Please explain why capacitor feed-thrus are bad from emf viewpoint. I admit to having a RF-noisy shop. Even turning on the Hakko soldering iron will cause a noticeable shift of measured voltage (uV to 10'2 of uV). These feed-thrus are directly soldered to PCB inside enclosure and are mounted upon a (hopefully) isothermal aluminum cover plate. Surely this is better than standard connectors whose pins are held by plastic. And unlike standard connectors, feed-thrus provide double-digit attenuation above 10 MHz (assumed 50 Ohm measurement). What do you recommend? CuTe jacks will have far greater RF leakage into the metal box.
Regarding shielding, left LTZ is powered by analog rectified and LM317-based regulator in its own metal box about a meter away. It's shielded cable is tied to its box which is connected to PE (Earth) of the AC mains. That's why the shield wire is pulled aside on the LTZ end (to eliminate ground loop). The right LTZ proto is powered via black jack from battery with linear PSU keeping battery in float condition. Right LTZ also has BNC jacks for zener voltage measurement whereby the shields of the coax were actively driven at guard voltage. Because I didn't have Triax cable to meter, this picked up noise and was abandoned. Therefore each LTZ has separate floating supply. Ground of each LTZ has single-point connection to case. Case is connected to shields of Cat-6 cable back to data logger where shields are tied to the PE screw of the 34970A data logger. All voltage measurements are differential via twisted-pairs within Cat-6 cable.
Regarding 100nF capacitors, my SPICE analysis shows this compromises phase margin. See attached where the Green traces depict the Bode Plot amplitude and phase of the Zener side of the regulator. Note gain peaking of LTZ transistor stage at 550 kHz (red trace) with 100nF B-E capacitor. I 'll show analysis circuit in another post (getting late).
Kleinstein already answered this question.
There are simply too many probable dissimilar metals and too many junctions involved.. better keep it simple, so not to loose the overview over the signal path.
Have a look in the LTZ 1000 datasheet, or the National Semiconductor application note for the LM199, both warn explicitly of thermal junctions, and simply to reduce the number of junctions.. that is the biggest problem in your setup.
Frankly speaking, I'm completely confused about your CAT 6 cable in series with the feed through capacitors plus clamped jacks, and so on.
Clamping the wrong metals, or oxydized metals also makes nice unpredictable thermocouples.
If think hat you have put too much focus on the shielding in comparison to e.m.f.s, that's true also for the absolute values of the different disturbance voltages.
I have seen similar shifts as you, then I used a big ferrite on the externally located DC supply cables, then these two 100nF, at least, maybe a simple grounding of the metal case, and now, no shift any more, even a switch mode LED light creates only minute changes on the the order of one ppm, instead of several ten ppm.
Maybe it's no good idea to ground the supply directly to mains earth. (I don't that, neither)
That's all a bit of vodoo, and fiddling around with the shielding (also keep it simple) , until the reference was nearly immune against RF or mains spikes.
Frank