Keithley 195a repair attempt [FIXED]

[giosif]

Hi,

I recently scored a Keithley 195a with a problem: in the 20 mVDC range, the meter always displays 0, no matter what is at the input.
For other ranges, the readings look fine.

Part of my troubleshooting, I checked the input impedance for the 3 lowest DCV ranges and I got the following:
 20 mV range --- input impedance fluctuating between values around 20 MOhms
 200 mV range --- input impedance fluctuating between values around 300 MOhms
 2V range --- input impedance reading OL on a meter which can measure up to 500 MOhms
So, at least for the first two ranges, the impedance is too low.

Next, I removed R12 from its socket and connected pin 3 of the socket to each of the other pins in the socket, one at a time, all while measuring input impedance.
And the result was that input impedance stayed on OL for all links between pin 3 and the rest of the socket pins, except for pin 10.
When connecting pin 3 to pin 10, I started to see the same type of readings for impedance as in the list above.
So, I desoldered Q104, plugged R12 back in and, now, input impedance looks proper for all lower DCV ranges (the upper ranges didn't have a problem to start with).
That said, if I feed 10 mV to the input, the meter still shows 0 in the 20 mV range, but I am starting to think this might be a calibration issue (that's because I can see the correct waveforms at TP5 for the 20 mV range and, for the next two ranges, the readings are correct).

However, I am not sure how to interpret the problem when Q104 is installed, as it seems to be working (not 100% sure, but it seems to measure the same as another PF5301, out of circuit).

[srb1954]

Hi,

I recently scored a Keithley 195a with a problem: in the 20 mVDC range, the meter always displays 0, no matter what is at the input.
For other ranges, the readings look fine.

Part of my troubleshooting, I checked the input impedance for the 3 lowest DCV ranges and I got the following:
 20 mV range --- input impedance fluctuating between values around 20 MOhms
 200 mV range --- input impedance fluctuating between values around 300 MOhms
 2V range --- input impedance reading OL on a meter which can measure up to 500 MOhms
So, at least for the first two ranges, the impedance is too low.

Next, I removed R12 from its socket and connected pin 3 of the socket to each of the other pins in the socket, one at a time, all while measuring input impedance.
And the result was that input impedance stayed on OL for all links between pin 3 and the rest of the socket pins, except for pin 10.
When connecting pin 3 to pin 10, I started to see the same type of readings for impedance as in the list above.
So, I desoldered Q104, plugged R12 back in and, now, input impedance looks proper for all lower DCV ranges (the upper ranges didn't have a problem to start with).
That said, if I feed 10 mV to the input, the meter still shows 0 in the 20 mV range, but I am starting to think this might be a calibration issue (that's because I can see the correct waveforms at TP5 for the 20 mV range and, for the next two ranges, the readings are correct).

However, I am not sure how to interpret the problem when Q104 is installed, as it seems to be working (not 100% sure, but it seems to measure the same as another PF5301, out of circuit).

Check the contact resistance of all relays. I had some very erratic readings on my Keithley 199 which I traced down to some bad relay contacts.

[giosif]

I desoldered both relays and checked them and contacts are fine.
So, it's not the relays.

[giosif]

Could someone please help me understand the roles of Q104 and Q112 in the attached schematic?
I went through the entire manual for the meter and, although it is quite informative and covers most of the functionality, it doesn't mention anything about these two JFETs (or I'm going blind).

Thanks!

[Kleinstein]

Q104 and Q112 are used as low leakage diode. These are likely 2N4117 (or similar) - diodes with comparable low guarantied leakage can be more expensive.

With to low an input impedance I would expect one of the JFETs be leaky or an error in the gate control. LM339 comparators seem to have a relatively high failure rate.

[giosif]

Q104 and Q112 are used as low leakage diode. These are likely 2N4117 (or similar) - diodes with comparable low guarantied leakage can be more expensive.

I see. These are, in fact, also JFETs, same model as on the input multiplexing side - PF5301.
From that perspective, they don't make much sense to me in the circuit (but my understanding of electronics is still pretty shallow).

