Teardown : Fluke 845A/AB/AR nullmeter/HZ voltmeter tweaks and mods (and repairs)

[branadic]

Frank, it works at the moment. What i did today is I lifted the neons a bit, as they are a bit shorter. I'm currently monitoring if anything has changed in respect of noise. Therefor I had the leads kept long enough before to do that.
However, I call it a fix, but don't call me a genius though. You know, luck is with the stupid ones. Would be great to see how other 845ARs perform (DMM set to 10NPLC, AZERO on), so that I can compare my results.

I was looking for a Keithley 155, but they go for a lot of gold, which I wasn't willing to pay.

-branadic-

EDIT: Looks like lifting the bulb slightly improved noise, since more light can couple into the plastic rods. The small variation is due to open windows here to get temperature somewhat down in the flat.

[dietert1]

Your plot shows  a noise signal of about 200 nV. Could you explain a little what you are measuring? I assume the monitoring output of the 845A is connected to a DMM for logging and the 845A has some gain between input and monitoring output. Is the vertical scale of your plot for the input or the output of the 845A?

Regards, Dieter

[branadic]

It's the noise measured with my R6581D in 1V range, with the input of the F845AR shorted. Therefore the output of F845AR was adjusted to deliver 1V at the output for a fullscale deflection (see also manual).

-branadic-

[dietert1]

If i understand that right, the 845A gain in this measurement is about 300 000 and its input noise was measured to be 200 nV / 300 000 = 0,7 pV. Is that correct?
I guess this is into a bandwidth of 1 Hz or so, limited by sampling rate of DMM.

Regards, Dieter

[branadic]

No dieter1, this calculation were about right if this was the noise on the 1V output voltage, but the 1V is already scaled back to 3µV range.
However, I wonder for what observation period the given noise values for K155 (150nVpp in 1µV range) and F845AR (max. 200nV in 1µV range) are valid. If it is related to the response time only, then my F845 is far below the maximum value and equals a K155.

-branadic-

[dietert1]

Thanks, i was trying to get a comparison with a modern chopper amplifier like ADA4522. Its noise spec is 117 nV typ into 0.1 .. 10 Hz. So i guess into a bandwidth of 1 Hz this would be about 10 or 20 nV - roughly a factor 10 better than the FLuke 845A.

Where the Fluke null meter really shines (different chopper technology!) is the low input bias current. As far as i remember someone measured about 75 fA . Compared to the 50 pA typ spec of the AD4522 this is a factor 700 better. The difference matters whenever source impedance is bigger than about 0,1 uV / 50 pA = 2000 Ohm. In fact, with the ADA4522 a source impedance of 200 Ohm may already cause a DC bias error bigger than noise.

Regards, Dieter

[Kleinstein]

The F845 and K155 have chopper inputs and should thus have little 1/f noise. So the length of the observation period (lower frequency limit) should not be that important, unless extremely long so it includes drift (e.g. thermal effects). The upper frequency limit should have more effect - the logical choice here would be to have the build in filtering function. So the direct comparison should include that number, especially if the filter is adjustable (e.g. extra filter at the output).

The low noise AZ OPs have a much higher bias current - this alone would not be such a big problem, as there could be a compensation, as the current is relatively constant. The other problem is quite some current noise. For the ADA4522 it starts at some 50 K that current noise is more important than voltage noise.

Some of the F845 noise is from the input filter / protection part. Just those resistor of 300 K in total give quite some noise.
The input filter can also be tricky with really high impedance sources - they will slow down the filter and there can be quite some current pulse from a charged filter cap.

[Dr. Frank]

I also measured the 1V monitor output of my 845AR, for 1µV and 3µV range.
At first I found out, that this output was extremely slowly responding, not following the needle deflection instantly, at all.
Reason for that, in my instrument, FLUKE had implemented a strong filter capacitor of 400µF over 4k7, instead of these 10µF, which you'll find in the schematics.

So I removed this capacitor completely, filtering time constant being determined by the DMM, a 34465A, at NPLC 10, i.e. tau = 200ms.

This gave a StD =  128nV_rms or 360nV_pp for the 1µV range, and 234nV_rms or 663nV_pp for the 3µV range.
A simulated NPLC 100 filter gave an additional damping factor of about 1.5 for all of these values.

As expected, that is about 2 times noisier than specified, due to the age of the neons, which effect branadic demonstrated faithfully.
The 1µV diagram was corrected for thermal drift, whereas the 3µV diagram shows the uncorrected -160nV/hr. drift.

