The LTFLU (aka SZA263) reference zener diode circuit

[dietert1]

Inside our two Fluke 8502A multimeters i found SZA263 references without temperature control.
Zener currents are fixed at 2,72 mA by 2753 Ohm 0.1 % resistors, collector currents of the references are adjusted for each reference chip, in our case 58 uA (107K + 13K6) and 19,99 uA (324K + 26K1).
Those Fluke multimeters were specified for a 4 hour warm-up time and operation at 23 +/- 1 °C (24h) and at 23 +/-  5°C (90 days and more).
For a range of 0 to 18 °C or 28 to 50 °C they gave an extra 2.5 ppm tolerance in the 10 V DC range.
That means a pretty low TC again - less than 0,05 ppm/K on average.

Regards, Dieter
 

[dietert1]

Meanwhile i implemented the two chinese LTFLU-1ACH references 2 and 4 with the circuit i proposed above (basic circuit + temperature measurement), with TC optimized NOMCA dividers for 10V/7V. Both references are adjusted for low TC at room temperature. The first plot is from HP 3456A with NPLC=1, one measurement every 10 seconds for almost 2 hours. This is a direct comparison of the references. I also logged the temperature output of one reference (200 mV/K).

Noise of measurement is about 0,2 ppm (2 uV), a promising result for the voltage difference between two references. Need to clean the boards yet and put them into their boxes. Also i want to see the noise of the HP 3456A with shorted inputs.

Regards, Dieter

[Kleinstein]

In the difference more the DMM should not be that critical. Still one should use more than 1 PLC if this is only for a slow measurement. One point every 10 seconds would allow at least 100 PLC.
To judge the temperature effect likely a larger temperature swing would help. This looks like only 1 C change.

[dietert1]

The 1°C temperature increase happened this afternoon (sunny weather here). I just wanted to show that test, because i was very happy with the result. I mean there was some doubt about the noise of the Nomca resistor arrays.
Meanwhile i repeated the two hour log with 10 NPLC. Now the Y scale is 1 ppm! Noise seems to be even lower now and there seem to be some waves in the curve. Hope that disappears once the references are in temperature controlled boxes. The boards fit into Hammond 1550 Z boxes, that i want to keep at constant temperature with Peltier elements.

Regards, Dieter

PS: Sorry, the first measurement was with 10 NPLC and this one is with 100 NPLC.

[branadic]

I measured the series of the outer two resistors (bottom and top) today (connected to second and third pad from the top row on the left). The sum of them is 200ohms, so each one is 100ohms.



-branadic-

[dietert1]

After observing two new LTFLU references for some days, i got a question to the experts.
For the direct comparison of the two voltages i made a 100x preamplifier with a ADA4522 ("null detector"). I am using two 470 Ohm resistors for the inputs and a 47 KOhm resistor for negative feedback, a very basic inverting amplifier circuit. I am measuring the amplified difference voltage using a HP 3456A in its lowest range. Then it has a resolution of 100 nV, so with the preamplifier the resolution is 1 nV.  The ADA4522 is specified with 117 nV noise peak-to-peak into 0,1 to 10 Hz. Using the preamplifier i am trying to reduce the difficulties of bringing sub uV voltages to the DVM without thermal EMF.

What would be a realistic expection for the noise in 100 PLC measurements that i am doing every 10 seconds? One measurement takes about 2 seconds, so the Nyquist limit would be 0,25 Hz and i would expect a small fraction of 117 nV, like 20 nVpp, about 2 ppb of a 10 V standard. Plus the noise of the two references.

I am asking this because today i got a sequence of about 300 measurements that exhibited less than 3 ppb noise after subtraction a the thermal drift. Is the reference noise negligible in such a  setup?

Maybe i should first measure the preamplfier+DVM noise with a short-circuit on the amplfier input.

Regards, Dieter

[Kleinstein]

The 100 PLC measurement should have a bandwidth of 1/2 Hz, unless there is extra filtering used and the actual time used is more than 2 seconds. I don't think the 3456 uses this. With AZ a measurement should take more like 4 seconds with the 3456.

1/2 Hz BW compared to 10 Hz would be a little less than 1/4 the noise, so maybe some 20 nV (peak to peak) from the OP.
There will be additional noise from the resistors though. 2 time 470 ohms in series give around 4 nV/Sqrt(Hz) or some 17 nV_pp for the 0.5 Hz BW.
For lower noise one could use a normal non inverting amplifier circuit and connect the amplifiers ground to one reference. This could use just one resistor of lower value.


The white noise of the reference can be quite low, possibly less than the described amplifier. The weak point of the reference is likely more in some low level of popcorn noise - so relatively rare jumps up or down.

[dietert1]

Thanks.

I had the references on top of the  3456A for several days and the stdev of the 100 PLC difference measurements was about 60 to 65 ppb. After repairing the  3456A fan (new oil) the setup is a little different now with very little vibration from the fan and the noise went down to about 3 ppb.
Have to put the preamplifier into a box yet, with thermal clamps for the flat cables from the references. Image shows thermal clamp inside a reference's box.

