• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Practical example of intersample peak greater than +6 dBFS

ayane

Active Member
Forum Donor
Joined
Dec 15, 2018
Messages
182
Likes
684
Location
NorCal
I've mentioned it before here on this forum, but it's entirely possible to have music that has intersample peaks beyond +6 dBFS, or one whole bit. It was mentioned again later on this thread by @LTig.

Here is a 30-second excerpt of "Get Away" from the Electronic Soundtrack album Code Lyoko by the Subdigitals. The audio is from track 9 on the CD I've ripped myself, and the excerpt is from the 3:10 mark in the track. This track contains a lot of clipping, as is common in the genre and modern music, unfortunately, but it is of note because unlike other clipped tracks, this track contains not one but multiple intersample peaks that exceed +6 dBFS.

I have over 12,000 tracks in lossless format in my music collection, and a few of them hit a peak above +6 dBFS, while hundreds have peaks above +3 dBFS - the headroom in the Benchmark DAC3. I chose this particular example because it exemplifies the issue multiple times in the same track. I can post more examples if necessary.

I'm posting about this because I'm pretty disappointed with DAC digital filter design. It is this forum which has highlighted this problem, Many thanks to @amirm's measurements. Most DACs on the market don't even attenuate properly at the Nyquist frequency for 44.1 kHz audio, which comprises the entirety of CD digital audio and many, many more digital music files. On top of that, all of these intersample peaks that exceed 0 dBFS means that a digital filter needs headroom to properly oversample. Lots of headroom, not a wimpy 3 dB, but at least 6 dB (one whole bit), and ideally 8-9 dB to cover these fringe cases too. So, I'm calling out all the manufacturers who don't do any of this properly: this isn't rocket science, folks. It's 2020 and the math has been beaten to death. #MakeDACFiltersGreatAgain
 
Last edited:

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
On top of that, all of these intersample peaks that exceed 0 dBFS means that a digital filter needs headroom to properly oversample. Lots of headroom, not a wimpy 3 dB, but at least 6 dB or one whole bit, and ideally 8-9 dB to cover these fringe cases too. This isn't rocket science, folks. It's 2020 and the math has been beaten to death. #MakeDACFiltersGreatAgain
As workaround you could get the RME ADI-2 and run it with -8 dB volume (with fixed reference level) to get rid of intersample peaks. Or use a playback volume of -9 dB in the software player.
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,204
Likes
16,986
Location
Riverview FL
The recording engineers (pick another title) are creating the intersample-over problem.
 
Last edited:

LTig

Master Contributor
Forum Donor
Joined
Feb 27, 2019
Messages
5,760
Likes
9,442
Location
Europe
I'm posting about this because I'm pretty disappointed with DAC digital filter design. [..] Most DACs on the market don't even attenuate properly at the nyquist frequency for 44.1 kHz audio, which comprises the entirety of CD digital audio and many, many more digital music files.
Full agreement. What happened to audio engineering?
 

q3cpma

Major Contributor
Joined
May 22, 2019
Messages
3,060
Likes
4,416
Location
France
Which is why using ITU-R BS.1770 replaygain with true peak should be compulsory for anyone with a little bit of knowledge. Thanks for highlighting this example, I wouldn't have thought that possible.
 
OP
ayane

ayane

Active Member
Forum Donor
Joined
Dec 15, 2018
Messages
182
Likes
684
Location
NorCal
As workaround you could get the RME ADI-2 and run it with -8 dB volume (with fixed reference level) to get rid of intersample peaks. Or use a playback volume of -9 dB in the software player.
I use ReplayGain within my player software, but only because I'm conscious of this problem and I also like having loudness normalized across albums.

The recording engineers (pick another title) are creating the intersample-over problem.
Absolutely. Recording "engineers" haven't a damn clue what they're doing these days. I wish people would just use 32-bit floating point for recording and mixing instead of whatever god-awful nonsense goes on in studios these days. Worst examples are the wonderful music that Rick Rubin ruined. Another vote for ITU-R BS.1770 awareness from me.
 
