Ultra Precision Reference LTZ1000

[essele]

I thought I'd post my exploits with building a battery-backed LTZ1000 reference unit ... this has been done with a huge amount of input from this forum -- so a massive thanks to just about everyone! I'm sure there are a long list of things I've done wrong, a few I've worked out myself, and probably hundreds I'm not aware of ... any improvements would really be welcome.

This is a board based on the 10V one from Dr Frank, and using a (very slightly) modified version from the one that Cellularmitosis posted (and hosts on GitHub.)

The reference board is contained in the Teko 393 tuner enclosure which you can see in the middle of the outer enclosure. I used modelling foam which should be a good insulator to build the internal compartments.

The power supply board, which is really meant to keep the batteries charged, is based on a tps7a4901 along with another LDO 3.3v regulator to supply the front panel. The batteries are charged in parallel pairs using MCP73831's powered from isolated DC/DC converters, all powered from a mains AC/DC converter.

The batteries should last three or four days for the reference, and the AVR uses an average of 0.5mA in non-alarm mode, and 1.5mA in alarm mode, so that will last for weeks (it has a separate battery.)



The rear panel is a PCB which I only used since it's easier than trying to cut holes in anything else.



The front panel is also a PCB, but has an atmega328p on the back along with a couple of resistor dividers for the analog inputs, two rear-mount LED's, and a small SMD piezo sounder. This keeps an eye on whether external power is connected or not, flashes the LED's to show battery levels when charging, and then alarms if the battery level starts to get low ... it also gives you a helpful set of tones when mains is applied or removed (yes, I had fun with that!)



I've made up a few reference boards with different resistor sets on (this one is Sfernice 2ppm ones from eBay, but I also have a 5ppm version, and one with Edwins PWW's), so my intent is to build out a few more enclosures ... but I've got a few lessons learned from this one to include...

- The LiPo chargers get a bit warm, so I will see if I can move them to the other side of the board so they are closer to the vent holes ... this isn't really an issue since it's only really a problem if the batteries are particularly low (I could also reduce the charging current, but it's only 170mA at the moment anyway.)

- I get some blips when turning power on or off ... they can be in excess of 10uV ... I'm not sure if this is related to changes in the 12V power supply or some kind of noise. I need to do some more experiments ... interestingly there aren't showing up since everything is in the enclosure??

- I used a simple header and crimp connector on the front panel with no polarisation, and 16.8v on one pin ... yes - one dead avr. Some polarisation needed, would probably switch to an IDC one with a box (also saves the crimping!)

- There are no power switches, so nothing can be switched off ... so to turn off the alarm once the battery gets low (assuming you don't want to plug the power in) is a big disassembly job. I think a couple of rear mounted slide switches on the front panel with cut-outs should work -- one for the 12V supply to the ref, the other to turn off the AVR.

- I soldered the wires onto the Pomona 3770 jacks ... it didn't go particularly well, will probably need to redo them ... do others solder, or use crimp terminals?

And finally, I can't really measure the stability of any of the references yet ... my 34470A seems to be awful, it wanders all over the place (well 1-2ppm in the space of a day, lots of noise at the start of any temp increase), on ratio mode (which hopefully eliminates any issues with the meter) with a Datron 4000A it seems much more steady, with only a very minor temperature based delta ... but I have no idea which of them is the cause yet. I have a Prema 6048 which is away for calibration with Keysight, but they are taking an eternity, I think it's the first one they've had. Next project is a switching unit, so I can do some sensible data gathering from multiple devices.

[MiDi]

It looks very nice 
Are there (invisible) earth connections directly screwed on all parts of metal case?
If not, you should add them for safety.

[essele]

It looks very nice 
Are there (invisible) earth connections directly screwed on all parts of metal case?
If not, you should add them for safety.

Thanks -- I missed that off of my list of lessons learned ... the rear panel actually has a copper pour and through hole plating on the four main fixings and one of the fixings for the power supply board, so that was supposed to do the trick, however it's not working. I'm not sure why yet ... I was just enjoying the thing being in one piece for a while before I dismantle it again!

Actually I hadn't really considered the top and bottom panels (which are separate) ... so maybe this isn't a good case choice, they do a single piece version which I think has the same dimensions, just may be more challenging for assembly.

EDIT -- so I mounted the back panel the wrong way around -- sigh, and so the plated hole that supposed to line up with the earthed stand-off doesn't ... I'll need to switch that around, but still doesn't solve the top and bottom issue. And actually the IEC socket fixings are earthed, so they would have been a better place to put the through hole plating (which I didn't!)

[MiDi]

And actually the IEC socket fixings are earthed, so they would have been a better place to put the through hole plating (which I didn't!)

That would have been my next suggestion 
I am not shure if the earth connections for the sides are appropriate, I guess not.

[alex-sh]

- I soldered the wires onto the Pomona 3770 jacks ... it didn't go particularly well, will probably need to redo them ... do others solder, or use crimp terminals?

Very nice! Well done mate
Pomona 3770-0 - I use these ones too for my LTZ1000 project and I solder wires to them. Make sure you have the correct heat temp and they absorb a lot of heat.

