[FT] xDevs.com KX LTZ1000A fairy-tale, or the story of little jumper.

[Andreas]

Hello Steve,

with 12K5/1K the nominal setpoint will be around 52 deg C nominal (+/-10 deg Tolerance).
I built all my 6 LTZ1000A with 12K5/1K (0.1% or 0.01%).
All are tested to environment temperature up to around 40 deg C. (38 deg C in the linked picture)
So for a maximum room temperature of 32 deg C I will be on the safe side.
( If I do not place the reference on a very hot calibrator. )

https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg928435/#msg928435

With best regards

Andreas

[Kleinstein]

It also depends on the thermal setup and zener current how much self-heating will rise the temperature above ambient. Especially with the A version, the self heating can be important - this is why it needs the higher set temperature. Working very close to the temperature limit also means low heater power.

Due to the resistive heater with a square law for heating power, the heater control loop will see a lower gain at low power. So a low heater power might want a slightly modified (e.g. more gain at low power) temperature control circuit to get the same level of temperature control as with higher set temperature. With the original circuit temperature control will be not as good close to the limit.

Of cause the is no problem using a slightly lower resistance for the divider.

[TheSteve]

Thank you for the advice guys. I'll probably stick with 12.5K/1K - my shop is generally quite cool.

[TiN]

VK5RC's KX tempco test ramp finally complete. 
I had to replace diecast box to twice bigger one, to fit his reference with it's box, to keep it as it was received. Still had to crook input LDO box on the top to have it all fit.



Concept is same, 50W TEC is attached on the bottom side of the box, cooled by CPU fansink. Whole thing controlled from K2510 in PID loop.



Feedback RTD, which is Honeywell HEL-705 Pt1000 jammed right against reference's inner metal box. Due to this I had to really crank gain of PID and make everything very slow, about 0.005C/minute otherwise whole loop oscillate like mad.



Reference's LDO power input was powered from K2400 SMU, set at +18.000VDC, 105mA compliance. Current consumption was around 15mA at +40C TEC BOX temp, and about 20.6mA at +22C. My KX reference design have two temperature sensors on PCB, which are MAX6610. One is near zener's opamp, second one is on bottom side next to LTZ. I've used handy K2002's to monitor temperature sensors. They were powered by 7805, which is in turn powered from 12V SLA. Due to long time I had to charge SLA from K2304 from time to time.



LTZ's output directly monitored by 3458A. ACAL DCV performed automatically on each 0.2C of ambient temperature change. Otherwise tempco of my meter is about 0.28ppm/K, so it was being corrected out.



Overall setup look:



And data graph plot. Blue and purple line are MAX6610 outputs, and one can clearly see how LTZ's oven temperature regulation reaching less and less margin, as temperature of far PCB end getting closer to center spot.



Final tempco of this run as result = -0.14 ppm/K ±0.02 ppm/K, which is quite bad for typical LTZ1000, to be honest. 

[Andreas]

Hello Illya,

a negative T.C. can be healed by either shortening the legs of LTZ1000A or by putting a (smaller 220-270K) R9 resistor.

with best regards

Andreas

[TiN]

There is nothing left to shorten on the chip, also you mean 400K R9? Afair, it's not on the module. However it can be added  .
Let's hear from the owner, what he want to do.

[MisterDiodes]

There is nothing left to shorten on the chip, also you mean 400K R9? Afair, it's not on the module. However it can be added  .
Let's hear from the owner, what he want to do.

TiN: Are you running 12k over 1k for heater resistor?  That might be a little cool for your setup - being all sealed up in boxes, etc. watch your heat flow away from LTZ - it can't be zero - your heater could be running out of operating headroom for that LTZ.  I wasn't sure if you've got a thermocouple on the LTZ itself.

There is something wrong there - that TC is much higher than what we've seen with PWW, both Edwin's and GR's - and Films.  It's usually closer to Datasheet TC value with recommend 13k / 1k to start...and then work to a lower value as your setup permits.

Do you have the new heater ratio resistors properly mounted on the PCB this time (not mangled like pictures before, sorry), and thermally close together?  I guess I didn't see the LTZ board itself, or I'm blind.

Have you done several thermal cycles?  Is this a brand-new LTZ or one you've used before?

