Ultra Precision Reference LTZ1000

[Kleinstein]

Looks like thermal regulation woks really well. The temperature of the chip itself seem to stabilize in just 1 second or so, even on the short step in between. Only the low temperature takes longer as the heater off state is reached with the rather large step down. There is a little bit of overshoot visible in the temperature (U_z and "Setpoint"), but not really much - so thats about how you want it when adjusting a regulator.

Its only the heating current that needs longer to reach a stable value - likely that is the time it takes for the case of the LTZ1000 to reach its equilibrium value.

[Andreas]

Hello,

in the meantime I did some measurements on LTZ1047B #3:

When comparing the unbuffered to the buffered output I recognized that the buffered
output had +24 uV more output than the unbuffered output.
-> much too much for the offset voltage of the LTC2057.

Putting 100nF across the buffered output made the both output voltages near equal.
-> obviously the output buffer picks up some noise from the wiring.

When soldering the 100nF across the buffered output I recognized on the unbuffered output that the
voltage drifted down by around -2 ppm (-14uV).
After soldering the voltage creeped slowly (around or less than 1uV/minute) upwards again.
Also a PSRR test with 14-18.5V instead of the battery showed around +0.5 .. 1 ppm drift.
Near the voltage regulator I measured a 2 deg C temperature increasement
The voltage regulator itself is stable at the output above 14.25 V input.
-> so most probably I have a thermal problem.

Is it the LT1013A with his hermetically housing (and KOVAR Pins)?
-> will have to do additionally tests.

With best regards

Andreas

[Kleinstein]

The OPs at the LTZ1000 are not that critical, when it comes to offset dirft or offsets from thermal EMF. So at least from the Offset it is not needed to use something like the LTC2057 here. The transistors in the LTZ1000 give a gain of about 100-200 - so something like a 100-200 µV offset from the OP would only result in 1 µV at the output or 1 µV at the sensing PN jinktion giving about 0.5 mK in temperature, which at 50 ppm/K would give someting like 0.2 µV at the output.

With AZ OPs, it might not be a good idea to use more than one in a circuit, as the AZ OPs cause some noise in the range of there chopping frequency and they are a little sensitive to noise in this range. So if usuing more than one AZ op - good filtering might be essential to isolate them. Also remember that HF interference (e.g cell phone, FM radio) might cause offsets. So to be really on the safe side, a closed case with proper HF isolation (ferrites+ (feedtrough)-caps would be the way to go.

The 22 Ohm at the output of the buffer for isolation is not that much - a higher value is likely a good idea. You don't want to drive much current from that OP anyway, as this would heat up the OP and can cause an extra Offset due to internal heating.

[Andreas]

Hello,

The OPs at the LTZ1000 are not that critical,

Good that you remind me.
between my old design LTZ#2 and LTZ#3 the only 2 differences are the
- LT1013 PDIP against CERDIP
- relative short legs against long legs of the LTZ1000A.
So either I have overlooked the sensitivity on #2 or the long legs are the reason.

With AZ OPs, it might not be a good idea to use more than one in a circuit, as the AZ OPs cause some noise in the range of there chopping frequency and they are a little sensitive to noise in this range. So if usuing more than one AZ op - good filtering might be essential to isolate them. Also remember that HF interference (e.g cell phone, FM radio) might cause offsets. So to be really on the safe side, a closed case with proper HF isolation (ferrites+ (feedtrough)-caps would be the way to go.

So its time that you do a tear down of your design here.
I am shure that we can learn a lot of it.

The 22 Ohm at the output of the buffer for isolation is not that much

Both outputs (buffered + unbuffered) are tested with capacitive loads from 1nF to 10uF.
Contrary to the datasheet design there are no oscillations.

