• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. 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!

ESS THD ‘Hump’ Investigation

peterq

Active Member
Joined
May 9, 2018
Messages
106
Likes
66
Benchmark is using LME49860 in their dac3 output stage. You could try both.
DAC3DX%2Binside%2B.jpg
 

peterq

Active Member
Joined
May 9, 2018
Messages
106
Likes
66
But according to khadas people, it seems the hump is not easy to solve, I also created a thread on ktb forum to track this issue,
https://forum.khadas.com/t/ess-hump-solution/3746/18

Hi All:
Our guys go to ESS Lab here in Shenzhen, and the hump issue do exist in there DAC chips, they also still don’t have a good solution at the moment.
They will update this issue to ESS Lab in USA, and yes, we will try to improve the issue on current KTB:
 

peterq

Active Member
Joined
May 9, 2018
Messages
106
Likes
66
Also KTB uses DFN packaging opamp which I think it's very difficult to change...
 

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,370
Likes
234,433
Location
Seattle Area
But according to khadas people, it seems the hump is not easy to solve, I also created a thread on ktb forum to track this issue,
https://forum.khadas.com/t/ess-hump-solution/3746/18

Hi All:
Our guys go to ESS Lab here in Shenzhen, and the hump issue do exist in there DAC chips, they also still don’t have a good solution at the moment.
They will update this issue to ESS Lab in USA, and yes, we will try to improve the issue on current KTB:
Great to see this.
 

gvl

Major Contributor
Joined
Mar 16, 2018
Messages
3,427
Likes
3,981
Location
SoCal
When I see word "audiophile" on a semiconductor spec sheet I start worrying.
 

Herbert

Addicted to Fun and Learning
Joined
Nov 26, 2018
Messages
527
Likes
435
Also KTB uses DFN packaging opamp which I think it's very difficult to change...
Didn´t notice this, as it looks like very short leads/pins on the board. But I guess soldering will be the same:
Heating the solder until it becomes fluid and expect it makes a bond to the chip. But removing the
original semiteh opams by an average DIYers means will probably destroy them...
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
Didn´t notice this, as it looks like very short leads/pins on the board. But I guess soldering will be the same:
Heating the solder until it becomes fluid and expect it makes a bond to the chip. But removing the
original semiteh opams by an average DIYers means will probably destroy them...
The suggested OPA1642 comes in 8-pin SOIC and VSSOP packages, and AD8397 is 8-pin SOIC only. These are not compatible with the 10-pin DFN footprint on the PCB. You'll have to find something else for this project.

DFN/QFN packages often have an exposed pad in the centre. Working with these requires a hot air station or reflow oven.
 

Herbert

Addicted to Fun and Learning
Joined
Nov 26, 2018
Messages
527
Likes
435
The suggested OPA1642 comes in 8-pin SOIC and VSSOP packages, and AD8397 is 8-pin SOIC only. These are not compatible with the 10-pin DFN footprint on the PCB. You'll have to find something else for this project.

DFN/QFN packages often have an exposed pad in the centre. Working with these requires a hot air station or reflow oven.

Thank you very much. So it will be only changing the resistors and maybe the Xtal - it was mentioned somewhere that lower jitter might help
a bit against the hump but I do not recall where. But I assume data is good anyway, no reason to replace it...?
https://www.sitime.com/products/lvcmos-oscillators/sit8208

I am not sure about the following: Did we come to the conclusion that all humpless Dacs had dedicated power supplies for the opamps?
The SRT6863 are suitable for a power supply from +/-2.5V up to +/-17V

http://www.semiteh.com/admin/upload/semiteh1-3812595.pdf
 
Last edited:

bbt8888

Member
Joined
Feb 8, 2019
Messages
7
Likes
13
... So it will be only changing the resistors ...
The calculated resistor values is based on Rs(Rdac) = 774ohm as listed in the spec. Since its tolerance is 11%, correct value for your board might be different. To verify it, you can measure voltages at RT6862D outputs --- the easiest access point will be from R32, R34, R45, R48 --- all should be around 46.77mV while not playing music. Basically the application note suggests that after correction all four voltages should be around 0V.
 

