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

[dietert1]

Removed the low pass filter and recorded the AC amplifier output TP7 averaging 1024 tracks. The instrument has 1 uV input in the +/- 1uV range with the integrator shorted, so the chopper really sees 1 uV difference. Channel 1 shows the 84 Hz signal used for trigger. Without low pass the signal is very noisy and my measurement on channel 2 is small. So i made a math track to produce a 4x vertical zoom. I think the measurement demonstrates that the photo resistors in my instrument have a risetime of about 1 msec and are good enough for 168 Hz.

Regards, Dieter

[Kleinstein]

The AC signal looks rather odd, far from what one would expect from a simple circuit.

There quite some fast transients during switching, that look like they are no coming from the LDR, but more like direct capacitive coupling or possibly an effect of Q106 effecting TP7. But also the slow part look not that ideal, as if the is some coupling capacitor or similar (e.g. C106, C110) too small, so that the gain goes down towards low frequency.
The rise time part looks not that slow, but it still is quite some fraction of the period. So I don't think a higher frequency would really help, more like make things worse.

[dietert1]

Are you trying to troll my contributions? Do you want to do the work of tuning the filters for the new chopper frequency? Sorry, your comments are good for nothing.

[Simon]

Are you trying to troll my contributions? Do you want to do the work of tuning the filters for the new chopper frequency? Sorry, your comments are good for nothing.

If you cannot stand comment then why post? If you don't like the fact that others can say things then: leave!

[Dave Wise]

You will lose some efficiency as the speed increases.  Make sure you still have enough open-loop gain.

[dietert1]

Your advice is based on your experience with HP gear. Why do you think Fluke was using the same photo resistors? Do you have a part id or datasheet of the photo resistors? The ones in my 845AR are sealed in glas tubes somewhat similar to the neon tubes, except with a flat window on the front.

Yes, i understood that changing the sampler frequency requires a lot more than just changing C203. Just wanted to provide some answer and did not have time to finish everything. Of course i will do more measurements.
I also read your remarks concerning the LED driving pulse shape. So today i put back a capacitor similar to C119, except somewhat bigger than before. What was important before for fast neon tube ignition (this trick missing in HP 740B circuit), now provides stronger LED pulses with some decay. Maybe the photo resistor turn-off time is even more critical than the turn-on time.

Regards, Dieter

[Dave Wise]

I'm familiar with the HP instruments from first-hand experience.  I don't own an 845A.  (Hey TiN, want to ship yours to me?)

I expect similar outcome because I believe the effort was similar.  HP made their own cells, using a proprietary blend of Cadmium Selenide and Cadmium Sulfide.  (They're molded in some kind of clear plastic, and they have individual serial numbers.)  The CdSe must have been included to achieve acceptable speed, at the expense of short service life.  I presume Fluke encountered similar obstacles and employed similar techniques to overcome them.  Part of the reason their instruments have continued working longer is that they used photocells only where necessary, and used transistor switches for demodulation, where microvolt accuracy isn't needed.

I will be interested to hear of your ongoing progress.

[Kleinstein]

Usually LDRs are faster turning on than they are turning off. There is a superposition of several time constants and the faster part is sufficient to get a reasonable on, while one has to wait for the slower parts to really get an off.
So chances are there is a time when both LDRs a conducting somewhat.
It may be a good thing that there can be a dead time when non of the neons  is lid. There is a resistor at the input anyway so some cross conduction is not that bad.

I don't see a real advantage of increasing the modulation frequency, as I don't expect the BJT based amplifier to have a lot of 1/f noise. The filter frequency at the input is already well lower than 80 Hz, so no problem there either.

[dietert1]

Dave Wise, if i understood correctly, the life span of those LDRs  in the chopper of our Fluke 845ARs may be over. Is this what you meant? Is there any reference explaining the higher speed and the shorter lifetime of CdSe elements in comparison to CdS? Couldn't find anything on the web.

Received an offer for FF627 fotofets, maybe they are a good replacement and pave the way to higher chopper frequency.

