Help explain intersample overs, please?

[theREALdotnet]

By just doing it right. Provide proper arithmetic and analog circuitry so there's no problem.

Exactly, this is part of a robust design that handles all reasonable inputs. Yes, the digital source engineers could do it, but they don't, so any competent DAC should account for it. The worst case is 44.1KHz so just give it 6 dB headroom.

I don’t think you’re arguing for the same solution. To me it sounds like @j_j wants a DAC that is specified for 4Vrms (5.65Vp) output to be allowed to peak up to 11.3V (6dB overshoot), and @MusicNBeer want’s to reduce the volume of the digital input signal to be reduced by 6dB before conversion.

The former solution might surprise amplifier input stages, the latter may be wasting dynamic range in situations where headroom is already provided (many software players, streamers, etc.).

 

[theREALdotnet]

BTW, what do CD players with built-in DACs do? CD playback is probably the worst-case scenario, do these players apply a digital volume reduction before conversion?

 

[j_j]

I don’t think you’re arguing for the same solution. To me it sounds like @j_j wants a DAC that is specified for 4Vrms (5.65Vp) output to be allowed to peak up to 11.3V (6dB overshoot), and @MusicNBeer want’s to reduce the volume of the digital input signal to be reduced by 6dB before conversion.

The former solution might surprise amplifier input stages, the latter may be wasting dynamic range in situations where headroom is already provided (many software players, streamers, etc.).

4dB is likely enough, with limiting rather than wrap-around, I think. It's hard to contrive a 6dB overshoot.

More to the point, level controls exist in the analog world, I have quite a few of them, as well as digital multipliers.

 

[MusicNBeer]

When did you last see SINAD results for a modern DAC producing actual 16/44 content? Amir tests DACs with 24 bit data. Knocking 6dB off to protect against poor recording/mastering practices seems to me the wrong way to go about it.

You just convert the 16 bit stream to 24 bit, then shift one bit right (down). You get 6 dB headroom and lose nothing because modern DAC chips have SINAD way over 100 dB.

With that said, the DAC chip manufacturers should be the ones ensuring the headroom.

 

[j_j]

When did you last see SINAD results for a modern DAC producing actual 16/44 content? Amir tests DACs with 24 bit data. Knocking 6dB off to protect against poor recording/mastering practices seems to me the wrong way to go about it.

Yeah, its very much NOT HARD to find out if you have intersample overs in the modern world. Just don't do them is another way to handle this. Oversampling by 4 and checking peaks suffices.

 

[restorer-john]

BTW, what do CD players with built-in DACs do? CD playback is probably the worst-case scenario, do these players apply a digital volume reduction before conversion?

It will depend on the digital filter ICs used.

I'm very interested to run some tests on a bunch of my players.

 

[MusicNBeer]

I think all of us agree that this is a failure of recording industry/hardware manufacturers. It's crazy that this even exists. It's such a simple phenomenon.

 

[theREALdotnet]

4dB is likely enough, with limiting rather than wrap-around, I think. It's hard to contrive a 6dB overshoot.

More to the point, level controls exist in the analog world, I have quite a few of them, as well as digital multipliers.

They do, and come to think of it, they exist in the digital realm, too. If you use software volume control, or the volume control of the DAC, you will have like dealt with the issue of inter-sample overs already, unless you routinely go over -4dB volume.

 

[MusicNBeer]

It will depend on the digital filter ICs used.

I'm very interested to run some tests on a bunch of my players.

Do you need files?

 

[MusicNBeer]

Do you need files?

I made three files, Fs/4, Fs/8, and Fs/16 fullscale at the correct phase. Fs/4 is 3 dB overloaded

 

[restorer-john]

Do you need files?

Sure! Post a link and I'll grab them.

 

[BeerBear]

I don’t think you’re arguing for the same solution. To me it sounds like @j_j wants a DAC that is specified for 4Vrms (5.65Vp) output to be allowed to peak up to 11.3V (6dB overshoot), and @MusicNBeer want’s to reduce the volume of the digital input signal to be reduced by 6dB before conversion.

The former solution might surprise amplifier input stages, the latter may be wasting dynamic range in situations where headroom is already provided (many software players, streamers, etc.).

Both of those seem like not insignificant compromises, then. Would dealing with just 4dB overshoots in the analog output of a DAC work okay with all amps?
TP overs of more than 4dB are rare from what I've seen, but they do exist in pop music. And that's just for lossless releases, with lossy codecs all bets are off.

IMO the responsibility of preventing ISP overs should be first and foremost on the side of music production. It's easy and foolproof (no need for new DACs, software or end-user intervention). And I think in the vast majority of cases it doesn't entail making any compromises with the sound.
That of course doesn't help for all the existing music out there, but it's good practice going forward.

 

[Sokel]

IMO the responsibility of preventing ISP overs should be first and foremost on the side of music production.

Agreed,but on the other hand I would prefer not to rely on other people's intentions and have a device that is foolproof.

