DAC measurements using DeltaWave

[pkane]

I thought you said we should use it (also I am using the settings on your gearspace tablet page to compare with those interface measurements ):

?
Non-linear EQ is meant to correct for differences in phase and filter frequency response so their effect can be eliminated when comparing two waveforms. Normally, you don't want to eliminate these, since their effect is not only measurable, but very often audible. The gearspace page I posted was to show what happens when you eliminate phase/frequency response differences due to filters.

 

[Music1969]

?
Non-linear EQ is meant to correct for differences in phase and filter frequency response so their effect can be eliminated when comparing two waveforms. Normally, you don't want to eliminate these, since their effect is not only measurable, but very often audible. The gearspace page I posted was to show what happens when you eliminate phase/frequency response differences due to filters.

Noted that. The default DW settings page is a good start for most cases?

 

[pkane]

Noted that. The default DW settings page is a good start for most cases?

The default settings are conservative, and should be a good start. These will closely match the results posted by Didier in the Gearspace AudioDiffMaker thread. The computation to apply to the comparison waveform to match it to reference should correct for level differences, fixed delay, and possibly clock drift. Anything else I'd consider 'advanced' manipulation and you should be very careful when applying these.

 

[Music1969]

The default settings are conservative, and should be a good start. These will closely match the results posted by Didier in the Gearspace AudioDiffMaker thread. The computation to apply to the comparison waveform to match it to reference should correct for level differences, fixed delay, and possibly clock drift. Anything else I'd consider 'advanced' manipulation and you should be very careful when applying these.

Now I want to capture my DAC output , recording with Cosmos ADC at PC176kHz. Then I will let DW do the downsampling to PCM44.1kHz because I am using Didier's 2min test sample.

I could use my own music but since I did Gearspace measureemnts before, I thought why not just stick to the same reference signal

So for DAC captures and nulls, the default settings is still a good start?

Or should I copy @manisandher's on page 1?

 

[pkane]

Now I want to capture my DAC output , recording with Cosmos ADC at PC176kHz. Then I will let DW do the downsampling to PCM44.1kHz because I am using Didier's 2min test sample.

I could use my own music but since I did Gearspace measureemnts before, I thought why not just stick to the same reference signal

So for DAC captures and nulls, the default settings is still a good start?

Or should I copy @manisandher's on page 1?

Default settings and Mani's settings are both fine. I normally also check the 'auto-trim start/end' option as that eliminates obvious lead-in/lead-out differences that may otherwise affect the result.

 

[Music1969]

Default settings and Mani's settings are both fine. I normally also check the 'auto-trim start/end' option as that eliminates obvious lead-in/lead-out differences that may otherwise affect the result.

If it says data appears to be simple waveform (measuring DAC output of Didiers 2min test sample) is it best to say yes, treat as simple waveform?

 

[pkane]

If it says data appears to be simple waveform (measuring DAC output of Didiers 2min test sample) is it best to say yes, treat as simple waveform?

Not if it's music. Something is wrong if DW thinks it's a simple waveform.

 

[Music1969]

Not if it's music. Something is wrong if DW thinks it's a simple waveform.

I never listened to Didiers 2min sample despite doing many loopback measurements LOL.

Is it music? I guess Ill hit play.

 

[Music1969]

Not if it's music. Something is wrong if DW thinks it's a simple waveform.

Ok DIdiers sample is actually classical music LOL.

Can you have a look for me to see why it might be thinking its a simple waveform?

My DAC output is recorded with Cosmos ADC at PCM176kHz. I set DW downsample it to match Didier's PCM44.1kHz test sample

Google drive with both Didiers sample and my DAC output recording:

pkane - Google Drive

drive.google.com

 

[pkane]

Ok DIdiers sample is actually classical music LOL.

Can you have a look for me to see why it might be thinking its a simple waveform?

My DAC output is recorded with Cosmos ADC at PCM176kHz. I set DW downsample it to match Didier's PCM44.1kHz test sample

Google drive with both Didiers sample and my DAC output recording:

pkane - Google Drive

drive.google.com

You're still trying to do more manipulation on the comparison waveform than is the default or conservative.

When downsampling, it's important to pick a proper filter (FIR, usually 1M is what I use), and you'll need to correct for clock drift since the two clocks are unlikely to be synchronized (in original2.wav and your 176k recording). These are the settings I'd try for this:

 

[Music1969]

@pkane here is what I am seeing, capture my Gustard AH90 DAC output at PCM176kHz with Cosmos ADC.

This is when I say NO to measure as simple waveform. It complains of clock drift but then uses another method and ends up with a good fit

 

[pkane]

@pkane here is what I am seeing, capture my Gustard AH90 DAC output at PCM176kHz with Cosmos ADC.