Regards, Dieter

PS: The feed-through capacity between the case and each signal line is 35 .. 40 pF, helps against RF entering the reference.

[iMo]

I've been using a diff amplifier with an opamp and 1 input resistor (with one of the Vreferences wired to the amplif's gnd), but nulling it is a nightmare. How do you null your references?
PS: .. for example Vref1=6.9978787V and Vref2=7.1723872V

[guenthert]

[..]
  The ADA4522 is specified with 117 nV noise peak-to-peak into 0,1 to 10 Hz. Using the preamplifier i am trying to reduce the difficulties of bringing sub uV voltages to the DVM without thermal EMF.

What would be a realistic expection for the noise in 100 PLC measurements that i am doing every 10 seconds? One measurement takes about 2 seconds, so the Nyquist limit would be 0,25 Hz and i would expect a small fraction of 117 nV, like 20 nVpp, about 2 ppb of a 10 V standard. Plus the noise of the two references.
[..]

  The Nyquist limit is .25Hz, that doesn't mean (noise) contributions of higher frequencies are lost, they just can't be distinguished from those of lower frequencies ("aliasing").  To get rid of (noise) contributions of higher frequencies (and prevent aliasing), you will need a low pass filter.

[dietert1]

Tuning the references was a pain. I ended up ordering various values of precision resistors over and over. Especially the adjustment of LTFLU collector current turned out to be difficult. First i used MF resistors but then i found that those have enough TC to screw up everything. In the end i set the same collector currrent of 40 uA for both references and the TCs are within +/-1 ppm/K, with a difference of 0,43 ppm/K.

To tune the 10 V output i added S102K resistors to the Nomca network. With the Nomca network alone i reached 10.002 V, so the tuning resistors get their TCs well attenuated. The final difference between the references is about 3 ppm now and both agree within some ppm with the standard i received at the Stuttgart meeting in June. 3 ppm means 30 uV, times gain=100 gives some mV, so no special nulling required for the amplifier. Each reference output has a 100uF + 3,3 Ohm snubber close to the difference amplifier. The HP3456A that measures the difference is connected between one of the references and the amplifier output, with Guard connected to the common minus of the references.

On the interface cable i have an input that tunes the reference voltage by about +/- 20 ppm, but i did not use that until now. I hope to use that together with the temperature signal to compensate residual TC (MSP430+ADS1256+DAC8554).

Regards, Dieter

[chuckb]

Other people have discussed and explained flicker noise so here is a visual to support the understand of flicker noise floor limitations in Voltage Standard measurements.

I used a K2182 Nanovoltmeter to monitor the 10V output voltage of two Fluke 732A zener references. This is similar to what you would do to calibrate a known Voltage reference to an unknown reference. For a real calibration the interconnect leads would be reversed to average the thermal EMFs. During this test the temperature was stable and I did not see any correlation between error voltage and temperature.

The raw data was sampled every 30 seconds and it has a 2.4 uVpp noise level. If that data is post processed with a 10 minute moving average 
the noise level only drops to 1.8 uVpp. With an extreme 2 hour moving average the noise is reduce to 1 uVpp. At these time levels the drift of the zeners starts to add to the noise. You can see a slight downward slope to the traces after 100 hours.

With 2 or 4 independent references you can get some noise reduction and you reduce the influence of one bad device.

A few years ago I compared two Fluke 732A 10V standards with a Nanovoltmeter. These older units use the SZA263 Zener. I don't know how the noise of the LTFLU compares but it should be over 1uVpp. Attached is a pdf of results.

[SilverSolder]

How stable is a nanovoltmeter?  -  isn't it a case of when a gnat breathes three floors up, it influences the reading?

[chuckb]

The K2182A and the HP34420A Nanovoltmeters have about 5nV rms noise at 1 second. When the inputs are auto-zeroed the noise gets better at longer duration.

I did some further stability testing of an HP3458A. I added the results to a plot with the HP34420A and the EM A10.

One HP3458A had half the noise of another HP3458A on the 100mV and 1V scales. I have a third unit I will check sometime. The data in the plot is the unit with higher noise.

The top line is the stability on the 1V scale with auto zero turned off. The line does not continue up forever. It levels off before 2uV. This will depend on the meter and ambient conditions.

There are measurements for other nanovoltmeters is the linked thread.

[dietert1]

Thanks for the link to the noise characterization thread. Meanwhile i found that the 3 ppb number i wrote above was a pretty stupid analysis error, so i am still left with 60 ppb as standard deviation for the time being. This compares well to the 3458a autozero noise curves presented by chuckb. From the curves  i read an Allan Variance of about 60 ppb at 2 sec. The 2x Fluke 732A comparison seemed to be 2,4 uV = 240 ppb at 30 seconds. Maybe noise wise the LTFLU with its four zeners is more like a LTZ1000.

Regards, Dieter

[SilverSolder]

The K2182A and the HP34420A Nanovoltmeters have about 5nV rms noise at 1 second. When the inputs are auto-zeroed the noise gets better at longer duration.

Looking at those graphs, it is clearly critical to zero the meter frequently when doing longer measurements...

