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

[AnalogSteph]

Ah, thanks, I completely forgot ReplayGain was still a thing! I actually used it years ago, but I'm glad to see it's still being actively developed.
As you can see from the discussion, I'm not in favour of just pushing users to solve the problem themselves, but it's nice to have another solution at hand. And it shows that it's not entirely unreasonable to think about it.

2 birds, 1 stone, basically. It's definitely one of my must-have features for any playback environment that wants to be taken seriously.

It would still be nice to know which DACs to recommend to the folks stuck with a traditional disc player with no digital volume anywhere in sight.

 

[KSTR]

Probably almost all current DACs do internal up-sampling. I think up-sampling ISP can lead to 3 results: true peak within headroom, hard clipping to 0dBFS or computing error like overflow. Do exist DACs, that belong to the last category?

See the links in post #86, especially the first one.

 

[edechamps]

Probably almost all current DACs do internal up-sampling. I think up-sampling ISP can lead to 3 results: true peak within headroom, hard clipping to 0dBFS or computing error like overflow. Do exist DACs, that belong to the last category?

Yes. Does this look like clipping to you?

Or this (yellow trace)?

Bottom line is, at least some DACs fail spectacularly when faced with intersample overs - they don't "just clip" as has been implied several times in this thread.

I think it would be insightful to play some real-world content with known ISO issues through a DAC that is known to be particularly bad at handling ISOs, and then ABX that against the same content played with enough headroom to avoid ISOs. This would definitely bring some real data to bear on the "but it is audible though" question.

 

[KSTR]

It would still be nice to know which DACs to recommend to the folks stuck with a traditional disc player with no digital volume anywhere in sight.

I'd rather start with a study what kind of misbehaved intersample overs actually are an audible problem so we can weed out the completely benign types (stable hard- or preferably, soft-clipping).

 

[AnalogSteph]

Bottom line is, at least some DACs fail spectacularly when faced with intersample overs - they don't "just clip" as has been implied several times in this thread.

Yikes. I thought overflow limiters were pretty much standard in digital filters. They've pretty much been around since the first CD players.

 

[popej]

at least some DACs fail spectacularly when faced with intersample overs

It's a bug. I would like to know, which devices are affected and I would appreciate testing being included in the standard procedure.

 

[kemmler3D]

Bad analogy. A better one would be finding the most comfortable fabric to make body bags so that the corpses will feel more comfortable in them

Intersample overs means the analog signals (original or intended for synthesized signals) is over full scale. Here are some illustrations of the often used fs/4 sine wave to demonstrate ISO's. Here we have our samples taken at the perfect timing. The reconstruction matches the original. Woo-hoo!
View attachment 322599

Shift the sampling time a little bit, the reconstructed wave has a different (reduced) magnitude from the original. Note that this waveform is "ideally reconstructed" as it is done mathematically with no headroom limit of intersample overs.
View attachment 322600

With a sampling timing shift 1/2 of a sampling period (1/8 of the period of the sine wave) from the 1st example, the reconstructed wave is a full 3 dB lower than the original. So, exactly the same input waveform, exactly the same method of reconstruction, just a little shift in the start time of the A/D sampling process and you get different results.
View attachment 322601

With real life signals (e.g. frequency ≠ fs/4 or a "nice even faction" of fs), the differences will be more prominent and will also include harmonic distortions. So for ISO's, we are talking about choosing which of the different reconstructions of signals irreversibly damaged at the source, with none of them being anything like the original.

Tell me this is not a tempest in a teapot.

Haha, good analogy. I get what an intersample over is, and I understand the point of view that it's not very worth worrying about. But, I think it's interesting just for the sake of it. Like a lot of stuff discussed on this site, it doesn't actually matter but the details can be engrossing.

Definitely a tempest in a teapot, but I don't really get the controversy. I think it would be interesting to know how different DACs handle the reconstruction and output, I don't know that it has to be significantly audible to be of interest.

 

[splitting_ears]

Just exactly what is the probability of encountering that particular signal in real life? You've got a number? As an engineer, I am perfectly happy with solutions that works almost everywhere.

Ok, you're a fellow engineer so I'm sure we can reach an agreement

The "intersample overs demo" signal is fs/4, well within the Nyquist limit.

