DAC measurements using DeltaWave

[manisandher]

There's a lot of speculation in the thread. Please can you explain what you did ASAP to clear this up. E.g. how do you do a null test with one DAC?

Done.

 

[MaxwellsEq]

Thanks for the explanation in post 2. Can you explain why you use variable output and how you compensate for the impact of this (e.g. is this an attenuator after the DAC or DSP volume contro)l?

 

[manisandher]

Thanks for the explanation in post 2. Can you explain why you use variable output and how you compensate for the impact of this (e.g. is this an attenuator after the DAC or DSP volume contro)l?

I use variable output from the DAC (using the DAC's internal digital volume control) so that the RME is always presented with a similar-level analogue signal. (Actually, there's currently a mistake in my diagram. I refer to the RME as 'AD/DA', when in fact I'm using it just as an ADC. I will correct this.)

The RMS of the 'compare' file is typically within 0.2dB of that of the 'reference' file (having been through a D-to-A, and then an A-to-D conversion stage). DeltaWave will adjust this so that the RMS value of the 'compare' file matches perfectly with that of the 'reference' file.

 

[antcollinet]

No, I mean in most DACs ESS chips freely spin off their own 100MHz crystal. And no, the DAC is certainly not required to recover the clock. Heard of ASRC, or perhaps FiFO reclock that some DACs use?

Even with ASRC, it is required to know the incoming clock rate so the resampling can resample the incoming waveform at the correct times, and so you don't get buffer under or overflow.

The DAC section doesn't use the incoming clock but the ASRC needs it.

 

[MaxwellsEq]

I use variable output from the DAC (using the DAC's internal digital volume control) so that the RME is always presented with a similar-level analogue signal

That's likely to make quite a big difference. These days digital attenuation is excellent, so much so that it should be inaudible, BUT in this case you are performing a null after setting the DAC to be non bit-perfect, which is unfair on the DAC. Use a proper analogue attenuator instead.

 

[manisandher]

That's likely to make quite a big difference. These days digital attenuation is excellent, so much so that it should be inaudible, BUT in this case you are performing a null after setting the DAC to be non bit-perfect, which is unfair on the DAC. Use a proper analogue attenuator instead.

I disagree. It's trivial to show that the digital attenuation in a DAC is totally transparent.

Instead of using the DAC's internal digital volume control, I can set the DAC to max. output, and adjust the level of the 'reference' using Roon's 64-bit DSP. The nulls are identical.

 

[MaxwellsEq]

I disagree. It's trivial to show that the digital attenuation in a DAC is totally transparent.

Instead of using the DAC's internal digital volume control, I can set the DAC to max. output, and adjust the level of the 'reference' using Roon's 64-bit DSP. The nulls are identical.

Do neither! Make it bit-perfect end-to-end and adjust with an attenuator. You are comparing nulls between input and output, so don't mess with the bits in between or your experiment is invalid

 

[Blumlein 88]

Related to what KSTR said already, what you'll find is the null level is dominated by filtering differences. That is why sometimes a DAC with considerably better THD+N will give a less deep null. You have a couple options. One is to use 96 khz files and restrict analysis to 20 khz using Deltawave's filter. This reduces greatly the DACs filter effect in the audible band. Another is to allow Deltawave to correct for phase and FR differences. Doing this will greatly improve results. As this only happens if FR and phase are the differences it shows nothing exotic or unexpected is responsible though some don't like that you are correcting the DAC. Still you know what you are correcting.

 

[mdsimon2]

Related to what KSTR said already, what you'll find is the null level is dominated by filtering differences. That is why sometimes a DAC with considerably better THD+N will give a less deep null. You have a couple options. One is to use 96 khz files and restrict analysis to 20 khz using Deltawave's filter. This reduces greatly the DACs filter effect in the audible band. Another is to allow Deltawave to correct for phase and FR differences. Doing this will greatly improve results. As this only happens if FR and phase are the differences it shows nothing exotic or unexpected is responsible though some don't like that you are correcting the DAC. Still you know what you are correcting.

I've never used the Deltawave EQ, but have used other EQ while doing Deltawave testing to greatly improve nulls. I assume it develops a single magnitude / phase correction filter and then it applies it to the entire analysis? Does it show the response of the applied filter?

Michael

 

[manisandher]

Make it bit-perfect end-to-end and adjust with an attenuator.

Why? Roon's 64-bit DSP, and the DACs internal digital volume control are demonstrably totally transparent. No analogue attenuator is going to be anywhere near as accurate.

 

[mdsimon2]

Why? Roon's 64-bit DSP, and the DACs internal digital volume control are demonstrably totally transparent. No analogue attenuator is going to be anywhere near as accurate.

While it won't have a huge impact, even a perfect digital volume control will only attenuate the main output signal from the DAC, it will not touch noise. As a result applying any digital attenuation will decrease SNR by the amount of digital attenuation, no way around this. Assuming all DACs have roughly the same output level I don't think a few dB of attenuation is a big deal, especially if you haven't corrected magnitude / phase response.

