E1DA 9038D performance according to df-metric

[solderdude]

Inaudible with music.
You could plot the spectrum of the the excerpt (original and recording).
What did PK metric show ? (as it is weighted)

 

[pjug]

PK metric was -60dBr. I have added these to my previous post. The blue curve is the spectrum of the original.

 

[pjug]

A section of 20Hz-20KHz white noise does not show much different.

 

[pkane]

Picking a 3 minute mixed music portion for analysis, 8Hz HP filter applied:

RMS of null: -49.2dB
df Metric: -36.9 median
PK Metric: -59.8dBr

 

[pjug]

Picking a 3 minute mixed music portion for analysis, 8Hz HP filter applied:

RMS of null: -49.2dB
df Metric: -36.9 median
PK Metric: -59.8dBr


View attachment 313616

View attachment 313617

View attachment 313618
View attachment 313619

View attachment 313620

Any guesses on why the DF metric is not as good as with the other headphone amps measured by the OP? Is the ripple and low frequency rolloff in the Spectra of Delta enough to explain it? And what might be causing the low frequency difference?

 

[pkane]

Here's the other recording using Shanling M0 Pro DAC that Serge posted. The same 3 minute portion as in my post above:

RMS of null: -71.5dB / -73.1dBA
df metric: -59.7dB
PK Metric: -76.8dBr

 

[solderdude]

Looks like some sharp subsonic filtering is messing things up ?

 

[pkane]

Looks like some sharp subsonic filtering is messing things up ?

Yes, and this happens with other DACs/converters. For example, RME ADI-2 Pro FS has a much better DF and RMS of null result when the DC filter is turned off in the unit. Which is why DF metric might be good when an overall threshold is known, but it's not a measure of audibility. It's a threshold of audibility (and that threshold still remains to be derived experimentally). It doesn't allow for a direct comparison between devices to determine which produces an error that is more or less audible.

 

[pjug]

Yes, and this happens with other DACs/converters. For example, RME ADI-2 Pro FS has a much better DF and RMS of null result when the DC filter is turned off in the unit. Which is why DF metric might be good when an overall threshold is known, but it's not a measure of audibility. It's a threshold of audibility (and that threshold still remains to be derived experimentally). It doesn't allow for a direct comparison between devices to determine which produces an error that is more or less audible.

Maybe not audible, but given that the PK metric is more than 16dB worse than the Shanling device, wouldn't it be good to have something to show this in the ASR review? Surely this is more concerning than something like a 110 vs 120 SINAD difference, or whatever these things have nowadays. Maybe just always including a frequency response curve with magnitude and phase would be enough.

 

[MaxwellsEq]

Yes, and this happens with other DACs/converters. For example, RME ADI-2 Pro FS has a much better DF and RMS of null result when the DC filter is turned off in the unit. Which is why DF metric might be good when an overall threshold is known, but it's not a measure of audibility. It's a threshold of audibility (and that threshold still remains to be derived experimentally). It doesn't allow for a direct comparison between devices to determine which produces an error that is more or less audible.

I find this fascinating. It would seem we have measures that can demonstrate a difference between two devices based on a filter action which I have always assumed is inaudible (i.e. what happens below 10Hz) and where is generally accepted there is no musical content.

There was a thread a while ago where somebody claimed their RME ADI sounded poor with the DC filter in!

But the real question is whether the RMS null and DF scores have merit if their score is heavily impacted by anomalies near DC...

EDIT: link to thread about ADI DC filter https://audiosciencereview.com/forum/index.php?threads/fix-your-rme-adi-2-fs-dac-sound.36741/

 

[Sokel]

I find this fascinating. It would seem we have measures that can demonstrate a difference between two devices based on a filter action which I have always assumed is inaudible (i.e. what happens below 10Hz) and where is generally accepted there is no musical content.

There was a thread a while ago where somebody claimed their RME ADI sounded poor with the DC filter in!

But the real question is whether the RMS null and DF scores have merit if their score is heavily impacted by anomalies near DC...

If it contains the famous 9038QM2 better use it with DSD .
Joke (not entirely...) aside this one is at least has it's famous hump reduced as it seems ,IMD is DSD levels low at around -30db level (about -80db) so what's left and most importantly:
Audible or not?