 

[DonH56]

As I see it, to prevent inter-sample overs during the creation/recording process, you'd have to reduce the level into the ADC (and DACs along the way) since you have no way of controlling the source material (musicians, singers). It also means it is almost impossible to completely eliminate all clipping since you cannot predict how loud things will be in the studio (let alone live recordings) and are typically trying to maximize dynamic range and signal-to-noise ratio by aiming for peaks on the order of -3 dB FS or more. Thus to my little pea brain the proper place to handle "overs" is in or at the output of the DAC with sufficient analog headroom to handle them. At least there it is straightforward circuit design... But again I am an old analog dinosaur so tend to think about solutions in that vein. - Don

 

[DonH56]

p.s. One problem I see is that the incessant (?) push for very high dynamic range leads to higher output voltages. I remember when 100~200 mV was pretty normal, and now we are testing at 4 V output. It is hard to reduce the intrinsic noise floor, virtually impractical at these high SNRs, so to get better numbers we need more output. As a result, handling overload and intersample overs becomes more difficult in a practical sense. Instead of needing a volt or two, or even 4 V to provide 6 dB headroom over a 1 V "max" 0 dB output, now we need to produce 15~20 V. That's a lot of voltage for drivers to produce with -100 to -120 dB or less distortion.

 

[BeerBear]

As I see it, to prevent inter-sample overs during the creation/recording process, you'd have to reduce the level into the ADC (and DACs along the way) since you have no way of controlling the source material (musicians, singers).

The standard recording practices already entail several dBs of headroom when recording and that's usually not where the ISPs we're talking about are created.
They're created in the last stages of production, with aggressive digital compression/limiting and with not leaving enough headroom on the finalized tracks. If the production includes external effects then of course the DA/AD conversions can add some clipping too, if the person responsible doesn't leave enough headroom. But in those cases the clippings are a permanent part of music anyway, aka "it's what the artist intended".
The more problematic ISPs and clipping is caused by the last stages (mastering) and it's something that can make the record sound different from what was heard in the studio.

 

[MusicNBeer]

Sure! Post a link and I'll grab them.

IntersampleOverTests.zip

1 file sent via WeTransfer, the simplest way to send your files around the world

we.tl

 

[RichB]

As I see it, to prevent inter-sample overs during the creation/recording process, you'd have to reduce the level into the ADC (and DACs along the way) since you have no way of controlling the source material (musicians, singers). It also means it is almost impossible to completely eliminate all clipping since you cannot predict how loud things will be in the studio (let alone live recordings) and are typically trying to maximize dynamic range and signal-to-noise ratio by aiming for peaks on the order of -3 dB FS or more. Thus to my little pea brain the proper place to handle "overs" is in or at the output of the DAC with sufficient analog headroom to handle them. At least there it is straightforward circuit design... But again I am an old analog dinosaur so tend to think about solutions in that vein. - Don

And to my pea brain, since there are music sources with inter-sample overloads, the DAC should handle the sources in the wild.

- Rich

 

[DonH56]

The standard recording practices already entail several dBs of headroom when recording and that's usually not where the ISPs we're talking about are created.
They're created in the last stages of production, with aggressive digital compression/limiting and with not leaving enough headroom on the finalized tracks. If the production includes external effects then of course the DA/AD conversions can add some clipping too, if the person responsible doesn't leave enough headroom. But in those cases the clippings are a permanent part of music anyway, aka "it's what the artist intended".
The more problematic ISPs and clipping is caused by the last stages (mastering) and it's something that can make the record sound different from what was heard in the studio.

All true, but aside from complaining about (or not buying) the source material that's out of my control. I keep hoping the loudness wars go away, but...

And to my pea brain, since there are music sources with inter-sample overloads, the DAC should handle the sources in the wild.

- Rich

Yah, that is what I think, too.

 

[little-endian]

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.

Although I share the (academic) opinion that a DAC should be able to cope with any valid input signal and thus provide enough headroom, not only it is still unclear to what degree that is a real audible issue (given that those mastered-to-the-max productions nowadays sound like shit already anyway, intersample clipping or not), it wouldn't be the first and only (theoretical) issue which is blown entirely out of proportion just like rare interpolations which CD-players perform when errors cannot be recovered anymore by the C1/C2-decoders, the infamous oh-so-evil jitter which needs to be either buffered by the finest of the finest reclocking techniques (Benchmark) or yet to be contrained by obscure super-cables and also the intersample peak thing (which Benchmark happens to have also a great solution at hand while knowing to omit the fact that they hadn't included that kind of headroom in their beyond-everything-DAC1 back in the days, either).

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.

Excellent point. Since most DACs now use DSPs for gain/volume control, the question arises with what kind of DACs this actually still is any problem. As far as I understand it, DAC's maximum volume setting - arbitrary headroom = good to go, despite the fact that one loses that little headroom's voltage equivalent peak to peak on the analog output, given the crazy SNR-values yet once again asking for any real-world disadvantage.

I think all of us agree that this is a failure of recording industry/hardware manufacturers. It's crazy that this even exists. It's such a simple phenomenon.

Well, I have the impression that with video, things are even worse. What is naturally taken into account for audio (pre-filtering to avoid anti-aliasing, post-filtering against imaging, proper dithering to uncorrelate the input signal from the resulting quantisation error) more often than not isn't obeyed when it comes to video. Countless Blu-ray releases showing severe banding, players and image viewers with lousy scaling producing aliasing, etc. Compared to that, things are pure gold in the audio domain.