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

[Herbert]

The only 8-Pin (if the datasheet of the SRT6863 is compatible to the 6862) is 8-Pin SOIC.
Same as 1612. So if it is SOIC without a thermal pad underneath and not DFN / SON,
replacement could be done with a soldering station iron with a fine tip...?

Once again the datasheets:

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

https://www.ti.com/lit/ds/symlink/opa1612.pdf?ts=1643830397145&ref_url=https%253A%252F%252Fwww.ti.com%252Fproduct%252FOPA1612

 

[mansr]

The photos up-thread clearly show both DFN-8 (RT6862) and DFN-10 (RT6863) chips.

SOIC-8 _can_ be replaced using a regular soldering iron, but the removal stage is no fun at all unless you cut the leads. You'll want to use a wide, though. If there's a thermal pad, you're obviously out of luck.

 

[KSTR]

The 10-pin RT6863 for the subtractor/output stage is pin-compatible with the standard dual opamp pinout, it just adds an Enable and GND Reference for that enable (the enable might be used for muting on this board), and of course a thermal pad. So, technically, it is not impossible to be replaced with a 1612 DFN-8 (with hot air and whatnot)...

 

[zenon]

...
SOIC-8 _can_ be replaced using a regular soldering iron, but the removal stage is no fun at all unless you cut the leads.

In fact, removing these SOIC-8s takes about 2 seconds if there is no thermal pad.

 

[mansr]

In fact, removing these SOIC-8s takes about 2 seconds if there is no thermal pad.

That's not a regular soldering iron.

 

[zenon]

That's not a regular soldering iron.

That's right. Anyone can make it from a piece of 2mm copper wire.

 

[solderdude]

The way I always removed those 100 pin IC's was using a spring from a ballpoint. I stretched it out.
Shove it under the pin.
Then heat the pin and put mild pressure (upwards) on the pin.
Pin desolders and goes upwards... next pin... etc.
Using a regular soldering iron and never peeled a trace of a board.
PCB's were of good quality though (Technics in the 80-90's)

 

[mansr]

The way I always removed those 100 pin IC's was using a spring from a ballpoint. I stretched it out.
Shove it under the pin.
Then heat the pin and put mild pressure (upwards) on the pin.
Pin desolders and goes upwards... next pin... etc.
Using a regular soldering iron and never peeled a trace of a board.
PCB's were of good quality though (Technics in the 80-90's)

That ruins the chip, though.

 

[Herbert]

That ruins the chip, though.

No problem as long as you do not need it anymore...
Another trick ist to bent a square from copper wire
that fits the square of the pins, put it on
and evenly heat the pins with a lot of solder
until the whole thing comes off. But a lot of heat
that could ruin the chip.
Then you'll need a plan for the replacement

 

[Grooved]

Actually, I'm not wrong, it's not really DFN as DFN were supposed to have up to 4 pins, not more and all under the chip, not going to the edge :

These ones have 8 and 10 pins, and it looks like they go on the edge of the chip, so they are more SON package, and should be compatible if each pin as the same function. It's certainly the difference with real SON package, and the reason they got called DFN-8 / DFN-10.

For the DFN-8, it look to be the same, so if the DT6862 is based on the DT6863 (without GND and ENABLE), it certainly has a pad under the chip, and can be replaced with OPA1612 (using hot air to desolder/solder)

So the main problem would be to find a SON/DFN chip with 10 pins to replace the DT6863, with the same pinout.
The only ones that I find called DFN-10 are Analog Devices and Microchip, they have a pad, but function of pin 5 and 6 look not to be the same. Analog Devices seems closer as pin 6 is Enable function, but pin 5 too instead of GND :

EDIT :
I found TI chip (not the right values) with the same pinout than AD DFN-10, and it's called VSON-10. But again, no GND like on the DT6863 called DFN-10 like the AD

If it's like the AD chip, if left unconnected, the amp if always enabled, which would allow to measure, but the pin 5 will be in contact with the GND of the Tone Board... Is it possible to bypass the GND connection ? cutting it on the board, or using something super thin to isolate it ?

 

[KSTR]

Package type names are not yet standardized so one company's WSON-8 is another's DFN-8.

Flat no-leads package - Wikipedia

en.wikipedia.org

So the main problem would be to find a SON/DFN chip with 10 pins to replace the DT6863, with the same pinout.

No. I checked and the enable pin isn't used anyway (tied to V+ --> ON). OPA1612 etc would fit with a bit of care wrt aligning the thermal pad.

----------

I'm currently checking with a D10B whether a better opamp really is the way to go.

I've soldered a lowly MC33078 (of ST manufacture, GBW=15MHz, SR=7V/us) in place of the LM4562 and to my surprise the performance did not suffer too much. At low levels and low frequencies (like 240Hz/-15dB) basically no change, and no sign of that characteristic "sweep" pattern in the residual at higher levels in the single-ended output like with the KTB (with D10B probed single-ended). The error signal is generally lower and is mostly common-mode and thus cancels quite well, very different to what I saw on the KTB. Bias voltage does not play any role (as expected), it just varies common-mode DC offset.

