the real world.

Be that as it may, as practical advice to the original poster: set Ozone according to what Alexey describes and turn down your levels -- or the threshold as the good Dr. prescribes. I have not had anything kicked back because of this, but I have turned things down a tad when looking at a reconstruction meter.

EDIT: Again to the original poster, this stuff tends to happen when you have an excess of High-Mid freq and High frequency material. Is your mix pretty bright? If so, try softening the high w/o muddying it up and see what that buys you.

Last edited by minister; 24th July 2011 at 06:44 AM.. Reason: Adding additional thought.
#66
24th July 2011
Old 24th July 2011
  #66
Lives for gear
 
Joined: Dec 2005
Location: San Francisco area
Posts: 3,348

philper is offline
Utterly bonehead question re: Elephant: how do I set it up as a brickwall limiter, say for -3 dbfs? I got the demo but the controls are unfamiliar enough to me that I don't think I'm getting to what I need.

thanks

phil p
#67
24th July 2011
Old 24th July 2011
  #67
Lives for gear
 
kk@jamsync.com's Avatar
 
Joined: May 2005
Posts: 928

kk@jamsync.com is offline
Quote:
Originally Posted by philper View Post
Utterly bonehead question re: Elephant: how do I set it up as a brickwall limiter, say for -3 dbfs? I got the demo but the controls are unfamiliar enough to me that I don't think I'm getting to what I need.

thanks

phil p
Are you using the "clip" algorithm and setting the output gain to -3 dBFS? Just curious what you're doing...I'm new to that plug as well.
__________________
___________________
K. K. Proffitt
President, JamSync®, Nashville
www.jamsync.com
http://jamsyncnashville.blogspot.com
(615) 320-5050
#68
25th July 2011
Old 25th July 2011
  #68
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
Quote:
Originally Posted by Alexey Lukin View Post
Actually any digital waveform, no matter how square, is a legit bandlimited signal (i.e. is a sampling of a certain bandlimited analog signal).
but not one that your d/a converter box can handle (?)
#69
25th July 2011
Old 25th July 2011
  #69
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
Quote:
Originally Posted by kk@jamsync.com View Post
In the real world of delivery specs and meeting them, especially when new standards throw new rules and limitations into the mix, the problem is binary:
pass or no pass. If you don't deliver what they say in exactly the way they say it, you fail. And unlike graduate school (yes a few of us went and have some grasp of mathematics), they don't grade on a curve.
that is different than reality
and what do you do if the spec is built on bullbleep
Quote:
I think your real argument is with the Recommendation ITU-R BS.1770-2 (03/2011). Perhaps you should write to them about being high school students who believe urban myths.
recently got a copy because it got linked to wrt intersample peaks. i have no need to tilt windmills.
Quote:
The rest of us will endeavor to pass the new specs the first time they are given to us, using the most accurate and economical means we can find.
but what if the spec is built on nonsense
and all you get is random attempts to try to do it

might as well push on a rope

there may be a real problem
i remain unconvinced that it is the so called interpeak sample issue.

and at least one d/a vendor agrees with me when they said that the only problem is a badly designed d/a converter.
#70
25th July 2011
Old 25th July 2011
  #70
Lives for gear
 
Alexey Lukin's Avatar
 
Joined: Dec 2007
Posts: 970

Alexey Lukin is offline
Quote:
Originally Posted by oldeanalogueguy View Post
but not one that your d/a converter box can handle (?)
Well, right, some (rare) digital signals may produce too high inter-sample peaks and be clipped during D/A conversion.
#71
25th July 2011
Old 25th July 2011
  #71
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
Quote:
Originally Posted by Alexey Lukin View Post
I'm talking about the windowed sinc low-pass filter - the most popular type of oversampling filter in D/A converters. It definitely has the ability to increase peak levels of the signal.
thanks for that info. i have been looking at a lot of d/a converters and chips and have not seen that one yet.
but then they didnt have fractals when i went to school.
who would have guessed the cantor set was one.
Quote:
This was true for non-oversampling DACs, which in fact quite poorly respect the Nyquist theorem (in part of reconstruction accuracy). But nowadays all music DACs are oversampling, and they use low-pass filters in order to comply with a Nyquist theorem. And this is where inter-sample clipping may occur.
i thought all a/d were oversampled.
i need to look more at d/a. have not seen an oversampled one (but maybe it is hidden in the chip).
actually oversampling or not -- with zoh you can filter it any way you want and it cannot peak over the highest sample.
Quote:

Let me give you an example. Generate the following digital signal at 8 kHz sampling rate:


This is a 2 kHz sine tone digitally peaking at 0 dB FS, but its analog levels go up to +3 dB FS.
er... um..... analog has no FS
its just db below the peak voltage the d/a is designed to use

you cannot legitimately mix the digital and analog domains
Quote:

Now put your DAC to the test: play this tone. My DAC outputs a pure sine, not clipped. It means that my DAC has the capacity for accommodating at least 3 dB of inter-sample peaks.
mine plays it too. while it ahs the same shape it is smaller and there are no peaks unless I twist the gain knob up.
Quote:

Now convert (resample) this signal to 44.1 kHz using a decent SRC and let it clip (by saving in a 16-bit format). Here's what you get:


Now listen to it. Clearly, it's different: distorted.

