Measurements: "ESS Hump" revisited (Khadas Tone Board V1.3)

[KSTR]

I do have to wonder whether it may be advantageous to reduce high-frequency output loading... sure, nobody likes inductors much. (That would obviously increase gain up there, too, which may be counterproductive in the light of potential slew rate issues. So exactly the opposite may be needed, output loading be damned. Aren't they already using R || R-C in feedback anyway?)
So you basically need to provide wideband low input impedance, right? What happens if you leave that part to some Cs at HF and include some L in series with the opamp input? Again, I know, nobody likes inductors much, but at these low impedances and well beyond audio frequencies, you couldn't need that much L?

EDIT: Note C9 in the above circuit, as well as the choice of R5-R6-C10 etc. ESS hump fixes generally involved fiddling with gains and feedback components, so we're probably on the right track.

Larger mods are not possible on this PCB and that little what is possible is PITA already, 0402 R's and C's, meh...
First try with Cf+Rs = 150pF + 51R (as per @Ben1987) gave no improvent over stock values.

Further experiments in the direction you've pointed out will need a seperate test PCB which is modding friendly. It's way easier to mod the D10B with SO-8 and 0805/1206 SMD's. And it's a pure I/V anyway, no subtractor present which eases hookup. I'm about to temporarily fit a bad OpAmp like 33078 to check if that degenerates performance in the same way as we see on the KTB. Stock is LM4562 btw.

 

[KSTR]

Also the THD vs frequency and IMD vs output level plot for the opamp is interesting. Looking at the CCIF plot vs SMPTE indicate towards HF IMD sensitivity.

I've been using the type of I/V analog filter specified in AD1955 data sheet with good results. (with additional cable capacitance compensated output buffers)
View attachment 182281
For high slew rate cases I've been using mostly AD844.

For ganged converter cases (like ESS 4-channel), I've been then doing mixing through separate active mixing stage after this kind analog pre-stage.

Thanks for that.

Here, and in AD797 spec sheet section for I/V, >= 100R ohmic network impedance is required to keep it stable with larger feedback caps (Aol is telling us).... now me thinks this is the origin of that 100R we find in many of those I/V-circuits? OPA1612 et al don't need it...

 

[Miska]

Here, and in AD797 spec sheet section for I/V, >= 100R ohmic network impedance is required to keep it stable with larger feedback caps (Aol is telling us).... now me thinks this is the origin of that 100R we find in many of those I/V-circuits? OPA1612 et al don't need it...

Likely yes, many high speed opamps need some similar kind of compensation on the feedback also to avoid excessive overshoot on step response. Which would be especially bad if you get close to rails. This is applicable to DAC I/V cases where the waveforms are typically more like high frequency square/stepped waves.

The error waveform you've been showing looks a lot like certain clipping behavior, either digital and analog. Where the clipped top "flips over" to negative side or inverse shape. I just have not figured out why it would happen this particular case. It may be either digital or analog origin...

 

[Pdxwayne]

After all this, any conclusion about audibility when playing music? If there is potential audibility, what type of music to use to best showcase the issue? What to listen for?

Thanks!

 

[KSTR]

Likely yes, many high speed opamps need some similar kind of compensation on the feedback also to avoid excessive overshoot on step response. Which would be especially bad if you get close to rails. This is applicable to DAC I/V cases where the waveforms are typically more like high frequency square/stepped waves.

The error waveform you've been showing looks a lot like certain clipping behavior, either digital and analog. Where the clipped top "flips over" to negative side or inverse shape. I just have not figured out why it would happen this particular case. It may be either digital or analog origin...

Made some progress.

Using 1nF + 0R (no degenerated integrator), --> fc = -3dB@200kHz. I had no NP0 so used a X7R/50V.

Original (470p + 100R) residual for -20dBFS, single-ended (directly after one I/V-stage):

Mod:

The "spikes" are the similar in both cases, these happen to be at the sinewave extrema when the signal changes direction. The D10B also shows it, single ended.
Balanced they will cancel quite a bit as they are giving even orders (in the end you subtract the second half of the waveform from the first). And with the mod they cancel better as they are a bit more equal.

The 64-segment modulation pattern also is visible in both cases but what really is important is that it is now more symmetric around the center at sampe #400 which again means it's mostly even orders --> will cancel in balanced.

