Are inter-sample overs the new audiophile bogeyman?

[antcollinet]

I know what they are - I know how they happen.

I've never knowingly heard distortion coming from intersample overs. When I first got my MiniDSP flex, I set the input gain to -3dB to avoid any risk. I recently set it back to 0dB and:

1 - Have not noticed any difference
2 - Have not ever heard anything I could attribute to intersample overs.

Is it not the case that only badly mastered music (basically clipping in any case) will cause them. And even then - if it only happens for an extremely brief period in the music it is going to be almost impossible to detect audibly in any case.

Yet there seems to be all sorts of FUD talked about it in conjunction with DACs. We even have a whole thread discussing how to test DACs for it. It feels to me like it is the "new jitter" Something that people can hang their hat of audble differences on - without it being an actual problem in reality.

What am I missing?

Has anyone else heard the effect of inter-sample overs - if so, which track, at what time? And what does it sound like?

 

[tw99]

I've also never heard the effect, but unlike you I've set my MiniDSP input to minus a couple of dB and kept it that way. It's a bit like Spotify lossless for me, I don't hear any difference from 320k Vorbis, but use lossless anyway. Might as well use the best technical performance that I have available.

 

[DVDdoug]

I've never heard them.

And what does it sound like?

it should be slight, occasional, short-duration Clipping.

Theoretically, there's no reason the reconstructed analog waveform can't go over 0dB. The 0dB digital limit is the highest you can "count to" with a given number of bits so you can't go over, and of course there is no inter-sample data. The analog side is only limited by voltage.

Something that people can hang their hat of audble differences on

Fine... But where are the controlled-blind listening tests?

Usually, the issue with DSP is boosting into clipping with EQ (or other boosting). Or resampling can sometimes cause clipping.



BTW - I have LOTS of MP3s that go over 0dB and I don't hear distortion, although I may not be clipping since I'm usually not playing at "full digital volume" into the DAC. (One of characteristics of MP3 is that it makes some peaks higher and some lower so 0dB normalized files (like a lot of CDs) go over 0dB. And, MP3 can go over 0dB, but of course DACs can't. Some people lower the level before converting to MP3, but MP3 is lossy anyway and I've never heard the clipping.

 

[Keith_W]

Read what @John_Siau said about intersample overs here and here. And @tmtomh has an example of an audio inversion at an intersample peak here. There is a good back and forth between John and Amir that mostly went over my head which you guys might find interesting.

 

[KSTR]

Years ago I had a YAMAHA CD-player which had severe and special IS-over-like errors in form of wraparound, coming from a re-dithering volume control chip ahead of the DAC chip proper. It took me years to finally pin down the occasional glitch I thought I heard with some recordings at full volume to that root cause.

I feel regular IS-Over effects of any well-behaved DAC (say, AKM) like simple clipping are almost impossible to hear if you don't have a reference.

 

[VientoB]

I set my DAC to -2db but recently set it back to 0 as my amp is low gain so all I was doing was reducing the maximum loudness for no real gain.

 

[kemmler3D]

I've had the same thought as @antcollinet ... If intersample overs were a big problem, we would have a track that would reliably produce audible distortion through at least a few different DACs.. Does such a song exist? Do the test tracks designed to produce ISOs actually even sound wrong on most DACs?

 

[antcollinet]

I guess wraparound errors (which I've not ever heard of before) might be an issue - but I'd call that a design fault.

 

[DVDdoug]

might be an issue - but I'd call that a design fault.

I agree. Any competent DSP programmer knows to avoid wrap-around/roll-over. (I'm not a DSP programmer and I know it. ) I've seen it showing-up in Audacity waveforms (not on my computer) but I don't remember what caused it.

 

[Sokel]

The funny thing about it is that we have wars going on -110dB THD+N vs -115dB THD+N about audibilitty and at the same time we're talking about ignoring stuff that can result at -20dB THD+N??? Even momentarily?
The story of the peaks, in general, re-branded?

My theory about lots of stuff in audio is that probably people have get used to clipping, all shorts of it.
Be it weak amps, be it ISOs, etc, it's there.

But is that hi-fi? Nope.
We either stick with the fidelity or we don't, we can't have it both ways.

 

[Nutul]

How long does an ISO last at 44.1 kHz...? Is that enough to perceive it as a disturbing element in the music flow? Is that even considerable as some unpleasant, out of place "noise" or "glitch"? Does it last long enough to cause an audible effect on the speaker movement? Like, I mean... it is not even like a flute player inadvertently hit a "key" (don't know how they're called, those things that flute player presses to produce notes...) and his nail hits the metal before the fingertip...

Let all this alone when speaking of 48 / 192 or even higher SRs.

 

[AnalogSteph]

I guess wraparound errors (which I've not ever heard of before) might be an issue - but I'd call that a design fault.

I do vaguely remember seeing indications of a cheap dongle DAC of sorts (Samsung or something?) suffering from this issue, so it's not entirely unheard of. But yeah, it's a pretty gnarly bug.

Having an ASRC in the signal path tends to generate some pretty nasty distortion on overs as well.

With that being said:
Any concerns about intersample-overs can be easily mitigated in any playback chain sporting some digital attenuation. You are most likely to run into trouble if you insist on partying doing digital audio like it's 1999 - CD player or similar into DAC at full volume. The current generation ESS DACs have finally moved their digital volume ahead of the built-in ASRC, and turning things down a bit will similarly help out the Cirrus CS43198 and the like (which if memory serves are not exactly drowning in analog headroom). Some DAC units will also leave a bit of headroom in general (e.g. Benchmark, RME, Topping E30 II (Lite)).