The summary: my DAC can successfully play inter-sample peaks up to 3 dB, but not all DACs are created the same. Some of them may clip on the first waveform. This brings the issue of limiting not only digital levels, but also analog levels (which has been described in many papers by TC Electronic).
I agree that the problem is in the design of the d/a converter but is not due to so called intersample peaks.
Quote:
It's definitely not a myth, but I'd say that the problem is somewhat over-emphasized. If someone's production is getting rejected because there's an inter-sample peak of +0.9 dB, there may be something wrong with their QA standards or their understanding of dBTP measurements.
I agree that there may be a problem but not that it is the so called interpeak sample caused problem.
#72
25th July 2011
Old 25th July 2011
  #72
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
Quote:
Originally Posted by Alexey Lukin View Post
They are related, but I'd rather say that it's the result of ringing in the impulse response of a reconstructing low-pass filter. Ringing is sometimes referred to as Gibbs phenomenon, but it's not strictly mathematically correct (and TC's explanation in page 3 is a bit confusing).
gibbs phenomenon is the overshoot at one point

ringing is something else
although as you sum more freqs the samples that appear to ring converge to the square type wave except for the one point at the leading edge where there is an overshoot at one point
#73
25th July 2011
Old 25th July 2011
  #73
Lives for gear
 
Alexey Lukin's Avatar
 
Joined: Dec 2007
Posts: 970

Alexey Lukin is offline
Quote:
Originally Posted by oldeanalogueguy View Post
i thought all a/d were oversampled.
i need to look more at d/a. have not seen an oversampled one (but maybe it is hidden in the chip).
Both A/D and D/A converters are now all oversampled, low bit-depth (1-5 bits).


Quote:
Originally Posted by oldeanalogueguy View Post
actually oversampling or not -- with zoh you can filter it any way you want and it cannot peak over the highest sample.
It won't peak after ZOH, but it will peak after the low-pass filter (be it analog or digital).
With oversampled DACs, this filtering (and rise in peak levels) happens in digital.


Quote:
Originally Posted by oldeanalogueguy View Post
er... um..... analog has no FS
its just db below the peak voltage the d/a is designed to use
you cannot legitimately mix the digital and analog domains
I just meant "the voltage level corresponding to a digital full scale".


Quote:
Originally Posted by oldeanalogueguy View Post
I agree that the problem is in the design of the d/a converter but is not due to so called intersample peaks.
Part of the problem is the converter design: they should be able to handle levels 1 or 2 dB higher than digital full scale (because of the filtering involved).
But another part is lack of awareness about these peak levels: some recordings that have been digitally clipped may produce these inter-sample peaks up to +1 dB and beyond. It may not be practical to design D/A converters with so much headroom, so mastering engineers should be aware of possible problems with such "clipped" content.
#74
25th July 2011
Old 25th July 2011
  #74
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
#75
25th July 2011
Old 25th July 2011
  #75
Lives for gear
 
Joined: Dec 2005
Location: San Francisco area
Posts: 3,348

philper is offline
Quote:
Originally Posted by kk@jamsync.com View Post
Are you using the "clip" algorithm and setting the output gain to -3 dBFS? Just curious what you're doing...I'm new to that plug as well.
That didn't seem to work...not sure how to set this up?

phil p
#76
25th July 2011
Old 25th July 2011
  #76
Lives for gear
 
kk@jamsync.com's Avatar
 
Joined: May 2005
Posts: 928

kk@jamsync.com is offline
Quote:
Originally Posted by oldeanalogueguy View Post
thanks for that info. i have been looking at a lot of d/a converters and chips and have not seen that one yet.
but then they didnt have fractals when i went to school.
who would have guessed the cantor set was one.

[/I]
Sure they did unless you went to school before the 1920's (in which case, I guess you really ARE old). Julia sets have been around for over 90 years and they are obviously fractals, even though Mandelbrot didn't coin the term until 1975. FWIW a disconnected Julia set is a Cantor set, so yes they had them when you went to school, unless you are a vampire or something.