We'll see what I can arrive at with regard to a feasible mod. I'm certainly willing to sacrifice bandwidth. With 2n2/806R it'll still be good enough, f3=100KHz, -0.2dB/-13deg error at 20kHz.

 

[KSTR]

After all this, any conclusion about audibility when playing music? If there is potential audibility, what type of music to use to best showcase the issue? What to listen for?

Thanks!

Piano Jazz would come to mind, or in general music with not too hot treble content and dissonant chords.
Audibility is sure debatable. I might do a comparison of modded vs stock, loopback recording and make it available for listening test/ABX.

 

[Pdxwayne]

Piano Jazz would come to mind, or in general music with not too hot treble content and dissonant chords.
Audibility is sure debatable. I might do a comparison of modded vs stock, loopback recording and make it available for listening test/ABX.

May I know when doing ABX, what would you listen for?

I have the orig KTB on hand. I did some comparison listenings yeaterday between it and Gustard x16 (via xlr), both going through Gustard H16, with headphones.

My impression is that, when playing music with lots of lows (like using the YouTube example I provided yesterday for metal music), the KTB's cymbals sounded a "little" less "natural" than the x16. This is just sighted voltage matched listening impression.

I look forward to your capture files for abx.

Thanks!

 

[KSTR]

Next try with 2.2nF+0R, -20dBFS:

A nice improvement. It also holds quite well for higher frequencies at these levels. With increased level I'm seeing rising H2/H3 vs stock, probably the X7R's (which also are only +-10% precision).

I think the direction is clear. Get rid of the series resistor and increase the feedback cap which would be an almost feasible mod, only components to be soldered on top of the existing ones (8 pcs total).

I'll play with the feedback R and C a bit more once I have a set of NP0 caps to fit in.

 

[Herbert]

@Herbert
Sadly, a piggy-back PCB with a new output stage won't be straight-forward to implement because the DAC chip's four output pins are not directly accessible. Also, one has to deactivate the existing I/V by cutting traces (ideally, the existing OpAmps should be removed). Plus we need to bring over the supply etc.
Very hard to do for less experienced DIY'ers. A lot of effort and likely not cost-effective, too.
Easier: Buy a cheap better DAC instead (like Tone2 Pro, D10B, note that both should only be used balanced).

As far as I can remember, when the board is powered by USB there is also 5V Power on Pin 1 of the headers.
(correct word? Germans say "Stecksockel")
As far as I understand, each differential output is followed by two paralleled resistors.
I.e. on the right channel R36/37 and R38/40. So though being SMD there is a little bit of room to put a small blob to solder enamel wire
without the need to directly solder to the pins of the IC. A hot iron and the Opamps are unsoldered.
The analog RCA Jacks can be soldered on the piggyback if needed.
EDIT:
But as I saw from your post above,
it could be even done on the board itself. GREAT WORK!

 

[Grooved]

Great job @KSTR !

@Herbert
... Buy a cheap better DAC instead (like Tone2 Pro, D10B, note that both should only be used balanced).

Regarding the Tone2 Pro, and it's not said anywhere that you have to use it balanced only.
I have one and used it both balanced and unbalanced, when unbalanced, the ring (of their specific balanced RCA) is not used.
I did some measurements on it with the first firmware and the lastest one, and it was improved, but I don't remember if the hump was lowered

 

[KSTR]

As far as I can remember, when the board is powered by USB there is also 5V Power on Pin 1 of the headers.
(correct word? Germans say "Stecksockel")
As far as I understand, each differential output is followed by two paralleled resistors.
I.e. on the right channel R36/37 and R38/40. So though being SMD there is a little bit of room to put a small blob to solder enamel wire
without the need to directly solder to the pins of the IC. A hot iron and the Opamps are unsoldered.
The analog RCA Jacks can be soldered on the piggyback if needed.
EDIT:
But as I saw from your post above,
it could be even done on the board itself. GREAT WORK!

It will not work this way.
You have to completely free the DAC chip's output pins, must not attach to anything.

Showing one I/V (of four), we would have to remove the X-'d parts and cut the trace at the red line. This free the DAC pin and keeps the OpAmp happy (feedback still closed).
Then we could solder a wire at the green points, connecting to both for more mechanical stability. The signal is only present at the left point, of course.

