Author Topic: LM399 based 10 V reference  (Read 334371 times)

0 Members and 2 Guests are viewing this topic.

Offline imo

  • Super Contributor
  • ***
  • Posts: 2378
  • Country: 00
Re: LM399 based 10 V reference
« Reply #1075 on: January 02, 2020, 05:27:00 pm »
I think the KSFs here are a) to make the oven with say <0.15Kp-p stability, b) to make <1ppm/K "ratio TC" 7->10V divider, and c) to find popcorn free LM399s (difficult). A lot of challenges, indeed  ;)
« Last Edit: January 02, 2020, 05:54:25 pm by imo »
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 6764
  • Country: de
Re: LM399 based 10 V reference
« Reply #1076 on: January 02, 2020, 06:02:23 pm »
I don't think there is much hope to find LM399 really free of popcorn noise. It is more about finding some where is popcorn noise is not too high.

The  7 to 10 V divider can be the real challenge: the TC matching may the easier part. The really hard part is to have the ratio long time stable.
Especially with an oven one may not even need good TC matching.
For a possible check one could also consider having the option to get the raw 7 V out, not just the final 10 V.
 

Offline 3roomlab

  • Frequent Contributor
  • **
  • Posts: 901
  • Country: 00
  • can you tell me how to cal sesame street meter?
Re: LM399 based 10 V reference
« Reply #1077 on: January 03, 2020, 12:35:36 pm »
spreadbury "theory" applied to LM399
i think just like LTZ drift has been experimented on by spreadbury
i suppose LM399 could have some gain in low drift from this theory of low/no oven operation
since the entire circuit is going to be overnized? might as well as make LM399 part of it?
« Last Edit: January 03, 2020, 12:38:10 pm by 3roomlab »
overclocked CPU and GPU are a waste of energy and time, it is highly energy + calculation inefficient vs watts. there is an entire influencer industry milking users off it, they call it "premium" but lifespans are short, oxymoronic crap , more like single use.
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2473
  • Country: de
Re: LM399 based 10 V reference
« Reply #1078 on: January 03, 2020, 01:55:13 pm »
Hello,

fortunately most of the LM399s cannot read the papers (nor data sheets).
Usually the ageing is only the first few weeks and turns to a "random walk" afterwards.
This first ageing is most probably equalizing the tensions of the cement between chip and housing.

attached a ageing curve LM399#2 and #3 are pre-aged devices.
LM399#CH6 and #CH7 are from day zero for the first 2000 days.
As reference 2 LTZ devices (unbuffered) with some "accidents"
(see the sudden shifts of the short cirquits).

All measured against a LTC2400 with temperature compensated AD586.

with best regards

Andreas
 
The following users thanked this post: 3roomlab, Svgeesus

Offline imo

  • Super Contributor
  • ***
  • Posts: 2378
  • Country: 00
Re: LM399 based 10 V reference
« Reply #1079 on: January 03, 2020, 04:34:37 pm »
@Andreas: my conspiracy theory would be AD/LT is selecting somehow (before a burn-in) the LM399s based on popcorn or 1/f noise level. They sell parts with "LM399" markings to the general public, and the parts marked with "weird numbers" to the selected customers. Guess who gets what..
« Last Edit: January 03, 2020, 05:16:42 pm by imo »
 

Offline guenthert

  • Frequent Contributor
  • **
  • Posts: 333
  • Country: us
Re: LM399 based 10 V reference
« Reply #1080 on: January 03, 2020, 05:17:11 pm »
@Andreas: my conspiracy theory would be AD/LT is selecting somehow (before a burn-in) the LM399s based on popcorn or 1/f noise level (or Vref<7V?). They sell parts with "LM399" markings to the general public, and the parts marked with "weird numbers" to the selected customers. Guess who gets what..
  Not quite and reality is not quite as nefarious.  There are companies who offer a selection service and it stands to reason that AD/LT is one of those.  HP (erm, Keysight) is one of those major customers who pay extra to get units with tighter/guaranteed specifications (and a customized label).  Those units are selected from the general pool.  The open market gets those which don't quite meet those tighter limits, but also - and that might be the overwhelming majority - all those units which haven't been tested at all.  You're free to purchase a set and select the best and sell the rest on Fleebay (which I think some do).
 
