Ultra Precision Reference LTZ1000

[bsw_m]

Are these hermetic? They don't look so.

The resistor body consists of 2 parts, which are hermetically glued together.
I would probably describe them as semi hermetic.

[mimmus78]

What's the size?

[bsw_m]

What's the size?

18.5x15.5x6.5 millimeters. Pin pitch 2.5mm

[maginnovision]

I have an LTZ1000A and right now I'm seeing random fluctuations at times. It'll run around 7.0903935V but I'm starting to see dips, sometimes suddenly, to 7.0903514V or ..611V, ..750V. These seem pretty extreme and way off the average. Is this thing dead, or is it my meter? I'm really considering buying another 7.5D to compare against because it seems odd to see these types of jumps especially with 100s integration times. Current is steady, and the ambient temperature is not changing that much. Maybe 10C over the course of today but many drastic dips. It has until now been very stable but also until today or yesterday my average was closer to 7.0904300V. That was a sudden change over about 4 readings and it has been low since but my other reference also suffered the same drop which makes me suspect the meter. 

[TiN]

And for those of us who don't have magic Palantir the meter is?
Sorry, but photos of setup, wiring and power source are mandatory for ppm healing by internet.. 

[maginnovision]

And for those of us who don't have magic Palantir the meter is?
Sorry, but photos of setup, wiring and power source are mandatory for ppm healing by internet.. 

So here is the setup and what binding posts and cables I'm using. In case it's hard to see... The Rigol DP832 is feeding both refs with 14V(readback is off, output is calibrated). The red one(1000A) has Pomona 3770 binding posts(copper/gold) and using Pomona 5291A copper/gold retractable banana plug cables straight into the Prema 5017SC. Inside the hammond boxes are teko boxes and inside those are the references, insulated from the boxes by non conductive foam.

[ap]

One thing that you may wish to observe is the Pomona cable/binding post combination. The cable plug insulation shells are spring loaded. I saw it happen that when inserted into the 3770 posts, the springs slowly pushed the plugs out of the posts. One would expect they don't, given this is same vendor.
What comes in mind secondly is external EMI interference. That is harder to trace though.

[maginnovision]

One thing that you may wish to observe is the Pomona cable/binding post combination. The cable plug insulation shells are spring loaded. I saw it happen that when inserted into the 3770 posts, the springs slowly pushed the plugs out of the posts. One would expect they don't, given this is same vendor.
What comes in mind secondly is external EMI interference. That is harder to trace though.

I just checked and the cables are still firmly in the binding posts. As for EMI, the Teko box is isolated from everything except for the green input banana jack. I have tried that earthed(not just to power supply ground) and it seemed to help some with noise(Can get 7-10 samples within a couple hundred nanovolts) but for some reason my kids have issues not playing with that cable so I do not have it connected right now. Also no, my kids have not been touching it since this started. The meter case is all earthed so it's fairly well insulated from outside noise. The cables and things could pick up some noise but I'd expect that to be more consistent since conditions don't change that much where these are really significant changes and it's not just 1 sample, it's 3-4 before it starts to return to normal.

[kj7e]

Are you willing to send your reference out?  If so, I could stick it on my DMM7510 for a few days and see if its your environment or the reference itself.

[maginnovision]

Are you willing to send your reference out?  If so, I could stick it on my DMM7510 for a few days and see if its your environment or the reference itself.

If nobody else has any ideas and it continues to spike like this I'd appreciate someone else checking it out so I might PM you about that.

[Dr. Frank]

Are you willing to send your reference out?  If so, I could stick it on my DMM7510 for a few days and see if its your environment or the reference itself.

If nobody else has any ideas and it continues to spike like this I'd appreciate someone else checking it out so I might PM you about that.

Hello,
if I understand correctly, you have 2 PCBs ordered from that external company, which probably copied TiNs design (or whatever).
And both PCBs show these downward dips of several ppm?
time constants of these dips are several seconds, up to 30sec?
(Sorry, but proper scaling is missing in your diagrams)

Then these references really suffer from external EMI disturbances, as discussed intensively before, in this thread!

TiNs design is mostly a copy of the original LT schematic, and this is by design susceptible to EMI.

I have seen such dips also in both of my prototype LTZ1000 references, I think I already have published these old measurements somewhere here.. and these look very similar.

You can try to isolate your PCB by a steel box, or whatever you like, but EMI enters via power supply, or output jacks, no chance..
Even some of TiNs measurements (some unfiltered ones) on his own references show such dips.