With to low an input impedance I would expect one of the JFETs be leaky or an error in the gate control. LM339 comparators seem to have a relatively high failure rate.

Well, given the fact that, when I remove Q104, I don't see the low input impedance anymore, I'd be inclined to think it is Q104 itself that may be leaky.
Either that or there is a de-synchronization between the switching of the input + zero + reference JFETs and the guard voltage switching, which makes the voltage difference between G and S of Q104 large enough to open Q104.
Would you agree?

On the LM339s, I suspected them as well and went ahead and replaced all 4 of them.
However, there is no change in behaviour.

[Kleinstein]

Using PF5301 as a low leakage diode does not make that much sense. AFAIK the gate leakage specs are not that specially low. It may be still a BOM simplification and internal leakage testing - the DMMs in some places use selected parts.
For a replacement I would consider something like BAV199 (dual low leakage diode) for both Q104 and Q122

If Q104 was bad, there is a chance the U138A (the OP to drive the guard could be bad), as current through Q104 would flow to the guard buffer. So one could check the guard level, if it still follows the input.

The other point to check would be the main amplifier, if it still woks OK if in the 20 mV mode.

[giosif]

Using PF5301 as a low leakage diode does not make that much sense. AFAIK the gate leakage specs are not that specially low. It may be still a BOM simplification and internal leakage testing - the DMMs in some places use selected parts.
For a replacement I would consider something like BAV199 (dual low leakage diode) for both Q104 and Q122

Ok. I might even have some BAV199's around somewhere.

If Q104 was bad, there is a chance the U138A (the OP to drive the guard could be bad), as current through Q104 would flow to the guard buffer. So one could check the guard level, if it still follows the input.

I did think of that, at some point, but I measured and the OP looks to follow the input (i.e. same signal as at pin 2 of U131).

The other point to check would be the main amplifier, if it still woks OK if in the 20 mV mode.

Now, this is interesting. I did notice this earlier, but wasn't sure how to interpret it:
* As per the first picture below, you can see that the input signal at any of the pins of R120 (which also means input at pin 3 of U131) is higher than the corresponding signal at the guard. There is roughly a 500 mV difference there.
  This difference is not present when switching to a range on the Keithley other than 20 mV.
* At the same time, in the second picture attached, you can see the signal at pin 2 of U131, as compared to the signal at the guard. As you can see, they match perfectly (also, further confirmation that U38A is working).

One detail about the above: these measurements were taken while feeding an approx. 1.8V signal (actually, I was measuring input impedance with another meter) while the range on the Keithley was manually set to 20 mV.
I don't know if this could explain the above behaviour, but I'd think not.
So, trying to interpret this, since the inputs of U131 are not at the same voltage, I would think it is either that U131 is (partially) bad, or something is pulling the feedback section for U131 down.
Would you agree?

Thanks!

[Kleinstein]

The signal shown is too large for the 20 mV input range. So the amplifier would run into saturation at some point. For this reason the signal at the guard is also smaller  - this should be about the limit the amplifier can do. The ADC likely will saturate even earlier. For a smaller test signal one could add a resistor to the input to get more like a 10 mV_pp signal.

Besides the saturation the scope trace shows the amplifier partially working. It may still be too much offset for U131, though it would take a lot (e.g. some 30 mV) to permanently keep the ADC in saturation.

So what is the voltage at R120 and the main amplifier output (TP5) when the input is shorted.

[giosif]

The signal shown is too large for the 20 mV input range. So the amplifier would run into saturation at some point. For this reason the signal at the guard is also smaller  - this should be about the limit the amplifier can do. The ADC likely will saturate even earlier. For a smaller test signal one could add a resistor to the input to get more like a 10 mV_pp signal.

Besides the saturation the scope trace shows the amplifier partially working. It may still be too much offset for U131, though it would take a lot (e.g. some 30 mV) to permanently keep the ADC in saturation.

You are right: I've run the test again, this time applying 15 mV to the input and, now, all traces are identical: guard, input 2 and input 3 of U131.
So, U131 seems to be working as expected.

So what is the voltage at R120 and the main amplifier output (TP5) when the input is shorted.