Why the '2057 instead of recommended LT1013?  Not that it causes that high TC, just curious. LT specifically recommends against the '2057 for it's high current noise across the LTZ - it can be a source of instability long-term.  We -never- use AZ amps for these LTZ current drivers.

You might want to run the test circuit as the standard, recommended setup.  That's what gives us lowest TC and best stability, time and time again - at least we haven't seen one with this high apparent TC.  Usually its much closer to datasheet specs or better, at least that's our experience.

You might run a few more cycles and see if there's any indication of settling down??

[TiN]

That's a good lot of questions, great!

TiN: Are you running 12k over 1k for heater resistor?  That might be a little cool for your setup - being all sealed up in boxes, etc. watch your heat flow away from LTZ - it can't be zero - your heater could be running out of operating headroom for that LTZ.  I wasn't sure if you've got a thermocouple on the LTZ itself.

That's right, 12K/1K. Previous test reveal that oven runs out of margin bit over +42°C, so as result TC ramp goes only to 40°C. It's very clear when there is no headroom left, by obvious +42ppm/K zener TC.

There is something wrong there - that TC is much higher than what we've seen with PWW, both Edwin's and GR's - and Films.  It's usually closer to Datasheet TC value with recommend 13k / 1k to start...and then work to a lower value as your setup permits.

I agree with this. However it's not mine, so owner has to decide if he wants surgery. I can try one minor thing, but that will be next week as gear runtime required for other stuff now. The original idea of me checking VK5RC's ref was just calibration test/comparison. But I feel of providing single data point, even median without knowing tempco is half-ass job, so much longer testrun is done as result.

Do you have the new heater ratio resistors properly mounted on the PCB this time (not mangled like pictures before, sorry), and thermally close together?  I guess I didn't see the LTZ board itself, or I'm blind.

Sorry for bit of confusion, you are not blind. Mangled board (no offence taken, it's true ) is my test module, which is waiting for LTZ chip now, so it's not the same ref. VK5RC posted photo of his module here.

Have you done several thermal cycles?  Is this a brand-new LTZ or one you've used before?

Yes, it had about of week of cycles prior to the ramp on post above. It was aged for a month at VK5RC site in Australia, then he shipped it to me for tests and it was compared vs my other 5 LTZs for another ~300 hours, before any temperature cycles even start.

Why the '2057 instead of recommended LT1013?  Not that it causes that high TC, just curious.

Long story short - it's by historical reasons. By that I mean my first volt-nuttery entrance 3 years ago, when I've designed KX PCBA, which takes two single-opamp SO8 packages for the LTZ circuit. Back then I had less understanding about opamps, their performance measures, etc. So PCB physically incapable taking LT1013 without doing bodgewires. And it was not expected that anyone order my gerbers or make modules, but here we are. I'd design things bit different today, having negative supply rail, larger footprint to support multiple options for precision resistors, not just tiny Z202's, use LT1013, provision larger capacitor footprints for films...

LT specifically recommends against the '2057 for it's high current noise across the LTZ - it can be a source of instability long-term.  We -never- use AZ amps for these LTZ current drivers.

I'm not going to argue on this, valid points. Just so far on hobby level I doubt difference between 2057 or 1013 or other opamps can be distingueshed with equipment we have available, be it 3458A or null-meters and 732B, looking at the output. I have 4 my modules with LTC2057, one with LT1097 and unmodified HP3458A's A9 STD REF module, which I compare from time to time over last few years. I was not able to tell the difference from data in short-term noise on any of them without actually looking at the board.

You might want to run the test circuit as the standard, recommended setup.  snip...

Yep, but that would mean making new design, new boards, ordering another boatload of expensive parts and spending another year on testing everything. With likely the very similar unmeasurable difference outcome, as down to ppm level there are so many things that can go wrong. At this point in my life, I'd rather spend efforts on few other projects. Perhaps new wave of LTZ-nuts would be willing to try this road, so we can see more different and interesting designs here.

You might run a few more cycles and see if there's any indication of settling down?

Having reference return to original voltage after ramp down suggest not settle/unstability issue? If it's unstable, we would see different end voltage. Power cycle for 30 minutes (moving VK5RC's REF from TEC box to near LTZ bank and K7168 scanner, reference voltage restored to -0.4 ppm of original assigned value, including 50% error from different 3458A.