The D10B's I/V setup is 1k6 // (~1nF + 10R). The 1nF is an upper limit eyeballed from the published frequency response, any larger and the -0.2dB goal at 20kHz is not met. It could be less, actually.
Will have to play more with the D10B I/V and see if I can provoke some similar error characteristics. Assuming those RT686x opamps are somewhat more comparable to that MC33078 than to a LM4562, that is.

 

[noi1996]

What about using a CFB op-amp like AD812 for the I/V stage ?
It won't be able to use the differential input capacitor and the feedback capacitor, but a wide bandwidth and fast slew rate can be obtained.
The internal resistance of the ES9038Q2M is 774 ohm, so the feedback resistor would be close to the recommended value on the datasheet, when the gain set to -1.

 

[mansr]

No. I checked and the enable pin isn't used anyway (tied to V+ --> ON). OPA1612 etc would fit with a bit of care wrt aligning the thermal pad.

The over-sized thermal pad could complicate soldering by pulling the chip out of alignment with the pads on the PCB. Should be manageable, though.

 

[mansr]

The internal resistance of the ES9038Q2M is 774 ohm

That value has a tolerance of 11%. For anything resembling predicable results, the output needs to feed into something with either much higher or much lower impedance.

 

[KSTR]

@noi1996,
CFB OpAmp seldom find their way into an I/V as it seems.

I'm more inclined to think a fast fully differential OpAmp like OPA1632 is best suited for the specialities of the ESS output (bipolar current source with paralleled AVCC-to-GND 1:1 resistor divider with 2x 1548 ohm)... or the "superbalanced" equivalent with standard type OpAmps (fast ones). Okto Research DAC seems to use the 1632, for example.

The advantage is that input common-mode signals are removed (no down-stream subtractor needed) and output common-mode voltage is zero by definition, plus the circuit is self-biasing. Extremely simple and efficient.
The higher common-mode input impedance is not a big penalty as the common-mode signals are small.

The scheme also could easily be extended to a second order filter. This increases differential input impedance but it has to be seen if ~10 ohms really would make a difference.

@mansr,
I've measured the DC output resistance on both channels for two chips and it is very consistent at 774 ohms.

 

[mansr]

I've measured the DC output resistance on both channels for two chips and it is very consistent at 774 ohms.

I was quoting the "data sheet."

 

[Herbert]

The D10B's I/V setup is 1k6 // (~1nF + 10R). The 1nF is an upper limit eyeballed from the published frequency response, any larger and the -0.2dB goal at 20kHz is not met. It could be less, actually.

Besides the digital filtering of the 9038 DAC, for basic understanding: Does it make sense to say:"Well, 99,9% of my source music material is between 44.1 and 48kHz sampling rate anyway. So I can "kill" anything in the analog domain starting above 24kHz with an analog filter and remove the hump as well? Because as far as I understand the hump in unaffected by digital filtering, but somehow seems to be triggered by noise outside the audio band...

 

[Grooved]

No. I checked and the enable pin isn't used anyway (tied to V+ --> ON). OPA1612 etc would fit with a bit of care wrt aligning the thermal pad.

Good news they aren't used! Gives some possibilities to try other chips

 

[KSTR]

Besides the digital filtering of the 9038 DAC, for basic understanding: Does it make sense to say:"Well, 99,9% of my source music material is between 44.1 and 48kHz sampling rate anyway. So I can "kill" anything in the analog domain starting above 24kHz with an analog filter and remove the hump as well? Because as far as I understand the hump in unaffected by digital filtering, but somehow seems to be triggered by noise outside the audio band...

No, filtering lower will not reduce the problem, the spray of harmonics is all within the 20kHz range.

To me things are now even less clear than ever.

What I see is:
- The D10B with low-tier MC33078 is as good as with the far better LM4562, in the 240Hz -15dBFS residuals. The two single-ended residuals for + and - I/V-paths cancel very well. Cancelling includes odd order distortions as much as even orders, that is, most of the single-ended error is common-mode signal. That makes parameters of OpAmp somewhat less likely to be the dominant factor in the KTB. I/V setup is not that important (I varied bias voltage and feedback resistor, only little effect seen).
- In the KTB the single-ended residuals (directly after the I/V-stage) are much dirtier and show that special modulation pattern strongly, and they do not cancel quite as well. With some mods in the I/V it gets better but only with the cross-capacitor I'm gettting halfway close to MC33078 in D10B.

Are those RT6862 really that worse than a lowly MC33078 or is it something else I'm overlooking?

 

[noi1996]

If you intend to "degrade" the D10B for comparison, how about add some impedance to the AVCC of the D10B ?
One of the differences may be that KTB uses the ADP150 for 3.3V and D10B uses the LM4562 as a 3.3V buffer for AVCC.

Edit: In D10B, It may be easy to connect the output of the 3.3V LDO (L33EK) around the ES9038Q2M IC directly to AVCC after removing the LM4562 for AVCC.