• 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!

How to Fix ESS Hump on SGD1 and LA-QXD1

Ben1987

Active Member
Audio Company
Joined
Oct 30, 2018
Messages
282
Likes
1,256
Location
ChengDu,PRC
Finally on this sunny Sunday, I scheduled a time to share the ESS hump repair solution.

The following method only verifies the two SONCOZ DACs SGD1 and LA-QXD1, both of them using ES9038Q2M DAC chips, and the others are only referrals.

To save you time, first put forward the method and conclusion: For the ES9038Q2M DAC chip, adjusting the combination of resistance and capacitance values of its IV converter and LPF circuit can fix the hump problem,and it belongs to the hardware repair scheme.

The following is the process of my verification:

2019-12-10-13:30
To start my validation and repair, first do nothing, look at the hump, it looks like this:
1210-1351.png


This hump is so ugly that I have overlooked this problem. Please forgive the unprofessional design process.The blue channel is the channel I started to adjust, and the red channel is the contrast channel, which is easy to distinguish.

I used two different OPAs in my design. First, I started changing the (IV and LPF) OPA’s placement. But unfortunately, the hump still stands out:
1210-1400.png


This is then verified using higher precision resistors. The resistance with 1% precision is used in the design. I have prepared for it and purchased a variety of resistance with 1‰ precision in advance.
1210-1415.png


The change of hump is not obvious, so it seems that the resistance precision is not to blame. Other work needs urgent attention at this time, so the verification work will be delayed a little.

2019-12-12 16:20
When I finally had time to continue my verification, I did some calculations, mainly the bandwidth calculations for the IV circuit. Normally, the larger the bandwidth design of the IV circuit, the more noise it may bring as a result, so the design should only meet the bandwidth of the audio signal. The design bandwidth of the IV circuit is around 130kHz, and I get good other parameters in this bandwidth, such as THD+N, linearity, SNR, crosstalk and so on.

I doubt that changing the bandwidth will bring the performance parameters I've already tuned to degrade the overall performance of the product.

SG.jpg


The following is a manuscript of my calculations, in which I calculate various combinations of IV resistance and capacitance values in a bandwidth range of 130khzkz-500khz.

FP.png


But the end result was a slap in the face, and the hump remained the same.

I decided to leave the IV circuit for a while and try to adjust the capacitance of the LPF circuit. The original design was 1nF and adjusted to 2.2nF, 3.3nF, and 4.7nF. It turned out that 2.2nF worked best, as follows:
1212-1823.jpg


The blue channel does go down, but that seems to be the limit of how far it can go down, and that's a big enough distance to be stationary.
Well, I think I'm hungry. I need to have dinner and clear my head.

2019-12-12 20:00
After dinner, I will continue to verify the bandwidth of the IV circuit. Who set the maximum bandwidth of 500kHz for me? Why don't I try it higher? I think there's a ghost here. I have to try it right now. I adjusted the capacitance of the IV circuit from 470pF to 100pF. Here's the result:

1212-2058.png


I couldn't believe my eyes and the hump had shrunk. I feel like I'm close to finding out the truth.

I then adjusted the capacitance on such a basis and found that the effect of the 150pF capacitor was optimized well:

1212-2153.png


However, the value of this capacitor is not some conventional value, and it may be difficult to purchase in mass production. Now that I have found the direction, and look at my messy laboratory, I think I need to sleep:

LAB.jpg


Well, get some sleep and continue the test tomorrow, 2019-12-12-22:30

2019-12-13 10:40
I kept the machine running last night to see how its thermal stability was:

1213-1056.png


The thermal stability is good, it doesn't affect the hump, which is good news.