The following users thanked this post: TiN

Offline imo

  • Super Contributor
  • ***
  • Posts: 2378
  • Country: 00
Re: LM399 based 10 V reference
« Reply #1081 on: January 03, 2020, 05:44:41 pm »
@Andreas: my conspiracy theory would be AD/LT is selecting somehow (before a burn-in) the LM399s based on popcorn or 1/f noise level (or Vref<7V?). They sell parts with "LM399" markings to the general public, and the parts marked with "weird numbers" to the selected customers. Guess who gets what..
  Not quite and reality is not quite as nefarious.  There are companies who offer a selection service and it stands to reason that AD/LT is one of those.  HP (erm, Keysight) is one of those major customers who pay extra to get units with tighter/guaranteed specifications (and a customized label).  Those units are selected from the general pool.  The open market gets those which don't quite meet those tighter limits, but also - and that might be the overwhelming majority - all those units which haven't been tested at all.  You're free to purchase a set and select the best and sell the rest on Fleebay (which I think some do).
An lawyer would say "KS is not using an LM399H/AH in their DMMs".. :)
 

Offline Andreas

  • Super Contributor
  • ***
  • Posts: 2473
  • Country: de
Re: LM399 based 10 V reference
« Reply #1082 on: January 03, 2020, 07:54:53 pm »
@Andreas: my conspiracy theory would be AD/LT is selecting somehow (before a burn-in) the LM399s based on popcorn or 1/f noise level. They sell parts with "LM399" markings to the general public, and the parts marked with "weird numbers" to the selected customers. Guess who gets what..

sounds familiar to me see:
https://www.eevblog.com/forum/metrology/lm399-based-10-v-reference/msg854683/#msg854683

so it must be true somehow.   >:D

with best regards

Andreas
 
The following users thanked this post: imo

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 10460
  • Country: us
  • DavidH
Re: LM399 based 10 V reference
« Reply #1083 on: January 03, 2020, 08:24:53 pm »
Not quite and reality is not quite as nefarious.  There are companies who offer a selection service and it stands to reason that AD/LT is one of those.  HP (erm, Keysight) is one of those major customers who pay extra to get units with tighter/guaranteed specifications (and a customized label).  Those units are selected from the general pool.  The open market gets those which don't quite meet those tighter limits, but also - and that might be the overwhelming majority - all those units which haven't been tested at all.  You're free to purchase a set and select the best and sell the rest on Fleebay (which I think some do).

Exactly because the noise and drift tests are very expensive because of the time they take.  So most common devices only had the short functional test.
 

Offline blackdog

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: nl
  • Please stop pushing bullshit...
Re: LM399 based 10 V reference
« Reply #1084 on: January 03, 2020, 09:15:28 pm »
Hi,

I want to show the oven that I'm going to use for this LM399 reference, this because I already have quite a few test hours in it and that the "gain" is above 150x.
So 15C temperature increase gives less than 0.1C change in the oven.
How good it really gets depends on how well the insulation is done for the oven.
By placing the furnace in a simple polystyrene box, I got more than 150x gain.

I showed this oven before on this forum, and the controller is equipped with a LM723 IC.
And the NTC is located in the 7mm thick bottom under one of the power MOSFets that heat the box.
The extra TO220 IC is a LM35DT, I thought it would be easy to measure the temperature of the oven, but its a bit of a drama temperaturesensor, noise as hell.  :-DD

The oven is made from an already used HF box from which I removed the connectors.
It is a beautiful box for an oven because, of the very thick walls, the bottom plate on which the MOSFets are mounted is 7.5mm thick.

