Help explain intersample overs, please?

[bennetng]

One thing I don't understand is why everyone mentions Benchmark's illustrations when talking about ISP. I mentioned it in 2006 here for example:

https://forums.dearhoney.idv.tw/viewtopic.php?t=52943&start=18

Yes, I mentioned the "0dBFS+ Levels in Digital Mastering" paper (at 2000) as well

analog volume controls often don't support presets, so it means you need to adjust the volume every time you listen to different music. On the other hand ReplayGain allows these things, plus you can edit the stored gain values if the default loudness algorithm don't fit your taste. Do it once and you can listen at your desired volume level without further intervention.

In hindsight, if people are accustomed to use digital volume management, loudness war would not be able to take root. If it still happens, blame those DAP/smartphones with weak output and require heavy compression and limiting to achieve the desired volume levels without clipping, intersample or not.

I don't know if any of you really Googled the paper or not so here is a screenshot:

So the Sony D50, a Discman, obviously has some headroom at least up to +1.25dBFS many, many years ago.

In fact John Siau himself also recommended others to use the digital volume control:
https://hydrogenaud.io/index.php?topic=98753.msg826323#msg826323

 

[MRC01]

If you're using an Infinite Impulse Response filter, maybe. If you're using a Finite Impulse Response filter, you can find the gain of the filter (taking the abs values of the FIR coefficients), and scale the coefficients to ensure the filter gain does not exceed unity. Then, regardless of the input stream, the output will never exceed FSD. The output is only dependent on the set of n samples currently passing through the n stages of the filter.

I don't doubt that. There are different ways to reconstruct the analog wave from the sampling points. I referred to the Whittaker-Shannon formula because it is the theoretically perfect/correct/ideal method, which makes it a good reference point for explaining intersample overs. That is, it clarifies that intersample overs don't arise as a side effect of a particular engineering approach, but are instead tied to DA reconstruction theory.

 

[j_j]

I don't doubt that. There are different ways to reconstruct the analog wave from the sampling points. I referred to the Whittaker-Shannon formula because it is the theoretically perfect/correct/ideal method, which makes it a good reference point for explaining intersample overs. That is, it clarifies that intersample overs don't arise as a side effect of a particular engineering approach, but are instead tied to DA reconstruction theory.

There are many "solutions" but all of them in essence are filters applied to the 'impulse' or 'staircase' coming out of a DAC.

As such, with a system with a defined bandwidth (say 20khz for a 44.1 sampling rate), a filter can be designed for most any reasonable interpolation level (number of times sample rate is improved, best done, but not limited to integer ratios) one chooses, and then the peak measured. If the filter is competent, there had better not be very much differences when you get past 8x oversampling. Having a big variation between two samples at 8x requires frequencies above fs/2, and if you see that, you screwed it up somehow!

So, different solutions may have slightly different answers, but they'd better be real (*&(&(ing close.

 

[RichB]

I don't know if any of you really Googled the paper or not so here is a screenshot:
View attachment 51941

So the Sony D50, a Discman, obviously has some headroom at least up to +1.25dBFS many, many years ago.

In fact John Siau himself also recommended others to use the digital volume control:
https://hydrogenaud.io/index.php?topic=98753.msg826323#msg826323

A blast from the past. I had this player

 

[blueone]

I just got the latest Norah Jones album released (Begin Again) on CD, and directly after I also listened to her first album (also on CD), Come Away With Me. Other than the surprise that Begin Again had less than 29 minutes of music on it (Excuse me?), I was also surprised by how much I had to raise the volume level when I listened to Come Away With Me. There wasn't any audible digital clipping on Begin Again, of course, though I have a Benchmark DAC3, but what in the name of good mastering practice makes these mastering engineers do this? I can't imagine why it would be good for vinyl production (which might explain the 30min run time) or MP3. A Norah Jones album is not a car dealer commercial on a TV channel.

 

[usersky]

I read all this with genuine interest. Consindering most if not all in the industry (except for one "illuminated" company) exhibit this issue, one expects that sound engineers monitor their productions under the effect of the very same issue that we as ordinary listeners do. If they, in their "creative" endeavour decide that this final effect is what they want to sell to us sheeps, that's how it should sound. Just sit and enjoy the art we pay for. If they would use magic Benchmarks and produce without hearing intersampling overs then I'd say this struggle would matter but as it seems, this is what they mix&master for and intended. I'd say on the contrary, if we try to fix intersampling overs, the sound would be different than what artists intended and approved. Then lets our dacs distort how can we argue with bleeding hearts and artists?? it's like polishing the paint on the face of Mona Lisa because it is so irregular.

Note: these days I find myself being unable to discern my irony from true talk anymore.