Attached full log graph, with rescaled ppm scale at +140/-2ppm. Clearly can see when oven failed to stabilize die temperature around 6pm April 10. Rest of the crazy stuff I was tuning PID for TEC controller before I got right magic and started test April 13.

[Andreas]

There is nothing left to shorten on the chip, also you mean 400K R9?

Hello Illya,

yes I mean the 400K resistor which is populated as R9 on my cirquit.
As a rule of thumb I calculate a 1 Meg resistor to compensate for -0.04 ppm/K.

Why the '2057 instead of recommended LT1013?  Not that it causes that high TC, just curious. LT specifically recommends against the '2057 for it's high current noise across the LTZ - it can be a source of instability long-term.  We -never- use AZ amps for these LTZ current drivers.

The 2057 has nothing to do with a bad T.C. My worst sample with -0.23 ppm/K uncompensated T.C. (LTZ#4) is built with a LT1013A.
Both 2057 samples have below 0.03 ppm/K T.C.

Do you have a application note or a serious source where LT recommends against the 2057?
Only notorious repeating without founded measurement data doesnt make it plausible to me.

with best regards

Andreas

[MisterDiodes]

Andreas:

Call LT analog apps engineering - compared to a '1013, that '2057 will have somewhere between 200 to 400 times current input noise as spikes on the inputs, @ switching freq  - and noise probably won't show up on the output due to a low speed amp, but will really show up on the inputs and power rails.

ON AZ amps: EVERY time the input pins switcher FETs change polarity, there is an injected gate charge current onto those inputs -That injected gate charge has to go to/ come from somewhere, and the only route open to it is to flow on the input pins, while even more charge has to be drawn from power rails.  That's the dirty little secret of all A-Z amps.  Yes the output won't show noise at switcher freq, but you'll find it everywhere else around the amp...be warned.  All A-Z datasheets will tend to gloss over this problem...AD has some A-Z amps with dithered switcher freq at least, and so they admit to the problem since they are trying to address it.

Look for short, tall input current spikes on '2057 inputs, at around 100kHz.  Also on power rails.  They can be in the uA (peak) range, and they are not included in the other noise specs at all.  The datasheet does briefly mention 0.7nA RMS input current noise at 100kHz...but that's not the story on current PEAKS: which are short and tall.

NOTE:  This effect is NOT modeled at all on LT-Spice.  In fact the switcher noise isn't modeled correctly for most of their AZ amps, '1043, etc.

That's not something LT measures for their datasheet, nor do they talk about it on their datasheet.  So measure for yourself - I have posted links to the measurement techniques here before, and you can see that effect for yourself - you don't have to find it in print or listen to me.  We have done those tests, but at client's expense and is not public - otherwise I'd be happy to show you our data.  Also, LT has confirmed to me that '2057 input noise spikes are certainly not the quietest for among common AZ amps either...but for some applications (resistor bridge input buffer) maybe the effect can be ignored.  Not so for LTZ, where the noise spike current is injected right onto the LTZ die - which you don't ever want.  Why would you choose to generate -more- noise across the zener and reduce LTZ stability for no real final benefit?

Will a '2057 work?  Sure.  Any better than '1013?  Nope.  So go with the much lower noise, recommend solution.  You have not invented any better wheel with a '2057 in this application.

For that reason, LT really doesn't want anything but LT1013 as the best op-amp to pair up to LTZ.  It really is the best middle-of the road compromise between  input current noise vs. voltage noise.  If you get an amp with low voltage noise you'll get high input current noise, and low current noise amps will have high voltage noise, etc.  Always a compromise.  '1013 sits right in that middle sweet spot, and that's why it's the amp that is the perfect fit for LTZ.

Oh - If you want to read more about it, check the Art of Electronics, 3rd edition (Horowitz & Hill), where you can find theory for you to learn more about A-Z amp noise.  Look in the chapters about Op-Amps, and Precision Amp circuits etc.  I think around chapter 8  - maybe.  I don't have page numbers for you but I know it's there - in fact the current input noise range they quote will be say 400 times that of a non-AZ amp...but our tests weren't quite that high, but in that ballpark.  The authors ran tests also, if you don't want to believe me.  Or just test for yourself.

Once you learn about input current and rail noise noise, those AZ amps used for LTZ current drivers aren't all as rosy as they seem, especially for long term stability.  They have their place for sure, but not as an LTZ current driver.