The battery is for the impression of the size of this oven.


.
Here you can see the extra 5K NTC measuring the temperature in the oven, this NTC was connected to a 34461A DMM.
It is clearly visible that I made sure that the wiring is mounted against the oven body.
This has been done to prevent you from making wrong measurements by leaking energy from the connection wiring.
During measuring, the lid was closed and the holes covered with tape.


.
This picture is for the impression how much space there is left for the Reference circuit, the parts around the LT1010 buffer do not come in this box.
That leaves some extra space to place the parts in an optimal way.
This because I'm still thinking about the remarks to leave out the bootstrap capacitor filter and only use the noise reduction of the 4x LM399 parallel.
I can then add more insulation to the LM399 IC in the oven, because no filter capacitors are needed.


.
I also thought about not connecting the oven in the LM399 IC's, but this oven is not good enough for that.
It requires an oven with a gain well above 1000x.
That's why this oven should be placed in a flask, or you could opt for a double oven.
As far as I'm concerned, both fall outside my project objective.

My first tests with the two types of Nichion VR(M) 470uF capacitors are good.
Eventually the 35V versions look good after first forming the capacitors at 15V and after that I started measuring leakage current around the 7V voltage which is the reference level.
One of these 35V types is at less than 1uV at 7V across the 10K resistor series.
So this is not the bootstrap filter but 1 single capacitor!

I've been thinking about "maat" his remarks about the strange drift of the bootstrap filter he tested.
My experience is that it works well but I also came across some weird effects today.
That wasn't the filter, but the cause was the power sources I used for the tests.

Those were two different power supplies, an old Harrison and the Rigol DP832.
Both can't be used properly because of the 1/F noise of these power supplies.

Today I picked up some different types of batteries I ordered from the wholesaler.
I first had to get these batteries to LAB temperature otherwise the drift of these batteries also caused too many errors...

Now the 4x penlite are quieter than my 10V reference PCB with the LT1021 on it.

This is a picture of the measurement on the 16V 470uF bootstrap filter with my old Philips PM3424 microvolt meter.
The DC value over the 10K resistor is less than 2uV.
Oh yeah, of course I regularly check if the "null" level of this meter hasn't changed.



And this is for the impression of my 10V Quad LT1021 LAB using a capacitor bootstrap filter.
The PostIt note is from June 2018 and the measured value on the 3458A is from this evening with a well warmed up 3458A and after an Autocal.


And below a link to the website of Rubicon with clear information about the properties of capacitors.

If you look at the leakage current between 22C and 42C, that difference is only small.
You have much more advantage in properly forming the capacitor and selecting the one with the lowest leakage currents than to worry about maybe 20% extra leakage current at 42C.


But, uh... still a lot to think about.
For the sake of completeness, I'm not looking for the perfect 10V reference, but try to do the best I can with the stuff I've got.

Kind regards,
Bram
« Last Edit: January 03, 2020, 09:21:45 pm by blackdog »
“Two things are infinite, the universe and human stupidity, and I am not yet completely sure about the universe.”
 
The following users thanked this post: 3roomlab, Andreas

Offline imo

  • Super Contributor
  • ***
  • Posts: 2378
  • Country: 00
Re: LM399 based 10 V reference
« Reply #1085 on: January 03, 2020, 09:32:12 pm »
Does an electrolytic (or tantalum) capacitor need a minimum DC bias in order to work properly (EDIT: to be properly "polarized")? I ask while the top capacitor in the bootstrapped combo sees "0V" best case.
« Last Edit: January 03, 2020, 09:43:51 pm by imo »
 

Offline blackdog

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: nl
  • Please stop pushing bullshit...
Re: LM399 based 10 V reference
« Reply #1086 on: January 03, 2020, 09:37:20 pm »
Hi imo,

Actually, I don't know...
I don't know what manufacturers say about this.
Maybe some forum members have more information on this?

