R2R vs Delta Sigma DACs - same results?

[solderdude]

That's the main 'problem' with SOTA performance of audio gear (near practical system limits).
It may look 'nicer' on paper (in certain numbers) but is inconsequential in real life when reproducing music.

And there we are also hitting on the remark from @pma.
SOTA performance.
There may well be less (shaped) noise in numbers but you hit another practical limit which is linearity which also is a kind of 'unwanted' signal (noise) which differs from what DS can reach (although drowned in shaped noise).

So in practice there is no benefit to multibit R2R (because of resistor tolerances even when compensated with smart calibration) certainly not with filterless (NOS) R2R.
The converter output may be more quiet when no signal is applied (and no muting is engaged in absence of a signal) but they have HUGE amounts of noise above Nyquist the moment a signal is present.

 

[antcollinet]

I realize that we are splitting hairs here, but I think that technically, the R2R DAC can have a higher SNR without intersample over 0db (for 16 bit pcm data).

Not really - because the volume control DSP of a delta sigma DAC is done in the 24 bit or 32 bit domain (after the 16 bit data has been padded into the 24+ bit data). So you are not losing any of the dynamic range of the 16 bit audio. Further, you are ignoring the fact that properly dithered 16 bit audio actually achieves a dynamic range of around 112dB, or up to 120dB with noise shaped dither.

24/192 Music Downloads are Very Silly Indeed

people.xiph.org

 

[Jazigo]

Not really - because the volume control DSP of a delta sigma DAC is done in the 24 bit or 32 bit domain (after the 16 bit data has been padded into the 24+ bit data). So you are not losing any of the dynamic range of the 16 bit audio. Further, you are ignoring the fact that properly dithered 16 bit audio actually achieves a dynamic range of around 112dB, or up to 120dB with noise shaped dither.

24/192 Music Downloads are Very Silly Indeed

people.xiph.org

I actually come from another thread where I was correct on 16 bit data. Due to quantization error, it is not possible to get better than -98.7db (even with 16bit being scaled to 32 with fir filters and all that).
There is no information lost of course, as the scaling to 32 bit will use all the data, but you are still limited to the original quantization error from the 16bit data. Since that is our noise-floor, we want to keep the SNR as high as possible, which means to map a full 16 bit to a full 32bit.

 

[Ropeburn]

Hi everyone!

Is there a case for R2R DACs being the better option for 16bit content?

Current oversampling DACs have the issue with intersample overs 0db, so one have to adjust the digital volume down to -6db.
The noise of 16bit audio is a best -98.7db.
This means that the SNR is at best 92.7db with a oversampling DAC, but we could get the full 98.7db with a R2R DAC.

You can choose: intersample overs you probably don't hear, or a bit less SNR you probably don't hear.

 

[Jazigo]

You can choose: intersample overs you probably don't hear, or a bit less SNR you probably don't hear.

Hahaha, yes. We can only dream of a speaker even coming close to these values. But still interesting technically.

 

[DSJR]

Hahaha, yes. We can only dream of a speaker even coming close to these values. But still interesting technically.

I was once told that our ears have a FIFTY dB dynamic range but on a sliding scale - a good while of high level monitoring as was once done and going outside, the general 'noise' of the surroundings is lost at least temporarily as the ears re-calibrate. In the same way, getting up in the ,morning to a quiet house and turning the sound system on full pelt will cause pain in my experience.

I'd also suggest that 'quiet details' in a typical music mix aren't much more than twenty dB down on th emean level in any case, so us all but wetting ourselves about sinad over 110dB is academic hobbyism really, although of course, it's wonderful that modern and some older digital gear can do this with ease.

I'd love a Sony 1610 system from the early 80s to be tested (maybe it has?) to see how it works today. I gather the real issue with very early digital releases was more in the workstations/editing and so on, than the converters themselves...

 

[voodooless]

I was once told that our ears have a FIFTY dB dynamic range but on a sliding scale

It's probably more like 30 to 40 dB for of instantaneous dynamic range. Then the ear is fast enough that, in practice, for most sounds, this is dynamically expanded to about 70dB. But the total range is about 120 to 130 dB, depending on frequency, conditions, and tolerance level.

 

[Ropeburn]

It's probably more like 30 to 40 dB for of instantaneous dynamic range. Then the ear is fast enough that, in practice, for most sounds, this is dynamically expanded to about 70dB. But the total range is about 120 to 130 dB, depending on frequency, conditions, and tolerance level.

70dB sounds about right. It's why vinyl and 12bit sampling instruments already sound really good to us.

 

[Scytales]

Ironically, what most people may not realize, is that the actual output stage of a Delta Sigma DAC also is R2R (or something to a similar effect) but only 4 to 7 bits or so.
This is much easier to make accurately than the difference between the MSB and LSB of a 16-20 bit R2R(ladder or other type) of converter.

I agree with your post, but I just want to add for the sake of completeness that, to my knowledge, there have been other techniques to convert multilevel sigma-delta modulation to analogue than something like an R2R ladder DAC: switched current sources, switched capacitors filter, and also DACs that do not produce varying voltages following the sigma-delta coded modulation but equal amplitude pulses of varying lengths (PWM or PLM converters, as Sony, which was the main champion of this method with Technics, called it).