Last edited:

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
Here is a 30-second excerpt of "Get Away" from the Electronic Soundtrack album Code Lyoko by the Subdigitals. The audio is from track 9 on the CD I've ripped myself, and the excerpt is from the 3:10 mark in the track. This track contains a lot of clipping, as is common in the genre and modern music, unfortunately, but it is of note because unlike other clipped tracks, this track contains not one but multiple intersample peaks that exceed +6 dBFS.

I was curious how your sample would look in JRiver in contrast to folky music I like.

I manually set my Reference Volume at 100% (actual playback SPL at listening position ~90 dB) for these samples. The vertical divisions on the analyzer graphic are scaled at ~2.5 dB per division (maybe, well, I'm not totally sure).


So I'm Growing Old on Magic Mountain by Father John Misty at 100% Volume
1592285776090.jpeg


So I'm Growing Old on Magic Mountain by Father John Misty at 80% Volume
1592285804900.jpeg


Get Away by Subdigitals at 100% Volume
1592285829569.jpeg


Get Away by Subdigitals at 80% Volume
1592285851246.jpeg


JRiver adds quite a bit of headroom for your particular track sample even at 100% volume, making it a bit more tolerable to listen to... Granted, I often listen to music at much lower volumes: my own program startup volume being at 40% with Loudness compensation enabled.

---

Max SPL Meter Reading at Listening Position (with volume correction):

So I'm Growing Old on Magic Mountain by Father John Misty at 100% Volume
1592368280869.png


Get Away by Subdigitals at 100% Volume
1592368408611.png


---

In JRiver playing with simulated pseudo 5.1ch... -- or no actual center channel

LUFS EBU R128 analyzer plugin (with volume correction enabled)

So I'm Growing Old on Magic Mountain by Father John Misty at 100% Volume
1592368559473.jpeg


Get Away by Subdigitals at 100% Volume
1592368678815.jpeg


---

LUFS EBU R128 analyzer plugin (without volume correction)

So I'm Growing Old on Magic Mountain by Father John Misty at 100% Volume
1592368733429.jpeg


Get Away by Subdigitals at 100% Volume
1592368810035.jpeg



I would say both tracks at maximum volume are just about close to equally loud (at their loudest parts) to my ears -- or at least that was my perception. I only played (with all speakers ON) both tracks when volume correction was ON.

No way am I risking fatigue or damage playing anything with zero correction.
 
Last edited:

majingotan

Major Contributor
Forum Donor
Joined
Feb 13, 2018
Messages
1,511
Likes
1,781
Location
Laguna, Philippines
I use UPnP to cast this track from my PC to my DAP, and while I didn't hear digital clipping at -0.0 dB o_O (probably my hearing is that bad but maybe the AK4499 DAC filters has more headroom), I was shocked on how squashed the dynamics are in this sample which is probably the most squashed track that ever reached my ears lol. Still baffled how I can't hear clipping or maybe I'm hearing it already but unaware of it other than no dynamic range whatsoever

Untitled.png

IMG_2012.JPG
 

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692
I've mentioned it before here on this forum, but it's entirely possible to have music that has intersample peaks beyond +6 dBFS, or one whole bit. It was mentioned again later on this thread by @LTig.
About your first link, John Siau is not the first one who talked about intersample overs:
https://www.audiosciencereview.com/...in-intersample-overs-please.11651/post-338342

About your second link, if you read the thread until the end you can see that I actually posted a link from John Siau and he already knew intersample overs above 6dB is possible.
https://www.audiosciencereview.com/...-best-audio-dac-in-the-world.8729/post-401550
Absolutely. Recording "engineers" don't have a fucking clue what they're doing these days. I wish people would just use 32-bit floating point for recording and mixing instead of whatever god-awful nonsense goes on in studios these days. Worst examples are the wonderful music that Rick Rubin ruined. Another vote for ITU-R BS.1770 awareness from me.
On end user's side, if ReplayGain/digital volume control/DACs with digital headroom eliminates audible distortion then it is a solved problem already, but the fact is doing so won't make Rick Rubin's stuff sounds better because most of the audible distortion is baked-in the audio data rather than clipped in the the playback chain, so it has very little to do with intersample overs.