Regards, Dieter

[Kleinstein]

The Photo Fets tend to have some offset-voltage, at least the H11F1 used in some newer version. Newer Photo-mos might be an alternative, though usually in a lower impedance range.

The main real advantage of a higher frequency is that filtering against aliasing is easier. However the present filter for the 845AR is way below 80 Hz.  The way to lower noise would be with lower resistance in the filter, maybe lower resistance in the chopper (i don't know the LDR resistance range), lower noise of the AC amplifier, more optimized waveform for demodulation (e.g. dead zone) and possibly going from a simple 2 switch chopper to a full polarity reversing chopper with 4 switches (this would also reduce the need for an anti aliasing filter).

Looking for photoconduction aging and CdSe, there seem to quite a bit of research done. From a first fast look things that happen may be oxide coming in and changes in the crystal structure (the thin films are far from equilibrium). At least for the oxigen effect there would be a slight advantage for the glass encapsulated version over epoxy. Increasing dark conductivity could be an issue. So the dark resistance might be used as a first indication of aging.

[Dave Wise]

Deiter: I would be somewhat suspicious of the photocells, but I don't assert that every one out there is dead.  But they may be dying.

Kleinstein, I believe the average resistance of one of these photocells while driven in a continuous on-off cycle is on the order of 10K-100K.

I can't remember where I read about CdSe, but it was faster than CdS but less stable over temperature.  And, I think, time.
What I observed in my old HP cells is loss of sensitivity but more importantly, loss of speed.  Most of their low efficiency as modulators is their slowness in turning off.  There is significant overlap between the turned-off one ceasing to conduct, and the turned-on one beginning to conduct.

H11F PhotoFETs generate quite a large offset!  I've seen everything from 0 to 200uV.  It varies with temperature and with LED drive current.  Some are positive (voltage on pin 4 with respect to pin 6), some are negative.  When I put two in a "totem-pole" topology for a modulator, I orient them so their offsets oppose so they at least partly cancel.  But after studying Fluke's H11F1 replacement board for the 845A modulator, I have concluded that I must manually cancel the remaining offset, and it's more important to match offset TC than offset amplitude.

Recently I played with a PhotoMOS relay, type ASSR-4114.  (Because I had one.)  It is unusable - it generates a large spike at turn-on, on the order of 20mV.  Yes, mV not uV.  I think it's due to gate capacitance imbalance, which means all PhotoMOS relays are disqualified.

The FF627/LS627 data sheets confuse me.  How do you operate one of the parts?  I'll be on the edge of my seat as you experiment with them.

[dietert1]

Meanwhile i checked the photocells (resistance measurements with static illumination) and they agree to each other within a factor two at various largely different illumination levels. With the orange LEDs i used the LDRs had 5,5 KOhm and 27 KOhm, dark resistance was above 20 MOhm.

One error i had made was selecting orange LEDs (611 nm). Their light appeared similar in color to the neons, but i learned that LDRs are rather selective at certain wavelengths, so they pick certain strong lines from the Neon spectrum.
I tried to use a white LED with monochromator to find the wavelength of maximum sensitivity, but without clear result after one or two hours. Trying a three color LED i found the LDRs were more sensitive to green and blue light, which means the LDRs are probably rather pure CdS. With those really bright green LEDs (e.g. HLMP-CM3A-Z10DD) the LDRs both reach below 1 KOhm.
In the Keithley 148 nanovoltmeter manual they explain a little what source resistance means for noise in the nanovolt region and the LDRs add to source resistance. Somehow i learned what we knew before from the Neon tube repairs: Instruments with low light intensity will be noisy.

FotoFets have been ordered, but will take some time.

Regards, Dieter

[Dave Wise]

When I experimented with LED illumination of HP photocells, I used surface-mount LumiLeds parts in the Luxeon product line.  I believe these are blue driving a red-green phosphor mix.  They are super bright compared to neon at the spec-sheet Typical drive current, and in designs where modulator input resistance is important, you have to decrease the current or the duty cycle to avoid excessive overlap.