Can '2057 be used as a low drift voltage boost amp / buffer?  Maybe.  But again, watch out for those danged AZ switching spikes that get everywhere else if you're not ready for it.  We're working on a 24 bit fast ADC now, and at those high resolutions ...boy, it gets very tough to keep the noise down in the mud.  It is sometimes more forgiving if you have a radiometric input for your hi-rez ADC, but this one isn't and needs a very quiet absolute Vref - so low noise is critical.  It turns out that a '2057 isn't the best choice for that either - watch out for very low drive current from these AZ amps, and adding an LT1010 isn't always the best either.  Traps everywhere at high bit counts!!

Here's that link I posted before on a chopper noise measurement technique - with some modifications this strategy can be used to explore '2057 noise and compare to other AZ amps.  Have to use shielded can.  '2057 will not have lowest or loudest input noise compared to amps described here, but it's up there:

https://e2e.ti.com/cfs-file/__key/telligent-evolution-components-attachments/00-14-01-00-00-70-21-03/Chopper-Noise.pdf

[TiN]

Since you mentioned fast hi-res ADC, you got me here though. Chopper amp definitely have source to generate extra noise, but LT still recommend using 2057 as input buffer if near DC signal need to be digitized accurately for new generation of their 32-bit SAR ADC, such as LTC2508-32. What would be better solution for DC accurate preamp for such ADC frontend/reference buffer, to digitize low signal levels in 0.1-1kHz band (e.g. as servo system for DAC).

Also 2057 datasheet covers charge injection a little, having DC average current included in bias specification.

[branadic]

@ MisterDiodes

And what is your recommandation and experience on LT1013, which one would you use:

- hermetic LT1013AMJG which is e.g. still available at RS Components made by Texas Instruments (http://de.rs-online.com/web/p/operationsverstarker/7320847/)
- one of the original available Linear Technology LT1013AC/LT1013C/LT1013D/LT1013I

- DIP (CERDIP) or SO8?
- if DIP (CERDIP) with or without precision sockets?

[MisterDiodes]

Andreas:

Info in print for you - this way you don't have to believe me, and a good discussion of the theory of what's happening.  Make sure you have a copy of Art of Electronics, 3rd edition at hand.

Go to page 337 and you will see an example of an AZ amp noise measured at 400X the datasheet value, and look how much higher that is compared to a low current noise bipolar amp.

Check section 5.8 thru 5.11 for a pretty thorough discussion on A-Z amps and various problems (especially current input noise spikes) and watch out what happens if output saturates.

You will find some real spectra plots of current input noise that run all the way up to the switcher freq and above - something you don't really see on A-Z amp datasheets.  '2057 is mentioned in text in a few places but it is the similar  design as other LT choppers listed.

Also check 8.9 and beyond.

After that, test your own - Input current noise density typically WON'T be on the amp datasheets, or - like LT - they just gloss over it and mention an RMS value, which completely hides the fact that the noise is in the form of a short, big spike.

[MisterDiodes]

@ MisterDiodes

And what is your recommandation and experience on LT1013, which one would you use:

- hermetic LT1013AMJG which is e.g. still available at RS Components made by Texas Instruments (http://de.rs-online.com/web/p/operationsverstarker/7320847/)
- one of the original available Linear Technology LT1013AC/LT1013C/LT1013D/LT1013I

- DIP (CERDIP) or SO8?
- if DIP (CERDIP) with or without precision sockets?

We've had good luck with plastic DIP as is done in 3458a.  If you really want a metal can, they do have those from time to time at LT's site - around $48 each.  We've never seen any cost-benefit from those, but whatever your spec calls for.

From what we've done, regular DIP seems to have less problems mechanically and from vibration pickup, and solder straight to board - no sockets.  If you're worried about very small humidity effects, go for a hermetic box with a nitrogen or C02 purge valve, or semi-sealed box with silica pack, or vacuum deposition coating.  Only if customer desires those features though.  The cost to go that route just to chase down a small fractional ppm is very high.

Normally we don't worry about just using just a plastic DIP pak in a protective (emi and air draft) enclosure, and protect the LTZ itself from air drafts - just like HP has always done in 3458a and all seems to be fine for most any use on a production line situation.  Our products do go through periodic cal and adjustment, and are normally within spec.