Herbert

Addicted to Fun and Learning
Joined
Nov 26, 2018
Messages
527
Likes
435
Thanks, I will do this, but it will take some time.
But I got another idea - but my knowledge is very limited
and I might be absolutely wrong here and talk rubbish.
In times when Burr-Brown PCM63 was king, I remember Parasound built DACS
with unbalanced and balanced output, like the DAC 1500.
For achieving balanced out, four PCM63 were used and
the inverted signal for the balanced output was achieved digitally,
I think it was IC902 / 74HC04AP in the attached datasheet inverting data
to Data-pin 21 of the PCM63.
So there was two identical (stereo) output stages on two PCB.
The "single ended" user only using RCA listened to two PCM63.
And the paralleled DACs each channel praised by reviewers were only true as long one used balanced XLR.
As the Khads Tone board is relatively cheap, couldn`this be done as well - and two Khadas boards
running in balanced mode would kill the hump as its noise would be inverted and cancelled?
All the best,
Herbert
 

Attachments

  • Parasound-DAC-1500.pdf
    2 MB · Views: 213
Last edited:

gvl

Major Contributor
Joined
Mar 16, 2018
Messages
3,427
Likes
3,981
Location
SoCal
You'd need to sync the clocks of 2 boards. Soldering and scope required.
 

gvl

Major Contributor
Joined
Mar 16, 2018
Messages
3,427
Likes
3,981
Location
SoCal
You probably could invert digitally before the one of the DACs with some sort of DSP software. The bigger question is if IMD is common mode or not.
 

peterq

Active Member
Joined
May 9, 2018
Messages
106
Likes
66
If my calculation is correct, the needed virtual ground voltage is 1.65*(806/(806+774))=0.84171V, and the current value is 3.3*3.3/(10+3.3)=0.8188V. To make it right, you can either change R40 from 3.3k to 3.424k or change R38 from 10k to 9.638k. The closest 1% resistor to 3.424k is 3.48k & for 9.638k is 9.53k. The other option is paralleling a 267k 1% resistor to 10k to make it 9.639k. Just my 2c.

@HououinKyoma
How precisely should the virtual ground match the recommended value? Within 1%? The tone board already have it within 3% and still have THD hump. And Rs (Rdac) of ES9038Q2M is specified to have 11% tolerance, so even if correct resistors value are installed individual board still can be far off your recommended value unless calibration on individual board is performed. Seriously I think this is a chip design issue from ESS's side to have such a big tolerance and also have critical design parameters based on this very loose value.

I didn't understand your equation(3.3*3.3/(10+3.3)), from the posting,
Rd = [1.65*Rf/(Rdac + Rf)]*10000 / [3.3-1.65*Rf/(Rdac + Rf)]
For KTB , Rf = 806, Rdac = 774, Then Rd = 3424 --- result is same :)

Also, I see C75 is less than the post, 470pf < 510pf

Actually, I don't know where 1.65 comes from, just my 2c.
 
Last edited:

bbt8888

Member
Joined
Feb 8, 2019
Messages
7
Likes
13
@peterq
(1) Virtual ground of TB setting (+In of RT6862D) = AVCC_R * R40/(R38+R40) = 3.3V * 3.3k/(10k +3.3k)
(2) Voltage mode output offset (Vocm) --- Bipolar zero out --- = AVCC/2 = 3.3V/2 = 1.65V, please refer to pp. 52 of ES9038Q2M data sheet.
Or from appnote "ES9018K2M Output Stage" V3=V4=0, then Vs(BT2) = AVCC/2, they both mean the same thing.
Anyway when dac output is bipolar zero (1.65V internally) & Virtual ground is set according to the formula in appnote, then it is easy to deduct the I/V output will be exact 0V.
(3) The 470pF will set the cutoff freq. @ ~420kHz, don't know why it is set so high. For comparison in the appnote, it is set @ ~116kHz.
 

peterq

Active Member
Joined
May 9, 2018
Messages
106
Likes
66
Thank you for the explanation.
So the first mod should be,
1. Change R53 and R40 to about 3.4k
2. Change C84 and C76 to 510pf

Those are easier than change opamp.
 

bbt8888

Member
Joined
Feb 8, 2019
Messages
7
Likes
13
@peterq
1. There is a long check list from @HououinKyoma 's post: layout, crystal, choice of opamps, Rf, etc. Virtual ground setting is just one of them. Will changing R53 & R40 solve the issue? Honestly I don't know. If it is this simple, Khadas should have already announced the solution. My gut feeling is there might need board level calibration: different R53, R40 for individual boards or even worse, different Rf values to match individual channel source impedance.
2. cutoff freq = 1/(2*pi*R54*C84) = 1/(2*pi*806*470p), it not only depends on C84 (or C82, C75, C76).
 

peterq

Active Member
Joined
May 9, 2018
Messages
106
Likes
66
Thank you, if we would like to set cutoff freq to about 110k, we should change C84 (or C82, C75, C76) to about 1700 ~ 1800pf ?
 
Top Bottom