With best regards

Andreas

[Andreas]

Hello,

did some noise measurement on LTZ#5 after 2 weeks of temperature setpoint cycling.

see also here as reference:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg833804/#msg833804

and here for he setup:
https://www.eevblog.com/forum/metrology/ultra-precision-reference-ltz1000/msg844021/#msg844021

so it seems that the popcorn noise has changed to something like softer "noise bursts"
But if it really helps ironing out the noise down to the typical 1,2uVpp?
Lets give a further try.
Or will the popcorn noise return after a longer time at normal temperature?

one measurement with 200nV/div and two measurements with 500nV/div.
all in all I did 52 measurements with 10 seconds duration so every 170 seconds a event.

with best regards

Andreas

Edit:
drift on LTZ#5 during temperature setpoint treatment was -29uV
compared to -5uV on LTZ#3 without treatment. (only running 24/7).

[alanambrose]

Hi,

Well some more LTZ drift and noise results to throw into the pot...

This is of a '94 vintage 3458A ref board (unmodified) as measured by two old 3458As. Internal temps, line frequency (as ppk from 50Hz), noise and voltage (as ppm of nominal 7.086V) were recorded at 2 NPLC. This test is still running but I will probably end it after 3 days. I deliberately introduced some ambient temperature variations to try and calculate the tempco. (Ambient was measured with a K2000 using a K thermocouple - this gives fairly nonsense readings because of the lack of a proper cold junction but it does show when the ambient temp changed). Both meters were autocaled at the start. The numbers in the panel in the middle btw are the regression coefficients.

The results so far seem to be:

(edit: updated at day 7)

+ one 3458A has a drift of 0.6 ppm per day, one is fairly flat and on spec at 0.1ppm per day.
+ the meters have opposite tempcos (#1074 is an 002 but they're both so old this shouldn't make a difference)
+ the relationship between internal temperature and ppm DCV seems to be immediate and fairly linear although opposite for each meter
+ there a positive relationship between DCV and line frequency

There are 3 moving pieces here of course including the ref board that is being measured, so it is possible that some other interpretation is possible. I'm particularly curious regarding the relationship with line frequency. Any thoughts?

Regards, Alan

[Kleinstein]

Line frequency is usually rather constant. So this should not be a real concern. It is well possible to have a correlation between line frequency and line voltage and than an influence of line voltage (e.g. through heat from regulators) - also a correlation with time of day is possible. There is also a small possibility that a change in the integration time, e.g. from rounding could also have a  minimal effect - but this should be small and could well be highly nonlinear up and down. With something like 0.2 ppm/Hz this is a rather small effect as the frequency is usually only changing by small fractions of a Hz.

[alanambrose]

Hi,

Yeah, line frequency apparently varied from 49.857 to 50.189 (i.e. -0.3% to +0.4% from nominal) over the couple of days of that test. So if the co-efficient of frequency impact on DCV is O(10^-7), as it appears to be, then it wouldn't be critical. (*1) I have on my list to run the individual regressions to see where the R^2 is. I can't quite get my head around the interaction between line frequency, the internal crystal, the integrator, and the dual-slope algorithm. Probably greater minds than mine have a better picture

*1 - (edit) of course ~0.3 ppm per Hz x ~0.3 Hz freq change ~= 0.1 ppm - so it might have some minor effect

On another subject - Andreas are you now preferring the LT1013 or LTC2057?

Alan

[TiN]

I like Alan's posts, so colorful
My TEC in resistor temperature chamber died on me, so I have meter "free" and will run my LTZ module in attempt to compare.
I have to ask though, why 2 NPLC ? If I read scales correctly, your source tempco is around 1ppm/C, which is bad unless you are not measuring LTZ output?
I'd think NPLC 10 or 50 to be used for such a test, to rule out ADC.

Also with 0.9ppm/day (similar to 1ppm/day I had on second A3 PCBA here) I'd consider A3 PCBA "broken". You can check if it's A9 LTZ drift or ADC by running ACAL. If ACAL brings readings back to initial value, your LTZ is stable, and it's A3 PCBA problem.

SN18 also mentions that if unit even with 002 option remained cold for long time, reference change it's condition to "pre-aged" and need to be run burn-in for few weeks to get stability back.

