Replacing LTZ1000 in an HP 3458A with a lower noise (?) reference

[zhtoor]

-

[TiN]

How do you plan to solve temperature coefficient issues of such an assembly?
Also LT1012A on it's own specified for 30 nV/sqrtHz already, which you will amplify 7 times?

I plan to try lower noise reference for 3458A, but it's going to be traditional approach.

[Kleinstein]

The Weston cells will have some noise by them-self. A JFET amplifier would not be good enough, as the critical part would be the very low frequency noise (e.g. 1-100 mHz range).  So if at all it would take an AZ OP to get the required low 1/f noise.  With a gain of about 6 and some 30-50 nV/Sqrt(Hz) to compete with this would require a noise in the below 5-10 nV/SQRT(Hz). So even an LTZ2057 would be more like too noisy.

The more obvious way would be using 2 - 4 of the LTZ1000 - that is probably what TiN is planing.

[mimmus78]

One can use 7 cells instead of one ... (just kidding) ...

Well I think the best way to go is to use 2 LTZ1000A running at double of the current at 1K/13K.

Anyway do you know what is max Vref that 3458A can safety handle?

[MisterDiodes]

Usually '5400 resistors networks are a real disappointment in this sort of application if you're after low noise - based on experience.  Remember, the more you're gaining up the Vref, the more noise you're going to deal with. And the '5400's ain't known for lowest noise, by design.

Just stick with an LTZ for better performance, or dual / quad LTZ's.  You're not improving a 3458a with this circuit.

Wet cells are really no use here.  Or anywhere.  If you touch them or move them or let the temp change at all you no longer have a stable Vref.  I know you're bound and determined to use those things, but zeners are simply the (much) better solution.

[TiN]

The infamous 2DW zener have lower noise, and will be much easier and better use then chemical cell.

Also you might even connect JJA itself as reference in HP 3458A, but you so pin-focused on noise alone, don't forget about reference gain stages on ADC, which convert 7.2V into +12V, -12V and +5V. These are the actual reference levels which used by ADC, not the 7.2V.

But either way, I'd suggest to build some boards and just go ahead with experimenting. You will find many not-so-obvious interesting stuff, while playing with 3458A A1/A9/A3 stuff.  No amount of chat on forum will replace actual hands on experience. 

[TiN]

Sorry if that sounded harsh, but that's not me who started the thread .

I have chemical cell, and I can see zero benefit of using it instead of LTZ reference. Noiseless readings (and circuits above at ballpark have more noise than stock A9 LTZ REF, not less). are practically useless, if you don't have the linearity and stability to hold the accuracy. Linearity is fundamental for better precision, and we already know 3458A is as good as it gets in that regard. I'm going paralleling route just because I'll get excess amount of resistors (due to MOQ). There are no hopes of magical 9.5 digits out of 3458A, no matter how many LTZs you put inside.

If my measurements would require low-noise data, it would be better to use bridge measurement arrangement or low-noise preamplifier to scale the signal accordingly. This way you get best of both worlds, low noise when you need it, and traceable and serviceable 3458A at same time.

[mimmus78]

This is an example of what you get with two lipos vs 3458a after many hours of settling time. If you ignore the jumps that are caused by LTZ1000A in my A9 you see noise is already pretty low, you can improve a little bit but as TiN said it is not going to add a further digit. This is the point when ADC TCR drift and A9 TCR drift, EMF, etc starts to became dominant on your measurements.

Maybe having a little less noise can make sense for evaluate other references noise ... but that is, even with my jumpy LTZ1000A I'm not loosing all that much.

[David Hess]

For ADCs which integrate the input signal which includes voltage-to-frequency converters and delta-sigma converters, flicker noise is a major problem and difficult to filter without adding drift so I would worry less about broadband noise which is much easier to filter and use a chopper stabilized amplifier for the first stage if high input impedance is necessary.  For lower noise than a chopper stabilized amplifier can provide, it can be used to remove flicker noise from another amplifier combining the best noise characteristics of both.

Low impedance references have a noise advantage from not requiring a high input impedance amplifier.

[Kleinstein]

The reference does not contribute very much to the noise, when measuring a zero voltage. It would be only the higher frequency part that is not fully filtered and than mixed back by the modulator to the low frequency range that would contribute -  however this would be the easy part anyway as filtering is possible in the range and partially used.

So the test with the short in the 10 V range is mainly measuring the noise of the ADC, with a little contribution of the input amplifier and possibly some (avoidable) higher frequency (e.g. 500 kHz range for the 3458) noise part from the reference.

The really problematic low frequency reference noise would be visible only when measuring a voltage that is significant different from 0, ideally close to full scale.

The LTZ1000 reference is already quite low on noise. The broad band noise is lower than the noise of the ADC in the 3458 itself, it is only the 1/f part of the reference noise that is really relevant.