I fully agree.

Just exactly what is the probability of encountering that particular signal in real life?

Zero. We agree on that, too.
Like the J-test, it is a specially crafted signal designed to examine the behaviour of any given device in a worst-case scenario.

The clipping occur during the digitization.

I think that is where our disagreement lies. Perhaps it's a poor choice of words on my part, sorry if that's the case.
If by digitisation you mean the interpolation that happens during conversion, then no wonder we can't understand each other!

For me, digitisation is the conversion from analogue to digital via an ADC. Taking an analogue 'real world' signal, as you say, and storing it in digital PCM values.
In this case, the anti-aliasing filter will prevent any weird behaviour like your graph shows - the system is band-limited. There's nothing to worry about in this scenario, and I think you'll agree?

The whole concept of inter-sample overs only makes sense when converting an encoded signal from digital to analogue, if and only if that signal has been:
1) digitally processed after capture or digitisation (think EQ, compressors, limiters, ...)
2) created digitally in the first place (think of digital synthesizers in a DAW, for example).

Does this make more sense to you now?

[EDIT : since @edechamps made a really good point in post #130. Technically, you'll always have a difference between the sample values and the sampled signal itself when going from A to D and from D to A. This is normal, since the only case where sample peaks and interpolated peaks can be equal is when you digitally generate a sine wave which frequency is an integer divisor of the SR, and which phase is either exactly zero or 180°. In practice, however, most people keep enough headroom and then adjust the gain digitally so you're unlikely to experience clipping.]

And so, when you say:

There is even less excuse with synthesized signals, since the maximum level is known in advance

Well, that's partly true.
Firstly, this is something I've been trying to point out. As I said, the production world is slowly moving away from peak meters (which do *not* interpolate like a DAC) to true-peak meters. In other words, by default, most audio workstations, software and plug-ins only display the PCM peak values, not the true upsampled/interpolated value. The consequence is that even a well-meaning engineer in the 2000s could deliver digital masters full of ISPs and never know it, because all the lights were green in the digital domain!

 

[BeerBear]

What do you think?

I would like to see that, because the more data the better.

But, since time is a limited resource, such testing has to be justified. And so far I've seen no evidence of actual music causing audible ISP clipping with typical DACs. If you find something let me know, because I'm curious.
A while ago I tried to hear it with a few songs that have ~4dB overs, but I couldn't. Although I used pop music, which has a lot of clipping and distortion anyway. It should be easier with some cleaner (classical) music, but that tends to not have a lot of overs. And it should be easier with some of those non-typical DACs that were mentioned, but then we should weigh their relevance accordingly.
We also need to consider that ISP clipping, audible or not, is easy to avoid, in production and in reproduction. Media players can handle this semi-automatically and streaming providers could do the same. So it's a case where the people who might care about this problem, can easily deal with it, with just some basic education.

Meanwhile, I can think of 3 or 4 other issues in audio electronics that have some real-world audible consequences, are harder to deal with, but are not currently tested by @amirm and most reviewers. So, until more evidence is presented, I would prioritize those instead.

 

[edechamps]

The whole concept of inter-sample overs only makes sense when converting an encoded signal from digital to analogue, if and only if that signal has been:
1) digitally processed after capture or digitisation (think EQ, compressors, limiters, ...)
2) created digitally in the first place (think of digital synthesizers in a DAW, for example).

I don't think you need to go that far: isn't it the case that you can end up with ISOs simply by taking the signal from an ADC, and then "just" normalizing it to 0 dBFS before releasing it? Just that simple, seemingly benign operation can produce ISOs that weren't there in the original signal, I would expect. That makes it very easy to release content with ISOs by accident, even by well-meaning people, if they are not intimately familiar with PCM theory (and let's face it, most people aren't).

 

[MRC01]

... When you say that ISPs are only a problem for 0.001% of the content out there, allow me to strongly disagree. In fact, I think the opposite is much, much closer to the truth. At least for music released after the 90s and older, badly remastered music. I certainly don't want to turn the industry upside down, only help people to get a better sound. In fact, I think writing an ASR article on ISPs to help people understand the issue and solve it on their end would also be beneficial. I would happily contribute to it.