This is when I say NO to measure as simple waveform. It complains of clock drift but then uses another method and ends up with a good fit

View attachment 439765View attachment 439766View attachment 439767View attachment 439768

I see. It'll complain about drift when the error exceeds a single sample, that doesn't mean that this is not accurate.

But, again, you're resampling comparison and then filtering both waveforms. The result seems OK to me, but I'd still record the comparison at the same sampling rate as the original file, 44.1k, and then run the comparison.

 

[Music1969]

You're still trying to do more manipulation on the comparison waveform than is the default or conservative.

When downsampling, it's important to pick a proper filter (FIR, usually 1M is what I use), and you'll need to correct for clock drift since the two clocks are unlikely to be synchronized (in original2.wav and your 176k recording). These are the settings I'd try for this:

View attachment 439763View attachment 439758

ok with your settings , it now doesnt complain about simple waveform and doesnt complain about clock dift

I think this is tricky to do with separate DAC and separate ADC (as many here already wrote). Unless you do it with SPDIF like manisandher did

Here is my DAC measured with Cosmos:

 

[Music1969]

but I'd still record the comparison at the same sampling rate as the original file, 44.1k, and then run the comparison.

ok for completeness, im trying this now

 

[Music1969]

I'm noticing the interfaces with better multitone measurement (loop of DAC to ADC) have better difference RMS performance with same loop

And vice versa - worse multitone measurement in loop, worse difference RMS performance

Probably not surprising really

ALso everyyone should go try latest DW update for DC filter:

DC filter effect on the null RMS metric

The point of the filter is to undo minimum phase across the spectrum, not actual DC blocking. The filter is constructed by first making a linear-phase/FIR filter, and then adjusting it for minimum phase. I understand the reason for applying DW's HP filter as a correction for the DC blocking...

www.audiosciencereview.com

 

[TNT]

The aim of this thread is to measure and compare DACs using DeltaWave, and not necessarily to get the best nulls.

This!!! Keep focusing on this - very interesting. Don't bother about "excuses" why no nulls. Please verify that your test setup with a "cable" produces a perfect null - then introduce DUT instead of "cable". Reconstruction filter and DC behaviour is a part of the DUT and should be identified why the test - it's not something that should be removed in oder to "get a good null".

Once you found a suitable process, I think it should be suggested to be added to the standard suite for DACs here... I believe it would forward the SOTA.

And please keep the analysis space as wide as possible that the test setup allows (freq and level) as we don't know what aspects really have an impact of our SQ experience - to limit any would be to perhaps hide reasons - which we believe can't count - but perhaps they do. Be open.

//

 

[pjug]

This!!! Keep focusing on this - very interesting. Don't bother about "excuses" why no nulls. Please verify that your test setup with a "cable" produces a perfect null - then introduce DUT instead of "cable". Reconstruction filter and DC behaviour is a part of the DUT and should be identified why the test - it's not something that should be removed in oder to "get a good null".

That's the problem -- the ADC's DC filter is contributing quite a lot to the error, and so the test is not getting at the measurement of DAC performance very well, at least not unless correction (for the ADC) is used. But I agree that I would like @manisandher to show the results using the new DW DC correction so we can see if all this then shows much difference in the nulling. My guess is it won't and we can let go of the idea that this should be a standard test for evaluating DACs.

 

[TNT]

Was the null showed with a cable as DUT? I don't think so....

//

 

[KSTR]

The quite significant reduction of Null depth from a 100.000% totally inaudible phase shift of, say 5degrees at 20Hz, should not enter the quality assessment of a DAC or ADC because it may significantly skew the result in the wrong direction.

But if we want to include it for whatever reasons, then we must also include the upper range end where the differences from different reconstruction filters have a similar strong effect on Null depth, albeit with a higher chance of audibility. This requires the measurement chain to run at at least 2x the sample rate of the DUT.

The drawback of this approach, however, is that the dull depth is mostly affect by simple, benign/irrelevant and notably mendable errors whereas the true and actually relevant distortion (that never can be fixed) may simply get swamped. I personally favor the alternative approach where the linear differences, notably at the range ends, are de-embedded so that the true distortion (of any type) is exposed. In a simple condensed figure-of-merit -- if that is a goal -- one should only consider linear differences when gross or likely audible, like for example the HF drop effects seen from unfiltered non-oversampling DACs

 

[TNT]

totally inaudible phase shift of, say 5degrees at 20Hz, should not enter the quality assessment

All scientists here can of course do their own assessments of presented data and should not be devoid of knowing all that possible details! (?) Given that there still doesn't seem to be adequate research in what is really detectable and what is not - better not reduce resolution in measurements... or? That would be the scientific approach.

I assume a proper measurement setup e.g. that the equipment used for measurements is significant better performing than any DUT and used in a proper manner.

Skip any one (1) "figure-of-merit" - it will not work. It is better to continue to educate us in the topic in order to be able to analyse a full set of measurements and draw conclusion from that. The correlation to SQ is left TBD.

//