Most of my cells had similar steady-state lit resistance but various different resistances at the normal duty cycle, I presume because they reacted at different speeds.

Slightly off-topic, the 740B front-end contains a special high-voltage photocell intended to isolate the input during overload or in STD mode.  It exhibited large photovoltaic response, 10-100 microvolts when lit.  I had to do away with it.  I installed current limiters based on the Supertex LND150 depletion-mode MOSFET, and an Ixys CPC1981Y SSR. </ot>

[notfaded1]

Somehow i learned what we knew before from the Neon tube repairs: Instruments with low light intensity will be noisy.

I have the box of new Fluke manual spec'd NE2U tubes for my 845AR now... I may try swapping them in this weekend.  I had to read that sentence a couple times but I get it now... you mean when our old neon tubes are sputtered with cathode material inside the glass envelope they don't produce as much light output and thus the photocells output more noisy resistance levels.  I think you're probably on the right track with the 3 color LED's and finding the wavelength that works best... I'll be interested to hear if you find an ideal solution.  It sounds like you're on the path to rebuilding the 845AR into a whole new animal now o.O!  Why not improve on 1969 technology in the year 2020 right?  God forbid it might work better than a 3458A?

Bill

[Kleinstein]

For the LED color chances are high that the green ones are good, yellow ones may also work well, but usually the efficiency is relatively low (about the worst wavelength in the visible range) at yellow and yellowish green.

I don't think the higher noise is about to little light. It is more about flickering light as not the whole electrode is lit evenly every time.

In principle the LDRs can also show some photovoltaic effect from the contracts, especially if unevenly illuminated. At some point more light would do more harm from PV effects and maybe thermal effects and even lower resistance when on would not help very much. So there is an optimum intensity. Some 5 -10 K resistance may be low enough.

[dietert1]

One problem with LEDs in the Fluke 845AR is its low power design, so the LEDs need to run at average currents of 1 or 2 mA. Higher LED currents can ruin once more the stability of the +/- 15V supplies. Currently i have the two LEDs anti parallel with 1 uF plus 5.6 KOhms in series, driven from the low voltage secondary (+/- 15.7 V at 84 Hz). That seems to work very well. Since i didn't want to drill the original neon mounts, i used some aluminum i found in the workshop plus black shrink-on tube.

Regards, Dieter

[notfaded1]

Another symptom of sputtered cathode material is it can also build up on the cathodes themselves and cause part of the cathode to not form the plasma completely (cathode poisoning) at least that's how I've always thought of it.  I'm a big collector of nixie tubes so I have a long history with cathodes in neon gas.  You can't do anything about the built up material inside the envelope (they're trash at this point) but for some of the larger more expensive nixie tubes we often run extra voltage through them for a while and it'll burn off the built up material off the cathodes.  I've rejuvenated many nixie tubes this way... it'll never be the same as NOS but for say a 300+$ tube it's nice to bring them back to seeing the entire digits on the cathodes.  Sometimes it takes running them hot for a day or two even for each cathode with cathode poisoning on it.

Bill

[Dave Wise]

Fluke's own LED retrofit (using H11F1's) appears to have specifically avoided using the +/-15V rails to drive the LEDs.  Instead, they repurposed the transformer winding that previously drove the neons.  A schematic sketch is in an earlier post in this thread, along with a couple of errata.

[dietert1]

When i look into the work that has been done to re-engineer that retrofit i am wondering who invented that circuit. I can hardly believe that they drive the HP2731 LEDs from the high voltage secondary of the 84 Hz converter. The risetime of that signal is about 20 usec in my  instrument, so there is no different timing than on the low voltage secondary. By the way: the resistors R14 and R15 are 1K as far as i have seen. 200R and 20K as in 845A_324_RevB_2-1.pdf don't make a lot of sense.

I checked that the LDRs do not generate significant photo voltage under the more intense LED illumination i proposed above. Interesting enough LDRs are best used with AC. If you let a LDR conduct DC current for some time, it becomes directional. It will generate a small photo voltage afterwards and needs some minutes to calm down again. Currently cross conduction absorbs about half of the input voltage in my instrument (test with short on integrator).