According to the thinking and direction of yesterday, there is still a resistance of 100R in the IV circuit. There should also be room for adjustment of this parameter. Let's try to adjust to 75R first, as shown in the figure below:

1213-1107.png


It's getting smoother, it's showing up, keep checking, adjust 39R, look at the picture:

1213-1117.png


It's a little smoother and smoother. Happy. Keep going. 51R. Look at the picture:

1213-1156.png


And finally, after a couple of tests. The capacitance value is set at 220pF and the resistance value is set at 51R. I feel that this is the result I want, so I modify the device of the two channels and adjust the output amplitude. The result is as follows:

1213-1332.png


So far, I think the hump problem has been solved, ignoring the channel difference caused by my manual welding. However, this hump is really so sensitive that it is hard to catch like a ghost. The adjustment of resistance and capacitance parameters is very small and hump changes a lot.

To verify my solution, I continued to experiment with the method on LA-QXD1. Again, the blue channel is the adjusted channel:

1212-1411-1.jpg


It looks amazing, it's smooth all at once, so I'm going to change the other channel, the red channel, and I'm going to look at the picture:

1213-1448.png


OK, I think the solution is feasible and verified here. I am very glad to share these experiences. It is a very interesting experience, and it also encourages me to truly achieve the product technology.

Finally, the circuit before and after the hump is pasted for comparison:

hump gone.jpg


SGD1 and LA-QXD1 correct the hump in the same way, so the comparison diagram of SGD1 can be posted here. Of course, someone mentioned the possibility of a software fix before. I only tried to turn off and turn on the register related to the THD+N compensation of ES9038Q2M, and the result was no different. I think this is a pure hardware repair.
(* ̄︶ ̄)
 

Attachments

  • 1213-1411.jpg
    1213-1411.jpg
    47.9 KB · Views: 453
Last edited:

JohnYang1997

Master Contributor
Technical Expert
Audio Company
Joined
Dec 28, 2018
Messages
7,175
Likes
18,292
Location
China
Very Nice!
I'm here to bring some more questions and perhaps thinking on this issue.
First of all, the obvious one is that the IMD is not eliminated, just being reduced. It's being buried in the noise then we can't see it. It's still an improvement. But in spectrum we should still see a sudden increase in imd.
Then, it seems like the story from everyone is a bit different. Maybe you just came across the solution with the specific peripherals and circuit layout and components. Yes, thank you for sharing. But I don't think this is the solution for everyone. It seems like different people solved the issue in different ways and they also seem to have done something that is preliminary to solve the hump before solving it and it's only when they found the last piece the hump is then reduced. And what's funnier is your solution doesn't seem to require some of the things from other solutions. It can well just be the resistor matching or the oscillator or the clock frequency even the pcb layout.

So i would like to see the spectrum of imd at that range from -50db to -30db(-45db and -35db). I think there still will be a sudden increase in thd and imd, where other dac chips only have the imd and thd decrease monotonically with the level. So if evidence proves this, imd jump(since no hump anymore) for existing ess chips can be reduced but cannot be completely solved.

BTW, by the circuit, you had to compromise a bit of noise performance didn't you?
 
Last edited:

WolfX-700

Addicted to Fun and Learning
Forum Donor
Joined
Jul 26, 2019
Messages
537
Likes
4,508
Location
China
Ben is an extremely hardworking electrical and electronics engineer! I like him very much, and his efforts have produced products that are incredible to me: SGD1 and LA-QXD1.

I respect all developers / manufacturers who bring good products, and I respect the engineering implementation behind good products.

But here I even see the spirit of open source that has moved me even more. (I have been in the Internet for more than 25 years. In recent years, I may have grown old. I always feel that some of the early Internet spirits are close to dying ...)

Thank you @Ben1987
 

JohnYang1997

Master Contributor
Technical Expert
Audio Company
Joined
Dec 28, 2018
Messages
7,175
Likes
18,292
Location
China
To push the question a bit further. What exactly caused this issue. The change of capacitors is just a surface. Why does it have to be like this? What happens if you increase the capacitance after the i/v stage. One opamp stage shall isolate the previous stage from the following stage. What about changing the gain of the second stage(i know it will change the output level but just to see).
And changing the resistor and capacitor in the i/v stage is a little bit hard to show the actual issue. Why not remove the resistor and see what effects it gives when you change the capacitor solely? And what about no low pass in the i/v stage? I'm just raising these questions. Maybe if Ben is done with this, I can jump on this and test all of these just to dig a bit further to find out what exactly caused this.