On the production side, when the distortion is intentional, floating point, BS.1770 and whatever technology/standard you can name are not going to "solve" any problem.
https://www.audiosciencereview.com/...in-intersample-overs-please.11651/post-336772

There was a time when unintended digital clipping is caused by technical limitation and carelessness, but obviously unrelated to the typical loudness war people are talking about:
https://www.audiosciencereview.com/...in-intersample-overs-please.11651/post-336609
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
Been doing some further comparisons and noticed that changing the mode from 5.1ch to 2.0ch stereo on my plugin meter changes the results a bit:

Get Away track sample to JRiver (DSP correction disabled) to Xonar U7 mII (1.13Vrms analog out) to miniDSP 2x4HD (2Vrms analog in & No DSP correction)

1593751063397.jpeg

Even though it's peaking and clipping when sent to the U7 mII, the voltage difference gives the miniDSP still a bit of headroom if I wanted to add some EQ filters in that later stage.

I was surprised to see REW's full-range random pink noise fairing better...
1593751449192.jpeg

... even though REW indicated peaks of up to +8.99 dBFS for this 0 dBFS generated noise, the loudness meter plugin only registers a +1 true peak... But the noisy signal is also so compressed that the dB RMS reading on miniDSP looks a little bit worse than the actual previous 'musical track' -- still not enough to reach 0dBFS due to voltage differences.


Now, with DSP correction enabled -- plus made a few a adjustments -- (in stereo mode) we aren't ever going to see clipping:
1593752014191.jpeg


*If you're curious why I even bother with having/including a miniDSP in the audio chain (adding additional noise), well, it's mainly because I need it's optical input for another external player -- besides the main PC.
 
Last edited:

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692
In fact the Get Away clip from @ayane has a lower dBTP value than my chiptune example posted on another thread:
https://www.audiosciencereview.com/...ents/40-space-traveler-stage-3-bgm-zip.45777/

As you can see both clips don't exceed 6dBTP if I use foobar SoX plugin's default settings (95%) with 4x upsampling.
sox.png


But wait... what if I set the passband to 99%?
sox 99.png


What if I use SSRC?
ssrc.png


That's why such a metric is unreliable and confusing. In case you wonder here are the white noise response of SoX's 90% vs 99%.
index.php


Now, if you read the datasheets of some DAC chips or AP measurements, which one is more similar to the actual response of these DACs?

AKM:
index.php


Cirrus Logic:
index.php


ESS:
index.php


BB:
index.php


All of them don't have a filter similar to 99%, more like 90%. The bottom line is whatever your true peak meter shows does not necessarily reflect what you DAC and anything before reaching the DAC (e.g. ASRC) do, and different true peak meters show different values, the errors can be higher than 2dB as shown above.

I've seen complains about using non-bandlimited test signals so I also have a post with a bandlimited test signal. Filters with typical steepness (as shown above) should not affect dBTP readings as the highest tone in the file is not higher than 18kHz, but different phase response will change the dBTP readings, feel free to play with it:
https://www.audiosciencereview.com/...ation-audible-intersample-clipping-test.2231/
 
Last edited:

Theriverlethe

Senior Member
Joined
Oct 20, 2019
Messages
493
Likes
344
Been doing some further comparisons and noticed that changing the mode from 5.1ch to 2.0ch stereo on my plugin meter changes the results a bit:

Get Away track sample to JRiver (DSP correction disabled) to Xonar U7 mII (1.13Vrms analog out) to miniDSP 2x4HD (2Vrms analog in & No DSP correction)

View attachment 71701
Even though it's peaking and clipping when sent to the U7 mII, the voltage difference gives the miniDSP still a bit of headroom if I wanted to add some EQ filters in that later stage.

I was surprised to see REW's full-range random pink noise fairing better...
View attachment 71702
... even though REW indicated peaks of up to +8.99 dBFS for this 0 dBFS generated noise, the loudness meter plugin only registers a +1 true peak... But the noisy signal is also so compressed that the dB RMS reading on miniDSP looks a little bit worse than the actual previous 'musical track' -- still not enough to reach 0dBFS due to voltage differences.


