Andreas:
It's OK if you don't understand the structure of a crystal lattice and aging effects, and some of the factors that might make that worse - or better. That's well beyond the scope of what I can show you here. I'll give you a hint below though.
From my experience after building hundreds of LTZ's, we've never needed to alter the basic LTZ circuit. Ever. Yes we've added shielding and Yes on the test bench we take pains to keep everything as quiet as possible. For units going into the field they include more ruggedized shielding techniques and that works in every situation we've encountered. Like how it works on 3458a. The reason we can't always add CPU's to the circuit for DAC controls is we're in a situation where WE can't be the ones making noise, and that's why sometimes pure analog is the way to go. Sometimes an assist with a CPU / FPGA is a good way. It just depends on application.
If you had to alter the circuit for your use, that's one thing - but I know from experience that doesn't apply everywhere. It doesn't apply to 3458a, it doesn't seem to apply to TiN's circuits, it doesn't apply to our circuits, LT/AD doesn't need or recommend they be used, etc. Now of course there might be -some- effect on YOUR board or test setup where the altered circuit got better results - but that doesn't mean it applies in -every- case, does it? If someone is having trouble, and adding the caps helps - then Yay!. But I suspect the problem really might lie elsewhere.
I would invite you to look at a new Re-Spin of an LTZ application, now that AD owns the Company - Look at the LTZ circuit here, and in this case you want a low noise Vref to show off the 20 bit AD5791 DAC they're selling - see page 20:
http://www.analog.com/media/en/technical-documentation/user-guides/eval-ad5791sdz-ug-1152.pdfAfter a chat with AD, I'm told there are some typos that will be corrected on the next doc release...But the note on C44 refers to
C13 C12, The note about R30 should be R48, and the note re: U3, U4 should refer to U1, U2.
The deviation from the LTZ1000 datasheet is they use SMT parts (board picks up vibration and stress changes), the '1013 is swapped for an AD4077-2 (Basically newer gen precision OP07) because AD wanted to showcase some of their own brand. The AD4077 current driver likes an extra cap compensation, and the usual 22nF cap has been upped to 100nf - otherwise the circuit topology is just really a basic LTZ.
Notice the RC between the LTZ and the 10V boost. That a good place to remove a little noise and isolate the LTZ from any downstream activity / current noise.
Notice the 15k over 1k heater resistor ratio. They went with this "Hotter" ratio because on an bare naked board they liked the warmer run temp of the LTZ - it would resist casual air drafts better. This may cause slightly lower long term drift but that is not the goal here.
Something else to notice is the standard resistors used for 10V boost: Because this circuit is feeding a DAC you'd like to have a little OVER +-10V cardinal output on this Vref, which means as designed you want the LTZ with a minimum 7.14VDC using the 25k/10k boost ratio. Since AD owns LTZ part, they can pick and choose the LTZ demo chips to make sure the LTZ demo board reference output has enough headroom to let the output of the '5791 output run to +-10VDC full scale (The AD guy was pretty sure on this when I pointed that out, but he is going to check).
Andreas
(BIG CLUE): After 36 years of LT / AD manufacturing LTZ chips and getting customer feedback, please take a look at that schematic topology and let me know how many low impedance caps you see straddling any single PN junction on the LTZ?. How many chopper amps are used? What could be the reason for that "Fairy Tale" ??
Now back to a real test: I haven't tested this board on my bench, but my colleague tried one of these new LTZ ref boards on his '5791 demo, and wanted to double check these test results - and he does have access to a .1 to 10Hz LNA (Jim Williams Wet Tant style):
http://www.analog.com/en/analog-dialogue/articles/high-precision-voltage-source.htmlSo I emailed him last night to see how the results went. In general his comments were positive. Testing the noise at the output of the LTZ was typical <<1.2uV p-p, and the '5791 output was fairly close to the published test report IF you pay attention:
A) Gotta add the goofy little afterthought foam LTZ draft cover in the kit. Looks like you're on your own for covering the bottom side (AD says they are changing that on next rev).
B) He couldn't do it with the USB cable from a laptop to the demo controller card - he used an isolated cable from a desktop PC box located on the floor a little farther away. The output of the DAC got cleaner when you unplug the digital controller cable, as expected.
C) The Keithley switcher power supply added a little more noise to the output (not a total disaster), but switching to batts or Protek 303 linear (transformer) supply worked fine.
D) This circuit didn't like the soldering iron cycling on and off on the same outlet strip, nor did it like the LED desk lamp.
E) BUT: Putting the circuit and LNA amp in a leftover Xmas Cookie tin got very good results, and after that the desk lamp was fine.
F) If he set the board on a soft foam bleacher pad, the board quit reacting to every mechanical noise on the test bench (Noise pickup from SMT caps). Without the pad, if he flipped a switch or tapped a pencil on the bench, that would show up on the output.
G) He saw more noise on the '5791 output at full scale - he was getting more like 4 or 5uV (or 7uV on occasion) peak. The datasheet DAC spec's 1.1uV added noise but that's at mid-scale, at 90%+ full scale you'll see more noise.
EDIT: He noticed that if you averaged the peaks from several 10 sec periods the peak noise could be called ~4uV peak. So he was able to get in the same ballpark as published results.
I realize that not everyone has access to LT / AD apps engineering. I realize people like to play.
Again: My only suggestion is - If you're finding that adding caps helps your circuit, you might sniff around a bit to find out what the true cause of the problem might be.
The standard circuit topology -should- work fine, and I suggest if you're having trouble with a modified circuit: Try stripping it back to the standard circuit and that might help you troubleshoot where the real problem is. You might find that correcting a thermal / noise issue will work as well or better as adding extra caps - and that in turn might give your LTZ the very best chance at aging gracefully.