The improvement for this problem is, either to use the Datron solution (for standalone reference 4910), implemented by Andreas.

Or you try adding 2 x 100nF foil capacitors in parallel to the 1k and 120 Ohm  resistors, which will short these EMI spikes, which affect the sensitive BE diodes in either transistors of the LTZ1000, and will unlatch the oven regulation .

This simple measure improved my mentioned references a lot.

Anyhow, get rid of any SMPSUs in your lab!

Frank

[maginnovision]

Are you willing to send your reference out?  If so, I could stick it on my DMM7510 for a few days and see if its your environment or the reference itself.

If nobody else has any ideas and it continues to spike like this I'd appreciate someone else checking it out so I might PM you about that.

Hello,
if I understand correctly, you have 2 PCBs ordered from that external company, which probably copied TiNs design (or whatever).
And both PCBs show these downward dips of several ppm?
time constants of these dips are several seconds, up to 30sec?
(Sorry, but proper scaling is missing in your diagrams)

Then these references really suffer from external EMI disturbances, as discussed intensively before, in this thread!

TiNs design is mostly a copy of the original LT schematic, and this is by design susceptible to EMI.

I have seen such dips also in both of my prototype LTZ1000 references, I think I already have published these old measurements somewhere here.. and these look very similar.

You can try to isolate your PCB by a steel box, or whatever you like, but EMI enters via power supply, or output jacks, no chance..
Even some of TiNs measurements (some unfiltered ones) on his own references show such dips.

The improvement for this problem is, either to use the Datron solution (for standalone reference 4910), implemented by Andreas.

Or you try adding 2 x 100nF foil capacitors in parallel to the 1k and 120 Ohm  resistors, which will short these EMI spikes, which affect the sensitive BE diodes in either transistors of the LTZ1000, and will unlatch the oven regulation .

This simple measure improved my mentioned references a lot.

Anyhow, get rid of any SMPSUs in your lab!

Frank

The biggest spikes are roughly 7ppm. I don't have any smpsu except in the PC's which are about 3 meters away. The spikes took place with only the logging PC, the meter and the power supply on. All the lab equipment is on it's own mains circuit with its own battery backup. I put a 2uF metallized polypropylene capacitor(Kemet C4G, 850VDC) across the PS output shortly after my last post. The reason I wasn't suspecting EMI is that I haven't seen this behavior in the month of continuous logging. If it continues, or happens again, I will add the 100nF foil caps across R7/R10. Thanks for the tip.

[TiN]

Theory about reference picking noise from power supply is easy to check - connect reference to +12V battery, like old one from car/motorcycle and log for few days.
Properly working ref should have stability within 0.2ppm. To remove DMM out of equation - use two references and connect them in opposition, so your Prema would only measure difference between REF1 and REF2 on most sensitive DCV range (100mV?).

[Edwin G. Pettis]

No, no and no, putting capacitors across semiconductor junctions only transforms voltage spikes into current spikes and the microscopic matrices does not like that at all, it can even create cracks eventually but mostly decreases long term stability and increases noise.  That is the last thing you want being fed into a reference zener, we have addressed this point before earlier in this thread.

Regarding the Linear Technology reference design, it is just that, it is a starting point from which the design engineer uses to refine, tweak and improve the design for their specific purpose.  Authors have often pointed out that app notes are not finished complete designs but starting points that you have to finish for your own specific use.  To wit, shielding a sensitive circuit from interference is something that you have to take care of and putting capacitors in the wrong places is not a fix, shielding requires the proper enclosure(s), often copper (or maybe aluminum) inside steel and sometimes mu metal around components, aluminum is not particularly effective in shielding on its own, I believe we also talked about this earlier.  Improper construction techniques only adds to the problems and this includes connections to the outside world.  Removing noise generating sources to farther away also reduces the garbage being injected into sensitive equipment, sometimes moving them further away won't work either if they're really noisy.....rule of thumb......always take care of noise at its source.

If you want a reference that provides the best possible long term stability then you're going to have to invest in doing things the right way and stop taking shortcuts that will affect long term stability even if you can't measure it today.

[maginnovision]

No, no and no, putting capacitors across semiconductor junctions only transforms voltage spikes into current spikes and the microscopic matrices does not like that at all, it can even create cracks eventually but mostly decreases long term stability and increases noise.  That is the last thing you want being fed into a reference zener, we have addressed this point before earlier in this thread.