[Andreas]

On another subject - Andreas are you now preferring the LT1013 or LTC2057?

Hello,

up to now no clear winner.
I will have to make more tests. (e.g. T.C. measurement)
 
The LTC2057 has advantages at very low power supply voltages (more headroom).
But I fear that it is more sensitive to EMI pulses (without the capacitors which I had to remove).
-> so another test to do.

but at the moment my largest problem are temperature gradients around the LTZ itself.
Depending on which side of the LTZ I apply my hot air gun the output voltage rises or falls.
So I will have to do better thermal isolation or at least a slot between the voltage regulator and the LTZ.
Perhaps also soldering the inner housing to the PCB to get better equalizing of the temperature.

With best regards

Andreas

[alanambrose]

>>> I like Alan's posts, so colorful

Thank you TiN, my Excel skills? I'm sure they stem from a childhood preoccupation with crayons.

[TiN]

Had my LTZ hooked and running overnight, reading stays well <1ppm window, with ambient change from 22.1C to 21.3C. Live data here.
Low noise ripple is ~4 uV, with pink noise jumps ~5uV sometimes. I'll heat up room when get back home tonight, to see the tempco in similar conditions to yours.
Meter sampling at 100 NPLC and AZ is on.

Hope this helps for your data analysis.

[alanambrose]

Hi TiN & Andreas,

>>> I have to ask though, why 2 NPLC ? If I read scales correctly, your source tempco is around 1ppm/C, which is bad unless you are not measuring LTZ output? I'd think NPLC 10 or 50 to be used for such a test, to rule out ADC.

I chose 2 NPLC to be able to measure ~DC-10Hz noise as well as drift. I agree with you that it picks up more meter noise that way - ~0.04 of RMS noise and ~0.15ppm of 'gain error' @ 7V reading if I read the specs correctly. I guess those will average out though and it is interesting for me to see when the noise spikes are - I have not located the source yet.

>>> Also with 0.9ppm/day (similar to 1ppm/day I had on second A3 PCBA here) I'd consider A3 PCBA "broken". You can check if it's A9 LTZ drift or ADC by running ACAL. If ACAL brings readings back to initial value, your LTZ is stable, and it's A3 PCBA problem.

Yes it looks suspect and I'm going to leave the test running for a while to see what happens to #1045. It's still got v1 of the slave processor firmware so that may have something to do with it. I have ordered the 8051 (which is where I think the slave firmware is) from Keysight but they are being a bit random - yesterday they sent me a box 50x30x20cm with one button in it < 1x1x1cm, ho ho ho.

Many thanks for running that similar test. I'm away for the weekend but will run the data through the same analysis for comparison when I get back. BTW my simple test setup below - no screening atm. The cover for the ref btw is 3d printed, double walled, and on the bottom of the board also.

Andreas re:

>>> The LTC2057 has advantages at very low power supply voltages (more headroom). But I fear that it is more sensitive to EMI pulses (without the capacitors which I had to remove).

Thanks for that feedback, I also found, on another board, that I had to remove decoupling from input and output to stop it acting up. In particular, a 100nF on the output resulted in a nice 250kHz oscillation. I'm trying to remember what I once thought I knew re xfer functions and stability I was thinking of trying the LT1013, which is why I asked.

>>> Depending on which side of the LTZ I apply my hot air gun the output voltage rises or falls.

Well that's a fairly severe test - that probably happens (I have not tried it) with say the HP layout?

Alan

[TiN]

Ouch Have direct connection to A9 terminals, without those terminal blocks. They can cause big thermal EMFs.
I set it 2 NPLC, will leave overnight. So far from data log I'm seeing variance ~0.5ppm or so:

Code: [Select]