How is autozero actually accomplished in practice?  Does the meter CPU disconnect the DUT from the input followed by shorting the inputs using internal switches, then reads the zero level from that internal short?

[SilverSolder]

A few years ago I compared two Fluke 732A 10V standards with a Nanovoltmeter.

This is a super interesting test.  -  One thing that strikes me is that one of the 732As could have been perfectly quiet, and the other one generated all the noise (at an extreme!)?

How big a factor would thermals have in this test? - the two boxes might have differing reactions to temperature changes, so the thermals don't actually cancel out "common mode style"?

One thing is for sure, making any absolute measurement better than 2uV is probably ambitious! - the graphs look like a live demonstration of the Heisenberg uncertainty principle...

[dietert1]

Meanwhile i have done  the ADA4522 preamplifier test with shorted input. Input offset voltage after thermal relaxation is -480 nV (spec < 5 uV), noise is 7.1 nV (standard deviation after subtraction of thermal drift). Noise is as calculated by Kleinstein, also the peak to peak estimate. The plot shows raw data. Vertical scale is 0.6 uV.

I know this is off topic, would fit better into the null detector thread. As far as i can see an ADA4522 makes a very good null detector if you put it into a box of constant temperature and adjust the offset. The problem will be low thermal connectors again.

Regards, Dieter

[Kleinstein]

The ADA4522 is a good amplifier for a low Z source. It has quite some current noise and is thus not really suited for a source with more than some 50 K. Another point is the bias current that can be quite high (more than 100 pA). So again not suitable for high impedance source.
If the high voltage range is not needed, one could use the 5 V supply brother ADA4528, that is supposed to have slightly lower noise.

For normal lab use one should have some higher frequency filtering at the input, to reduce the current spike going out and maybe more important to make it insensitive the RF properties of the input.  Without filtering, the source impedance in the 10-100 MHz range may have an effect and this can add some uncertainty.

[dietert1]

Your reply is certainly correct and very general (as always), but does not really apply here. When comparing two voltage references, the source impedance is less than 1 Ohm. So i don't quite get why we should consider 50 KOhm. Of course i can reduce the feedback divider from 470R/47K to 10R/1K to reduce the 7 nV noise to 4 nV, as you proposed above. A 30 uV difference between the two references will then generate a current of 3 uA, so a nV measurement requires a source impedance in the milliOhms.
This ADA4522 preamplifier i made is a cheap and rapid method of getting reference comparisons in the low nV. I did not have to do lengthy experiments modifying input terminals on the DVM or getting into expensive used equipment from ebay or low thermal EMF equipment. From the graph you can see that resolution is about +/- 60 nV, which is already unusual, as most of the readers here know from own experience. My post was meant as a reply to SilverSolder.

I will certainly put the preamplifier into an aluminium box to keep RF away. The input filters you propose are already there. Will have to add a MUX and some low thermal EMF ports to enable comparison to other references, though.

Regards, Dieter

[SilverSolder]

Not sure I am understanding this right:  If the noise of the ADA4522 is 117 nV p-p typical from 0.1 Hz to 10 Hz (from the data sheet),  and the gain is set to 100, does that mean the resulting amplifier baseline noise becomes 11.7uV in that pass band? (disregarding resistor noise, thermals, etc for the moment)

The x100 amp idea seems like a practical solution that leverages existing equipment for comparing references.

[Kleinstein]

The noise specs are relative to the input.  With AZ OPs the noise is essentially white noise, so that reducing the bandwidth to a more DMM typical 1 Hz would reduce the number from the 0.1 - 10 Hz range by about factor of 3   (Square root of 10).
Adding such an x 100 amplifier is a way to do low level measurements, like the difference of 2 refs. or thermocouple voltages. The ADA4522 is lower noise than most 6 digit meters  (except special low level DMMs like 34420, K2182).

In most cases I would prefer a normal non inverting amplifier configuration over the difference amplifier type, as it has a little lower noise and high input impedance (though still not good for high impedance signal sources).

[dietert1]

Yes, the inverting amplifier isn't the best configuration. In my case it resulted from using the same 15 V power supply for both references and for the difference amplifier - a very simple test setup. As soon as the preamplifier gets its own power supply (maybe batteries or taken from the 3456A front end), a different amplifier configuration can be used with very small currents into either reference. Then the difference measurement will be less dependent on the reference source impedance.

Regards, Dieter

[SilverSolder]

What is a good target for bias current?  -  are there many metrology situations where one would be looking into much more than a 100K resistance e.g. a divider of some kind?   1pA would give an error of 100nV in that case -  it might even be possible to "null out" this error current on the DMM by adding a constant?

[Magnificent Bastard]

What is a good target for bias current?  -  are there many metrology situations where one would be looking into much more than a 100K resistance e.g. a divider of some kind?   1pA would give an error of 100nV in that case -  it might even be possible to "null out" this error current on the DMM by adding a constant?

This thread is about the LTFLU and SZA263--  Isn't this thread getting a little off-topic?

Measuring things with null meters is definitely an important metrology related topic, but should we move this to an existing or new thread?