Ultra Precision Reference LTZ1000

[hwj-d]

OT: One can also, generally speaking, exaggerate. If that goes so far as to create a certain monoculture of opinion, our Forum Jewel has lost. In my opinion, the limit here is to represent what is technically reasonable without taking other opinions to absurdity from the outset. The learning effect for all the fellow readers is to justify this difference and not to prevent it. And this learning effect is the meaning of the whole, even above technical correctness, because it is only in the ensuing discourse that it becomes accessible at all.

[alex-sh]

For example the smd capacitor's V shaped impedance (https://www.avx.com/docs/techinfo/CeramicCapacitors/parasitc.pdf) below.

While the X7R 1206 10nF will block "best" say from 20-200MHz, the 10nF Wima MKS2 will block "best" from  12-28MHz.

Yeah, this is a very good point. I found it a hard way that one 10nF cap not equal to another 10nF cap. Or in other words depends on type/make/manufacturer.

[3roomlab]

we need the "jaxbird analyzer"
was any of you using it ?
https://www.eevblog.com/forum/projects/impedance-analyzer-build-and-experiments/?all

[eplpwr]

Do you have a link to those mu-metal-boxes?

Here you go:
https://www.ebay.com/itm/Mu-metal-nickel-iron-alloys-shielding-case-for-transformers/282939791744?hash=item41e0869980:g:S4cAAOSwt51auf2g

Rather small since they are made for transformers; not a box in the generic sense. I like the pre-made holes since I've read that Mu-metal may need re-annealing after being formed or machined.

The aluminiumbox doesnt need to be thermally coupled to the LTZ-board as no component generates so much heat, that it needs to be cooled. Also you generally want a good thermal insulation of the LTZ-board to the case and mechanical flexibility of the board to avoid strain-induced offset-voltage.
I suggest following AN159 on shielding box construction, for generally good shielding: https://www.analog.com/media/en/technical-documentation/application-notes/an159fa.pdf Page 7.

I guess I was somewhat unclear regarding the purpose of my aluminium "box" - it is to be a heated oven. I'm planning to custom-machine an aluminium slab to fully encapsulate a voltage reference, a heater of some sorts (like the '723-based version described in another thread) and possibly a step-up/-down board to get a specific voltage like 5V or 10V as output. Thus, my goal is to maximize thermal transfer to each component and have a large, hopefully isothermal, mass containing all the electronics. In practice, it will be at least three "slabs" that make up the whole enclosure, I'm thinking a top part with MOSFET heaters, a middle part with heater control and Vref, and a bottom part with MOSFET heaters again. I'm aware that a fully custom encapsulation will mean that every dimension, specifically for resistors, will be unique so there will be a need to machine new parts if I change from WW to BMF, for example. Circulating air is supposedly a bad thing, so I'm planning on very narrow encapsulation of the components (say, 1mm). My question is if it's good to leave it like that - with an air gap surrounding each component - or try to actually make thermal contact using some method?

I had some thoughts of other materials, like Ag or Cu, but I've read that machining is very difficult and would require casting. Since I've worked with aluminium before with good results I'll stick with that - I have a couple of CNC machines so I won't have to rely on my unsteady hands.

Hopefully, I've explained what I'm aiming for in understandable terms. 

[branadic]

There is no need to machine a custom aluminium case as simple aluminium square profile will do the job. You should put your effort more into the circuit design and components characterization.
If you don't know where to start take a look at videos like

and you can see the oven construction and how 732B is constructed and designed, also read the circuit diagrams on it. This will give you a brief idea how they made it.
The best solutions is the one that was shown to be working with low effort and a one others can duplicate to reproduce the results made with it.
If you would ask me, it's not worth the effort machining a custom aluminium case, even though it's a nice project from the perspective of CNC machining.

-branadic-

[exe]

we need the "jaxbird analyzer"
was any of you using it ?
https://www.eevblog.com/forum/projects/impedance-analyzer-build-and-experiments/?all

It's too low freq (500kHz). I used to use a signal generator and oscilloscope to do this. Basic arrangements are described here: https://meettechniek.info/passive/capacitance.html .

I did even simpler than that: I just connected oscilloscope to signal gen. Then I shunted the output and observed signal amplitude vs frequency. At some point amplitude stops falling. Afaik this is the resonant frequency of the capacitor and it cannot be used anywhere closer or above this frequency.  The amplitude should tell the ESR, as capacitor is essentially a voltage divider with 50Ohm signal gen. But I'm not 100% sure if this is a right arrangement because there was not termination, etc. This also does not give information about signal phase.