 

[mansr]

I just got the latest Norah Jones album released (Begin Again) on CD, and directly after I also listened to her first album (also on CD), Come Away With Me. Other than the surprise that Begin Again had less than 29 minutes of music on it (Excuse me?), I was also surprised by how much I had to raise the volume level when I listened to Come Away With Me. There wasn't any audible digital clipping on Begin Again, of course, though I have a Benchmark DAC3, but what in the name of good mastering practice makes these mastering engineers do this? I can't imagine why it would be good for vinyl production (which might explain the 30min run time) or MP3. A Norah Jones album is not a car dealer commercial on a TV channel.

No, it's an audiophile showroom demo. Has to be heard over the crowd.

 

[bennetng]

A blast from the pas. I had this player

Archimago also measured an old Sony SACD player and it does not clip the 0dBFS white noise as well.
http://archimago.blogspot.com/2018/04/retro-measure-2001-sony-scd-ce775-5.html

He tested it with a CD-R, therefore 16-bit limited.

Here is Lynx L22, the yellow plot is obviously clipped, but of course, it has a much cleaner spectrum when there is no clipping.
http://archimago.blogspot.com/2017/04/retro-measure-2002-lynx-l22-pci-audio.html

 

[blueone]

No, it's an audiophile showroom demo. Has to be heard over the crowd.

I haven't seen a crowded audiophile store outside of New York City since the 1990s.

 

[MRC01]

... I'd say on the contrary, if we try to fix intersampling overs, the sound would be different than what artists intended and approved. Then lets our dacs distort how can we argue with bleeding hearts and artists?? it's like polishing the paint on the face of Mona Lisa because it is so irregular. ...

To follow your Mona Lisa example: with audio recording we're not creating the Mona Lisa, we're making a copy of it so people can enjoy it in their homes without traveling to the Louvre. If the purpose of the copy is to preserve it as closely as possible to the original, so people can see what it actually looks like and appreciate its artistry in full, then the "copy engineers" should use the highest resolution, with no exposure compression, ensure white balance is perfectly neutral, etc. The loudness wars suggest intentionally distorting the copy, oversaturating it so it visually "jumps out at you", at the cost of masking subtle shades of light and color so doesn't resemble the original. This copy makes it impossible to fully appreciate the art because it has squashed detail, subtlety and life out of it.

 

[RichB]

My take away from this, is that there are music files that generate inter-sample overage.
I'd prefer an implementation that can handle most, if not all, overages without requiring external attenuation.

- Rich

 

[L5730]

Not perfect, but just ram all audio through George Yohng's W1 limiter free VST plugin. Set the output a little lower and set the ceiling to -3dBFS or something.
We've pushed this plugin hard and it still sounds like audio rather than crapping out. Compare a null test on the processed audio to the original, and you'll see how the waveform is unchanged, except for the peaks. Does exactly what it should, rounds the peaks to limit their amplitude.

If one is prepared to do a 2-pass (and maybe script/automate it) the ISP could be calculated on a 1st pass, and then the exact numbers for the plugin could be worked out with simple math.

 

[j_j]

I'd say on the contrary, if we try to fix intersampling overs, the sound would be different than what artists intended and approved.

The problem is that different DAC's respond in VERY different ways. But yes, I understood the irony.

 

[RichB]

I read all this with genuine interest. Consindering most if not all in the industry (except for one "illuminated" company) exhibit this issue, one expects that sound engineers monitor their productions under the effect of the very same issue that we as ordinary listeners do. If they, in their "creative" endeavour decide that this final effect is what they want to sell to us sheeps, that's how it should sound. Just sit and enjoy the art we pay for. If they would use magic Benchmarks and produce without hearing intersampling overs then I'd say this struggle would matter but as it seems, this is what they mix&master for and intended. I'd say on the contrary, if we try to fix intersampling overs, the sound would be different than what artists intended and approved. Then lets our dacs distort how can we argue with bleeding hearts and artists?? it's like polishing the paint on the face of Mona Lisa because it is so irregular.

Note: these days I find myself being unable to discern my irony from true talk anymore.

They might use Benchmark DACs
Slamming and clipping the recording may not be the artist's decision but made by the "expert" mastering it.

When someone waxes on about the "art", I like to substitute the word "food" for "art" to gain the proper perspective.

- Rich

 

[j_j]

I think this screenshot will illustrate what is going on.
View attachment 51495


Top waveform hits max digital level one bit per cycle. It is a max level 11,024 hz tone at 44.1 khz. Over time the phase of when samples occur vs the waveform slowly shifts. It is the same exact wave at the same exact level shifted in time in the bottom track. None of the bits are at max level. Now in the light blue area I've applied 2.44 db of gain which is allowed by this and most sound editors. It can cause two problems.