Again, thanks ben for the contribution and transparency.
 

JohnYang1997

Master Contributor
Technical Expert
Audio Company
Joined
Dec 28, 2018
Messages
7,175
Likes
18,292
Location
China
No performance sacrifices have been made so far in the tests.
The noise rised a little bit right? It's common to have gain in the i/v stage and reduce back down to get better noise performance. Like i saw in the original circuit.
1212-1411-1.jpg
 
OP
Ben1987

Ben1987

Active Member
Audio Company
Joined
Oct 30, 2018
Messages
282
Likes
1,256
Location
ChengDu,PRC
The noise rised a little bit right? It's common to have gain in the i/v stage and reduce back down to get better noise performance
I think it was a lot of hand welding.Multiple heating and cooling, etc
 

Attachments

  • SMPTE-Ratio.jpg
    SMPTE-Ratio.jpg
    107.4 KB · Views: 536

SpyB

Active Member
Joined
Sep 3, 2019
Messages
142
Likes
103
Thank you Ben for your excellent work,and transparent contributions to ASR.

And of course for 2 truly excellently engineered DAC's

I will have more to say after my purchase
 

VintageFlanker

Major Contributor
Forum Donor
Joined
Sep 20, 2018
Messages
4,940
Likes
19,692
Location
Paris
Finally on this sunny Sunday, I scheduled a time to share the ESS hump repair solution.

The following method only verifies the two SONCOZ DACs SGD1 and LA-QXD1, both of them using ES9038Q2M DAC chips, and the others are only referrals.

To save you time, first put forward the method and conclusion: For the ES9038Q2M DAC chip, adjusting the combination of resistance and capacitance values of its IV converter and LPF circuit can fix the hump problem,and it belongs to the hardware repair scheme.

The following is the process of my verification:

2019-12-10-13:30
To start my validation and repair, first do nothing, look at the hump, it looks like this:
View attachment 43299

This hump is so ugly that I have overlooked this problem. Please forgive the unprofessional design process.The blue channel is the channel I started to adjust, and the red channel is the contrast channel, which is easy to distinguish.

I used two different OPAs in my design. First, I started changing the (IV and LPF) OPA’s placement. But unfortunately, the hump still stands out:
View attachment 43300

This is then verified using higher precision resistors. The resistance with 1% precision is used in the design. I have prepared for it and purchased a variety of resistance with 1‰ precision in advance.
View attachment 43301

The change of hump is not obvious, so it seems that the resistance precision is not to blame. Other work needs urgent attention at this time, so the verification work will be delayed a little.

2019-12-12 16:20
When I finally had time to continue my verification, I did some calculations, mainly the bandwidth calculations for the IV circuit. Normally, the larger the bandwidth design of the IV circuit, the more noise it may bring as a result, so the design should only meet the bandwidth of the audio signal. The design bandwidth of the IV circuit is around 130kHz, and I get good other parameters in this bandwidth, such as THD+N, linearity, SNR, crosstalk and so on.

I doubt that changing the bandwidth will bring the performance parameters I've already tuned to degrade the overall performance of the product.

View attachment 43303

The following is a manuscript of my calculations, in which I calculate various combinations of IV resistance and capacitance values in a bandwidth range of 130khzkz-500khz.

View attachment 43304

But the end result was a slap in the face, and the hump remained the same.

I decided to leave the IV circuit for a while and try to adjust the capacitance of the LPF circuit. The original design was 1nF and adjusted to 2.2nF, 3.3nF, and 4.7nF. It turned out that 2.2nF worked best, as follows:
View attachment 43305

The blue channel does go down, but that seems to be the limit of how far it can go down, and that's a big enough distance to be stationary.
Well, I think I'm hungry. I need to have dinner and clear my head.