[martinr33]

It is really hard to measure the performance of a good reference. For example, you could use a 3458a, but it doesn't have the stability of a dedicated 10V reference, so it is hard to tell what you are looking at.

If you build a second unit, the delta between them can give you some ideas.

[essele]

Ok, so I finally got my Prema 6048 back from Keysight, all calibrated and happy.

I'm still working on my USB switching unit, so I've just done some overnight measurements comparing the 34470A and the 6048 measuring the reference from the posts above.

The top orange line is the measurement from the 34470A and the blue line is the 6048. The lower orange line is the temperature ... I pretty much melted the BME280 when trying to get the bloody thing to solder onto my board, so I'm not entirely convinced about the absolute temp, relative seems to be ok though.

The scale for the main measurements is ((v/first_v)-1)*1000000. So this should show ppm deviation from the original reading in each case.

You can see the 34470A is a bit all over the place (well, from +0.4ppm to -1.2ppm) although still easily within its 24h spec, it does show the marked different to the 6048 which is sitting in +/- 0.3ppm range.

The 34470A is in 100PLC, AZ on, fixed 10V range, no filter. Powered on for weeks.
The 6048 is on 20s sampling (8.5 digit), 20V range, no filter. Powered on for about 90mins at the start, so may still be warming.

Measurements are in a loop with no other delays, so that works out at approximately 1 measurement every 24 seconds.

Clearly there isn't enough information in this to make any concrete assertions, and the data set is a bit too short, but I thought it was an interesting initial set of data.

I should be able to get the switching going in the next few hours, then the intention is to sample two LTZ1000's and a Datron 4000A on both DMMs.

[HighVoltage]

Very interesting, essele!

The 34470A has some known flaws.
My biggest complaint was the ACAL drift.
Actually the 7 1/2 digit 34470A is much worst than the 6 1/2 digit 34465A, even worst than some of my older 34401A.

Keysight is working on a solution....
https://www.eevblog.com/forum/metrology/keysight-34470a-calibration-and-acal-problems/

[maginnovision]

That looks pretty good! I wish Keysight could handle the 8017. I'm guessing the lack of english manual is why they "can't". In fact the only place I know of that can is Prema which means paying 350$ return shipping alone. Does it get more stable at higher integration times or nearly the same output?

[essele]

I’ve now got my usb switch working but I saw some very strange behaviour.

I was switching between two 10v refs, and reading one 20s reading from the Prema each time ... the readings slowly converged, very weird ... I initially suspected my switch (since I built it!) but after some testing it looks like there is some kind of filtering going on in the Prema (despite the digital filter being off) ... it really does look like a rolling average.

It seems to be related to the number of readings and not elapsed time, so I now run it in 4s continuous mode for 20s before switching to a 20s single reading and it seems much better.

I’ve just rehoused the second reference in a similar case as above (but v2 with some improvements) and I’m about to kick off a new set of measurements, I will try some variations of the above and longer integration times as well as it seems a bit more jittery than the single measurement above.

[notfaded1]

Which of the LTZ1000 reference PCB do you think is the most evolved and includes ability to do the 7V to 10V?  Or... do you think the 7.x to 10V is better done outside the reference container or inside it?  It seems a stable 7.x volt reference with boost to 10V trimmed output is the ideal. From your experience what do you suggest?

Bill

[maginnovision]

I’ve now got my usb switch working but I saw some very strange behaviour.

I was switching between two 10v refs, and reading one 20s reading from the Prema each time ... the readings slowly converged, very weird ... I initially suspected my switch (since I built it!) but after some testing it looks like there is some kind of filtering going on in the Prema (despite the digital filter being off) ... it really does look like a rolling average.

It seems to be related to the number of readings and not elapsed time, so I now run it in 4s continuous mode for 20s before switching to a 20s single reading and it seems much better.

I’ve just rehoused the second reference in a similar case as above (but v2 with some improvements) and I’m about to kick off a new set of measurements, I will try some variations of the above and longer integration times as well as it seems a bit more jittery than the single measurement above.

If it runs like the others then yes the filter includes a rolling average of 10(or more) readings. Ideally you should watch one for a time, then switch over.

I haven't looked but if you can force a trigger that resets the average and starts anew.

[Kleinstein]

The LTZ1000 reference is usually more stable than the 7 to 10 V step. So it makes absolutely sense to also have a 7 V output, not just 10 V. The 10 V stability is mainly dependent on the 7 to 10 V step. For the 7 to 10 V step there are several possibly option besides the choice of very stable resistors - so this would be better a separate thread.

In most cases there is no problem to have the 7 to 10 V step in the same container. It can make things even a little easier and could save on an extra buffer or wire compensation.

[notfaded1]

Here are my aged babies I want to build with.  They're the oldest ones I could find available.  My hope is they're more stable now and aged already.

Essele that's a beautiful build you put together there!  What batteries did you use there?