The middle track is upsampled to 352,800 hz sampling rate. It shows more closely the actual waaveform. As you can see the actual peak of the reconstructed waveform will be between samples and above the sample values. In the left side that would be fine, but in the right side it means the waveform is higher than what a zero level wave would be. If the analog section has this much headroom then fine no problem for the left half. For the right half already at near max levels upsampling will cause peaks to be above max digital level and the reconstructed upsampled waveform will be squared off. Like this one picture below. This was the bottom track above with 2.44 db gain upsampled to 352,800 hz sample rate.

View attachment 51497

So this waveform was a tone sampled in a place that left room for digital gain, but if upsampled it both would require headroom in the analog output, plus it would have pushed the digital values into clipping and you'd still get a clipped output waveform. Even with analog headroom.

Some noise like waveforms will require 10 db headroom to prevent this occurring. However they happen so rarely you likely wouldn't even hear the momentary clip even if such an uncommon noise-like signal occurred. So 3 or 4 db headroom can prevent this effectively from ever being audible.

Even then, this hardly ever matters. Modern retarded mastering means the waveforms are pushed way up and squared off even if not high enough to cause clipping. So they sound like clipping even if just hard limiting and high level compression.

And if it's properly reconstructed, the reconstruction filter will also go above +-1 which is wrong. Some DAC's handle it, some saturate the digital filters inside the delta-sigma, and one or two (ack!) actually do an integer wrap. The effects can be, um, spectacular.

 

[MC_RME]

And if it's properly reconstructed, the reconstruction filter will also go above +-1 which is wrong. Some DAC's handle it, some saturate the digital filters inside the delta-sigma, and one or two (ack!) actually do an integer wrap. The effects can be, um, spectacular.

Here is an example of a fully failing DAC with those 'spectacular' integer wraps, around the middle of the measurements. Didn't expect that in 2023...

Denafrips Pontus 2 (12th Anniversary Edition) Measurements - GoldenSound

Measurements of the Denafrips Pontus 2 12th Anniversary R2R DAC, and testing to see whether claims of NOS capability are true.

goldensound.audio

 

[restorer-john]

Here is an example of a fully failing DAC with those 'spectacular' integer wraps, around the middle of the measurements. Didn't expect that in 2023...

Denafrips Pontus 2 (12th Anniversary Edition) Measurements - GoldenSound

Measurements of the Denafrips Pontus 2 12th Anniversary R2R DAC, and testing to see whether claims of NOS capability are true.

goldensound.audio

Why is Goldensound testing for intersample overs at ~22kHz?

"However the Pontus 2 does not clip, but instead when a sample value reaches above the maximum, it ‘wraps around’ to the minimum negative value, causing a huge sudden transient which will be very audible and may appear as crackling/popping."

Who a going to hear it? Bats maybe?

 

[j_j]

Why is Goldensound testing for intersample overs at ~22kHz?

"However the Pontus 2 does not clip, but instead when a sample value reaches above the maximum, it ‘wraps around’ to the minimum negative value, causing a huge sudden transient which will be very audible and may appear as crackling/popping."

Who a going to hear it? Bats maybe?

1) if it's 4 samples per cycle, it's 11+ kHz.
2) IM or other things caused can splatter all over the output spectra, including at very much lower frequencies.

 

[restorer-john]

1) if it's 4 samples per cycle, it's 11+ kHz.
2) IM or other things caused can splatter all over the output spectra, including at very much lower frequencies.

You're right- about 92uS on the AP's scope display. I read it as 10uS per division.

 

[earlevel]

Probably obvious to a lot of people, but perhaps not everyone. More and more, modern convenience has made inter-sample overs simply a moot point.

What I'm getting at is that I'm sitting at my computer, where I listen via my Topping DX7 Pro, using its digital gain knob. It's never, ever close to the 0 dB setting, which would be painfully loud even it it didn't overdrive my iLoud MTM amps into saturation. And of course it's a similar situation with people using a laptop or other computer, with headphone or outboard speakers—you're probably using digital control. When I'm listening on my iPhone, the digital gain is rarely full blast except when I'm relying on hearing a conversation on the tiny built-in speakers—for audio listening through headphone and IEMS, it's backed off a few dB.

When working on music, I have a choice of always being at 0 dBFS and relying on analog pot somewhere, but it's more convenient to allow a bit of digital headroom and control it from sitting at the computer anyway. My 20 year-old Acura, which I fitted with a USB interface, is the only place where the phone feeds the USB with no ability to adjust the gain, and the only volume control is the analog one on the built-in stereo (yet it still manages to sound great—my music friends love to check out their mixes in my car, LOL).

But like I said, "more and more", it's unlikely for listeners to even generate such an over in normal listening.

NB: This is just an observation, I'm not saying that music production shouldn't strive to eliminate them, or that DAC manufacturers shouldn't pay attention. And of course I understand giving up a little dynamic range, but in general turning down is just dropping more into the noise floor anyway, and since virtually everything I listen to is 24-bit, there is no practical penalty for "throwing away bits" with digital gain control.