Kind regards,
Bram
“Two things are infinite, the universe and human stupidity, and I am not yet completely sure about the universe.”
 

Offline David Hess

  • Super Contributor
  • ***
  • Posts: 10460
  • Country: us
  • DavidH
Re: LM399 based 10 V reference
« Reply #1087 on: January 03, 2020, 10:35:27 pm »
Does an electrolytic (or tantalum) capacitor need a minimum DC bias in order to work properly (EDIT: to be properly "polarized")? I ask while the top capacitor in the bootstrapped combo sees "0V" best case.

No, electrolytic capacitors work fine with 0 volts of bias.  Noise and leakage depend on the capacitor so they may need to be tested for it.
 

Offline blackdog

  • Frequent Contributor
  • **
  • Posts: 607
  • Country: nl
  • Please stop pushing bullshit...
Re: LM399 based 10 V reference
« Reply #1088 on: January 04, 2020, 11:57:25 am »
Hi 3roomlab,

Do you want to tell us where the picture of the drift of the LM399 at different temperatures comes from.
Thanks

Bram
“Two things are infinite, the universe and human stupidity, and I am not yet completely sure about the universe.”
 

Offline FransW

  • Regular Contributor
  • *
  • Posts: 96
  • Country: nl
Re: LM399 based 10 V reference
« Reply #1089 on: January 05, 2020, 08:54:12 pm »
The drift rate at that temperature of 95°C is nowhere specified, but it is an Arrhenius calculation, according to e.g. P J Spreadbury: 'The Ultra-Zener.. is it a portable replacement for the Weston cell?' , Meas. Si. Technol. 1 (1990).

They demonstrated, that the drift of the LTZ1000 doubles with each 10°C increase of the oven temperature. At 55°C they measure typically -2ppm/year. 95°C would yield a 16 times higher drift.
....
So this mean one can expect about a 2 times higher drift rate for the LTZ1000 A due to the about 10 K higher set point needed. This could be a bit less if good insulation is used around the non A version.

Due to the square law for the heater, half the power needed means about 70% of the current needed. So there is some power saving for the A version possible, but not that much if external insulation is added.  A first point to save power and keep the drift low is choosing an oven temperature that is not too high for the planed use.

Just as a crazy idea: The heat lost from the transistor to drive the heater could in theory also be used to heat the reference - so maybe have the transistor on top of the LTZ instead of somewhere far away. This may need a modified compensation however, to really work at the low power end.

From: https://www.eevblog.com/forum/metrology/ltz1000ch-or-ltz100ach/msg2341581/#msg2341581
 

Offline 3roomlab

  • Frequent Contributor
  • **
  • Posts: 901
  • Country: 00
  • can you tell me how to cal sesame street meter?
Re: LM399 based 10 V reference
« Reply #1090 on: January 06, 2020, 12:37:03 am »
Hi 3roomlab,

Do you want to tell us where the picture of the drift of the LM399 at different temperatures comes from.
Thanks

Bram

it is my assumption (it is mutilated plot from LM399) based on the experiment of Mr PJ Spreadbury's work/data on finding drift relation of LTZ1000 vs heater temperature (ie : temp drop by 10C ~ half drift rate).
in that data, there are 2 LTZ samples for 35C, so my further assumption is 25C maybe even -10C are possible. and could be applicable to LM399 for low drift using external active thermal management

making low temperature circuit i think is worse than MSL safety rating ... i think RH need to be 0% and everything need to be made artificially dry ... RH0% baking 99C for 48hrs? bury everything to "hell" with 1kg silica for 10days?

but it is all assumption which i think could work
« Last Edit: January 06, 2020, 01:06:01 am by 3roomlab »
overclocked CPU and GPU are a waste of energy and time, it is highly energy + calculation inefficient vs watts. there is an entire influencer industry milking users off it, they call it "premium" but lifespans are short, oxymoronic crap , more like single use.
 

