Thanks for your feedback.
syn08: I measured THD, not THD+N. Anyway, the noise is relatively low, so I am primarily concerned about the THD for now. The noise can probably be reduced slightly by using a lower noise power supply. I will probably use an op-amp based AVCC to get very low noise.
I have tried with different IV's. The first one with 2xLME49990. The present one with a socket, having tested OPA1612 and OP1656. The results are very similar. I don't have a balanced to single ended op-amp on the output, relying instead on the differential input of the analyzer (RTX6001).
IVX: I am certainly far away from the -117dB without compensation. And that is what's worrying me a bit. My unit is 12dB worse. Could that be due to the circuit/layout or due to variation between chips?
altor: Your results look quite good. Do you have a number for the THD without compensation?
do not trust Spectraplus for THD+N ar any noise-related calculations, it is wrong.
Interesting, how much is 0dBFS at your DAC output? For me, 0dB is always about 4Veff. It would also be interesting to see your numbers at -0.5dB, -3dB is quite low.
Cannot find a screen shoot. Will measure with Cosmos.
2.8Vrms.
Picture above was at -3dB.
No he's using it to heat the ADC. Or rather, I'm assuming, connecting some of the IO ports to ground and then driving them to cause heating in the micro that's glued to the top of the ADC. A thermister is then wired up to an ADC input pin on the micro and used to sense the temperature, with the micro turning the IO ports on and off as required to maintain a specific temperature.
It's probably a very cheap way to do this actually but I'm guessing rather unnecessary.
Can we have curves for performance vs temperature?
It's probably a very cheap way to do this actually but I'm guessing rather unnecessary.
Can we have curves for performance vs temperature?
The curve, especially the realistic one would be quite a difficult task for me. Try to imagine how you can sync the temperature rising with 3rd harmonic dropping down to -145db, or even how to measure the temperature of 5x5mm QFN precisely if any thermocouple+its thermal interface will sink half of the heating power out. Cut the window in the aluminum case against ES9822 and place there a thermal camera? Uhh..
BTW, ESS eng confirmed that poly-resistors in that chip simulating the same way i.e. hotter means better linearity. Of course, if my target was -120db, no reason to care about the thermostat.
BTW, ESS eng confirmed that poly-resistors in that chip simulating the same way i.e. hotter means better linearity. Of course, if my target was -120db, no reason to care about the thermostat.
This is what I call a “bang-bang” temperature regulator, and is a rather bad idea (to put it mildly) at under micro volt level. To my experience it generates spurious each time the controller goes “bang”, by induction, radiation, power supply lines, etc… Last time I tried this and learned the lesson the hard way was with a Peltier cooler for a LNA.
Only an analog regulator would be quiet enough to heat the ADC (if that is required for performance, which I doubt is worth the trouble), unless you care only about the SINAD and can ignore the random spikes.
Hmm, would like to exchange the small oscillators or at last one only with a SMA connector, so to glue the SMA on board
and connect with thin wires to the ADC 
and may terminate with an 50E resistorThe question rises of the purpose golden & the red cycled area...
So some support required
Hp
Attachments
So you can tweak a part from -120 to around -145dB just by increasing the temperature?
I am surprised to see such a huge difference with modern IC's.
I had a similar experience with the (old) AK5394A. It has an optimum temperature, where distortion is reduced considerably. One of the advantages of the AK5394A is that it has a built-in heating function
(also known as a rather high power consumption ). The problem was that in general it had to be cooled, which is more difficult than heating a device. I did consider Peltier elements, but ended up using an aluminum bracket to transfer the heat to the chassis.With the ES9822 I guess it might be possible to heat the device with a resistor or two adjacent to the device and a control circuit to keep the temperature at a preset point.
I also tested the newer AK5397 (now also obsolete). With the AK5397 I could not see any THD variation over temperature.
@HpW
Perhaps it would be easier to use a crimp SMA connector with a (very) thin coax cable, like the cables often used inside routers etc.? Of course with AC coupling and termination if needed. The thin cable (perhaps around 1 mm diameter) would remove the risk of mechanical stress.