05/02/2016-20:23:23;[     829]: 7.136742640 , dev -0.5409 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.16 hPa
05/02/2016-20:23:23;[     830]: 7.136742906 , dev -0.5036 ppm, T:33.7 , EXT_T:21.43 , RH:74.32 , Press:1019.18 hPa
05/02/2016-20:23:23;[     831]: 7.136745218 , dev -0.1796 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.17 hPa
05/02/2016-20:23:24;[     832]: 7.136743529 , dev -0.4163 ppm, T:33.7 , EXT_T:21.42 , RH:74.31 , Press:1019.20 hPa
05/02/2016-20:23:24;[     833]: 7.136744685 , dev -0.2543 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.16 hPa
05/02/2016-20:23:24;[     834]: 7.136744062 , dev -0.3416 ppm, T:33.7 , EXT_T:21.43 , RH:74.32 , Press:1019.18 hPa
05/02/2016-20:23:24;[     835]: 7.136745041 , dev -0.2044 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.17 hPa
05/02/2016-20:23:25;[     836]: 7.136744240 , dev -0.3167 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.18 hPa
05/02/2016-20:23:25;[     837]: 7.136744774 , dev -0.2418 ppm, T:33.7 , EXT_T:21.43 , RH:74.32 , Press:1019.15 hPa
05/02/2016-20:23:25;[     838]: 7.136742284 , dev -0.5907 ppm, T:33.7 , EXT_T:21.43 , RH:74.32 , Press:1019.16 hPa
05/02/2016-20:23:25;[     839]: 7.136743618 , dev -0.4038 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.20 hPa
05/02/2016-20:23:25;[     840]: 7.136743796 , dev -0.3789 ppm, T:33.7 , EXT_T:21.42 , RH:74.31 , Press:1019.16 hPa
05/02/2016-20:23:26;[     841]: 7.136744685 , dev -0.2543 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.17 hPa
05/02/2016-20:23:26;[     842]: 7.136744151 , dev -0.3291 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.20 hPa
05/02/2016-20:23:26;[     843]: 7.136745041 , dev -0.2044 ppm, T:33.7 , EXT_T:21.42 , RH:74.31 , Press:1019.19 hPa
05/02/2016-20:23:26;[     844]: 7.136743084 , dev -0.4786 ppm, T:33.7 , EXT_T:21.42 , RH:74.31 , Press:1019.16 hPa
05/02/2016-20:23:27;[     845]: 7.136742906 , dev -0.5036 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.16 hPa
05/02/2016-20:23:27;[     846]: 7.136744596 , dev -0.2668 ppm, T:33.7 , EXT_T:21.42 , RH:74.31 , Press:1019.16 hPa
05/02/2016-20:23:27;[     847]: 7.136744418 , dev -0.2917 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.21 hPa
05/02/2016-20:23:27;[     848]: 7.136744329 , dev -0.3042 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.17 hPa
05/02/2016-20:23:27;[     849]: 7.136742995 , dev -0.4911 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.16 hPa
05/02/2016-20:23:28;[     850]: 7.136744062 , dev -0.3416 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.17 hPa
05/02/2016-20:23:28;[     851]: 7.136744062 , dev -0.3416 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.18 hPa
05/02/2016-20:23:28;[     852]: 7.136742817 , dev -0.5161 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.18 hPa
05/02/2016-20:23:28;[     853]: 7.136744329 , dev -0.3042 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.17 hPa
05/02/2016-20:23:29;[     854]: 7.136744062 , dev -0.3416 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.17 hPa
05/02/2016-20:23:29;[     855]: 7.136744151 , dev -0.3291 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.16 hPa
05/02/2016-20:23:29;[     856]: 7.136743973 , dev -0.3541 ppm, T:33.7 , EXT_T:21.43 , RH:74.31 , Press:1019.15 hPa

I need to change my 12V battery, so I could run ref with clean power to see if it can be better.

V1 firmware? What slave processor you talking about? MCU on ADC?

[alanambrose]

Yes, my thermal emfs are all over the place , I've been researching low emf cables etc but no conclusion so far.

>>> V1 firmware? What slave processor you talking about? MCU on ADC?

To put it another way, the answer to REV? is 9,1 which I believe is the masked rom in the 8051 on A3, which is what I ordered a week or so ago. If Keysight are consistant with their shipping 'system' they'll probably deliver this huge part in its own container.