Offline dietert1

  • Frequent Contributor
  • **
  • Posts: 364
  • Country: de
    • CADT Homepage
Re: LM399 based 10 V reference
« Reply #1091 on: January 06, 2020, 08:50:46 am »
The measurements presented are not that conclusive to use them for extrapolation below 25 °C. Aging at 25 °C may be small because the parts used had been on shelf for years at room temperature (20 .. 30 °C). When you go below room temperature, aging may start anew.

Regards, Dieter
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 6764
  • Country: de
Re: LM399 based 10 V reference
« Reply #1092 on: January 06, 2020, 09:31:15 am »
The normal theory suggests that aging is slower at low temperature. However the data shown for the LTZ refs are not very conclusive.
Another difficulty with aging is that there are different processes / parts that show aging. At a high temperature the faster processes may have reached the final values and one only sees the slow processes, while at low low temperature one would see the effect of the faster processes with the slow part essentially at a stand still. So this can be more of less independent effects, especially after some burn in process.

The LM399 is made with a fixed temperature - if not using the temperature control, one could as well use a different more conventional zener with higher current, like the old 1N829A or the Chinese 2DW232. The main point of the LM399 is that it makes it easy to use a heated zener. That lower temperature slows down aging is no a new finding. So when designing the LM399 they probably had a reason to choose some 95 C and not 70 C or 50 C. If it would be so easy to improve, I would expect a slightly different version with lower temperature setting to be available too.
 

Offline dietert1

  • Frequent Contributor
  • **
  • Posts: 364
  • Country: de
    • CADT Homepage
Re: LM399 based 10 V reference
« Reply #1093 on: January 06, 2020, 09:35:37 am »
Ease of use is also the reason for choosing a temperature near 100 °C, otherwise you need humidity control.

Regards, Dieter
 

Offline imo

  • Super Contributor
  • ***
  • Posts: 2378
  • Country: 00
Re: LM399 based 10 V reference
« Reply #1094 on: January 06, 2020, 10:08:41 am »
The LM399 and LTZ1000 have different internal wiring. While LTZ is basically a naked zener, the LM399's zener is embedded into a circuit with many transistors and resistors. That may influence the temp vs. aging too.  The zener's current in LM399 is ~0.25mA, afaik.

PS: the LM329 datasheet shows the Ref's schematics identical as the LM399's one.
PPS: Dobkin's paper shows the 30k and 10k resistors are trimmable, it could be they can set the zero TC for 329 and 399 accordingly.
« Last Edit: January 06, 2020, 11:59:06 am by imo »
 

Offline stijena1973

  • Contributor
  • Posts: 8
  • Country: hr
Re: LM399 based 10 V reference
« Reply #1095 on: January 09, 2020, 04:19:26 pm »
I doubt that one can calculate individual noise from pairs.

Indeed you can :) Click :

www.intersil.com/data/an/an177.pdf

The calculation is not difficult!

The measurement is a pairwise sample DUT noise density over frequency. Because the assumption is that the samples are statistical independend the noise power at sa specific frequency is simply the addition of the noise power of the DUT pair.

If you have n=4 samples you can make n*(n-1)/2=6 pairwise measurements. I will call the reference sample spectral noise power at a specific frequency simply a,b,c,d.

What you measure with the pairs is the sum of the noise powers

a+b
a+c
a+d
b+c
b+d
c+d

You get the noise power of the first sample reference by the following expression:

a=(1/3)*(((a+b)+(a+c)+(a+d))-(1/2)*((b+c)+(b+d)+(c+d)))

a=(1/3)*((3*a+b+c+d)-(1/2)*(2*b+2*c+2*d))

a=(1/3)*((3*a+b+c+d)-(b+c+d))

a=(1/3)*(3*a)

a=a

For the general case n the expression is:

a=1/(n-1)*(sumwith(a)-1/(n-2)*sumnot(a))



The application note from LT does not give a calculation example. It is only stated that you have to use a much better reference as a DUT partner.