2019-12-12 20:00
After dinner, I will continue to verify the bandwidth of the IV circuit. Who set the maximum bandwidth of 500kHz for me? Why don't I try it higher? I think there's a ghost here. I have to try it right now. I adjusted the capacitance of the IV circuit from 470pF to 100pF. Here's the result:

View attachment 43306

I couldn't believe my eyes and the hump had shrunk. I feel like I'm close to finding out the truth.

I then adjusted the capacitance on such a basis and found that the effect of the 150pF capacitor was optimized well:

View attachment 43307

However, the value of this capacitor is not some conventional value, and it may be difficult to purchase in mass production. Now that I have found the direction, and look at my messy laboratory, I think I need to sleep:

View attachment 43308

Well, get some sleep and continue the test tomorrow, 2019-12-12-22:30

2019-12-13 10:40
I kept the machine running last night to see how its thermal stability was:

View attachment 43309

The thermal stability is good, it doesn't affect the hump, which is good news.

According to the thinking and direction of yesterday, there is still a resistance of 100R in the IV circuit. There should also be room for adjustment of this parameter. Let's try to adjust to 75R first, as shown in the figure below:

View attachment 43310

It's getting smoother, it's showing up, keep checking, adjust 39R, look at the picture:

View attachment 43311

It's a little smoother and smoother. Happy. Keep going. 51R. Look at the picture:

View attachment 43312

And finally, after a couple of tests. The capacitance value is set at 220pF and the resistance value is set at 51R. I feel that this is the result I want, so I modify the device of the two channels and adjust the output amplitude. The result is as follows:

View attachment 43313

So far, I think the hump problem has been solved, ignoring the channel difference caused by my manual welding. However, this hump is really so sensitive that it is hard to catch like a ghost. The adjustment of resistance and capacitance parameters is very small and hump changes a lot.

To verify my solution, I continued to experiment with the method on LA-QXD1. Again, the blue channel is the adjusted channel:

View attachment 43317

It looks amazing, it's smooth all at once, so I'm going to change the other channel, the red channel, and I'm going to look at the picture:

View attachment 43315

OK, I think the solution is feasible and verified here. I am very glad to share these experiences. It is a very interesting experience, and it also encourages me to truly achieve the product technology.

Finally, the circuit before and after the hump is pasted for comparison:

View attachment 43316

SGD1 and LA-QXD1 correct the hump in the same way, so the comparison diagram of SGD1 can be posted here. Of course, someone mentioned the possibility of a software fix before. I only tried to turn off and turn on the register related to the THN+N compensation of ES9038Q2M, and the result was no different. I think this is a pure hardware repair.
(* ̄︶ ̄)
Outstanding.o_O

I'm shot. Why @Ben1987 doesn't have a Technical Expert badge already ?!
 

badboygolf16v

Active Member
Forum Donor
Joined
Jul 24, 2018
Messages
285
Likes
356
Interesting. So in theory, this kind of fix could be retrofitted to a KTB?
 

JohnYang1997

Master Contributor
Technical Expert
Audio Company
Joined
Dec 28, 2018
Messages
7,175
Likes
18,292
Location
China
Interesting. So in theory, this kind of fix could be retrofitted to a KTB?
That's a maybe. As the output impedance of single chip is different from two chip in parallel, the component values should be different. The 390ohm in the i/v stage will reduce the gain. Sure you can increase the gain back at lpf stage but that may increase noise. Someone has to redo the process of testing values to be able to apply on KTB. And since it's not hard to do, I think some people will do it in short time. Even with su-8 etc.
 

daftcombo

Major Contributor
Forum Donor
Joined
Feb 5, 2019
Messages
3,687
Likes
4,068
We can now do blind tests of the very same DACs with and without the hump.
Some people try that before doing some funny but probably useless DIY optimization.
 

pos

Addicted to Fun and Learning
Forum Donor
Joined
Feb 13, 2018
Messages
572
Likes
718
Could this ghost manifest in another form in different measurements?
I mean, there is definitely something strange going on in this chip.
 
Top Bottom