Interested in your results on clean power, I'm assuming for the moment that my big noise spikes are power-related.

A.

[TiN]

So you ordered 8051 chip alone?
Mine REV? says 9,2. You got me thinking...

Meanwhile, 2 NPLC seem more noisy but span is very similar. No wild jumps as of now
Only if I moved cable, readings jump. Like now, just did it, deviation changed from ~0.6ppm to ~3.2ppm. So it's important to leave setup alone and have it located away from main working table which get vibrated/stressed often.

Use usual UTP ethernet cable with single strand wires, it's copper and very good for such connections. Clean copper wire fresh, without any oxidation, solder one end to A9 PCB test points. Crimp other end right to DMMs terminals. Keep them tight.

[Andreas]

>>> The LTC2057 has advantages at very low power supply voltages (more headroom). But I fear that it is more sensitive to EMI pulses (without the capacitors which I had to remove).

Thanks for that feedback, I also found, on another board, that I had to remove decoupling from input and output to stop it acting up. In particular, a 100nF on the output resulted in a nice 250kHz oscillation. I'm trying to remember what I once thought I knew re xfer functions and stability I was thinking of trying the LT1013, which is why I asked.

>>> Depending on which side of the LTZ I apply my hot air gun the output voltage rises or falls.

Well that's a fairly severe test - that probably happens (I have not tried it) with say the HP layout?

Hello,

most amplifiers will oscillate with 100nF (more than 100pF-1nF) directly at the output. (except some special OPs like AD826).
You will need a isolation cirquit with a 10-100 Ohms resistor and a feedback capacitor. (See my LTZ cirquit for the buffered output).

The heating test sounds more dramatic as it is.
The temperature that I use is 100 deg C and I do apply only for a few seconds over a larger area.
Unfortunately it is not good reproducable.

With best regards

Andreas

[branadic]

My board is now assembled too. Need to check each solder joint tomorrow and clean the board before powering it up for the first time. 

[Andreas]

My board is now assembled too.

Photo?

From my heatgun experiment I have made a diagram for LTZ#5
The heatgun is a hot air heater from a SMD rework station set to 100 deg C (minimum temperature).

21:59 heating south (below Pin 8 ) of LTZ
22:08 heating east (right from Pin 2) of LTZ
22:14 heating north (above Pin 4) of LTZ
22:19 heating west (left from Pin 6) of LTZ

peak amplitude is around 50uV = 7 ppm output change per heating.
heating west gives a negative jump. (all others positive).

22:25 heating the Z201 resistors north of the LTZ.
(not shure if that jump is from the resistors or more likely from LTZ).

The NTC north east from LTZ shows temperatures from 26-32 deg C (averaged temperature over 1 minute)

With best regards

Andreas

[Galaxyrise]

Well the setup wiring is ok so that leaves a strange operation of the chip.

The Vbe of Q1 is 20 mv greater than Q2 when they are all check by themselves. In a full, active circuit the Vbe of Q1 is 30-110mV below Q2. That's something to look into.

Together with the statement of Ken
it would explain that the Vbe on Q1 is higher than on Q2 with base + collector connected.

 The two transistors in the LTZ are *not* matched-- Q1's geometry is significantly smaller than Q2, and so much less current will be going through Q2 than the Zener when the operation point reaches equilibrium. 

So a possible explanation would be that the hfe of Q1 is significantly higher than that of Q2 for the active cirquit.
And the large stray of 50 mV from your samples in active cirquit could indicate that the hfe is very different between samples.

I attempted to replicate chuckb's results.  I used p-channel jfet+resistor current sources to drive Pin1, Pin5, and Pin8; and 2057s to drive Pin4 and Pin6.  I also provided for easy attachment of a resistor to each base via jumper, to act as the "regular circuit" for comparison.