This method is called "three-cornered hat" method. Of course it can be four-cornered etc.

 But minimum is three, and pre-condition is that they are statistically indepedent.
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 6764
  • Country: de
Re: LM399 based 10 V reference
« Reply #1096 on: January 09, 2020, 09:30:18 pm »
Calculating the noise of individual reference from measurements in pairs is rather difficult, as the noise is not constant in time  :-DD. So the individual noise amplitudes will scatter quite a bit. So the cornered hat version is not really practical, as the individual measurement will contain quite scattering.

There are usually 2 practical method to check the noise:
One is to amplify the noise with an AC coupled amplifier. So there will be lower frequency cut off, at maybe some 0.1 Hz.

The other method is using a known lower noise reference to subtract. A first approximation could be using the average of several (e.g. 5) references as the point to compare too. Once really bad units are out, one can assume the average is lower than the individual units.
 

Offline ch_scr

  • Contributor
  • Posts: 39
  • Country: de
Re: LM399 based 10 V reference
« Reply #1097 on: February 14, 2020, 08:50:57 pm »
I have finally built my east-german-themed burn-in rig for my czech LM399-clones. Some western parts sprinkled in as per tradition.
928404-0
It is based on a B589(AD589 clone) Ref / OP07 / 1k shunt current sink and one quad NPN array / three six-fold p-fet arrays used as current mirrors to supply 15 current sources tied to one reference.
928400-1
The B341 and U105 IC's were made in east germany but there are still some on ebay. I tried my best to have star ground / supply where applicable. Kicad project zip attatched.
Maybe the best thing that has come from this folly is that I have found a socket that really grabs the references. And I mean grab, no voltage fluctuation because the reference does not move after insertion. I believe they are "precision sockets" (like the turned ones but with additional inserts). As they are from the parts bin, I have no name or number, but the pictures of precision sockets look like it. As I tried to match the pitch of the references on the pcb the sockets have to be broken apart, soldered in one-by-one and insulated with heatshrink...
How would you set the measurement cycle on a 34970A? I thought to dwell on one reference(call it #1) for like 8 hours, then switch through all other ref's (#2 to #15), measure voltage there for short amount of time and stop on #15 for the next 8 hours. Now measure #1 - #13 for a short amount and dwell on #14; and so on, to minimize relay actuation while having long and short term data on all references.
 
The following users thanked this post: Andreas

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 6764
  • Country: de
Re: LM399 based 10 V reference
« Reply #1098 on: February 14, 2020, 09:09:59 pm »
The current mirrors from discrete transistors usually don't work very well. even for less precision needed, one should add some resistors at the emitter / drain.  Even with transistor arrays the mirrors tend to be not the precision wanted to test references.

The more normal circuit would be to wire one of the LM399 with an OP to create a reasonable stable 10 V and than use resistors from that 10 V level. Even if just an TL431 is used for the 10 V level, this may be more accurate than the current mirrors.
The heater part usually likes to have more than 10 V anyway.

For the reading cycle, I would consider just switching to the next reference at a regular rate, like every 10 or 100 seconds.
It kind of depends on the effort spend in separating the data later and the MUX used. The obvious choice here would be CMOS MUX chips like DG408 (or even old  CD4051), that are good enough for a 7 V level.
 

Offline ch_scr

  • Contributor
  • Posts: 39
  • Country: de
Re: LM399 based 10 V reference
« Reply #1099 on: February 14, 2020, 09:24:47 pm »
The heaters are supplied by two 7815 . Good point about the emitter / drain resistors. I have some CD4067, that would allow fast cycling without stressing the relays in the 34970. Did not think of that, just took the relays as given.
Edit: Just realised, U105 has the drains connected inside the ic - no resistors there.
« Last Edit: February 14, 2020, 09:49:38 pm by ch_scr »
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf