• Welcome to ASR. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

DAC measurements using DeltaWave

DeltaWave will certainly time align the two files. And yes, it has a very good "clock drift" function. However, the best nulls are still achieved by synchronising the DAC and ADC clocks.

ESS chips typically run in async mode, can this have an effect on the test results?
 
ESS chips typically run in async mode, can this have an effect on the test results?
Are you meaning async USB? If so it doesn't' apply since @manisandher is using Toslink as the input to the DAC.

The dac is required to recover the clock from the input stream. What method is uses, and how well it does this will have an impact on the test result
 
In terms of accuracy, simple frequency response will tell you the most about DAC accuracy, unless THD+N is really bad.

Not always, I don't think.

Take the 'fast linear' and 'fast minimum' filters in the SU-10. As you can see from Amir's measurements, they have a virtually identical frequency response, well beyond 20kHz. Their respective THD+Ns look like this (0dB to -160dB scale):

1736099563918.png


The 'fast minimum' actually performs slightly better than the 'fast linear'!

And yet, my null tests suggest it is less accurate with real music... as I will show :).
 
As you can see from Amir's measurements, they have a virtually identical frequency response, well beyond 20kHz.
That is only magnitude. Phase is frequency response, too.
The best null will probably reached, if the DAC frequency response (EDIT: Magnitude AND phase) looks like the inversion of the ADC frequency response used for the nulling.
I suppose my purpose in this thread is to show how accurately a DAC produces the music signal it is presented with. And that THD+N is perhaps not the best measure...
THD will have very little to do with the result. It is more another way of looking at FR (again: magnitude AND phase).
 
Last edited:
All good stuff Michael. Thanks.

I suppose my purpose in this thread is to show how accurately a DAC produces the music signal it is presented with. And that THD+N is perhaps not the best measure...

100%. THD+N is not the best measure, but it's a pretty convenient one. Even when doing musical comparisons, the source content matters.

Here, we compared a Topping D90 vs. a Topping D90 into a 22dB SINAD 300B SET into a non-inductive resistor

In terms of accuracy, simple frequency response will tell you the most about DAC accuracy, unless THD+N is really bad.
But translating that frequency response into "what does that mean for music" which has a musical spectrum that rolls off with higher frequencies is a different comparison.

Have you tried using the Deltawave comparator to see if you can reliably differentiate the recordings from each other (or from the original) via listening with your "best" DAC? That would be the most interesting comparison IMO.
Once you record the analog result (through speakers and then a microphone) as opposed to an electrical result (from wires/cables) the ambient noise is so high that it's hard to get really perfect nulls. Still, the negative and positive controls are important and electrically, a good ADC does a great job with run-to-run consistency even from day to day when dealing with pkMetric.

Electrically, there is "electrical weather" where some days the electricity seems clean and other days it isn't. But it's well outside the audible range.
 
But translating that frequency response into "what does that mean for music" which has a musical spectrum that rolls off with higher frequencies is a different comparison.

Here's the spectrum of the track I'm using:
1736102049246.png
 
Not always, I don't think.

Take the 'fast linear' and 'fast minimum' filters in the SU-10. As you can see from Amir's measurements, they have a virtually identical frequency response, well beyond 20kHz. Their respective THD+Ns look like this (0dB to -160dB scale):

Virtually identical is not the same as identical. Both FRLP and FRMP will not extend all the way to 22 kHz, both will have some magnitude response ripple and FRMP will not have linear phase response.

What is your hypothesis? What do you think affects the null more than magnitude and phase response?

If FRMP has a better null my guess would be that it is because it has slightly extended magnitude response compared to FRLP and that this matters more than phase response (at least for pkmetric).

FRLP vs FRMP.png


Michael
 
Are you meaning async USB? If so it doesn't' apply since @manisandher is using Toslink as the input to the DAC.

The dac is required to recover the clock from the input stream. What method is uses, and how well it does this will have an impact on the test result

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?
 
Last edited:
We had a long discussion a while ago why comparing the DAC output to the original is extremely difficult, the residual is most of the time fully dominated by the roll-off of any filters present (DC filter if present, and of course the combined analog and digital reconstruction filters)... and note that's the combined response of the DAC Under Test and the ADC used to digitize the output.

In the end this means the residual is dominated by simple linear errors in regions that are basically outside of what is relevant. These linear errors also include ever so slight gain drifts, again both in the DAC and ADC used.

DW's results are very much influenced by the exact settings... I've spent days to fine-tune the settings to show the best results. The most basic thing is to restrict the analysis range to 20Hz...20kHz, exact numbers, via pre and post filtering. In general the matching parameters (gain and delay) usually need to be manually tweaked for lowest residual.
 
Here's Amir's result:

1736102939637.png


The 'fast linear' and 'fast minimum' are virtually identical up to 20kHz. But as you say, the latter has an extended frequency response. However, the track I'm using is already -100dB at 20kHz, and attenuates very rapidly beyond that.
 
reserved for equipment and method
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?
 
We had a long discussion a while ago why comparing the DAC output to the original is extremely difficult, the residual is most of the time fully dominated by the roll-off of any filters present (DC filter if present, and of course the combined analog and digital reconstruction filters)... and note that's the combined response of the DAC Under Test and the ADC used to digitize the output.

In the end this means the residual is dominated by simple linear errors in regions that are basically outside of what is relevant. These linear errors also include ever so slight gain drifts, again both in the DAC and ADC used.

DW's results are very much influenced by the exact settings... I've spent days to fine-tune the settings to show the best results. The most basic thing is to restrict the analysis range to 20Hz...20kHz, exact numbers, via pre and post filtering. In general the matching parameters (gain and delay) usually need to be manually tweaked for lowest residual.

I'm using pretty much the simplest settings possible. The key is synchronising the DAC and ADC clocks. If this isn't done, then all bets are off.

What encourages me is the tiny variability between samples, as I showed in the OP.
 
The key is synchronising the DAC and ADC clocks.

I was earlier trying to hint that ESS solutions run on their own internal clock with ASRC, idk how much effect this has on the results but the ESS DAC chips are not clocked by SPDIF, unless the device specifically provides such option.
 
What is your hypothesis? What do you think affects the null more than magnitude and phase response?

These are clearly the two determining factors. However, in 'traditional' tests, magnitude only seems to be considered using non-modulating tones. And I don't see much consideration of phase response at all.

I'll continue with sharing my setup/method and some further measurements. If nothing comes of it, that's fine - it's something I've wanted to do anyway. And I'm sure I'll learn a lot (from people like you Michael) too :).
 
I was earlier trying to hint that ESS solutions run on their own internal clock with ASRC, idk how much effect this has on the results but the ESS DAC chips are not clocked by SPDIF, unless the device specifically provides such option.

If I ran the DAC and ADC with their internal clocks, there's no way I'd be getting the nulls I'm getting... without a lot of tweaking in DW, at least.
 
If I ran the DAC and ADC with their internal clocks, there's no way I'd be getting the nulls I'm getting... without a lot of tweaking in DW, at least.

I suspect the effects of ASRC are negligible given sufficient internal clock stability and frequency match, but it should be pointed out that DACs can use their own clock which is independent from the SPDIF clock.
 
I suspect the effects of ASRC are negligible given sufficient internal clock stability and frequency match, but it should be pointed out that DACs can use their own clock which is independent from the SPDIF clock.

With a ASRC the DPLL will still lock to the incoming signal. REW will report that the DAC output is sync'd to the input (0.0 ppm clock drift) and you can use a rectangular window FFT. Therefore I don't think it should have any impact to the Deltawave analysis.

Michael
 
We had a long discussion a while ago … The most basic thing is to restrict the analysis range to 20Hz...20kHz, exact numbers, via pre and post filtering. In general the matching parameters (gain and delay) usually need to be manually tweaked for lowest residual.

@pkane ?

those settings may matter for the simple no. But the main advantage of using PK metric is that it is perceptually rated so that the differences in the areas that you can hear are weighted higher than the areas that you cannot hear. Even though the PK metric has not been fully independently, validated, it is a consistent mathematical comparison, and we definitely see that there are products where you can cross compare with very good nulls and products where you cannot cross compare with great nulls.
 
Back
Top Bottom