In the long term, perhaps we should do the opposite: whenever ISOs or clipping is detected, emit a blast of noise/static, or stop playing the track due to "invalid audio data". Studios would have to stop their irresponsible behavior of squashing the recordings to death.

Put differently, by accommodating these improperly encoded digital recordings, we would only enable studios to continue intentionally abusing recording technology in the name of the loudness wars, and to make it even more egregious. Or, as LTig said:

I bet that when finally all DACs can faithfully reproduce +6 dB ISOs music will be recorded 6 dB louder to get the desired distortion.

I follow restorer-john's approach. After "returning" terrible sounding heavily squashed downloads for refund too many times, I just stopped listening to that kind of music. If we all did this, Studios would have to do better or go out of business. Even when streaming, if I encounter a heavily squashed recording I make sure to send a note to support to tell them so they can pass the word to the studio.

... I don't care one iota for poor recordings. If they are crap, I don't buy them or if I do, I return them for a full refund.

 

[kemmler3D]

Does anyone know of any combinations of music clips and DACs where ISOs are clearly (or at least marginally) audible? I think we need to start with that, if there's any action to be taken here.

Also, ignorant question: For an intersample over to be audible, the DAC needs to render the peak at >0dB AND the amp needs to audibly clip with that input from the DAC - correct? If I have understood correctly, since we are talking analog clipping and we don't usually (hopefully) drive our amps to their limits, I really wonder whether this is a real-world problem or just a curiosity.

I'm personally interested in this for the technical aspect alone, but I'm also doubtful that it matters for playback quality.

 

[splitting_ears]

Or this (yellow trace)?

Bottom line is, at least some DACs fail spectacularly when faced with intersample overs - they don't "just clip" as has been implied several times in this thread.

Good grief! That's almost like turning a single sine wave into white noise. Even a guitar distortion pedal doesn't behave like that

Just for comparison purposes (although axes and scales do not match), here's what happen when you hard clip a 10kHz sine tone by 6dB with proper oversampling:

 

[MRC01]

Just for comparison purposes (although axes and scales do not match), here's what happen when you hard clip a 10kHz sine tone by 6dB with proper oversampling ...

Shouldn't we use a lower frequency for testing, since clipping introduces high frequency harmonics and at 10 kHz even the 2nd harmonic is inaudible? Also, in music lower frequencies tend to have higher amplitudes; music is not likely to have 10 kHz tones near full scale.

 

[splitting_ears]

Shouldn't we use a lower frequency for testing, since clipping introduces high frequency harmonics and at 10 kHz even the 2nd harmonic is inaudible? Also, in music lower frequencies tend to have higher amplitudes; music is not likely to have 10 kHz tones near full scale.

Here you go!

 

[MRC01]

Here you go! ...

Great, now we can see (and hear) those harmonics. You can even go 2 octaves lower. It's the low frequencies that are most likely to clip, and when they do, the harmonics are sprayed all over the mids and treble where our hearing is most sensitive.

 

[pkane]

Here you go!

View attachment 322780

This appears to be a nearly perfect square wave. Not unexpected with hard clipping.

 

[edechamps]

For an intersample over to be audible, the DAC needs to render the peak at >0dB AND the amp needs to audibly clip with that input from the DAC - correct?

Again, and because that bears repeating: most DACs will not "render the peak at >0dB", because most DACs are not designed with ISOs in mind. It's "undefined behavior", to borrow a term from software engineering. When presented with ISOs, some DACs will clip inside the DAC itself; other DACs will break horribly and output a waveform that may be corrupted beyond recognition. The amp doesn't even factor into this equation.

 

[Adis]

@splitting_ears
Just don't forget about artistic intention and that when *we push the shite so hard that it hits the fan, we do it on purpose - we're counting on it. Want to give us more headroom? Thanks, we'll use that too. Just because we can.

*we - the music producers.

 

[splitting_ears]

Put differently, by accommodating these improperly encoded digital recordings, we would only enable studios to continue intentionally abusing recording technology in the name of the loudness wars, and to make it even more egregious.

Please. This has nothing to do with production quality. The decoded peaks will always be higher than the encoded peaks, minus one *very* specific constructed case.
This is not up for debate or judgement, it is a logical consequence of the sampling theorem.