Regarding the Linear Technology reference design, it is just that, it is a starting point from which the design engineer uses to refine, tweak and improve the design for their specific purpose.  Authors have often pointed out that app notes are not finished complete designs but starting points that you have to finish for your own specific use.  To wit, shielding a sensitive circuit from interference is something that you have to take care of and putting capacitors in the wrong places is not a fix, shielding requires the proper enclosure(s), often copper (or maybe aluminum) inside steel and sometimes mu metal around components, aluminum is not particularly effective in shielding on its own, I believe we also talked about this earlier.  Improper construction techniques only adds to the problems and this includes connections to the outside world.  Removing noise generating sources to farther away also reduces the garbage being injected into sensitive equipment, sometimes moving them further away won't work either if they're really noisy.....rule of thumb......always take care of noise at its source.

If you want a reference that provides the best possible long term stability then you're going to have to invest in doing things the right way and stop taking shortcuts that will affect long term stability even if you can't measure it today.

As far as I know the Teko 273 tin plated steel boxes are made for RFI and EMI noise so I'm not sure if that's a shortcut? Do you have another suggestion? I'll go back and read up on the capacitors.

Theory about reference picking noise from power supply is easy to check - connect reference to +12V battery, like old one from car/motorcycle and log for few days.
Properly working ref should have stability within 0.2ppm. To remove DMM out of equation - use two references and connect them in opposition, so your Prema would only measure difference between REF1 and REF2 on most sensitive DCV range (100mV?).

I have run it on a battery and it was .14ppm better but as of now I'm just trying to get a basic idea of it running(looking for issues like these) so a power supply is easier. Although that was when I was just running them in a cardboard box waiting for other boxes to arrive so it's not really good data which is why I've deleted all but 1 of those logs per ref. After I've logged for a few more days, assuming it looks stable, I'll switch back to battery and log a week, then try measuring using both refs.

[Edwin G. Pettis]

Is your Teko case well grounded?  As far as being steel, that is a good first step.  If you are still experiencing excessive noise on your reference output, then the EMI is likely being conducted by the cabling to and/or from the power supply or DVM/scope.  The noise can be coming from the power supply (also conducted through from the power line) or the EMI is being radiated by something nearby.  Since you say that you did not notice these glitches in the previous month then something has changed since, something you added in the room perhaps or something new generating noise on the power line.  Sometimes large appliances such as refrigerators will generate power line hash or transmitters for instance.

The LTZ circuit is incapable of generating such noise itself aside from little 1/f noise spikes so when experiencing unusual noise on the output of the LTZ circuit look for outside sources and do no try to band aid the problem by putting capacitors where they don't belong.

If the spikes are still present on the output of the reference when operating on a battery, you've just eliminated one source, the line operated power supply.  Since you say there was only a 0.14PPM drop in noise then that is mostly just 'normal' noise your power supply is generating, then you need to look for the source of the spikes, start turning things off and seeing if the spikes go away.  What kind of lighting do you have, that is another wideband noise source if it isn't incandescent or Halogen.

[maginnovision]

The case is well grounded but on a hunch I decided to check the input shorted with a gold plated copper shorting plug. This is what I got and may well be what I'm seeing. Integration time is 10 seconds so this is only slightly over an hour but I still think it's pretty revealing. I guess I need to start a repair thread. 

[TiN]

I have to admit, I'm in Edwin's camp here on LTZ vs caps discussion. 3458A's A9 don't use magic caps either, for what it's worth (yes, not great design for long stability otherwise).
More often than not spikes and other instability is caused by measurement gear or wiring or environment (EMI/RFI/power coupling) than reference itself. Trying to make reference more robust against external world influence is kinda going backwards.

maginnovision.
If your references are indeed based on my design, I can only remind that my KX calibration/testing programme is still applies to these, if you willing to have 2nd set of eyes and dataset for reference.

[maginnovision]

Well I fixed the meter and now I'm roughly 2 uV from the ideal 7.0904622V varying +/- 3-4uV(max 7.090463, min 7.090456) over last hour or so without earth connected. Not bad if it continues like this since the meter is +/- that many uV with a short. I'll have to get a meter that doesn't use an lm399 at some point to really check stability or send it to one of you. I guess I could try my 1072, I don't know if that's more or less stable than an lm399. Thanks for all the suggestions and offers to double check.

