ESS THD ‘Hump’ Investigation

[JohnYang1997]

what do you mean "try"? Do you want I check FFT with your wav files?

Yes. FFT.

 

[IVX]

in "25MHz" mode:

 

[JohnYang1997]

in "25MHz" mode:
View attachment 25732View attachment 25733

What the hack. I need to check if my files are clipping.

 

[JohnYang1997]

in "25MHz" mode:
View attachment 25732View attachment 25733

https://drive.google.com/file/d/1-c1FpPFKUH3OXwy_GVGJyOQkBTMGpYB4/view?usp=sharing try this one
https://drive.google.com/file/d/1L7sxkrRNjOX-JeAlAHsrh7NWNd5mkGTg/view?usp=sharing and this

 

[JohnYang1997]

The two tone in the first link was fine. The multitone has some issue but the craps should be under -88db.

 

[IVX]

it was my fall, the player had extra gain, sorry. You know what, I using iso-trafo and ADC for that FFT, I don't think if the resolution will be enough(at 20Hz trafo for sure is bad, @1KHz THD .0002%) to see 9038 artifacts yet. My AP is blind SYS2702 and has no FFT at all.

 

[JohnYang1997]

it was my fall, the player had extra gain, sorry. You know what, I using iso-trafo and ADC for that FFT, I don't think if the resolution will be enough(at 20Hz trafo for sure is bad, @1KHz THD .0002%) to see 9038 artifacts yet. My AP is blind SYS2702 and has no FFT at all.

Supposedly there should be modulation noises/ artifacts in the spectrum, which are not shown with other dac chips, like cs4398. Something like this.

 

[IVX]

yes, I see that and without 30Hz these nails disappear too. I see these artifacts in both mode i.e. with/without hump.

 

[JohnYang1997]

yes, I see that and without 30Hz these nails disappear too. I see these artifacts in both mode i.e. with/without hump.
View attachment 25736

Does one mode has higher modulations? I mean, the hump doesn't seem to completely gone with the other mode just less prominent/ lower than noise. I think that's crafted in the ess architecture.

 

[IVX]

I didn't notice any difference yet, I want to try the test AK4490 as well.

 

[JohnYang1997]

I didn't notice any difference yet, I want to try the test AK4490 as well.

Sure. In my measurements they don't present on ak4493 or cs4398.

 

[IVX]

Hah, my 4490 has it )) Need to chek if iso-trafo affect that or ADC(CS5381 with OPA1612 frontend)

 

[JohnYang1997]

Hah, my 4490 has it )) Need to chek if iso-trafo affect that or ADC(CS5381 with OPA1612 frontend)

Probably try -6db output gain.

Spoiler: 4493

Spoiler: 4398

 

[IVX]

BTW, I have 4398 as well.

 

[maxxevv]

Hi there, I do not follow that thread but if anyone interested in the root cause of that "hump" phenomenon I have to show something. The attached plot is power sweep vs THD+N my tiny ES9038Q2M USB->DAC->HPA dongle on real planar headphones loaded Fs=44.1kHz 24b. 100MHz named curve correspond reg#0 value 2’b00: MCLK = XI (default), 25MHz at 2’b11: MCLK = XI / 8. As you can see the blue 100MHz curve has no hump at all, and you can ask me why some DACs uses lower MLCK value? Check this out(note: FFT after 40db notch), harmonics distortions
distribution is quite different for 100MHz vs 25MHz, however THD+N very close for both modes. 25MHz mode has nearly perfect harmonics set - 2, 3, 5.. and nothing more, then 100MHz has lot's of small nails at 15, 17 and 19khz. I made some buttonless "easter eggs" hidden interface for my #9038S USB->DAC->HPA dongle to let user switch between 3 modes:
1) Normal-mode 100MHz
2) Performance-mode 25MHz
3) SE-mode 100MHz+.25% of 2nd harmonicView attachment 25726View attachment 25729View attachment 25730

So your measurements suggests that a 100 mhz clock would smooth out the 'hump'. But a 25Mhz clock would give a better THD+N measurement ?

Would it be possible to give a run using odd multiples of 25Mhz ? Like 3x or 5x or 7X or even 11x 25Mhz and see if you get something in between or better? As prime multiples may help eliminate the harmonics in the THD+D while the higher frequency seems to smooth out the 'hump'.

Its just a gut feel from an engineering point of view, no real basis in electronics but its something we do when it comes to vibration elimination in mechanics when we do not have enough information to completely model the frequencies within a system outside of similar measurements, and sometimes it does work.

 

[IVX]

no, it would be not easy to replace oscillator..

 

[Herbert]

For the time being I did not try any modifications. But to my knowledge a new Tone Board was "promised" for Q2/2019 that might adress the humo. May is almost over now. Any news on that?

 

[IVX]

Herbert
under the "humo" do you mean hump? Actually, I believe that the first post of that thread wasn't correct about the I2C dump. No idea why but looks like toneboard sending that dump for nothing to the wrong address because transactions have no acknowledges. Maybe by mistake (forgotten to comment old code), maybe to fool up chinese copy/pasters(hard to believe because it is quite simple design and cheap chinese boards on ES9038Q2M widely available) but I2C transactions without acknowledges weren't received by ES9038Q2M, hence, there are should be other short transactions. So, if anyone interested to get actual I2C dump needs to try again for a longer scanning period. Or just replace there XU208-128-TQ64 (about $10) for a brand new and make your own firmware for that, all source code available here https://www.xcore.com/viewtopic.php?t=6375 need some minor modifications.

 

[Herbert]

Yes, humo was simply misspelled. But I I was asking if a re-designed board is to hit the market after Q2 of 2019

 

[IVX]

And you expected they will fix the hump which actually ESS issue? If toneboard has a noticeable hump at Fs 44.1-48, then they intentionally use max clk_gear divider in the reg#0 to get less high-order harmonics. The hump is the compromising.