Please see attached picture (inputs were shorted using a 4-way short).
There seems to be some ripple at TP5.
The picture was taken right after powering on the Keithley; now that the meter has been running for a few minutes, the ripple is smaller, but still there.

[Kleinstein]

The ripple is odd, as it is only after the amplifier. However it looks like the amplifier is not always working in the x 100 mode. the -2 V part suggests that this is a measurement of the -2 V reference at a gain of 1. There probably is also a phase with zero at a gain of 1 (e.g. the last part with no ripple).
The ripple could be capacitive coupling with the case open.

This makes it rather unusual to have the 20 mV range not working. The voltage at TP5 looks OK for the ADC, so it is odd to get a constant zero reading.

[giosif]

So, right now, my suspicion is that the readings in the 20 mV range are 0 due to wrong calibration constants for that range.
The self-tests are all good, so it's not that the SRAM IC would be bad, but I am considering the possibility someone has been playing with calibration.
What I would want to fix at this stage is the low input impedance for the 20 mV and 200 mV ranges.
If I sort that out, I can try a calibration (I hope I can run that only for the 20 mV range) and see if I get proper readings after that.

[giosif]

The ripple is odd, as it is only after the amplifier. However it looks like the amplifier is not always working in the x 100 mode. the -2 V part suggests that this is a measurement of the -2 V reference at a gain of 1.

So, I too found this rather puzzling, at first, since table 7-12 in the manual says the input buffer gain for N2 should be X100, yet I get -2 V at TP5.
And I think this is just the manual being wrong, as I doubt one should expect -200 V at TP5 for N2.
Also, checking for all DCV ranges, I get the same behaviour with regards to readings at TP5:
   N1: input value times the gain value specified in table 7-12
   N2: 0 V
   N3: -2 V
   N4: 0 V
And I think these are correct values, since Q114 (which is ON for N2 and N4) is for the 0 reference and Q103 (which is ON for N3) is for the -2 V reference.

There probably is also a phase with zero at a gain of 1 (e.g. the last part with no ripple).

That's my understanding as well.

The ripple could be capacitive coupling with the case open.

That did cross my mind, although it going lower in amplitude as the meter warms up would imply it might not be coupling from outside of the circuit.

[Kleinstein]

The readings look like the parts of the auto zero cycle. This meter only has a -2 V  reference and not separate 200 mV and 20 mV ref.. So the 4 readings would be Input with gain, Zero reading with gain, -2 V with gain 1 and zero with gain 1.
This would allow to compensate for drift of the ADC gain.

[giosif]

Sorry, I was rushing with my previous response and had a brain fart.
You are correct: the x100 amplification for N2 in the 20 mV range would be 0 V, since the input is the zero reference.
And, yes, the meter is taking different readings over time, switching between input signal, zero reference and the -2 V reference.

Back to the input impedance issue for the 20 mV range, any other suggestions?
Just to provide another detail which I've just realized I haven't mentioned up to now: when measuring input impedance, the resistance meaurement fluctuates around a given value, and I can see on the multimeter bargraph that it is pulsating / cycling.

[Kleinstein]

Measuring the input impedance may not work with an external meter: the external meter may exceed the 20 mV range quite a bit, and the high input impedance is only guaranteed while inside the valid +-20 mV range.
With more than some 120 mV the amplifier will reach saturation the the guard amplifier can no longer follow the input. At some point (e.g. 400 mV) the protection "diodes" (FETs) will start conducting.
The pulsating part is, because the input would only be connected short for 1/4 the time.

For short transients when switching from 0 V to the input, there may be even a transient problem in the design.

[giosif]

Hmm... I cannot 100% disprove the idea with measuring input impedance in the 20 mV range, but this would be the first meter I've ever seen with this behaviour (i.e. for other meters I've tried, all measured high impedance in the low ranges, irrespective of the level of the input signal).
Do you know this for a fact or is it more something you suspect?
If the latter, could someone with a working 195A confirm this on their meter, please?