We have used PWW resistors from G.R., Pettis, Riedon, Vishay and all work well with LTZ - Sometimes customers really want hermetic film and we'll do that also, but no real improvement in Vref operation cost per ppm stability.  If you're temp range is small enough PTF56's work on the cheap <Laughing> ..but at the cost of LTZ you might as well use good resistors.  For DC work PWW work well and cost is very reasonable compared to expensive film units, plus you get the exact values you want - but if you have a place where lots of high freq noise around you might have to use better box shielding or film resistors to reduce pickup.

Just depends on what you're going after. If you're chasing absolute lowest possible 1/f noise effects down to DC, then PWW resistors are where it's at.

[MisterDiodes]

Since you mentioned fast hi-res ADC, you got me here though. Chopper amp definitely have source to generate extra noise, but LT still recommend using 2057 as input buffer if near DC signal need to be digitized accurately for new generation of their 32-bit SAR ADC, such as LTC2508-32. What would be better solution for DC accurate preamp for such ADC frontend/reference buffer, to digitize low signal levels in 0.1-1kHz band (e.g. as servo system for DAC).

Also 2057 datasheet covers charge injection a little, having DC average current included in bias specification.

TiN:
That's funny you mentioned 2508-32, and I was JUST on the phone with LT.  I have to tred very lightly here on public forum, and I was asking them the same questions about Vref noise / '2057, etc.  I am under NDA with LT since we do some work for them BUT:

1) The 2508-32 is NOT a -true- 32 bit ADC, as they are quick to point out.  It has 32 bits max -resolution- with all filters turned on and only at max decimation rate, but that is NOT 32 bits true DC resolution.  This is NOT a poor man's 3458a voltmeter.  If your input signal or Vref noise is not Guassian, forget it - those SINC / Averaging filters won't help you at all.

2) They will discuss techniques trying to use a 6655 as Vref source, but putting a multi-stage '2057 LP filter at 0.8hz, and then you still need a buffer to buffer / drive Vref input on the 2508.  Which still leaves 1/F noise problems.  The only trouble is that only gives you a realistic measurement time span of some minutes, not even a 10 minute's worth of stability for repeatable measures.  OOps. where did that 100kHz show up from??  Ooops...that '2057 is getting noise into the Vref driver buffer.  Ooops...maybe a Bipolar buffer amp is better... Maybe we deal with a little bit of drift somehow.

OOPS - don't let that '2057 saturate in the filter, or it'll take some time to recover....Ooops...when it recovers it's got a little hysteresis... and so on.

I'm not saying '2057's are bad, but these are really tough problems in general at high bit counts.  Every bit of noise becomes a problem.

3) The clue on these ADC chips with very high sample rates (1MHz) and associated noise don't make them the best choice for an absolute DMM-style voltage measure device.  These are generally more for weigh scale / pressure transducer / seismic transducer where your sensor is typically a resistor bridge and driven by the same Vref the ADC is running from - so your input signal is always a RATIO of Vref.  That way your Vref can wander around a little bit and you've still got some fairly decent signal data.

All in all that 2500 / 2508 are probably good 16~18 bits chips if you need faster sample speed, but be careful:  Notice that the SNR and distortion aren't even good for 20 bits or so even at lower freqs?

Still faster than a 2400 / 2408 BUT one thing with those chips is the 4ppm INL is fairly predictable and curves trimmed out in software.  The 25xx. chips have lower INL but not as trimmable.

Yeup:  That still exposes the elephant in the room with Vrefs:  For stable absolute measures down into the sub-20~22bit range, you have to have a quieter and quieter Vref and NO 1/f noise (or as low as possible) if you want to make repeatable measures over longer time spans.  This is at the point where even LTZ's and 732a's aren't going to be good enough for a true, quiet 24 bits flicker-free at DC.  That ADC system is only going to be as good as the Vref in terms of noise and stability.  And that's one of the reasons why when all is said and done even the best 3458a is only good for several low ppm for DC accuracy, and they are not longer-term DC transfer accurate instruments.  Notice there is also a 24hr accuracy / uncertainty also that is much higher than what 8.5 digits seems to be offering.

