Help explain intersample overs, please?

[AudioEnthusiast]

Again not necessarily. But there's some truth to that. One counter example is using linear phase in AD and minimum phase in DA there is chance that this may still happen. Also if the filter of the AD is steeper than the DA it also may happen.

So then I’m back to my original question...

Let’s say the original signal is 16/44.1. To avoid the overs, we reduce the “digital volume” by 3.5 dB (say) before we upsample. Doesn’t that “squash” the input signal? If two samples are separated by 1-2 bits, there are now fewer bits to fit all the samples into, so those two samples may end up the same value after attenuation. Do I understand that correctly? And if so, isn’t that distortion of a different kind?

Or...is the input sample word size widened to 32 or 64 bits before the attenuation? Is that what I’m missing?

 

[AudioEnthusiast]

Ok, I did some reading on digital volume. I guess what I’m asking is, doesn’t digitally attenuating the input signal decrease the SNR? Seems like it would unless the input word length was increased and the attenuation done in a larger bit space.

 

[JohnYang1997]

Firstly yes it will truncated some bits. What matters is the noise at the output or the SNR of the actual DAC. Then some dacs like ESS based can preserve most of the SNR when reducing volume internally. For AK based I think(correct me if I'm wrong) will have a trade off between SNR and else. Most likely reducing a little bit of SNR can give better SINAD because distortion may also decrease.
So in the end you may avoid the intersample clipping and gaining SINAD at the same time. So this may be worth the trade off. More over, one can always use two dacs with each for parallel to gain back the 3db of SNR.

 

[pjug]

The Benchmark "explanation" is self serving and should be taken with a grain of salt.

Classic marketing.

Create a problem, mis-identify it (say it's not clipping when it clearly is), say everyone else has the same problem and then solve the problem in your products. Use an artificial, worst case overloaded signal at a chosen frequency to "show" the problem. Benchmark have been called for their not exactly transparent website content before.

I find this paragraph from that Benchmark blog entry very hard to believe. I really would not expect any modern DAC to have problems with intersample overs. Can anyone here confirm or refute this?

FAULTY D/A AND SRC CHIPS
Every D/A chip and SRC chip that we have tested here at Benchmark has an intersample clipping problem! To the best of our knowledge, no chip manufacturer has adequately addressed this problem. For this reason, virtually every audio device on the market has an intersample overload problem. This problem is most noticeable when playing 44.1 kHz sample rates.

 

[restorer-john]

So then I’m back to my original question...

Let’s say the original signal is 16/44.1. To avoid the overs, we reduce the “digital volume” by 3.5 dB (say) before we upsample.

If your original signal is 16/44.1 with no clipping, what are you expecting to gain (no pun) by reducing the level and upsampling in the first place?

FAULTY D/A AND SRC CHIPS
Every D/A chip and SRC chip that we have tested here at Benchmark has an intersample clipping problem!

First of all they say it isn't clipping, it is "intersample overs". And then it's "intersample clipping".

 

[AudioEnthusiast]

If your original signal is 16/44.1 with no clipping, what are you expecting to gain (no pun) by reducing the level and upsampling in the first place?

That’s a good question and I don’t know the answer. I think it’s because most DAC chips upsample internally as part of the conversion process? I admit I don’t have a firm grasp of the whole delta-sigma-decimation concept used in modern DACs.

So it’s not me upsampling the original signal, it’s the DAC.

...I think? /scratches head

 

[AudioEnthusiast]

My point being, if I feed 16/44.1 into a modern DAC, I believe it’s going to upsample anyway right? Unless it’s a ladder DAC? So that’s where the overs come from, and that’s why Benchmark claims they must provide the headroom.

I’m totally a dilettante here, although my background is physics and software. I don’t understand a lot of the details of DSP, hence my question.

 

[mansr]

I find this paragraph from that Benchmark blog entry very hard to believe. I really would not expect any modern DAC to have problems with intersample overs. Can anyone here confirm or refute this?

