Trollocks; I thought this had posted a couple of days ago but just realised it didn't.
It's been a bit overtaken by events, but since I went to the trouble I may as well post it anyway.
Presumably all the later measurements are with the extra caps added which limit the bandwidth to around 1 to 2 Hz - which will significantly reduce the resistor, op-amp voltage and current noise (none of which should have much 1/f noise), but rather less of the 1/f noise from the LTZ1000 references. But I can't be bothered to try and calculate the impact.
Recently made some measurements of my equipment using a Pipelie LNA (s/n 1157A). Values below are RTTI. Should be comparable to Zlymex results column 6.
Fluke 731B #1 avg = 2.5 uV p-p
Fluke 731B #2 avg = 1.7 uV p-p
Malone DMMCheck 5V output avg = 26.8 uV p-p
Ensemble of 7 LTZ1000 7.115V output = .39 uV p-p
Ensemble 5V output (custom resistive divider) = .35 uV p-p
Ensemble 10V output (opamp X2) = .87 uV p-p
My measurement setup is pretty basic. In particular, I have not shielded everything very well and I notice that readings increase when I am nearby in the room. Still, the Fluke 731B measurements suggest that I am not wildly wrong.
However it seems that my circuits do not preserve the low noise of the ensemble very well. Straightforward ratios would suggest
7V output = 1.2uV/sqrt(7) = .45 uV <ideal> <--did ok here!
5V output = 0.39/1.43 = 0.27uV <ideal> <-- marginal
10V output = 5V*2 = 0.55 uV <ideal> <-- very marginal, but good compared to zlymex table right?
Assuming ensemble measurment is at output of U23A, noise over .1 to 10Hz:
1) 7.1 to 5V stage:
a) .39/1.43 = .27uV
b) U23B, Vn = 117nV pp x noise gain (1x)
c) 2112//5000 = 1484 ohms = 102nV pp thermal noise
d) U23B In- 16pA x 2k ohms = 32nV pp
e) U23B In- 16pA x 1484 ohms = 24nV pp
RSS noise = .32uV pp. Not too far from .35uV from your earlier test - it's not clear if your revised measurements are including C27 and C28, so I ignored those.
2) 5 to 10V stage:
a) 5V stage noise .35uV pp
b) U24 Vn 117 nV pp x gain (2x) = 234nV pp
c) 10k//10k = 5K ohms = 188nV pp x gain (2x) = 376nV pp
d) U24 In+ x 10 ohms - negligable
e) U24 In- 16pA x 5k x gain (2x) = 160nV p
RSS noise = .84uV; again not far from your .87uV.
Better, from a noise perspective, would be to generate the 10V from the 7.1V avoiding the added noise of the 7.1V to 5V stage. The drift of the 10V output relative to the 5V would likely be greater as the two stages can drift in opposite directions, but the 10V drift would be lower than the current scheme (given similar performance scaling resistors) as a) the ratio is lower and b) the 7.1 to 5V stage drift is eliminated.
Using lower value resistors would help - but DSMZs don't come cheap so filtering with added caps is a pragmatic solution. But when you are looking at low level, very low frequency noise signals, should you worry about dielectric absorption and temperature dependence of those caps adding their own 'noise'? (ie. not actually noise, but hard to distinguish changes from the low frequency noise and drift of the references).