It would have been tough to graph them on an IBM 360 using Fortran IV and multi-stamped ASCII chars, and even the Amiga took several hours to draw a simple-yet-proper Mandelbrot in the mid-80's. Still Julia sets must have been around when you went to school.
#77
25th July 2011
Old 25th July 2011
  #77
Lives for gear
 
kk@jamsync.com's Avatar
 
Joined: May 2005
Posts: 928

kk@jamsync.com is offline
Quote:
Originally Posted by philper View Post
That didn't seem to work...not sure how to set this up?

phil p
Yeah, we could use some help. Not sure if the "width of the knee" is actually addressing the attack in clip mode. It's like where is the real transient attack adjustment?
#78
25th July 2011
Old 25th July 2011
  #78
Lives for gear
 
Joined: Dec 2005
Location: San Francisco area
Posts: 3,348

philper is offline
Another side-note: why would a broadcaster be so bent about an over of some tenths of a db?

phil p
#79
25th July 2011
Old 25th July 2011
  #79
Lives for gear
 
kk@jamsync.com's Avatar
 
Joined: May 2005
Posts: 928

kk@jamsync.com is offline
Quote:
Originally Posted by philper View Post
Another side-note: why would a broadcaster be so bent about an over of some tenths of a db?

phil p
Why would everyone in radio want stuff pushed to the max even though it sounds like crap?

Because they "hear" with their *eyes* and they do everything by committee and the lowest common denominator wins. If there's a new spec, they'll get the equipment budget to follow it and assign the newest intern to watch the meter and write down when it goes over. If someone accidentally knocks it out of calibration and no one notices, stuff can go on for months until enough people complain.

Even in the audio world, you see absolutely silly stuff. One time I was at the unveiling of a large board at a very large studio with a very important audio engineer who said over and over how proud he was that this new board did "convergence" for surround. Didn't even understand the concept, obviously, and a reporter for the local rags dutifully quoted him. Jaded, who, me?

Do I think it's dumb to reject for .9 dBTP? Yes. Do I think it will continue to happen? I think it's likely.
#80
25th July 2011
Old 25th July 2011
  #80
Lives for gear
 
Joined: Dec 2005
Location: San Francisco area
Posts: 3,348

philper is offline
Quote:
Originally Posted by philper View Post
That didn't seem to work...not sure how to set this up?

phil p
Here's what Aleksey@Voxengo said re: a support request I made yesterday--re using Elephant as a brickwall peak limiter

Greetings,

Simply lower the Out Gain to -3 or -10 and then adjust the In Gain as
you need.

Best regards,
Aleksey Vaneev
VST plugins, AU plugins, Professional audio plugins - Voxengo
#81
25th July 2011
Old 25th July 2011
  #81
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
Quote:
Originally Posted by philper View Post
Another side-note: why would a broadcaster be so bent about an over of some tenths of a db?

phil p
especially when the commercials are at least +20 over the average program content
#82
25th July 2011
Old 25th July 2011
  #82
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
Quote:
Originally Posted by Alexey Lukin View Post
Both A/D and D/A converters are now all oversampled, low bit-depth (1-5 bits).



It won't peak after ZOH, but it will peak after the low-pass filter (be it analog or digital).
With oversampled DACs, this filtering (and rise in peak levels) happens in digital.



I just meant "the voltage level corresponding to a digital full scale".



Part of the problem is the converter design: they should be able to handle levels 1 or 2 dB higher than digital full scale (because of the filtering involved).
But another part is lack of awareness about these peak levels: some recordings that have been digitally clipped may produce these inter-sample peaks up to +1 dB and beyond. It may not be practical to design D/A converters with so much headroom, so mastering engineers should be aware of possible problems with such "clipped" content.
i think we agree its the design of the d/a system that is the problem not some imaginary interpeak in the digital domain nyquist guarantees that there will be lots of analog "higher" than the digital values were. expecting the peaks to line up somehow is internet wizdumb at its best.
#83
25th July 2011
Old 25th July 2011
  #83
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
Quote:
Originally Posted by Alexey Lukin View Post
Well, right, some (rare) digital signals may produce too high inter-sample peaks and be clipped during D/A conversion.
i can show you examples where the analog "peaks" are way higher than the digital samples. that is nyquist at work. the more samples you have the less this happens. as samples go to infinity the "peak" goes to zero. with a properly designed d/a converter system any digital data will never cause an analog voltage that is too high. and if your does such then you should be adjusting it after conversion in the analog domain where you can see it not guessing in the digital domain where you do not know what the output will really be after conversion.
#84
25th July 2011
Old 25th July 2011
  #84
3 + infractions, forum membership suspended.
 