FAULTY D/A AND SRC CHIPS
Every D/A chip and SRC chip that we have tested here at Benchmark has an intersample clipping problem! To the best of our knowledge, no chip manufacturer has adequately addressed this problem. For this reason, virtually every audio device on the market has an intersample overload problem. This problem is most noticeable when playing 44.1 kHz sample rates.

They are correct in that few, if any, DAC chips handle arbitrary signals without clipping. A simple test it to play white noise at a few different levels and record the output spectrum with a wider bandwidth. Typically, you get something like this:

With no clipping, the three curves would all have the same shape. The elevated levels above the filter cut-off frequency are evidence of saturation (clipping) in the digital interpolation. This particular recording is from a TI PCM1794A chip recorded at 192 kHz (the rising level above 60 kHz in the green trace is modulator noise from the ADC). AKM and ESS chips give similar results.

The sin committed by Benchmark here is overstating the implications of this. When playing actual music, even highly compressed modern productions, this clipping occurs only where the recording has already been peak-limited (i.e. clipped). The distortion added by the DAC chip isn't going to be noticeable compared to what's already in the recording except perhaps in a few extreme cases.

If you are paranoid and want to be on the safe side, use a software volume control with 6 dB attenuation. There is no need to pay Benchmark prices for this. (That said, their products do seem to be well made.)

 

[JohnYang1997]

My point being, if I feed 16/44.1 into a modern DAC, I believe it’s going to upsample anyway right? Unless it’s a ladder DAC? So that’s where the overs come from, and that’s why Benchmark claims they must provide the headroom.

I’m totally a dilettante here, although my background is physics and software. I don’t understand a lot of the details of DSP, hence my question.

It's all up to implementation. Just put it that way.
In the opposite inside is it better if all music is limited to -3db and below?

 

[AudioEnthusiast]

It's all up to implementation. Just put it that way.
In the opposite inside is it better if all music is limited to -3db and below?

Right, you’re trading one thing for another, that was my question.

Anyway, it seems the answer isn’t as clear cut as I thought at the beginning, and it comes down to engineering decisions rather than right way or wrong way.

 

[MC_RME]

I must say I am astonished about the negative vibe here on Benchmark's claims. And yes, I can fully confirm them. I am not aware of any SRC that takes care of intersample overs, so you need to attenuate its digital input signal. There is no way around that*. And ESS chips do need digital attenuation at their input , as they over-react on the slightest intersample over (+0.5 dBFS). Apart from Benchmark the only other company that got this right was Oppo in their UDP-205, where they attenuated 0 dBFS to -6 dBFS before entering the ESS.

Also there is a big difference between AKM and ESS due to differences in their internal architecture. AKM chips typically provide about +2 dB headroom at their analog output. And when it gets too much all you see in a wideband FFT is out-of-band harmonics - no big deal. ESS on the contrary also pollute the audio band with spikes - a totally different behaviour.

So Benchmark is right in both cases, and most DACs tested here will fail in the second case. Just that nobody ever looked for that.

Attenuating on the digital domain will then reduce the DAC's SNR by the same amount. Point is - why care? Nobody hears noise anymore, that stuff has become much too good. Better make a device fool-proof than providing noise figures which won't help anyone.

* The distortion caused by SRCs fed with intersample over signals is dramatic and fully within the audio band.

 

[MC_RME]

My solution is entirely hypothetical. I don't know how feasible it is.

At some point you have to enter the fixed point DAC with your 32 or 24 bit data. Then you will have to shift back everything = digital attenuation required.

 

[RayDunzl]

Overs example:

44.1/16bit file, white noise with 0.999 amplitude (no full scale samples). No "clipping" reported.

Resampled in Audacity to 352k/32float

Original samples don't intuitively describe the "wave" that the original would create when converted to analog. I think the resampling at least approximates that wave visually. Correct me if I'm wrong on that.

Here, I see clipping where I might not expect it, and no clip where maybe I would expect it.

And, in this one second example, I didn't have to look far to find this example, the result is loaded with clips...