Michael

 

[manisandher]

Related to what KSTR said already, what you'll find is the null level is dominated by filtering differences. That is why sometimes a DAC with considerably better THD+N will give a less deep null. You have a couple options. One is to use 96 khz files and restrict analysis to 20 khz using Deltawave's filter. This reduces greatly the DACs filter effect in the audible band. Another is to allow Deltawave to correct for phase and FR differences. Doing this will greatly improve results. As this only happens if FR and phase are the differences it shows nothing exotic or unexpected is responsible though some don't like that you are correcting the DAC. Still you know what you are correcting.

Well... we listen to DACs with filters, so, I don't want to eliminate their effect. I'm finding the null tests incredibly insightful, especially the 'Spectrum of Δ' plots. The differences between filters, well down below 500Hz in some cases, is amazing.

In any event, I'm interested in knowing how accurate DACs are with 44.1 content.

 

[manisandher]

While it won't have a huge impact, even a perfect digital volume control will only attenuate the main output signal from the DAC, it will not touch noise. As a result applying any digital attenuation will decrease SNR by the amount of digital attenuation, no way around this. Assuming all DACs have roughly the same output level I don't think a few dB of attenuation is a big deal, especially if you haven't corrected magnitude / phase response.

Michael

It's certainly no big deal, provided the input sensitivity of the ADC remains the same.

FWIW, here's a comparison between Roon's DSP and the DAVE's digital volume control:

Well within run-to-run variations.

​

​

 

[olieb]

Do neither! Make it bit-perfect end-to-end and adjust with an attenuator. You are comparing nulls between input and output, so don't mess with the bits in between or your experiment is invalid

There will be conversion to analog, digitizing and digital volume adjustment (in Deltawave).
or
Digital volume adjustment + conversion to analog, digitizing and digital volume adjustment again.
This additional digital operation happens is rendered in 24bit (<-140dB). I do not think it makes any kind of meaningful difference. The impact from noise when using an additional analog attenuator is (probably) higher.

The differences between filters, well down below 500Hz in some cases, is amazing.

Why?

The null comes from (in order of relevance):
linear (filter) effects in magnitude and phase • • • • • • • • • • • • • • • • • • • • • • • • noise • • • • nonlinear effects • • • • digital effects

So the null will tell hardly anything about SINAD.
Your file has a lot more energy in mids than in highs.
Does this impact the spectra? ∆ is low in highs because the there is no signal?
How do the ∆-spectra look like with white noise?

 

[manisandher]

So the null will tell hardly anything about SINAD.

Pretty much the whole point of this thread is to get away from SINAD .

Your file has a lot more energy in mids than in highs.
Does this impact the spectra? ∆ is low in highs because the there is no signal?

It's a music signal, so yes, far more energy in lows/mids than highs.

This certainly impacts the 'Spectrum of Δ' plots. But it's the case for all the DACs. And yet, some DACs/filters are substantially worse in the lows, mids and highs - and not just the highs as a simple FR might suggest.

 

[mdsimon2]

And yet, some DACs/filters are substantially worse in the lows, mids and highs - and not just the highs as a simple FR might suggest.

I've shown that pkmetric can be improved from -63 dB to -117 dBFS by implementing EQ which only affects FR (magnitude / phase), see here.

What support do you have for your statement that the differences you are seeing are not due to simple FR? Would you mind sharing your reference file? I'd like to try it out.

Michael

 

[manisandher]

As an example of what I mean, let's take another look at the effects of the 'fast linear' and 'fast minimum' filters in the SMSL SU-10 DAC.

Here are the 1kHz sine plots (again):

What might one conclude from these? Well, that the 'fast minimum' is slightly 'better', right?

Perhaps the FR plots of the filters provide more insight? Here are Amir's plots:

The two filters in question look pretty much identical below 20kHz.

But now, look at the 'Spectrum of Δ' plots:

The effects of the 'fast minimum' go all the way down to 30Hz!

Maybe this was already obvious to everyone here, but it certainly wasn't to me!

 

[pjug]

If you just upsample with different filters you can do the same kind of Deltawave comparisons and see the effects without any analog conversion. Just use the different Roon filters and make straight digital captures.

 

[Sokel]

If so many things are needed for a very well constructed test and we still have doubts imagine the wild west of a real-world ABX with DACs as they sit on top of the spaghetti,at greatly reduced levels (if tested straight or with the handicap of the most attenuated one so to match levels),with different output impedances,etc .

One would think that an anechoic chamber comparison would be the next logical step for apples to apples.
Thanks for this!

 

[Blumlein 88]

I've never used the Deltawave EQ, but have used other EQ while doing Deltawave testing to greatly improve nulls. I assume it develops a single magnitude / phase correction filter and then it applies it to the entire analysis? Does it show the response of the applied filter?

Michael

It does not show the filter. It does show the change in phase with phase EQ and without.