NOW:  For one solution, I know how its done for critical absolute measures at an aerospace company for space vehicle instruments:  You don't rely on ONE 20 or 24 bit ADC.  You average not from one ADC / Reference but you have to average across 5ea. ADC's each with their own Vref.  So five different -systems- involved feeding the digital math filters.  And that will get you down to around a good 1 or maybe 1/2ppm stable absolute measure: say 20 or 21 bits.  It's hard to get past that.  It can be done... but you have to get more creative for reading an absolute voltage value without having a JJ-Array on hand.

You can try filters on the Vref with really, really low cutoff freqs, but then you realize how long it takes to stabilize after power up <Laughing>  and that still doesn't address 1/F noise.

If you look back at Jim William's Linear Tech app notes on true 1ppm voltage refs - notice there are always 3 or 5ea. 3458a's involved?  That's what it takes to get into those low uncertainties.

Did you notice the price of 250x chips?  What happens to the cost when you need 8 SIMULTANEOUS inputs?  Yikes!  The customer isn't going to like that...

[Kleinstein]

It should not be important which version of the LT1013 is used. So even the SO8 version should be OK - though the DIP version has advantages (less sensitive to board stress). The LT1013 is very convenient because it is single supply. If one has a negative supply available anyway, there might be alternatives (still BJT base and otherwise quite similar to the LT1013)  with slightly better specs, but the OPs don't make a big difference anyway. The amplifier following the reference circuit will have a much higher (e.g. 200 times) influence - so that choice is important.

The choice of resistors and the thermal layout (board and case) is much more important than a little less noise or drift of the OPs.

For compensating residual TC with R9  (the 400 K resistor in the LTC1000 circuit), this effect depends on the temperature setting and the thermal design. It uses the change in heater voltage as a temperature dependent part. This is more like indirect and nonlinear in temperature, especially with a low temperature setpoint. So there is no such universal number saying an extra 1 M gives a certain change in TC.

[VK5RC]

Hi TiN et al,
Thank you for doing so much work and publishing the results - it seems to be providing some interesting discussion points.
I would be happy TiN if you want to try anything with that unit e.g. R9 etc. Let me know the parts costs and I will happily reimburse you.
The LTZ is mounted a little high (I used forceps as a heat sink when soldering) so it is about 3-4mm above the board. I have a quote request in place for some of Edwin's 13k pww resistors to replace the 12k.
Robert

[Andreas]

Here's that link I posted before on a chopper noise measurement technique - with some modifications this strategy can be used to explore '2057 noise and compare to other AZ amps.  Have to use shielded can.  '2057 will not have lowest or loudest input noise compared to amps described here, but it's up there:

https://e2e.ti.com/cfs-file/__key/telligent-evolution-components-attachments/00-14-01-00-00-70-21-03/Chopper-Noise.pdf

Thanks for the link.

It confirms that I have it done all right in my design (mainly against EMI).
Every input of a OP-Amp is at least filtered with 22-100nF so that the poor LTZ
is not scared if I take a chopper OP-Amp instead of the LT1013.

With best regards

Andreas

[Andreas]

It should not be important which version of the LT1013 is used. So even the SO8 version should be OK -

Hello,

I have to disagree.
The SO8-version is only available in D or I grade which is specced with double value in offset drift.
When you regard the worst case with 5 uV/K and a temperature span of my lab of 18- 33 deg C (15 deg span):
this may give about 75uV drift in tempererature setpoint voltage corresponding to around 2 ppm drift in output voltage.

so the max 2 uV output voltage drift due to the LT1013 over a 50 deg C span is valid only for the "typical" value of the A-grade of the LT1013.

with best regards

Andreas

[Andreas]

I would be happy TiN if you want to try anything with that unit e.g. R9 etc. Let me know the parts costs and I will happily reimburse you.
The LTZ is mounted a little high (I used forceps as a heat sink when soldering) so it is about 3-4mm above the board. I have a quote request in place for some of Edwin's 13k pww resistors to replace the 12k.

Hello,
if you plan to do any changes on the key components it does not make sense to experiment with R9 now.
The T.C. sensitivity might change with a different setpoint.

The 3-4 mm above the board also let not much room for improvement.
I would let at least 0.5 - 1 mm play between the board and the body of the LTZ. (no board stress directly on the body).
From full length to nearly zero length I got about 0.12 ppm/K difference.