 

[restorer-john]

I must say I am astonished about the negative vibe here on Benchmark's claims. And yes, I can fully confirm them.

Benchmark are simply guilty of making their claims and illustrations entirely self serving and yes, deceptive.

Let's make the "good" D/A converter green and the "bad" D/A converter red. Let's actually hide the red D/A converter's fundamental altogether and just show the "bad" stuff. Let's have that nice green spike up there all by itself over the top of that nasty red one. Let's pick a particularly nasty 11.025KHz frequency too.

The distortion caused by SRCs fed with intersample over signals is dramatic and fully within the audio band.

Great. It should be really easy for us to hear. Please provide an example track of a CD (16/44.1 original) where this "dramatic" issue can be demonstrated on all the non RME and Benchmark D/As we might own.

 

[blueone]

I think John is right. Inter-sample overs is a marketing term adopted by the recording industry to describe a situation where certain DACs improperly handle digital clipping, and in reality the digital clipping is the result of poor recording technique. I'm glad the Benchmark DACs properly handle digital clipping, especially since I own one, but I think the positioning of this problem is being somewhat misrepresented as an intrinsic issue with digital recording. It isn't. It's an issue with DAC design.

 

[MC_RME]

Great. It should be really easy for us to hear. Please provide an example track of a CD (16/44.1 original) where this "dramatic" issue can be demonstrated on all the non RME and Benchmark D/As we might own.

I clearly wrote the SRC (!) distortion is 'dramatic'. How many DACs include an SRC? So how can this be demonstrated when the majority of DACs don't have SRCs?

Also you are starting to change my statements by intentional misinterpretation. All DACs with AKM chips and an output stage with a bit analog headroom have the mentioned (positive) behaviour, so it's not only RME. And I never said it's only RME (and Benchmark). If discussions with you are going this kind of route then I stop right here.

Let's make the "good" D/A converter green and the "bad" D/A converter red. Let's actually hide the red D/A converter's fundamental altogether and just show the "bad" stuff. Let's have that nice green spike up there all by itself over the top of that nasty red one. Let's pick a particularly nasty 11.025KHz frequency too.

Both signals obviously had the same level. Then one of them will be painted in front of the other, so how can one see the red one? That's normal view in AP. And using 11.025 kHz (fs/4) is not 'particularly nasty'. That is the 'standard' test signal for intersample peaks, as this one has +3 dBFS of level (due to its intentional phase shift of 45°).

 

[restorer-john]

So how can this be demonstrated when the majority of DACs don't have SRCs?

With respect, you not only backed up Benchmark's claims but doubled down, if anything. Here are their claims:

Every D/A chip, every sample rate converter and virtually every audio device has the problem, according to them. That's a very bold claim. But not true according to RME? Which is it?

Also you are starting to change my statements by intentional misinterpretation.

No, I am calling for evidence, that's not unreasonable is it? Evidence of the "dramatic"* audio-band issues you speak of. Evidence we can all hear if we take a commercial CD (at 16/44) and play it into a typical D/A converter we may own. I have several here, including some really vintage ones. The entire discussion/article on Benchmark's website is about CD 16/44 intersample overs/clipping when running through a D/A converter that over-samples/interpolates.

And all of this so-called problem is caused by recording "engineers" who in this day and age can't produce content without constantly running out of bits...

* your word

 

[j_j]

I think John is right. Inter-sample overs is a marketing term adopted by the recording industry to describe a situation where certain DACs improperly handle digital clipping, and in reality the digital clipping is the result of poor recording technique.

Not at all. Intersample overs are a real thing, they demonstrably exist, and some DAC's do not handle them very well at all.

A proper production, of course, wouldn't have any, in my book, but that's a separate issue.

 

[j_j]

And all of this so-called problem is caused by recording "engineers" who in this day and age can't produce content without constantly running out of bits...

* your word

The DAC can not reproduce something in its input range. That is a flaw.

 

[Blumlein 88]

I think this screenshot will illustrate what is going on.

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.

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.