R2R vs Delta Sigma DACs - same results?

[Jazigo]

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

In this testing (https://www.audiosciencereview.com/forum/index.php?threads/smsl-pl200-review-cd-player.66199/) of a cd player with a kick ass DAC, we see that the SNR decreases when measured at the XLR output.

 

[voodooless]

In this testing (https://www.audiosciencereview.com/forum/index.php?threads/smsl-pl200-review-cd-player.66199/) of a cd player with a kick ass DAC, we see that the SNR decreases when measured at the XLR output.

You'll have to be more precise in pointing out what you mean... In any case, this player does not have an ESS DAC, so volume control implementation might be different.

 

[Jazigo]

You'll have to be more precise in pointing out what you mean... In any case, this player does not have an ESS DAC, so volume control implementation might be different.

You may be right. I the second test you can see that with a -6dbfs he loses SNR, but that may actually be since the signal is -6dbfs lower and not the volume turned down -6dbfs.

 

[gwing]

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.

Well I can't argue with that. If you have a system with a noise floor any volume reduction, even analog, is going to worsen the SNR but what do you think this has to do with digital attenuation of a 16bit signal in 24/32 bit form?

 

[Jazigo]

what do you think this has to do with digital attenuation of a 16bit signal in 24/32 bit form?

We want the highest SNR as we can at the lowest signal as possible, not matter if we feed the DAC one format or the other.
I was initially arguing that we would lose SNR when we turn down the volume (not the content), but @voodooless has a counter argument I am currently researching that may prove otherwise.

 

[bmc0]

but that may actually be since the signal is -6dbfs lower and not the volume turned down -6dbfs

Yes, this is correct. You seem to have initially misinterpreted the measurement.

Here's a small demonstration—
1kHz sine at -0.01dBFS:

Signal attenuated 6dB before quantizing to 16-bit:

You can see that the noise floor stays at the same level. If we instead convert the 16-bit data to a higher bit depth before doing the gain reduction, we get this:

The 16-bit noise floor goes down accordingly (until we reach the limits of the higher bit depth).

 

[solderdude]

I'd like to point out that the noise floor of the plots above is not the actual noise floor in any DAC.
The S/N ratio (from 0dBFS) is what's important for the signal fidelity and even that will be lower than any recording where non-gated mic recordings are present.
So in practice this is all moot and a pointless discussion.
Sadly there are such things as physical limits and noise floor is one of those.

 

[voodooless]

I was initially arguing that we would lose SNR when we turn down the volume (not the content), but @voodooless has a counter argument I am currently researching that may prove otherwise.

Well, in a normal digital volume control before the DAC, you do lose SNR, because the SNR of the DAC does not move. But here with 16-bit content, you have still plenty of headroom before you reach the DAC SNR, which others already pointed out.

Now if the volume control is integrated into the DAC, there is somewhat more wiggle room and bend the physics a little.

Either way though, it generally is not an issue. If you don’t heat the DAC noise floor, either implementing will be transparent.

 

[bmc0]

I'd like to point out that the noise floor of the plots above is not the actual noise floor in any DAC.

The unweighted SNR in the first plot is 93.5dB. Plenty of real DACs can achieve that. Perhaps you're confusing the level of the individual FFT bins with the total integrated noise.

 

[solderdude]

The unweighted SNR in the first plot is 93.5dB. Plenty of real DACs can achieve that. Perhaps you're confusing the level of the individual FFT bins with the total integrated noise.

That's exactly what I was pointing out. SNR is not shown nor mentioned in the FFT's (post #46).

 

[bmc0]

That's exactly what I was pointing out.

I don't quite see the point you're trying to make. The absolute level of the integrated noise and/or the frequency bins is irrelevant to the point I was demonstrating.
For completeness, I used a 64k FFT and rectangular window. The output from a good DAC analyzed with the same FFT length and window would be nearly the same.

 

[Jazigo]

Yes, this is correct. You seem to have initially misinterpreted the measurement.

