E1DA 9038D performance according to df-metric

[solderdude]

How does the null sound ? That could give some clues.

 

[Serge Smirnoff]

That's quite a difference in number. Seeing that PK metric is weighted it makes more sense as to how audible it is.
I mean... I have not seen any negative comments about the E1DA when it concerns sound quality.
-40dB is bordering on audible where -80 is inaudible.

Maybe the one Serge had has an issue with the clock or other issue.
For the sake of science it should be repeated with another device. Maybe even some ABX testing ?
Otherwise it is just a meaningless number skewed by something that may not even be audible.

We have two #9038 devices, both show the same results.

I also need to check DeltaWave results for DF. Could be some difference there as well.

 

[Serge Smirnoff]

How does the null sound ? That could give some clues.

Good idea )), will do

 

[pkane]

The null-difference has one disadvantage - it depends on the level of a reference signal. Df is a ratio of the null RMS to the level of reference signal. So, Df is a relative parameter and has exactly the same physical meaning as simple RMS of null difference.

Right. Not a major disadvantage when measured with a full-scale orchestral piece But that's why PK-Metric has the option of dBFS or dBr units, where dBr is relative to the signal level.

 

[NTK]

The “efficient resampling method” was developed. The time warping algo does it with any predefined accuracy. For example White Noise can be shrinked/stretched with df=-100dB accuracy in the current version. So, the origin of the time inconsistency in #9038 is not the time warping algo.

Has the details of your time warping algorithm been disclosed? It is kind of important as it has a substantial influence on the outcome of the Df numbers.

 

[pjug]

Do headphone amps typically use a HP filter to remove DC? With white noise even a 1Hz HP filter gives -51dB DF metric. 5Hz HP gives -41dB.

@IVX does the 9038D have a HP filter to remove DC?

 

[Serge Smirnoff]

Has the details of your time warping algorithm been disclosed? It is kind of important as it has a substantial influence on the outcome of the Df number

I didn't write a paper about the algo but the Matlab code is open (http://soundexpert.org/articles/-/blogs/visualization-of-distortion#part3) and you can always check that algo by feeding exactly the same signals as input and output into diffrogram utility. In order to have 0.1dB accuracy of Df computation the ADC+algo must provide 18dB better accuracy than a DUT.

 

[Serge Smirnoff]

We will provide all used/required signals for verification of the results.

The test vectors used for building df-slides of DAPs:

t-signals + m-signal(Diversity) @44.1/16 - http://soundexpert.org/vault/media/combo-vector-7.2(44-16-S)-play.flac
t-signals @96/24 - http://soundexpert.org/vault/media/t-vector-7.2(96-24-S)-play.flac

 

[pjug]

The test vectors used for building df-slides of DAPs:

t-signals + m-signal(Diversity) @44.1/16 - http://soundexpert.org/vault/media/combo-vector-7.2(44-16-S)-play.flac
t-signals @96/24 - http://soundexpert.org/vault/media/t-vector-7.2(96-24-S)-play.flac

So these are the input signals only? Can you also give us your ADC recordings? I doubt that many of us have the 9038D.

 

[Serge Smirnoff]

So these are the input signals only? Can you also give us your ADC recordings? I doubt that many of us have the 9038D.

#9038D, brick-wall filter into 32 Ohm load:

9038D_Brickwall_32Ohm.flac

Shared with Dropbox

www.dropbox.com

#9038D, brick-wall filter, no load:

9038D_Brickwall_InfntyOhm.flac

Shared with Dropbox

www.dropbox.com

Shanling M0 Pro, 32 Ohm load:

Shanling_M0_Pro_LinearPhaseFastRollOff_BAL.flac

Shared with Dropbox

www.dropbox.com

 

[IVX]

Do headphone amps typically use a HP filter to remove DC? With white noise even a 1Hz HP filter gives -51dB DF metric. 5Hz HP gives -41dB.

@IVX does the 9038D have a HP filter to remove DC?

The sample I keep has <1mVdc, -40db is 1/100, so if that was DC it should be 27mV. As I said, I have only one idea of what may affect the result it is the residual 1.3MHz noise from the internal 2 SMPS. Actually, any audio-related test is better to limit within an audio BW 20-20K or 10-30K. IMO, the new method has to prove its precision isn't worse than the previous one first, and next offer an additional test set which can not see the previous method.

 

[solderdude]

Good idea )), will do

Can you post the null (as well as the null amplified +40dB)
If that has the same content as the source file than it won't be audible.
If there is some beating with the 1.2MHz clock interfering with some clock in the ADC leaking in somewhere it should not have music content and you might see it in the 1kHz spectrum (FFT).

Maybe @pkane can run the recordings and original through his program. It seems to work differently here and there in signal processing.

 

[MaxwellsEq]

Do headphone amps typically use a HP filter to remove DC? With white noise even a 1Hz HP filter gives -51dB DF metric. 5Hz HP gives -41dB.

@IVX does the 9038D have a HP filter to remove DC?

That's interesting. Such low frequency changes should be inaudible. That implies that DF is giving a score based on something that's inaudible.

 

[Sokel]

That's interesting. Such low frequency changes should be inaudible. That implies that DF is giving a score based on something that's inaudible.

Isn't it's purpose the exact opposite?
Or audibility is of no concern here?

 

[MaxwellsEq]

Isn't it's purpose the exact opposite?
Or audibility is of no concern here?

I'm curious whether the score is a composite of things such as DC transfer, clock-to-clock variations that don't appear in a J-test. These things should not have an impact on sound. But there's a claim on the table about correlation with subjective listening tests. Does this mean that 5Hz HPFs and clock breakthrough, which can be measured, are audible?

 

[Rottmannash]

Be nice for someone at ESS to chime in about now.

 

[solderdude]

Be nice for someone at ESS to chime in about now.

This does not seem to be an ESS or chip issue.
What this needs is some more scrutiny and an ABX test (recording vs original) or some listening to the null to get more light on this.

 

[pjug]

#9038D, brick-wall filter into 32 Ohm load:

9038D_Brickwall_32Ohm.flac

Shared with Dropbox

www.dropbox.com

#9038D, brick-wall filter, no load:

9038D_Brickwall_InfntyOhm.flac

Shared with Dropbox

www.dropbox.com

Shanling M0 Pro, 32 Ohm load:

Shanling_M0_Pro_LinearPhaseFastRollOff_BAL.flac

Shared with Dropbox

www.dropbox.com

Looking at a 2 min section of the music portion, I also get a poorer than expected match. Does this delta of the spectra give anyone an idea of what might be going on?

 

[solderdude]

16Hz -0.2dB ?
I have no idea how low the BW goes for white noise (and if it is long enough) but if there is signal there below 16Hz the result could be skewed with white noise.
With nulling music -0.2dB at 16Hz is inaudible but might be measurable.
The small variations > 2kHz (due to the filter) are in the 0.03dB range (-40dB difference) so could also play a role.

 

[pjug]

16Hz -0.2dB ?
I have no idea how low the BW goes for white noise (and if it is long enough) but if there is signal there below 16Hz the result could be skewed with white noise.
With nulling music -0.2dB at 16Hz is inaudible but might be measurable.
The small variations > 2kHz (due to the filter) are in the 0.03dB range (-40dB difference) so could also play a role.

This was with a section of the music. When I get a chance I will look at it with a section of noise.