Frank

PS: It's necessary to check all electrolytics, as these may have high ESR, or even high leakage.. I've already had replaced many of these before performing this test.

[Kleinstein]

Even if using a DMM at the output, it could help to have some analog filtering. AFAIK most of the modern DMMs have no analog filtering before the ADC and are thus sensitive to some aliasing noise (e.g. 25 Hz range for 1 PLC AZ mode at the meter).
So it would make sense to have something like the 5 K *10 µF filter from the diagram. The minimum would be to dampen out residual signal at the chopper frequency. Some 50 ms time constant would not really hurt even if one want's to wait some 5-8 time constant. Another option could be to operate the DMM in non AZ mode.

[Dr. Frank]

Even if using a DMM at the output, it could help to have some analog filtering. AFAIK most of the modern DMMs have no analog filtering before the ADC and are thus sensitive to some aliasing noise (e.g. 25 Hz range for 1 PLC AZ mode at the meter).
So it would make sense to have something like the 5 K *10 µF filter from the diagram. The minimum would be to dampen out residual signal at the chopper frequency. Some 50 ms time constant would not really hurt even if one want's to wait some 5-8 time constant. Another option could be to operate the DMM in non AZ mode.

 My argumentation is different.
The 845AR noise is specified for a response time of 5 * tau = 3 sec for the 3µV range (250nV pp noise) and 5 sec for the 1µV range (200nV pp noise).
Obviously this is related to the needle response.

As this 400µF capacitor delivers a totally different response time for the monitor output, I tried to approach the specification by an appropriate integration time.
NPLC 10 should give about 5 tau = 1sec, better comparable to the spec.

But I don't know, how this compares to branadics measurements.

This 400µF, C3 sits directly on the monitor output jacks, so it's easy to identfy it, by opening the outer shield.

Frank

PS: technology of 1968..fascinating

[Kleinstein]

With the different time constants (.6 and 1 s) the different noise specs for the 3 µV and 1 µV make sense. The square root of .6 is close to 200/250.
For a first order filter the noise BW is  at 1 / (4 tau), while the integration over a time window gives a noise BW of 1/T (in non AZ mode). So 10 PLC integration (200 ms) without AZ would correspond to an analog 1st order filter with .8 seconds - just between the two ranges.
So the 10 PLC are indeed a good choice in non AZ mode. Brandic used this too.

The AZ mode would add some extra noise BW from aliasing. Thus my idea to keep a little analog filtering. Anyway just using non AZ at the DMM is the easier way, though it does not help that much with chopper residuals. There could be additional filters for this anyway.
The 845 provides so much gain, that extra noise / drift of the DMM in non AZ mode would not matter. So there is no need to use AZ mode at the DMM.

[branadic]

Thank you Frank for your results.

-branadic-

[lukier]

I did a similar test, but short on the input terminals (not zero function), covered with some foam, 1uV range. Meter is 3458A, 1V range, NPLC 10, AZ ON.

The plot is raw data, so 1V = 1uV. Pk-Pk noise seems to be ~440nV, but there was some upward drift (temperature?)

[Dr. Frank]

Lukier, Does your instrument have 10 or 400uF on the monitor out?

[lukier]

Don't remember off the top of my head. Next time I'll be opening it up I'll take a look.

[branadic]

It's a 10µF/25V capacitor in my unit.

BTW: There are some indications that my unit is from 1974.

-branadic-

[branadic]

So the neon bulbs arrived today. Here is a direct comparison of the new AC3/NE2U (left), the originals bulbs that were installed (middle) and the Barthelme 82305 (right). However, the Barthelme 82305 do work for me. Enough spare parts for the future though

-branadic-

[Dr. Frank]

I followed branadics hints, , and also replaced the neons inside my 845AR, with the Barthelme type.
The first picture shows the old neons, the one in front is only halfways lit, the rear one by 3/4 only.
Both evidently show random flickering, and the glass tubes are black inside.

You also see nicely, that due to the rectification of the 200V square wave, only the lower rods, adjacent to the plastic light guide, will glow.

The 2nd picture shows the new neons, the rods evenly lit, no flickering, and much brighter.
The apparent difference in brightness is caused by the cameras shutter speed.

You can also see two black markings on the wires of the front neon.
I first held the neons outside, in front of the hole of the metal blocks, to measure and mark, how high the glass bulbs had to be soldered, so that the light emission would be optimal.

