Project KX : DIY calibrator / reference source/meter

[TiN]

Other two boards are LTC2057 as well.

[MK]

I made a model in LT spice using a standard small signal transistor for Q1 and 2 and an ordinary 6.2V zener for the reference zener, that ought to be fairly close to the true behaviour and I found it marginal for stability and difficult to tame completely, you first need to find if it is the heater circuit or the main zener loop that is oscillating.

[Andreas]

Hello,

I think that the current regulator is oscillating.
A current schematics (or link) of the cirquit would help.
In my cirquit I have added C8+R19 to avoid oscillations
together with (large) capacitive loads at the output of the reference.

with best regards

Andreas

https://www.eevblog.com/forum/projects/ultra-precision-reference-ltz1000/?action=dlattach;attach=52101;image

[TiN]

Schematics of revision board : PDF file.

Also bought some connectors and components, going to try tonight. Found triax connectors in local shop, got two pairs male and female ones. 15$ a piece not cheap, but still way better than digikey's. Plan to use them for noise amplifier test rig setup.

Load on reference outputs should not be that heavy, its direct connection to scan card and K2002 rear terminals after card. I suspect C4 capacitor, as thats one of changes via old board. Ground is different on this one now.

[Andreas]

Hello,

C4 is already in the datasheet.
But LTC2057 has much larger open loop amplification than the LT1013
so oscillations are more likely to occur.

Interestingly C4 has different values for the positive reference cirquit
in the datasheet (2nF) and the application note (22nF).
I did not mention this before-> so what is the "right" value?

With best regards

Andreas

[TiN]

I am using 22nF at moment.
Adding C8 fixed oscillations. Added two more boards, to 5 total, and will see how it goes in few days.

Just for reference note here:

Channel 2 - Rev B01 PCB, LTZ1000CH, LTC2057, 12K/1K heater -  7.1304749 VDC reference, 5ppm span on graph
Channel 3 - Rev A01 PCB, LTZ1000ACH, LTC2057, 13K/1K heater - 7.1367946 VDC reference, 5ppm span on graph
Channel 4 - Rev B01 PCB, LTZ1000ACH, LTC2057, 13K/1K heater - 7.1585775 VDC reference, 5ppm span on graph
Channel 5 - Rev A01 PCB, LTZ1000ACH, LT1097, 13K/1K heater -  7.1366038 VDC reference, 5ppm span on graph
Channel 6 - Rev A01 PCB, LTZ1000CH, LTC2057, 12K/1K heater -  7.1296732 VDC reference, 5ppm span on graph

[Andreas]

C8 of your schematics or from mine?

With best regards

Andreas

[TiN]

From yours Datron piece

[TiN]

Minor update as in aging progress for first five modules.

Logged data.



Setup photo:



References photo:

[Andreas]

Hello,

nice setup.
just one question for the data.

What does 5 seconds scan time mean?

I see only around 1200 measurement points for the 42 hours.
(144 values for 1 hour)
The min and max values of the 1 hour window have a different range than the std dev and the mean value.

with best regards

Andreas

Edit: By the way: are the measurement values done interleaved or sequentially (regarding temperature)

[TiN]

It's just a sampling rate in sequence. Switch channel, wait 5 seconds, trigger measurement, switch channel, wait 5 sec, trigger next measurement, and so on. K2001 does same, except it's not switching anything.

[Dr. Frank]

Here's again my 35h measurement on an LTZ1000 (@ 45°C) against a 3458A.

Noise seems to be similar to your references.
But I found < 0.2ppm drift, whereas your references all drift about 0.7ppm relative to your DMM.

Strange thing, is it possible that your DMM is not stable enough?
Maybe you use one of your new LTZ1000 as the master reference and calculate or better measure  the momentary difference to all all others.
 

Frank

[TiN]

Strange thing, is it possible that your DMM is not stable enough?
Maybe you use one of your new LTZ1000 as the master reference and calculate or better measure  the momentary difference to all all others.