With best regards

Andreas

[Echo88]

@ MisterDiodes: Always a pleasure to read about your experiences, especially since i have the 2508-32-Evalboard sitting in a corner!

Maybe you can have a look at the proposed voltage-reference-filter-design mentioned here and share your opinion: http://electronicdesign.com/energy/filter-trims-ultra-precision-voltage-reference ?

Also: do you know any good ref-filter-design for a LTZ1000 regardless of the stabilizing-time?

Edit: Since the LTZ1000/Ref-Amps are the best voltage references out there apart from the superior JJA and companies like Fluke/Keithley surely like to improve their stuff: Do you know of any successful projects which improve stability/noise of those references or does it just not make sense and the only good way is to handselect and parallel those references like it is done in the Datron 4910?

[VK5RC]

Hi Andreas,
The other parts (13K) will be mainly for my other two LTZ references, re R9, if TiN is interested and has the time I thought it might be an interesting experiment. I can't get easily such nice data as he is able.
Robert.

[TiN]

I'll play with compensation resistor. Best case would be after this to leave this reference be, and use it as comparison vs rest of the references.
But that will be next month, as all my gear is tied up now, running other python snakes, and I have few more items to clear out first.
For time being it's attached as is to my scanner setup (2x3458A/K7168) with rest of the LTZ crowd for logging.

If we get tempco fixed (<0.05ppm/K), then VK5RC could use this known tempco ref to test his own 3458A tempco, so he could later "transfer" better stability into rest of his LTZ's.

[Edwin G. Pettis]

@echo88,

I remember this article, while there is some reduction in 1/f noise from this filter, the application of it with a LTZ reference is questionable.  First, the LTZ has significantly lower noise than any of the Maxim Vref chips, the lowest appears to be about 4uV P-P typical vs. the LTZ's 1.2uV P-P.  In the case of the LTZ, the filter would also increase the DC drift of the output noticeably, there are always trade offs to be made and frankly I doubt this filter will reduce the 1/f noise of the LTZ all that much.  In this case, the trade offs of perhaps slightly less 1/f noise vs increased DC drift is likely a washout at best.

I also question the measurements being done with a DVM, it is a sampling/averaging device by nature and therefore is not the best instrument for measuring P-P waveforms, an analog o'scope or strip chart is much better or perhaps a spectrum analyzer.  1/f noise averaged will not produce an accurate result and that is all a DVM can do.  Of course, comparing a DVM's averaged noise measurement has some small indication of 'total' noise but nothing more.  You simply cannot average out 1/f noise.

The best noise reduction technique is to use the lowest noise components in the circuit, since the LTZ represents the lowest in noise available, that leaves the other components, the LT1013 is intended to be used with the LTZ, it was designed with that in mind and the lowest resistor noise components are wire wound.  The next level of noise reduction is the circuit layout and power supply noise.  There are always trade offs in circuit designs and you have to pick the places in the circuit where you can reasonably reduce the noise, unfortunately filters in this case are very difficult to implement when you're working at such low frequencies and the DC output is also very important.

[Kleinstein]

The main source of 1/f noise in the LTZ reference it the LTZ1000 itself. So there is not much to gain from using super high quality OP or resistors. So the obvious choice would be a slightly higher current for the LTZ1000 or two of the references in parallel. Also an other, lower noise reference and slow compensation of the long term drift of the low noise reference might be an option.
 
This filter circuit adds twice the noise of the OPs. So for a low noise reference it needs lower noise OPs, which will cause more current noise and possible drift. The listed MAX4238 is specified at 1.5 µV_pp für 0.1 - 10 Hz. So you trade in a reduction of the 1.2 µV_pp by an additional about 2.1 µV_pp from the 2 OPs. So the OPs uses should be much lower noise, more like LTC2057.

The 0.1 -10 Hz frequency range is used in datasheets, because it can be measured in reasonable time and with analog AC coupling. However in many applications of a stable reference the actually interesting noise could be the 0.001 -0.1 Hz range. Filtering is even less practical here.

Using a DMM to measure the LF noise is not that bad. However one has to include the averaging of the meter as part of the filter that sets the bandwidth. With fast enough readings (e.g. 1 PLC) this should not be an issue. The more important point could be noise of the input amplifier - so it might need an extra amplifier stage or a low level meter like the 34420.