 

[danadam]

My theory about lots of stuff in audio is that probably people have get used to clipping, all shorts of it.
Be it weak amps, be it ISOs, etc, it's there.

Maybe. But if ISO problems are inherently audible then people should still be able to hear it in direct comparison, clipped vs non-clipped, right? The OP is asking for an example of a track where such thing is possible.

Or are you saying that people got used to clipping so much that somehow they lost the ability to hear the difference? But if so then the problem solved itself, right?

 

[jkim]

The current generation ESS DACs have finally moved their digital volume ahead of the built-in ASRC

Where did you get this information? From which generation?

EDIT. Found it.

ES9020:

Volume control is placed BEFORE the reconstruction filter.

ES9039Q2M / ES9069:

Volume Control is BEFORE the reconstruction filter.

ES9038Q2M / ES9028Q2M / ES9018K2M (and all DAC chips with Hyperstream II or previous gen):

It is unclear which comes first, but it is likely that volume control is AFTER reconstruction filtering.

ES9219 (w/ Hyperstream III):

Volume Control is AFTER the reconstruction filter. I also confirmed this with the volume control of Neutron DAC V1. EDIT. It seems that the information on the ES9219 datasheet is not accurate. Its digital volume control prevents ISOs.

So, it seems that only the current generation (w/ Hyperstream IV) ESS chips have that desirable structure. Of course, even with old gen chips, RME and Qudelix wisely chose to implement software volume control, rather than feed the purists crowd with the silly "bit-perfect playback" misconception...

 

[Sokel]

Maybe. But if ISO problems are inherently audible then people should still be able to hear it in direct comparison, clipped vs non-clipped, right? The OP is asking for an example of a track where such thing is possible.

Or are you saying that people got used to clipping so much that somehow they lost the ability to hear the difference? But if so then the problem solved itself, right?

What I'm saying is that is clearly measurable, and causes mayhem on the chart.
Does it matter if it's audible?

Are any of the stuff we're talking about usually audible, at the abyss of -100dB THD+N?
But combine them all, think of all the stuff people do with their gear (where everything that can go wrong usually does) and you end up with people endlessly searching for the next "upgrade" and ways to get rid of the headache their gear is causing them.

Peace of mind is better, one thing at a time.

 

[Donuts123]

Inter-sample overs are not inherently audible, it is only if your signal chain doesn't handle them that they may be audible.

You could obtain some lossless files that have large ISOs. Play using foobar2000 with bit-perfect ~6dB pad, compare with the raw output (matching volumes). The fb2k True Peak Scanner component can tell you how many ISOs there are in a given track.

It would be nice if there were some synthetically generated test sequences to compare. Audible differences could be much easier to notice there than if your sample is (say) a modern dance track which was mastered with a lot of compression and clipping anyway (where a bit more distortion may not be too obvious).

 

[DVDdoug]

Does it matter if it's audible?

Yes! Audibility is EVERYTHING!

Are any of the stuff we're talking about usually audible, at the abyss of -100dB THD+N?

-100dB "THD" usually isn't audible. But the "N" exists when there's no signal to mask it and no signal to measure or calculate a dB difference. "-100dB" is relative to the signal so at low signal levels it's no longer -100dB. Generally, low-level noise is more of a concern than low-level distortion.

But I'm not worried about -100dB of noise either.

There is no "one number" where distortion and noise become audible because there are different kinds of distortion and noise, and some people have better hearing than others.

 

[jkim]

Just confirmed that ES9039Q2M's volume control is placed before reconstruction filtering and hence can create headroom for ISOs.

Tested E1DA 9039S which has no DSP chip. And the Windows volume control adjusts the DAC chip's volume control register.

Fed the 11025 Hz ISO test signal (saved from Multitone Loopback Analyzer) at full scale and with the device volume set to 0 dB:

The same full-scale signal with the device volume set to -3 dB:

PCM volume control on other DAC chips (e.g., ES9219 or CS43198/CS43131) does not work this way. Heard that some AKM chips also place volume control before OS filter but I don't have any DAC with an AKM chip on my test bench...

 

[JSmith]

A long thread on this topic;

Let's develop an ASR inter-sample test procedure for DACs!

Hi there, It's been a while since I thought about this, but why aren't DACs tested for their ability to handle inter-sample overs (or ISP, inter-sample peaks) in ASR reviews? Here are some established facts: ISPs are very real, and almost any material that's been released after the 90's can...

www.audiosciencereview.com

JSmith

 

[jkim]

A long thread on this topic;

Let's develop an ASR inter-sample test procedure for DACs!

Hi there, It's been a while since I thought about this, but why aren't DACs tested for their ability to handle inter-sample overs (or ISP, inter-sample peaks) in ASR reviews? Here are some established facts: ISPs are very real, and almost any material that's been released after the 90's can...

www.audiosciencereview.com

JSmith

A shorter but older thread on the topic as well :

Help explain intersample overs, please?

The Benchmark DAC3 webpages explain the concept of intersample overs and how they can lead to distortion when upsampling from 44.1 or 48 kHz to higher rates. I get that, and I think I understand the math behind it. What I’m less clear about is the solution. Benchmark talks of designing in...

www.audiosciencereview.com

But it seems that the OP was aware of these...