Here's a small demonstration—
1kHz sine at -0.01dBFS:
View attachment 487888
Signal attenuated 6dB before quantizing to 16-bit:
View attachment 487889
You can see that the noise floor stays at the same level. If we instead convert the 16-bit data to a higher bit depth before doing the gain reduction, we get this:
View attachment 487890 View attachment 487891
The 16-bit noise floor goes down accordingly (until we reach the limits of the higher bit depth).

Thank you for the visuals! Are those measurements or products made to illustrate the point? Either way, what is proposed means that a dac with 120db SNR can turn the volume down as much as -21.3db before it start eating into the original 16bit data(98.7db SNR)(?)

 

[bmc0]

Are those measurements or products made to illustrate the point?

The latter. The plots show what you'd get from a theoretical, perfect 16- or 24-bit DAC.

Either way, what is proposed means that a dac with 120db SNR can turn the volume down as much as -21.3db before it start eating into the original 16bit data(98.7db SNR)(?)

Yes (more-or-less). Note that the unweighted SNR of 16-bit for a full-scale sine wave with simple flat dither is about 93.3dB. The SNR without dither is a bit higher, but the quantization distortion that results is generally more objectionable than dither noise.
As has been mentioned, the perceptual SNR of 16-bit can be very high with noise-shaped dither—in the neighborhood of 120dB.

 

[Jazigo]

Thank you to all who have contributed to information, so that I can stand corrected!
Which brings me back to my first post; no there is not a case for R2R DACs having better SNR than DS DACs.

 

[antcollinet]

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.

How? Given my point 2 above.


Ignore -just seen your post above this one.

 

[earlevel]

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

If I followed the conversation correctly—and I think I did—@Jazigo was referring to starting with 16-bit conversion from R2R. If you start with 16-bit, you can't get better by dithering. Your source needs at least one bit more than your target for dither. Dithering to a same or larger word length target just adds noise. So I'd say your statement makes more sense if you had said "to 16 bit" (from larger), but you seem to be saying "from 16 bit" (to larger? or same?). No?

 

[antcollinet]

If you start with 16-bit, you can't get better by dithering.

You can if the 16 bit source data has been dithered when it was created - as it should be.

 

[earlevel]

You can if the 16 bit source data has been dithered when it was created - as it should be.

Look...the person you responded to was clearly talking about a 16-bit R2R DAC. Meaning you have 16 bits to work with. You can't dither 16 bits and have it gain something. (Technically, dithering is adding noise—the only practical reason to dither is if you intend to reduce the word length. I don't think we're talking about going down to 15 bits, hence my assertion.)

If you mean a different thing, please take time to explain it to me, don't just repeat.

 

[antcollinet]

Look...the person you responded to was clearly talking about a 16-bit R2R DAC.

Right. So the record company produces a CD. 16 bit. They do the mastering in 32 bit. Then when they create the CD master they dither that down to 16 bit.


I buy it, take it home, put it in my CD player connected digitally to my external DAC. The dac is now converting 16 bit dithered samples. The dynamic range produced can be as high as 120dB. I was pointing out that the R2R DAC does NOT get better dynamic range compared to a Delta Sigma Dac with 16 bit data - so the discussion included both architectures.

 

[earlevel]

Right. So the record company produces a CD. 16 bit. They do the mastering in 32 bit. Then when they create the CD master they dither that down to 16 bit.


I buy it, take it home, put it in my CD player connected digitally to my external DAC. The dac is now converting 16 bit dithered samples. The dynamic range produced can be as high as 120dB. I was pointing out that the R2R DAC does NOT get better dynamic range compared to a Delta Sigma Dac with 16 bit data - so the discussion included both architectures.

Ah, thanks for the explanation. Yes, agreed, the R2R DAC can't get better resolution, and in fact the unrealistic resistor precision requirements for the ladder to be precise to 16-bit guarantees it's going to be worse than Delta Sigma with 16 bit data.