Meanwhile i studied a little bit thermal voltages on the input resistors. R111 and R112 were carbon resistors that generated about 20 uV each in a simple test. I replaced them with UPF50 that measured about 3 or 4 times lower. In the same test R114 (wire wound) generated only 1.8 uV, so it will stay. Will have to check those huge resistors R110 yet (2x 300K).

Regards, Dieter

[Kleinstein]

Possible thermal EMF from the resistors is a good point. under normal conditions there should not be much thermal gradient at the resistors though. Replacing them with other types one has to keep in mind the maximum voltage. R110 is the main series resistance to limit current from ESD. There should be no need for precision / low noise resistors, but low thermal EMF may help a little. So the cheap thin film ones should be well good enough.

The rise / fall times for the LEDs should not be a big deal, as the LDR is way slower anyway.  A more important factor / possible point for improvement could be a dead time with both LEDs off. For a speed up one could also consider to drive the LEDs initially with more current and less current later. The series capacitor already does this to some degree. Because of the cross conduction / delay I see no real advantage in a high chopper frequency. The AZ OPs use a higher chopper frequency, as there CMOS amplifiers have a low of 1/f noise and one can not include large capacitors on chip - so fast chopping makes sense there. With discrete parts one could even consider going lower to reduce the cross conduction and charge pumping effects. However the transformer would make this difficult.

I had a quick though about the input current. As far as I understand the circuit there could be some input bias if the DC amplifier (and demodulator) has an offset and thus significant ADC voltage is needed to compensate. The AC current though R115/C105 would than cause DC bias. So it may help to trim the offset of the DC amplifier.

Are the LDRs used very special (extra fast) parts ?
I have 2 old equal LDRs in glass case - would it make sense to build an LDR chopper with these ?

P.s. to answer the question myself: the common LDRs are slow and hardly useful. 20-30 ms seem to be normal and the ones I have are not much faster.  So the LDRs for the chopper are special faster types.

[Dave Wise]

Partial quote.

By the way: the resistors R14 and R15 are 1K as far as i have seen. 200R and 20K as in 845A_324_RevB_2-1.pdf don't make a lot of sense.

Dieter, after studying TiN's photographs again, I think I have to agree.  R14 and R15 are marked 10011 and 11001 respectively, with one simply facing backwards.  Not 20022 and 22002.  If those stripes are brown not red, 1K makes more sense than 110 ohms, the alternate interpretation.

In that case, my explanation of the offset trim is bunkum.  I don't know why R14 and R15 exist, unless it's to swamp out U4 and U5's small but variable on-resistance.
Fluke took the trouble to install them, along with their teflon terminals - it must have been worth it.
And if R14 == R15, then U4 and U5 must be selected so their mostly-canceled offsets result in the correct net polarity for R7 to trim back to zero.

DISCREDITED PARAGRAPH
When lit, U4 and U5 produce on the order of 100uV of
offset voltage, with source negative and drain positive.
Asymmetrical resistors R14 and R15 forming the common
point between U4 and U5 transmit the respective offsets
unequally, resulting in a net positive, which current
from U3 via R6 and R7 trims back to zero.  The trim
current is adjustable from 1.6uA to 4.4uA, developing
16 to 44uV across R114.

NEW PARAGRAPH - February 2020 - ALSO DISCREDITED - January 2021
When lit, U4 and U5 produce on the order of 100uV of
offset voltage, with source negative and drain positive.
U4 and U5 are selected so their mostly cancelling offsets
result in a net positive, which current
from U3 via R6 and R7 trims back to zero.  The trim
current is adjustable from 1.6uA to 4.4uA, developing
16 to 44uV across R114.

Oh, and one more mistake I made.  R3 is marked 10031, not 20011.  Must be 10K, since 130 ohms is absurd.  That means current into U3 is (15-2.5) or 1.25mA .
I don't know what I was thinking.

I sure wish I knew U3's part number.  I want to say LM336, but 162K for R5 sounds absurd.