Unsoldering the OpAmps in the leadless DFN-8 package with their bottom GND solder pad is not so easy, this is not your standard SO-8 which is easy to remove. That's for experienced SMT repair personnel only, as you will need special tools.

5V USB is not the problem but we would like to use the +-5V rails which are already present on the board. And a GND, of course. That's seven wires, plus the preparation as above, 4x.

The KTB has other slight shortcomings that question the total effort a bit: large clock drifts (probaly supply-related, the 100MHz oscillator needs a very low noise supply at low frequencies and DC. The 9038's AVcc needs a rock-stable low-noise supply as well.
The used ADP150 regulators are fair in this reagrd but nothing more, quite noisy/drifty at LF/DC.

My summary would be that the simplest possible patch is the only thing that's realistic. Handling 0402 SMT parts is already enough of a challenge...

 

[KSTR]

As far as I can remember, when the board is powered by USB there is also 5V Power on Pin 1 of the headers.
(correct word? Germans say "Stecksockel")
As far as I understand, each differential output is followed by two paralleled resistors.
I.e. on the right channel R36/37 and R38/40. So though being SMD there is a little bit of room to put a small blob to solder enamel wire
without the need to directly solder to the pins of the IC. A hot iron and the Opamps are unsoldered.
The analog RCA Jacks can be soldered on the piggyback if needed.
EDIT:
But as I saw from your post above,
it could be even done on the board itself. GREAT WORK!

It will not work this way.
You have to completely free the DAC chip's output pins, must not attach to anything.

View attachment 182357
Showing one I/V (of four), we would have to remove the X-'d parts and cut the trace at the red line. This free the DAC pin and keeps the OpAmp happy (feedback still closed).
Then we could solder a wire at the green points, connecting to both for more mechanical stability. The signal is only present at the left point, of course.

Unsoldering the OpAmps in the leadless DFN-8 package with their bottom GND solder pad is not so easy, this is not your standard SO-8 which is easy to remove. That's for experienced SMT repair personnel only, as you will need special tools.

5V USB is not the problem but we would like to use the +-5V rails which are already present on the board. And a GND, of course. That's seven wires, plus the preparation as above, 4x.

The KTB has other slight shortcomings that question the total effort a bit: large clock drifts (probaly supply-related, the 100MHz oscillator needs a very low noise supply at low frequencies and DC. The 9038's AVcc needs a rock-stable low-noise supply as well.
The used ADP150 regulators are fair in this reagrd but nothing more, quite noisy/drifty at LF/DC.

My summary would be that the simplest possible patch is the only thing that's realistic. Handling 0402 SMT parts is already enough of a challenge...

 

[Herbert]

Thanks for the explanation! But the four ADP150 (SOT23-5 Package) could be replaced with better ones...?

 

[KSTR]

Regarding the Tone2 Pro, and it's not said anywhere that you have to use it balanced only.
I have one and used it both balanced and unbalanced, when unbalanced, the ring (of their specific balanced RCA) is not used.
I did some measurements on it with the first firmware and the lastest one, and it was improved, but I don't remember if the hump was lowered

Unbalanced, the KTB2 performance degrades, at least measurable, see https://www.audiosciencereview.com/forum/index.php?threads/khadas-tone2-pro-review-dac-amp.20684/

Balanced:

Unbalanced:

Assuming the the additional dirt is constant in level it will degrade SNR big time at lower levels.

Also, it could be like it is with the D10B where the unbalanced feed actually is tapped off of only one I/V converter, without the required subtractor to afford the cancelling of even-order harmonics and other signal-related dirt. That's of course speculation for now but the spectrum suggests that as well (higher even order harmonics) -- as far as the plot even allows to spot that (a way suboptimal plotting style, IMHO).

 

[KSTR]

Thanks for the explanation! But the four ADP150 (SOT23-5 Package) could be replaced with better ones...?

No. At least there is nothing pin-compatible that is significantly better. LT3042 is the benchmark regulator these days when it comes to low LF noise.