 

[antcollinet]

I actually come from another thread where I was correct on 16 bit data. Due to quantization error, it is not possible to get better than -98.7db (even with 16bit being scaled to 32 with fir filters and all that).
There is no information lost of course, as the scaling to 32 bit will use all the data, but you are still limited to the original quantization error from the 16bit data. Since that is our noise-floor, we want to keep the SNR as high as possible, which means to map a full 16 bit to a full 32bit.

1 - with dither it is possible to get much better than 96dB dynamic range from 16 bit (see the link I posted above)

2 - Once you've expanded the 16 bit date to 24 bits, you can turn the volume down by 48dB without losing *any* of the 16 bit data. If it is expanded to 32 bit for the DSP, than you can turn it down by 96dB before you lose any of the 16 bit data. Believe me when I tell you neither of these are a problem.

 

[solderdude]

I agree with your post, but I just want to add for the sake of completeness that, to my knowledge, there have been other techniques to convert multilevel sigma-delta modulation to analogue than something like an R2R ladder DAC: switched current sources, switched capacitors filter, and also DACs that do not produce varying voltages following the sigma-delta coded modulation but equal amplitude pulses of varying lengths (PWM or PLM converters, as Sony, which was the main champion of this method with Technics, called it).

Indeed, the same is true for the multibit converters where different methods were used than the 'classic' resistor ladder networks.

 

[Jazigo]

1 - with dither it is possible to get much better than 96dB dynamic range from 16 bit (see the link I posted above)

2 - Once you've expanded the 16 bit date to 24 bits, you can turn the volume down by 48dB without losing *any* of the 16 bit data. If it is expanded to 32 bit for the DSP, than you can turn it down by 96dB before you lose any of the 16 bit data. Believe me when I tell you neither of these are a problem.

Sorry to burst your bobble (i was in the same bobble until recently):

SMSL PL200 Review (CD Player)

At least as a CD player, the OPPO was reproducing the digital content of the CDA nearly unchanged, when measured from analog outputs, and that was a first in my tests. As of now, only the SMSL PL200 did as good. So yes, an ESS or AK seem to be required to get there. Of course remains all other...

www.audiosciencereview.com

Basically a CD player with a kick ass DAC. Why do we not measure better values? See the comments.

 

[solderdude]

We do not see better values because CD format is the limiting factor.

Not so much the DAC. It is possible to reach 22 bit resolution but of course... with high noise in the smallest resolution.
R2R being worse in linearity but has different noise(character).

The S/N ratio simply decreases with output voltage because of physics.

 

[Jazigo]

We do not see better values because CD format is the limiting factor.

Yes and what I am pointing out is that technically an R2R DAC can achieve higher SNR on 16bit data (cd quality) than an DS DAC.

 

[gwing]

Sorry to burst your bobble (i was in the same bobble until recently):

SMSL PL200 Review (CD Player)

At least as a CD player, the OPPO was reproducing the digital content of the CDA nearly unchanged, when measured from analog outputs, and that was a first in my tests. As of now, only the SMSL PL200 did as good. So yes, an ESS or AK seem to be required to get there. Of course remains all other...

www.audiosciencereview.com

Basically a CD player with a kick ass DAC. Why do we not measure better values? See the comments.

I did have a look at that thread but the bubble isn't burst

If you have a 16bit signal and upsample to 24 or more it is true that you are still limited by the quantisation of the original 16 bit recording and that subsequent replay will not improve on the original (let's assume here that we are just upsampling and not actually trying to mess around with dither etc. to actually improve the signal).

However as Ant says we have at this point a 24 or 32 bit signal, even if the true information content is just that of the original 16bit signal, and therefore we have 32-16 or 24-16 bits of attenuation we can use for volume control before we get to the original 16bits i.e. we are not losing any signal content.

 

[Jazigo]

I did have a look at that thread but the bubble isn't burst

If you have a 16bit signal and upsample to 24 or more it is true that you are still limited by the quantisation of the original 16 bit recording and that subsequent replay will not improve on the original (let's assume here that we are just upsampling and not actually trying to mess around with dither etc. to actually improve the signal).

However as Ant says we have at this point a 24 or 32 bit signal, even if the true information content is just that of the original 16bit signal, and therefore we have 32-16 or 24-16 bits of attenuation we can use for volume control before we get to the original 16bits i.e. we are not losing any signal content.

Yes, you do not lose the content but the SNR is still the same, -98.7db at full volume. If you adjust the volume, you lose SNR, but not the content. It is the SNR we are after.

 

[voodooless]

Technically, this is not the case for ESS Dacs:

 

[danadam]

Yes, you do not lose the content but the SNR is still the same, -98.7db at full volume. If you adjust the volume, you lose SNR, but not the content. It is the SNR we are after.

No. When you adjust the volume, after extending the 16-bit data to 24 or 32 bit, then both the signal and the 16-bit quantization noise get reduced by the same amount and the relative difference between them stays the same. You start losing SNR only when you reduce the volume so much that you hit quantization noise of 24 or 32 bits respectively.

 

[Jazigo]

Technically, this is not the case for ESS Dacs:
View attachment 487866

I see your point! This raises a question why we don't see this in the measurements of the CD player...

 

[voodooless]

I see your point! This raises a question why we don't see this in the measurements of the CD player...

See what? What are you expecting to see with a CD player?