 

[pjug]

I think some of the problem has to do with the known issue with large signals. Amir noted it in the ASR review, and @Archimago also discusses it in his review. There is a fix for that, as Archimago writes in the tweak section in http://archimago.blogspot.com/2022/12/review-e1da-9038d6k-dongle-dac_11.html

There's a volume controller up at the top of the main screen along with adjustable "Max" and "Min" limits. For example, in Part I when we were looking at maximum resolution, we know that there's a bit of saturation above -0.5dBFS, so we can click on the "0.0dB" and pull it down to "-0.5dB" as a limit if we want to use this as a pure DAC and want to avoid the risk of that extra distortion up there. By the way, pulling this down to "-3.0dB" would limit the max output to 2Vrms which would be a typical "line level" voltage like many DACs.

When I compare a lower level section of music from @Serge Smirnoff 's files, I still get a low DF number, but the PK factor improves to -70 something.

edit: Archimago also has a recording of the 9038D and his own test track, available for download and comparison.

 

[pjug]

Comparing from Archimago's recording using a lower level section (peaks about -6dB), I get a PK factor of -78dB. DF factor is still low, though, about-35dB.

 

[AnalogSteph]

Fascinating. That's like 0.03 dB (+/-0.015 dB) of passband ripple clear as day. Which in turn would be consistent with using the filter named "Brick Wall" in the DAC datasheet, which reaches stopband at 0.5fs, matching Amir's findings.
(Mind you, I'd rather have the regular old linear phase fast rolloff filter, which drags things out until 0.55fs - still not at all bad - but sports only +/-0.002 dB of passband ripple instead. In practice there is so little going on between 0.45fs and 0.5fs, or 19.845 kHz to 22.05 kHz at 44.1 kHz, that what little aliased output is generated between 0.5fs and 0.55fs is not going to bother any amplifier, let alone be audible in any way.)

 

[csglinux]

The teeny pass-band ripple and very small low-frequency roll-off shouldn't be contributing much to any error metric. @pkane's posts seem to confirm the OP's results that the main error from the 9038D is in the phase at lower frequencies. That isn't going to show up on amplitude vs frequency FFTs of sine tones (or even multiple sine tones).

@pkane also mentioned some RME kit has a similar problem, and I've seen low-f phase errors with my own RME equipment. Some questions for RME experts: I'm not sure why a DC filter being on or off should affect an RMS value - is RME's filter somehow flawed? Does anybody how to tweak or disable RME's DC filters? Can this be done in TotalMix?

 

[pkane]

Does anybody how to tweak or disable RME's DC filters? Can this be done in TotalMix?

it’s a new setting in the recent firmware releases. Probably a few years ago. Can be changed from the unit itself. Or with the latest remote app (awesome!)

 

[Serge Smirnoff]

Here are the results of testing #9038 and M0pro with the multitone signal:

DFwf = -34.12 dB	DFmg = -52.27 dB	DFph = -45.10 dB
DFwf = -75.68 dB	DFmg = -75.57 dB	DFph = -63.91 dB

... and with the classical piece from the GearSpace forum:

DFwf = -36.21 dB	DFmg = -55.70 dB	DFph = -30.29 dB
DFwf = -80.33 dB	DFmg = -83.89 dB	DFph = -48.86 dB

You can listen the null signal from the both devices amplified to the same LUFS (the middle part of the classical piece above):

null-#9038.zip
null-M0pro.zip

The degradation of the low freqs in #9038 is due to the phase inaccuracy, not the magnitude one (Mg diffrograms do not show this). As the low freqs are not in phase with the reference signal they are exaggerated in the null signal. DC filter could be a reason.

The time inconsistency of #9038 is visible in both Mg and Ph diffrograms. It is audible as a “codec-like” degradation. It's hard to say what is the reason - the clock instability and/or instability of the power supply of the circuits.

 

[solderdude]

Had a listen to the null and the 9038 is by far the better performing DAC.
The numbers seem skewed by the LF response not so much the sound quality. The 9038 null is 'muffled' the M0 null is far less pleasant.
The null is amplified it seems so might well be below audible thresholds anyway.
The problem is the null is not attenuated with a fixed amount of dB so can't really say which one would be better.

 

[Veri]

Had a listen to the null and the 9038 is by far the better performing DAC.

thanks for doing the test.

 

[Serge Smirnoff]

The problem is the null is not attenuated with a fixed amount of dB so can't really say which one would be better.

Yes, I attenuated the null to the same perceived loudness in order to evaluate the type of the degradation. The M0's null is attenuated much more (you can listen the noise).