Probably, a better approach is just open the datasheet and see impedance curve. For smd caps reputable manufacturers prive this info. But this project doesn't use smd components .

I also considered buying this: https://www.sdr-kits.net/introducing-DG8SAQ-VNWA3 . But a bit too pricey.

[FriedLogic]

Hi,

I think that the real issue that Edwin and MisterDiodes were referring to is the general approach of putting any capacitor on the IC pin to kind of clamp down on the noise there instead of stopping it getting to the IC in the first place. 'Optimising' the capacitors might only make that aspect of things worse.

The other thing is: The lattice distruction theory using caps is not logic in this application. Every ac voltage across an impedance causes a current. If the voltage is essentially shorted by the cap (as it seems to be as the cap helped), most of the current is now flowing through the capacitor, and the voltage accross it (and accross the junction in parallel) is lower than without the capacitor as the capacitor presents an AC short (thats why it is used). So the junction is seeing a lower AC voltage (ideally close to zero) with capacitor than without, so the lattice should also be less affected by the disturbance. In this way of using the capacitor, the voltage is also transferred in a current (as the claim was), but the current generated through the capacitor is not through the lattice/junction, and any potential current through the junction, caused by the noise voltage applied, is reducted as the voltage is reduced /shorted.

That assumes that the noise is coming into the IC on the pin that has the capacitor on it, which may not be the case.

Now the extent to which any if this might affect the IC in a specific case is another question. Using capacitors like this may not be ideal, but it's possible that in some cases it does not have major negative effects - and references which see very little noise anyway may be one case.

[Kleinstein]

It is definitely better to keep out RF interference from the whole circuit - this also needs the right caps. However this is more like a separate topic better not discussed in this already long thread. It not only applies to the reference bus other circuits too.

It may sill be a good idea to have the circuit tolerant to some RF, so it is not upset so much. Here extra caps may help - however they should be more likely rather small as it's only the really higher frequencies that matter. From external shielding the RF power should be low enough that possible damage to the chip due to RF should not be an issue.
The problem with a cap just added base to emitter is that it also interferes with the stability of the control loop. Even if not oscillating is can cause a problem during turn on / turn off.  AFAIK the control loop is already in the normal form not unconditionally stable - adding extra caps only makes thinks worse and might cause significant overshoot in turn on or with an external short.  A problem with EMI is also that it can depend on details of the layout / parasitic inductance - so even a measurement is only valid for the one board and the DMM side to detect interference might also be susceptible.

[bsw_m]

First TCR test for resitors from Ajax: https://xdevs.com/ajax_tcr_test1_may19/
So 10kOhm (blue line) - this is old broken resistive element from resistor.

P.S.
I began to receive a lot of questions in a personal, related to the price of resistors and their acquisition.
I do not sell these resistors. You can write directly to the manufacturer: e-mail: rezist@ajax.by
But, if you have any problems when interacting with the manufacturer, I can theoretically assist in placing an order and purchasing resistors.

[Pipelie]

First TCR test for resitors from Ajax: https://xdevs.com/ajax_tcr_test1_may19/
So 10kOhm (blue line) - this is old broken resistive element from resistor.

P.S.
I began to receive a lot of questions in a personal, related to the price of resistors and their acquisition.
I do not sell these resistors. You can write directly to the manufacturer: e-mail: rezist@ajax.by
But, if you have any problems when interacting with the manufacturer, I can theoretically assist in placing an order and purchasing resistors.

FYI, here is the TCR result of TiN's data.

[Pipelie]

some results from the TCR measurements I did recently,  for comparison .

[Pipelie]

continue...

[3roomlab]

Update the HP A9 reference board clone (03458-66509):

does anybody know what to call the (gold plated?) J400/401 part in digikey? or know the part ref name/model? (the thru hole socket thingy)
*mmm ty @ MB (the post after this), I later saw CM post #2060. which had the link  Molex 22-17-2052
https://octopart.com/search?q=Molex%2022-17-2052

[Magnificent Bastard]

Update the HP A9 reference board clone (03458-66509):

does anybody know what to call the (gold plated?) J400/401 part in digikey? or know the part ref name/model? (the thru hole socket thingy)

I believe these are made by Molex (see attached PDF).