 

[Grooved]

Unbalanced, the KTB2 performance degrades, at least measurable, see https://www.audiosciencereview.com/forum/index.php?threads/khadas-tone2-pro-review-dac-amp.20684/

Balanced:
View attachment 182362

Unbalanced:
View attachment 182363
Assuming the the additional dirt is constant in level it will degrade SNR big time at lower levels.

Also, it could be like it is with the D10B where the unbalanced feed actually is tapped off of only one I/V converter, without the required subtractor to afford the cancelling of even-order harmonics and other signal-related dirt. That's of course speculation for now but the spectrum suggests that as well (higher even order harmonics) -- as far as the plot even allows to spot that (a way suboptimal plotting style, IMHO).

Right, but is this unbalanced performance better or the same than on the Tone1 you are working on?
I supposed the path is not the same at all even if using the Tone2 Pro as unbalanced since the Tone1 doesn't have any OPA1612

 

[KSTR]

Right, but is this unbalanced performance better or the same than on the Tone1 you are working on?

Hard to tell without having the board at hand for measurements at lower levels.

I supposed the path is not the same at all even if using the Tone2 Pro as unbalanced since the Tone1 doesn't have any OPA1612

Yes, it certainly is very different.
The question is whether the second stage -- the lowpass filter stage -- actually re-balances the signal (a subtractor with a balanced output) or not. I would assume (and hope) so but it's not self-evident.
The Tone1 has the subtractor (with an unbalanced output) as it should be.
The Topping D10B, as a very minimalistic design, doesn't not have a subtracting re-balancer and relies on the receiver (amp) to do the subtraction in order to remove the commom mode signals. Unbalanced its distortion is very significantly higher than balanced.

 

[Herbert]

@KSTR BTW I have this TOS7a4700 based power supply,
I have no idea how good it really is.
But if this a cleaner power source, I could lend it to you.

Single Rail TPS7A4700 Ultra Low Noise</br>Power Supply

Discover the Best Low Dropout, Ultra Low Noise Voltage Regulators Trusted by Renowned Authorities such as NASA, NIST, and Scientific Labs. Ensure Precision and Reliability in Your Power Supply Solutions, Best Power Source for your HiFi and HiEnd equipment

www.ldovr.com

 

[KSTR]

Hi,

I think I now have found a solution that behaves well and gives nicely balanced improvement for all frequencies and levels.

The original idea of increasing the I/V capacitor and removing the series resistor helped at low levels/frequencies but failed in that it produces too much distortion at higher levels and higher frequencies, worst-case even 20dB more than stock....

It turned out that it best to leave the 470F as is and only reduce the series resistor from 100R to 10R. Then the key trick (also seen in older DAC chip's datasheets, like AD1853) was to add a 220pF "transient killer" capitor accross the inputs of the I/V which shorts out fast differential-mode transients which are at and (way) above above the bandwidth of the opamp here.
Sanity checking specs in a simulation showed that the increase in differential noise gain is not causing severe issues wrt stabilty, etc.

C4 (220pF 5% NP0) , R4 and R5 (10R) are the added components (ignore the double R2 designator, btw).

Patching 0402-sized SMT components, actually adding one with a patch wire... puh, you will feel like a watch-maker must feel. ;-)
The chip below the patch wire is 3x3 mm size, those 0402's are like a grain of sugar.
But at least nothing has to be unsoldered.


Throughout a wide level and frequency range I now see a constant improvement, enough to remove the hump and give much better distortion spectra. Only at the very extreme end, 10kHz @ 0dBFS the perfomance degrades to actually worse distortion vs. stock by ~10dB. At 10kHz @ -10dBFS we have a draw. Looks acceptable to me.

240Hz @ -20dB distortion spectrum:

THD is now way lower than THD+N which means noise dominates, no chance anymore for "humping"

I'm running out of board channels by now (not wanting to solder around too many times, given the fragility) and might have spoiled one channel on my own board already. I'll have to restrict any further experiments to the minimum (build back one channel to original while doing the full suite of final A/B measurements all with the same board channel.

 

[Herbert]

Bravo! A lot of work but a simple fix. If you need an unaltered stock board for reference,
I have a second Khadas board!
Question:
The choice of 220pF is not sampling frequency dependend, as the DAC puts out
the same sampling rate no matter the source?
I think, a normal 220pF capacitor with a width 5mm betwinn pins might fit as well.
It just has to be glied on before soldering...