Pretty possible, it was never calibrated since factory cal.
Data values in excel are linked in my previous post already, you can play with those if you want to.

Here is result graph if I use average of CH5 reference as a master, normalize other references to it's voltage, and output all on single graph (excluding MV106).
Vertical axis scale = 0.1ppm, so seem to be correlating to your graph pretty well. CH6 somewhat larger off, but rest are within +/-0.2ppm.

I plan to calibrate 2002 to CH5 reference with ratio amp (to get 2V and 20V), so that will be as a nulling point to reduce offset error from own 2002's LTZ1000A.

May worth to note, that CH5 reference is using VHP202Z 0.01% 70K resistors, while all four others are VHP203 0.1% 70K

[Andreas]

Pretty possible, it was never calibrated since factory cal.

Stability has nothing to do with calibration (accuracy).

I still do not know why a 1 hour standard deviation takes 144 measurement points
and the 42 hours (which would give more than 6000 values) give less than 1200 points.

You are using 10 NPLC and some filtering. (What kind is it?)

I think that Frank uses 100 NPLC (2 seconds integration time) with his HP3458A.

With best regards

Andreas

[Andreas]

Hello KX,

I have calculated stability (Allan deviation) from your measurement values.
Best value is CH2 (column4 in data) with around 0.25uV best stability
(corresponding to stddev) if averaging around 16 measurements.
The other values are around 0.3uV stability.

The largest stddev is on ch6. (column 3 in data)
If I look at the classified data I see that there are 2 overlapping distributions.
So you cannot really speak of "standard deviation"

When I compare your setup to my 24 Bit ADC with 2:1 precision divider I get the following picture.
Measurement is 8.5 hours with around 1 deg C temperature span. Around 6.5 measurements / sec.
Noise is a horror: around 15uVpp over all with respect to 5V ADC input or 30uVpp span at 10V divider input.
(with except of some outliers one of it by switching the fluorescent lamp).

Allan deviation gives best stability of 0.255uV at 5V ADC input or 0.51uV at 10V divider input when averaging around 256 measurement values (nearly 1 minute integration time).

Standard deviation of single measurement is 1.9uV(5V) or 3.8uV at 10V

With best regards

Andreas

[Andreas]

Hello,

from Franks data with HP3458A and LTZ1:
around 25000 measurement values within 35 hours (5 seconds each) so obviously with NPLC 100:

@Frank: Thank you for sending the raw data.

First raw data: Y axis scaled in uV (instead of ppm).

2nd: Allan deviation delivers best stability of setup = 0.1uV with averaging 6-7 measurements.
(0.1uV is best noise level within 3458A spec).

3rd: Standard deviation of setup for single measurement over all 35 hrs including drift = 0.485 uV

with best regards

Andreas

[TiN]

Hi Andreas,

I had post raw data too, so you can compare as well. Its excel file linked in post #161

I timed sampling rate again, and my apologies, actual rate is 12sec per value. And since i have 6 values (5 references and scan card temperature sensor) to read, it takes 72 seconds for one sample on X axis.

[Andreas]

Hello KX,

your diagram for the "best" Allan deviation (ch2) is already done in post #166 = KX_col4_ch2 diagram.
The other Allan deviations are slightly worse going from 0.25uV to 0.3uV short term stability.
The worst distribution is with ch6 showed as KX_col3_ch6 diagram.

so if you actually have 50 measurement values per hour,
then your 1 hour standard deviation calculation is actually made over a 3 hour period (144 values).
And your total measurement time is not 42 hours as mentioned but 24 hours. (1189 values).

So in comparison
Franks setup: 0.1uV short term stability
KX setup: 0.25-0.3uV short term stability
Andreas setup 0.51 uV short term stability.

Interestingly is that my setup tends to be stable at this value (0.51uV) also for larger averaging times.
This is most probably due to the temperature compensation that I use.