-MB

[3roomlab]

I was playing around with the outline of 03458-66509
I just realized the LTZ is not really sitting in the middle of the 2 screw hole
I checked on other photos and it seem to be "not an error" ?
the plastic cap however looks symmetrical ... hey whats going on 
Im tempted to put it straight in layout 

[Andreas]

Hello,

perhaps it has to do with the Pin8-marker (nose) on the LTZ-package?

with best regards

Andreas

[3roomlab]

Hello,

perhaps it has to do with the Pin8-marker (nose) on the LTZ-package?

with best regards

Andreas

to have the screw avoid the tab, the LTZ needs to be really way out approx 3-4mm more (blue X is original position). right now its about 1mm off. if the LTZ is turned, it sits center. I dont think we will get what the original story is about this 1 haha

[alex-sh]

I know 6 1/2 DMM should not be used to track LTZ1000, but given that I do not have any better one yet, what are the best settings used for monitoring LTZ1000 (or LM399)? Currently, I have it set-up as follows:

1) Filter on (Moving average, count - 100, window 2%)
2) NPLC: 12 (maximum) 

What do you think?


Thanks
Alex

[Kleinstein]

Reading a reference with a DMM will give a result that depends on the ratio of the external and meter internal reference. 6.5 digit meters tend to have something like an LM399 reference, so one would mainly see the noise of that reference.
A few Fluke meters have an  SZA263 / LTFLU based reference, even with only 6.5 digits. So this case may not be so bad.

What setting works best depends on the meter. It also sets which frequency range is seen most. Digital filtering can to a large part also be done later. So using less digital filtering will give more data an one can decide afterwords how much averaging is suitable.

[TiN]

I know 6 1/2 DMM should not be used to track LTZ1000, but given that I do not have any better one yet

Ideally you would want two LTZ1000 references and then monitor difference between them using 100mV (or whatever minimal range there is) with DMM. This will give you magnitude less error from DMM's own reference. You can ship then one of the references to somebody with calibrated high-end setup, to provide you accurate voltage of the reference. This way you can cheaply obtain ppm-level accuracy and resolution on your LTZ1000 boards, without spending thousands for 8.5-digit DMM.

[bsdphk]

No: Ideally you would have three references and 3 DVMs, so you can do 3-cornered-hat style modelling.

[alex-sh]

I know 6 1/2 DMM should not be used to track LTZ1000, but given that I do not have any better one yet

Ideally you would want two LTZ1000 references and then monitor difference between them using 100mV (or whatever minimal range there is) with DMM. This will give you magnitude less error from DMM's own reference. You can ship then one of the references to somebody with calibrated high-end setup, to provide you accurate voltage of the reference. This way you can cheaply obtain ppm-level accuracy and resolution on your LTZ1000 boards, without spending thousands for 8.5-digit DMM.

 

Ok, its time to place yet another Digi-key order
How long do you think the reference has to be burned in on the battery power before calibrating it with 8 1/2 DMM?

[alex-sh]

What setting works best depends on the meter.

I have Keithley DMM6500

[TiN]

burned in on the battery power before calibrating it with 8 1/2 DMM?

Depends on the reference, really, and resistors used. Usually community here determined 1000 hours is reasonably low number. Some reference examples drift even years after assembly. Some calm down and settle within a month. It is always good idea to build more units than you need, cherry pick the "golden" ones for your use, and then share the rest to community Cal Clubs .
For real good calibration at cardinal point, you would not even use 8 1/2 DMM , but something like verified other reference (F732 and alike) and nullmeter such as Keithley 155 or Fluke 845.

[alex-sh]

The build has been completed.
Now the reference will be burning for a few months. By the way, is there any advantage using batteries vs a PSU? I understand that batteries produce noise by the chemical reaction. Unfortunately, the PSU is not much better. I do not have a dedicated lab so during the day there are all kinds of intereference introduced via the PSU through mains supply during the day. At night it's much better. What's the best choice please? Any recommendations?
I went through AN124, but the cost of the circuit (Vishay XTV138M030P0A alone is £340+ and its not available) is absurd. FYG - NASA had some issues with this type of capacitor. There were tantalum wet slug capacitor failures in the Apollo telescope mount charger battery regulator modules.


Another question I have is on the temp oven. I have branadic's board fitting into the Hammond case and its not designed for oven.
What kind of inaccuracy may be introduced if I do not keep LTZ1000 in the oven at contract temp?