[Dr. Frank]

(Zener based) References only age over temperature (~Arrhenius), but practically not over time at R.T.
Therefore, just fire them up and monitor, how they perform.
50°C oven ~ -1..2ppm/year, doubles each 10°C. In continuous working mode, drift rate usually decreases over time, until Random Walk is main drift phenomenon.
Frank

[essele]

Hi notfaded1,

Thanks - I’m pretty pleased with them (there are two now!)

The batteries are from HobbyKing...

https://hobbyking.com/en_us/turnigy-2000mah-1s-1c-lipoly-w-2-pin-jst-ph-connector.html

... the first time I used them they were quite a bit cheaper, I do think it would be better to use a 4S pack now, but I’m still struggling to find an easy way of charging a pack. I also still need to look at the noise created when charging as there are a variety of switched mode supplies in there.

[essele]

Some more data...

I’ve switched to running at 20s integration times for 200 seconds before then grabbing a reading, this seems a much better approach, the filter is hopefully then not ‘contaminated’ with old readings.

The manual for other Prema’s suggests setting the integration time will clear the filter, so I think more experimenting is needed here as I’m fairly sure that’s not what I was seeing.

So here is a graph for overnight ... the orange and blue lines are the Prema readings for two LTZ1000’s and the grey and yellow are the 34470A.

After an initial settling down the two references seem to track each other pretty well for both meters — my interpretation is that the Prema has a negative temp coefficient and the 34470A has a positive - and I can’t really tell what’s going on with the LTZ’s other than they seem to be reasonably well matched. Does that seem like a reasonable interpretation?

I’m going to add a Datron 4000A into the mix and probably the 7v outputs as well (once I make up some more cables), although at close to 4mins per sample it’s going to reduce the data points a fair amount.

[babysitter]

The wrong foam could remove parts of the magic smoke, not as in "all magic smoke", more like local depletion. White foam doesn't increase my trust.

[essele]

Final graph for the moment ... I'm going to need to find/build a temperature chamber now!

So I've swapped all the dodgy test leads with gold plated Pomona double banana plugs and used copper test lead wire, this seems to have stabilised the 34470A quite a bit.

The attached graph is the 7v and 10v of reference 1, and the 10v of reference 2 ... I did run a test with a Datron 4000A but I was still experimenting with the Prema so it's not particularly useful, I'll need to do it again at some point (it had quite a few spurious multi-ppm jumps though!)

Very roughly the Prema seems to have a -0.2ppm/C temperature coefficient, and the 34470A seems to have a +0.3ppm/C, although clearly as I don't know the reference coefficients they could be better or worse than that, hence the need for the temperature chamber.

Confusingly the 7v output (dark blue, and grey) seems to be better on the Prema and worse on the 34470. I will run some analysis on the numbers when I get some time.

I have discovered a few things about the Prema as well ... I'll start another thread on that shortly .. basically it's not as granular as the manual suggests and relies on averaging unless you get to the top end of the integration times. So 4s for example (on 20V range), which is supposed to be 7.5 digits, the first reading after a filter clear is always a multiple of 5uV from any other first-reading. The included graph is with 80s integration times .. which seems to be ok.

[niner_007]

Working on my LTZ1000 reference I spent quite a bit tweaking the schematic and getting it nice

[Andreas]

C9 is wrong
C11+C12  too
Andreas

[niner_007]

Thanks, you're quite right, I haven't double checked the values yet.

[dietert1]

Yes, have been using a similar preamplifier for our HP 3456A, yet with lower gain (100x). Since the HP 3456A has a native resolution of 100nV, the preamplifier gave me a resolution of 1 nV. Our Fluke 8600 has a resolution of 10 uV, so 100x would result in 100 nV resolution, still about 0,01 ppm of 10 V. So i am not sure you need 10 000x. When using an OpAmp at 80 dB gain i would select a device with more than 120 dB open loop gain. And you may want chopper amplifiers with input filters in order to get a stable null.

Regards, Dieter

[MiDi]

Which Null-/nV-Meter has input current in fA range?
Usually they are in the pA range, in contrast to electrometers which need to be in the low fA range.
For measuring to the uV you want a front-end with low 1/f voltage noise and a very low TC including T-EMF.
LMC662 or other fA input bias current op amps have quite high 1/f voltage noise and TC.

[Kleinstein]

A DIY nullmeter is off topic in this thread, that is very long already. It may still be an interesting topic for a different thread.
The null-meters usually have a higher input current, more like pA to nA. It depends if the aim is more lowest voltage noise (e.g. nV range and below) or low input current, possibly to the single digit pA or possible also below. With a floating supply (e.g. batteries) it only take 1 amplifier at the input.  The rest is more like a low power ADC and LCD to keep the power consumption down.  For low bandwidth measurements digital filtering can also have advantages, like an FIR type response. So it may be an interesting project, but not in this thread.

Back to topic with the LTZ1000 reference: does the divider to set the temperature have to be  1K and some 12.5 K to 14 K  or is it possible to use higher resistance, like 2 K and 25 K.  I ask because it could than be possible to use a resistor array for the divider. A ratio of 1 : 13  (voltage divider to 1/14) could be made from 8 equal resistors, also in a way to use 2 arrays of 4 resistors without need for very good matching.