I measured a hfe of about 200 for both transistors, consistent with the datasheet and an indication I haven't damaged the transistors (yet).  Pin4 was around 650mV and Pin6 was 623mV.  Then I tried adding a resistor from Pin6 to Pin7.  If this produced a change, I couldn't see it.  I removed that resistor and put in the resistor at Pin4.  Pin6 showed a slow drop from self heating, but Pin4 had 100mV immediate drop and the slow heating drop.  Removing that resistor had the opposite effect.  I tried roughly doubling the resistance, and the drop was now 50mV.

That seems to be consistent with Andreas running at 440mV heated+complete circuit vs chuckb's 580mV unheated + no zener current.

Could it be that die traces + bond wires add up to enough resistance to pin 4 to cause that shift?

[Andreas]

Hello,

wow. you made a own PCB only to measure this. Impressive.
What FETs do you use. P-Channel are relatively rare.

Would be nice if you could provide a larger image of the cirquit diagram. The numbers are hard to read.

Could it be that die traces + bond wires add up to enough resistance to pin 4 to cause that shift?

Mhm. roughly 4 mA change give 90 mV voltage drop.
As resistance this would be 22 Ohms.
I am not a expert but 22 Ohms is a value which seems too high for me.

I know that if you activate the undocumented heater (400 Ohms) in
a LT1019 reference the ground level on the chip (measured as output voltage change)
 is shifted by around 1.25 mV (at 30 mA heater current).
This would give around 40 mOhms for the pin.

So for me there must be a different explanation for the 90 mV.

With best regards

Andreas

[branadic]

Photo?

Sure, here we go...

[Galaxyrise]

What FETs do you use. P-Channel are relatively rare.

MMBFJ177LT1G. I got 10, characterized them for 100uA, and choose the closest match for Q1 and Q2.

Would be nice if you could provide a larger image of the cirquit diagram. The numbers are hard to read.

Done.  I turned on part names, used a clearer font, and left if at the native resolution.  Updated the image in my earlier post  (the _Flip is because I swapped the positive and negative inputs on the schematic symbol.)

So for me there must be a different explanation for the 90 mV.

Using chuckb's images I computed about 42 "squares" along the trace to pin 4, which would be a resistivity around 0.5 ohm/sq to cause that shift.  That seemed high but not impossible, and it fit the nice linear relationship of the voltage shift to current.  Ah well

Thanks

[Andreas]

Sure, here we go...

 

which would be a resistivity around 0.5 ohm/sq to cause that shift.  That seemed high but not impossible, and it fit the nice linear relationship of the voltage shift to current.

So if I understand you right you should be able to measure the 22 Ohms of pin 4  in series to the zener and also in series to the base/emitter of Q1.
The zener would be easy to measure. The datasheet states a typical zener voltage change of 80 mV for a 1 to 5mA difference. (so around 20 Ohms for both sides of zener + the zener itself).
But usually discrete zeners (directly contacted) in that voltage range have also dynamic resistances around 10 Ohms.

The base / emitter would be more difficult. In normal operating the base current should be around 1uA.
But I think that up to 50 or even 100uA you will be on the safe side.

With best regards

Andreas

[alanambrose]

Hi,

Interesting results when we were away for the weekend (the whole graph is updated in my previous message above):

+ the temperature was very stable
+ there were still a few noise excursions - I'm suspecting mains noise
+ there were a couple of interesting artifacts (below) - I'm suspecting mains voltage droop. Exactly how this gets into the reading I'm not sure - atm the ref heater is powered by a E3631A and the ref by an Advantest R6144. I'm not logging these right now.

Otherwise the coefficients were similar to before -



- although you can see the time drift starting to mean revert be more square root like (rather than linear). I guess its possible to repeat this analysis this with a square root model for time drift. Interestingly whereas we're normally interested in high R squared, low R squared is what we're looking for here i.e. ideally the readings tend towards random processes with the coefficients all zero.

TiN - if you give me a link to your data I can do a preliminary analysis - you could copy the existing file without stopping the test. Were you impacted by the earthquakes?

p.s. >>> So you ordered 8051 chip alone? Mine REV? says 9,2. You got me thinking...

?

A.