With best regards

Andreas

Edit: attached all stability diagrams (ch2-ch6)

[Dr. Frank]

Hello Andreas,

many thanks for that Allan deviation calculations!
It's the first time that I'm looking on the stability of my 3458A and LTZ this way.. time-nuts like 

Well, that demonstrates, that the 3458A really delivers fully stable 8 digits = 0.01ppm, to the last count, if you use around 12 sec sampling time.. including the LTZ1000, which is also that stable in this time domain.

Using a less noisy source, e.g. the 5442B, one may even reach the 9 digits region on the 3458A (9 digits output over the GPIB).

The lesser stability for longer time period demonstrates drifts in the 3458A gain resistors (these in in the A/D ASIC) mainly, I assume.

Reason is, that during these 32h of testing, the instrument was never autocal'ed, and the internal temperature had slowly risen about 1°C. According to the specification, that may lead to a maximum of 0.1ppm 0.2ppm drift in the 10V DC range, i.e. Transfer Accuracy parameter, including +/- 0.5°C change ambient, but it's obviously less than that, namely 0.05ppm.

I assume both LTZ circuits (my external one and the 3458As reference) being much more stable on this time scale.

Frank

[TiN]

What software was used to make those graphs?
I am still looking for easiest way to get statistical analysis without much manual fiddling.
Excel is no good when it gets hundreds of data samples to display.

I made a little tool using LabView to collect data, but it's used just for aquiring samples and quick overview on channels, not any analysis.

As of references, I got pair of two heatsinks and nice styrofoam box. Will sandwich a TEC between heatsinks and install this on top of the box.
This will act as internal thermostat for internals of box. Sensor is Pt RTD, controller - Keithley 2510.

Not sure if I will be able to get any subambient temps inside, but heating everything up should not be much of an issue.
Keeping LTZ's stock heater control it will be possible to measure effect of temperature change on rest of parts, like resistors, opamps, etc.

[macboy]

Still thinking about idea of converting K7001 into calibration switch box, with custom reference/resistance/current add-on card

I am very interested in the above. Keep us posted.

[branadic]

What software was used to make those graphs?

Andreas uses DF6JB’s Plotter by Ulrich Bangert http://ulrich-bangert.de/html/downloads.html but you can use Matlab/Octave for data analysis too.

branadic

[TiN]

Time to make things little more complicated.. 
All photos clickable.

Now isolated box have thermal control. Top cover embed sandwich of two heatsinks and TEC module.
TEC powered and controlled from Keithley 2510 SMU. Temperature sensor for 2510 feedback is Honeywell HEL-705 1kohm RTD,
located around middle of box internals. There is little 40x40 5V 0.08A fan in box as well to mix air inside.



Copper heatsink facing inside, blue alu fansink facing outside

There are total 5 LTZ modules in the box, three old ones, and 2 new. Same units we were capturing data before, I have no others.

I was curious so see if gradual slow temperature entry affect LTZ reference stability differently, so I put two newer revision modules into separate metal cast chamber,
with own internal RTD (same Honeywell HEL-705).



Two newer revision modules with RTD sensor (black+white wire).
Internal arrangement:



Measurement setup:



Keithley 2400 - powering fans
Keithley 2001 - measuring RTD in metal box
Keithley 2510 - power TEC and measure RTD in chamber
Keithley 2002 with 2001TC-SCAN card gathering all 5 reference voltages.

First I get overnight data with setpoint +40°C on 2510. Now it's collecting data with setpoint +20°C.

[TiN]

About a week of data from 5 references in thermostat box.
Multiple temperature setpoints were tested, +30c, +50c, +55c, brief +21 and +60C.

[Andreas]

Hello,

if you really want to operate the devices at those high temperatures you would have to put the setpoint of the temperature higher. (you are out of regulation at higher temperatures).

What happens when there are those "negative spikes" on the chart?
Do you have problems with the power supply of the references?

With best regards

Andreas