[bsw_m]

I can not wait and perform dirty test for 13k/1k divider:
Applied from calibrator 7.000000V to input of divider and measured output by 6.5digit voltmeter. Measured output voltage from divider is 0.499988V with ambient temperature 24 Celsius. Then warmed divider to 60 Celsius, output voltage the same. Voltage changed to 0.499987V when temperature rise the 75 Celsius. When the temperature dropped to 24 Celsius output voltage is 0.499988V

[branadic]

No, no and no, putting capacitors across semiconductor junctions only transforms voltage spikes into current spikes and the microscopic matrices does not like that at all, it can even create cracks eventually but mostly decreases long term stability and increases noise.  That is the last thing you want being fed into a reference zener, we have addressed this point before earlier in this thread.

Unless I haven't read a few scientific publications presenting real world correlations of latice structure of silicon vs. current spikes I put this into the drawer "myths, fairy tales and fake news". Even though it sounds reasonable it's not truth in the same moment.

-branadic-

[iMo]

All you need is g..gle:
The Belorussian precision foil resistors:
http://www.ajax.by/rpf-2.html
http://www.ajax.by/rezist.html
http://www.ajax.by/rez_set.html

Update:
The "RFP-2 datasheet" with tempco measurement details:
http://www.atkp.narod.ru/PDF/RFP_2passport.pdf

[Dr. Frank]

I have to admit, I'm in Edwin's camp here on LTZ vs caps discussion. 3458A's A9 don't use magic caps either, for what it's worth (yes, not great design for long stability otherwise).
More often than not spikes and other instability is caused by measurement gear or wiring or environment (EMI/RFI/power coupling) than reference itself. Trying to make reference more robust against external world influence is kinda going backwards.

Hello Illya,

you compare A9 board, which sits double shielded inside the 3458A, to 'our' standalone reference application, which all are definitely exposed to external EMI.

That comparison is not feasible, as it's definitely a different use case for the LTZ circuit.

Datron, as one of only two designers and suppliers of an LTZ based 10V reference (e.g. 4910) definitely has applied several components for hardening the circuit against external EMI. One feature are one or two 100nF parallel to the BE of the temperature sensing transistor, the other is an additional RC filter at the 1013 opamp for the reference amplifier.
I assume, that you still have that 7000 reference, maybe you can have a reverse engineering look inside that circuit, and I bet you will find similar circuit (as Mr. Pickering was working for Datron, as well as on the 7000, afaik).

Your argument, NOT to make the LTZ circuit more robust against EMI ('going backwards') is kind of strange, what do you propose, instead?


I can only contribute my own experimental experience, that these additional components really suppress external EMI pulses, but w/o any degradation of the long term drift.
All of my 7 LTZ circuits show an annual drift of less than 2ppm from start, and less than 1ppm after an initial run-in time of about half a year, despite all having these capacitors fitted.


The claims of Mr. Pettis, aka MisterDiodes, are not plausible to me, because he/they did never give any experimental proof for their claims, neither any detailed description, how/where their LTZ circuits were used, i.e. whether inside a hermetically shielded environment, or as a standalone application.
Also, any links to scientific papers about such supposed 'lattice degradation' are missing.
Therefore, I still consider Mr. Pettis objections to this case as being baseless, irrelevant and even misleading.

Frank

[bsw_m]

http://www.ajax.by/rpf-2.html
http://www.ajax.by/rezist.html
http://www.ajax.by/rez_set.html

Yes, it is they.

[TiN]

My approach is rather to keep ref circuit with minimal support parts necessary. Shielding is important for ref like LTZ, with or without additional circuitry. Double shielding like in 3458A is not so special. If we found time and money to build multiple LTZ boards, going that last step to ensure happy environment for it isn't unexpected.
Even Fluke recognize this as normal with much more robust 732 and suggest doing critical measurements only on floating battery power.

Going by experimental data, my own results with 7000 are questionable, it produce some wierd data so I don't have confidence that overcomplicated 7000 design is actually good.

Adding caps feels like taking cheap 10ppm/K resistor, putting it in oven and calling result 0.1 ppm/K standard. Valid approach for low cost applications but it does not cancel resistor own errors, and we rely on assumption of oven stability instead of inherent resistive element quality. Maybe wrong analogy, but hope I shared the idea.

After all I never intended KX modules as finished product, it have zero protection and quite beginner unfriendly. Having direct unbuffered LTZ output to external world isn't good idea either. Adding output amp with bipolar discrete stage like on my FX design resolves many if not all of the issues from external loads. My main LTZ ref is triple shielded and on battery power.

My final note would be : the more designs, the better Let's not just focus on saying design A or design B is worse or better, as either approach has own design quirks.