Now, with DSP correction enabled -- plus made a few a adjustments -- (in stereo mode) we aren't ever going to see clipping:
View attachment 71703

*If you're curious why I even bother with having/including a miniDSP in the audio chain (adding additional noise), well, it's mainly because I need it's optical input for another external player -- besides the main PC.

It looks like you have Jriver’s clip protection enabled in every example. Check the bottom left.
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
It looks like you have Jriver’s clip protection enabled in every example. Check the bottom left.

Ah, yes... True. The only other alternative which is 'flat line overlows' is still a form of 'hard' clip protection. I haven't looked into whether there is an 'overide' setting for the built-in protection, but from what I understand, the above 0dBFS peaks should still register in the meter.
 

Theriverlethe

Senior Member
Joined
Oct 20, 2019
Messages
493
Likes
344
Ah, yes... True. The only other alternative which is 'flat line overlows' is still a form of 'hard' clip protection. I haven't looked into whether there is an 'overide' setting for the built-in protection, but from what I understand, the above 0dBFS peaks should still register in the meter.

You're right, I think the clip protection is applied last in the chain before dithering.
 

ernestcarl

Major Contributor
Joined
Sep 4, 2019
Messages
3,106
Likes
2,313
Location
Canada
In fact the Get Away clip from @ayane has a lower dBTP value than my chiptune example posted on another thread:
https://www.audiosciencereview.com/...ents/40-space-traveler-stage-3-bgm-zip.45777/

As you can see both clips don't exceed 6dBTP if I use foobar SoX plugin's default settings (95%) with 4x upsampling.
View attachment 71854

But wait... what if I set the passband to 99%?
View attachment 71855

What if I use SSRC?
View attachment 71865

That's why such a metric is unreliable and confusing. In case you wonder here are the white noise response of SoX's 90% vs 99%.
index.php


Now, if you read the datasheets of some DAC chips or AP measurements, which one is more similar to the actual response of these DACs?

AKM:
index.php


Cirrus Logic:
index.php


ESS:
index.php


BB:
index.php


All of them don't have a filter similar to 99%, more like 90%. The bottom line is whatever your true peak meter shows does not necessarily reflect what you DAC and anything before reaching the DAC (e.g. ASRC) do, and different true peak meters show different values, the errors can be higher than 2dB as shown above.

I've seen complains about using non-bandlimited test signals so I also have a post with a bandlimited test signal. Filters with typical steepness (as shown above) should not affect dBTP readings as the highest tone in the file is not higher than 18kHz, but different phase response will change the dBTP readings, feel free to play with it:
https://www.audiosciencereview.com/...ation-audible-intersample-clipping-test.2231/

Thanks for checking it. But yes, I have not disabled my 'room correction' which is just xo set for upmixing to MCH.

Here is it with everything set to stereo:
1593968361797.png


The total 'loudness integration is at -6 so that's what the OP was actually pointing at and not the true peak? I'm just guessing...

Duh... of course, you're probably right with what you showed in your tests.
 
Last edited:

Pluto

Addicted to Fun and Learning
Forum Donor
Joined
Sep 2, 2018
Messages
990
Likes
1,631
Location
Harrow, UK
Recording "engineers" don't have a fucking clue what they're doing these days
I think you might be wrong there. This is such an extreme example that, I suspect, has been deliberately (and quite cleverly) engineered to cause this extreme inter-sample excess (which, incidentally, I measured at about +4½dBTP). There is a totally unnatural extreme of energy between 15 and 20kHz and reducing this by quite a modest (totally inaudible) amount reduced the inter-sample peak to about +1dBTP.

I wouldn't be at all surprised if this “music” messed up more than a few tweeters through overheating, not to say completely ****ing up some people’s hearing for the rest of their lives.
 

KSTR

Major Contributor
Joined
Sep 6, 2018
Messages
2,690
Likes
6,013
Location
Berlin, Germany
With 20x (filterless) upsampling I can see a +6.6dBFS IS-Over at 0:10.488 but as @Pluto said this such a messed up music sample that the clipping (or worse, wrap-around like the YAMAHA CDX-595 CD player does) is irrelevant. The IS-overs that I tend to find in my collection of music seldom go beyond +1...+2dBFS which is handled by most DACs pretty well (either faithfully or with benign soft-clipping).
 
Top Bottom