Thanks,
Dave Wise

UPDATE 2021-Jan-25

When I built the H11F1 modulator for my 740B #1, I matched offset.  This turned out to be a mistake, and I rebuilt it with a new round of selection and matching.
Using a toaster oven environmental chamber, I recorded the offset voltage of nine H11F2, all the same production lot.
I recorded V at 15mA, my chosen drive current, and 20C, 30C, 40C, 50C, and 60C, then calculated TC.  I also varied the drive current and calculated drive current coefficient IC.

Offset was all over the place, some positive some negative, but all had positive (and fairly constant) TC and IC.  I found two good matches.
TC was approximately +0.5uV/C and IC +0.5uV/mA.  (Pin 6 with respect to pin 4.)  In my chosen pair, one had positive offset, one negative.  The sum when wired in series pin 4 to pin 4, i.e. cancelling TC and IC, was around 50uV.
This is more than the zero adjust range, so like Fluke I added a coarse trim.  In the 740B this is easy, just insert a resistor at the appropriate end of the control.
In the oven, my chosen pair drifts only a couple uV from 20C to 60C.

Note from TiN's picture in post #21 that U5 has a paint dot.  I believe this reflects selection and matching, and dictates install order so the net offset is positive.  Which is then cancelled by Fluke's negative-only coarse trim.

NEW PARAGRAPH - January 2021

When lit, H11F optoisolators produce significant offset voltage.
U4 and U5 are selected so their temperature coefficients cancel.
Secondary selection rejects pairs with excessive total offset.
Then the pair is installed such that the offset - U4 pin 6 minus
U5 pin 6 - is positive.  Current from U3 via R6 and R7 trims this
back to zero.  The trim current is adjustable from 1.6uA to 4.4uA,
developing 16uV to 44uV across R114.

[dietert1]

Today i tried once more to measure the thermal EMF of the large 150 KOhm input protection resistor of the Fluke 845A (2x 300K DALE parallel). This time i put a capacitor in parallel (6.8 uF foil) and shielded the whole thing. Then a HP 3456A read:
low thermal short: -0.3 uV
Fluke resistor: -1.2 uV
Fluke resistor reversed: -0.5 uV

These readings are with 100 PLC with Autozero and reproducible +/- 0,1 uV. So reversal of the resistor generated 0.7 uV and i think it has a thermal EMF of 0.35 uV. Then there seems to be some offset current into the HP3456A input that generates -0.55 uV over the resistor, gives 3.7 pA.

So it means that Fluke 845A resistor generates a fraction of a uV, which could be disturbing in the +/- 1 uV range. Thermal EMF will cause very low frequency noise. On the other hand when i see the input protection resistors of that HP 3456A (a chain of 8x 7.5 KOhm carbon resistors) and how stable they work, maybe it doesn't matter and the thermals in the Fluke 845A stem from the LDRs. Maybe the LDRs need to be inside a metal block, similar to a HP 419A.

Regards, Dieter

[Kleinstein]

Thermal EMF at the resistors only comes up when there is a temperature gradient. The main point is the 845 is likely that the resistors will likely see essentially no temperature difference. The same is true for the resistors at the input of the 3456.
Reversing the resistor would likely also change the temperature gradient - so while it is a good idea to measure both directs, one not easily combine the 3 readings. I would take the 0.35 µV only as an order of magnitude estimate. The terminals of the relatively high power DMMs tend to be quite warm and thus susceptible to produce  thermal EMF problems.

To really check resistors one should expose them to a defined temperature gradient / difference, with pure copper wire as the return path. So more like intentionally one side some 10 K hotter than the other side.

[dietert1]

To answer your questions: The measurements i reported were taken observing low thermal EMF methods, i.e. using pure copper wires and lugs. Otherwise an accuracy of +/- 0,1 uV is not possible, of course. In the Fluke 845A that resistor is not mounted on a board, so it should get packed into some thermal insulation, even better with some heat spreader around it.

Regards, Dieter

[Dave Wise]

I have updated my post #245 to point out my previous errors and add a new, more plausible description of the H11F offset trim.