Joined: Jun 2011
Location: at home
Posts: 2,401

oldeanalogueguy is offline
Quote:
Originally Posted by kk@jamsync.com View Post
Sure they did unless you went to school before the 1920's (in which case, I guess you really ARE old). Julia sets have been around for over 90 years and they are obviously fractals, even though Mandelbrot didn't coin the term until 1975. FWIW a disconnected Julia set is a Cantor set, so yes they had them when you went to school, unless you are a vampire or something.

It would have been tough to graph them on an IBM 360 using Fortran IV and multi-stamped ASCII chars, and even the Amiga took several hours to draw a simple-yet-proper Mandelbrot in the mid-80's. Still Julia sets must have been around when you went to school.
yuppp we had the sets they had not yet invented fractals i didnt put them together until a student in an honors math class (in the room before i taught the intro to systems class) many years later told me they were doing fractals and then mentioned cantor. bingo!! i hadnt seen a cantor set for at least 30 years -- but then it hit me that those old cantor sets were fractals we just didnt know it yet.
#85
25th July 2011
Old 25th July 2011
  #85
Lives for gear
 
Alexey Lukin's Avatar
 
Joined: Dec 2007
Posts: 970

Alexey Lukin is offline
Quote:
Originally Posted by oldeanalogueguy View Post
with a properly designed d/a converter system any digital data will never cause an analog voltage that is too high.
You are probably right. Unfortunately, since in some special cases these analog peak levels can go very much beyond digital peak levels, these "properly designed" converters most likely do not exist.
And mastering is all about making recordings sound good on real-world playback systems, not on some ideal systems.
#86
25th July 2011
Old 25th July 2011
  #86
Lives for gear
 
nucelar's Avatar
 
Joined: Jan 2007
Location: Barcelona
Posts: 795

nucelar is offline
I think that the main concern of The EBU /ITU is not the problems that may ocurr during DA conversion, but during downstream coding/ transcoding / recoding, etc. of the audio stream. PCM audio which causes intersample peaks beyond full scale is known to cause artifacts during perceptual coding (mp3, ac3, aac, etc.), so a -1 dBTP limit is just a precautionary measure.

BTW I'm testing the Sonnox Limiter and I cannot find the way to tell it that the true peak ceiling should be at -1. Tom, have you succesfully used the sonnox for that purpose?
Cheers
#87
25th July 2011
Old 25th July 2011
  #87
Lives for gear
 
kk@jamsync.com's Avatar
 
Joined: May 2005
Posts: 928

kk@jamsync.com is offline
Quote:
Originally Posted by nucelar View Post
I think that the main concern of The EBU /ITU is not the problems that may ocurr during DA conversion, but during downstream coding/ transcoding / recoding, etc. of the audio stream. PCM audio which causes intersample peaks beyond full scale is known to cause artifacts during perceptual coding (mp3, ac3, aac, etc.), so a -1 dBTP limit is just a precautionary measure.
Good point. YouTube audio illustrates this fact. Many amateur videographers run the audio into the red. It may sound passable to their ears when they play it back at home, but once YouTube encodes it (and most people have no clue how to pass through their encoder), the audio is simply dreck. The video is often bad, but the audio is far worse.
#88
25th July 2011
Old 25th July 2011
  #88
Lives for gear
 
Alexey Lukin's Avatar
 
Joined: Dec 2007
Posts: 970

Alexey Lukin is offline
This is not necessarily related to inter-sample clipping and happens even when dBTP levels are perfectly limited.
The explanation of this effect is simply in the nature of lossy encoding: the waveform gets changed, which can randomly raise and lower peak levels.
#89
27th July 2011
Old 27th July 2011
  #89
Lives for gear
 
espasonico's Avatar
 
Joined: Jan 2005
Posts: 664

espasonico is offline
I´m confused.

So what´s the point of peak limiting if in the AD stage it´s gonna sound broken ?

It´s possible that those high itersample peaks appear when somebody applies a heavy limiting relaying too much on the limiter ?

I mean, if you limit a peak of 3dB, then, the intersample peak is gonna be around +3db ? That makes sense to me.
#90
27th July 2011
Old 27th July 2011
  #90
Lives for experience
 
minister's Avatar
 
Joined: Nov 2004
Location: minneapolis, mn
Posts: 2,301
My Recordings/Credits

minister is offline
Quote:
Originally Posted by nucelar View Post
BTW I'm testing the Sonnox Limiter and I cannot find the way to tell it that the true peak ceiling should be at -1. Tom, have you succesfully used the sonnox for that purpose?
Cheers
Select "Recon Meter". Now it tells you the effect of reconstruction.
New Reply Submit Thread to Facebook Facebook  Submit Thread to Twitter Twitter  Submit Thread to LinkedIn LinkedIn  Submit Thread to Google+ Google+ 
 
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump

SEO by vBSEO ©2011, Crawlability, Inc.