The noise went down dramatically, back to specification, what can be observed in the comparison noise chart, new neons on the right.
The 845AR is quite good regarding noise performance, at NPLC10, it's 94nV_rms for 3µV range, and 48nV_rms for 1µV.

For NPLC100 @ 1µV, which reflects better the needle response time of 5sec, the pp noise is really < 200nV.
The movement is now rock stable, as good as new.

Frank

[Kleinstein]

From the pictures is looks that just turning the neons by 180 degrees and thus swapping the pins could give them some new life, using the other electrode and other half of the glass.

[Dr. Frank]

From the pictures is looks that just turning the neons by 180 degrees and thus swapping the pins could give them some new life, using the other electrode and other half of the glass.

I don't think, that just 'from looks ' the neons could get them new life.

You better may test on your own 845, if the gas inside, or only the rod is exhausted.

The blackening was inside the whole bulb, most prominent on the illuminated side, though.

[Martin.M]

you can be happy not to have the NE2 with front lense, (by example: used in the Philips PM2421),
they are really obsolete now.

I don`t like the repair of optical choppers   

Martin

(my nullmeter is the 883A)

[Dr. Frank]

Well, I'm sick and tired of assumed 'self-evident' conclusions, which on closer investigation turn out to be lukewarm only.
Such assumptions are currently quite popular here in Germany, e.g. concerning climate protests and effectiveness of CO₂ reduction measures.   

So instead of such speculations, I made some real physics experiments on the worn out and new neons.
Setup is simple, a variable 0.. +/-200V source, 100k current limiting/measuring resistor, across that the Dave-O-Meter, in series with the neon bulb, either exposed to the illuminated room, or inside a light-tight box.
I cranked up the voltage, until either the neon turned on, or the DMM showed a voltage drop.

The old and new neons were both specified to have an ignition voltage of < 135V_dc. In addition, the old NE2U contained some radio active substances to mitigate the dark effect, that is to let the neon ignite even if these are used in total darkness, like in this chopper application.

The old neon at its worn out bar ignited at about 120V, likewise inside the box, but the emission was really bad and irregular.

Using the pristine bar, by reversing the voltage, it also ignited at about 117V with light shining on it, and it really showed a very regular glow around the bar, but at > 150V / 1mA only.

Inside the box, the neon did not ignite before 170V, so very probably not inside the 845AR any more.
Therefore, these bulbs were really worn out, probably the neon gas, and/or the radioactive substance.


Now I tried that same experiment on the Barthelme neons.

With lights on, they ignite already at ~ 65V, independent from polarity, and sample. Only 500µA @ 110V were needed to get a nice and full glow.

To my big surprise, they even ignite inside the box at exactly the same voltage as outside, so they show absolutely no dark effect.
That's maybe due to new gas mixtures, like adding some Krypton gas.

So that's the reason, why this replacement really works, and why I now really love such optical choppers with neons inside.

Frank

[Pipelie]

hello,
I ran some test today.

but instead of short on the input terminals I used the zero function,  1uV range. Meter is 3458A, 1V range, NPLC 10/100, AZ ON.

The plot is raw data, so 1V = 1uV. Pk-Pk noise seems to be ~100nV when using 10 NPLC.

edit:
20190921-3458_k155_ADC#3_100NPLC.csv.zip  is a CSV format file, delete the ".zip" before use.

[Dr. Frank]

Hi Pipelie,
thanks for the measurements!
By specification, the K155 has about half the noise of the 845AR, and that's obviously the ballpark what we both do measure, only 18nV_rms for the K155, and ~41 nV_rms for the 845AR, both at NPLC100.

Would it be possible, that you also determine the bias current of the K155?

It's not specified for neither instruments, but for the 845AR, we measured < 100fA, I think that can be found far upwards in this thread.

Thanks. Frank.

[TiN]

For sake of pointless comparison, adding 20uV range datalog with EM A10 head.
Input is measuring zero, with copper wire extension, soldered by cadmium solder to copper version LEMO input, terminated with 34420-81603 short.
DSV raw data file. 2048 samples from raw data used to plot first attached graph.

And yes, cadmium solder have huge difference on stability.
With normal 60/40 solder I had poor thermometer instead of nanovoltmeter. See second graph. 

I have also ran log on Keithley 182-M, 3mV range with Keithley 1488 short at the input.
All filters OFF.



P.S. /me about to get banned to mentioning RoHS prohibited cadmium solder. Toxic members beware!