For the pulsating part (to be precise, this is the resistance value on an external multimeter that I'm talking about, when measuring Keithley's input impedance), I am not sure about that: if my understanding of how input multiplexing works, when the JFET for the 20 mV range is turned on, all other input JFETs should be off and the input signal would reach only ping 3 of the U131 OP (which should have high impedance).
At the same time, when any of the other input JFETs turn on (i.e. for zero and -2 V references), the 20 mV JFET should be off (which also has high impedance).
So, if my understanding is correct, there should be no pulsating on the external meter.

[Ordinaryman1971]

My question is... do you have another PF5301or equivalent to test in circuit?

[giosif]

My question is... do you have another PF5301or equivalent to test in circuit?

I do not. What was your thinking behind the question?

[Kleinstein]

Not that many other meters have a 20 mV range. From the circuit diagram it is clear that the input would no longer be high impedance if the external voltage is high than some 400-500 mV. The guard amplifier can go up to some 120, maybe 150 mV. The rest is than the drop at the diode.

It is normal that the high impedance claim is only valid inside the range and maybe a little beyond, but not very much.

For the pulsating part, 3/4 of the time the input is not connected to the amplifier and thus can be high Z even up to some 12 V. Only for 1/4 the time the amplifier is connected and thus the voltage limit applies.
How bad the extra switching spikes are is hard to say - I would not expect too much trouble, as there is a 910 K series resistor. So 910 K for maybe some 10 µs every 100 ms would still look more like 9 G Ohms.

[Ordinaryman1971]

I would just replace this transistor and see where the road takes me.
I have working unit if you need some testing done.

[giosif]

Not that many other meters have a 20 mV range. From the circuit diagram it is clear that the input would no longer be high impedance if the external voltage is high than some 400-500 mV. The guard amplifier can go up to some 120, maybe 150 mV. The rest is than the drop at the diode.

It is normal that the high impedance claim is only valid inside the range and maybe a little beyond, but not very much.

I will take your word on that.
So, this might mean that, all along, the issue could have been just software related (i.e. calibration constants stored in SRAM).
I will try to calibrate just the 20 mV DC range and see what I get.

For the pulsating part, 3/4 of the time the input is not connected to the amplifier and thus can be high Z even up to some 12 V. Only for 1/4 the time the amplifier is connected and thus the voltage limit applies.
How bad the extra switching spikes are is hard to say - I would not expect too much trouble, as there is a 910 K series resistor. So 910 K for maybe some 10 µs every 100 ms would still look more like 9 G Ohms.

Not sure I fully understand your point here.

I would just replace this transistor and see where the road takes me.
I have working unit if you need some testing done.

Thanks for the offer!
Just before I try to calibrate the 20 mV range, as a definitive confirmation to what Kleinstein is saying above, could you please measure input impedance in the 20 mV DC range?

[Kleinstein]

For the input it is often not so much the input impedance that matters, but the input bias current. With such an input stage the current at the input is usually not proportional to the input voltage - especially not necessary zero with zero voltage. It is more like the input impedance if the derivative of the input bias.

The usual methods to measure the input bias is with a high resistor (e.g. 10 M) to ground or watching the drift with a capacitor (e.g. 1-10 nF) at the input. However this need to voltage reading in the effected range to work. Usually there should be very little difference in the input current between the 20 mV and 200 mV range - except the guard amplifier saturating earlier in the 20 mV range.

One may be able to see the switching spike, by connecting a scope directly to the DMMs input. This may need some shielding to keep 60/120 Hz hum out as it is high impedance.

[TurboTom]

@giosif:

I guess that I get pretty similar input resistance readin on mine. Measured with a Rigol DM3068 with an open input voltage of 2V in 200MR range:

20mV range ~20MR, fluctuation by about +-3MR
200mV, 2V ranges OL
20V, 200V ranges 10MR

Tested with the diode test at minimum current (200nA), I get a voltage reading across the input teminals of the 195A:
20mV range min. ~0.8V, fluctuating heavily due to the measurement sequence.
200mV range min. ~2V, fluctuating
2V range OL
>= 20V range 2V (obviously)

So I'ld say the input resistances that you measured appear reasonable.

Over the last two decades or so, I repaired a few of these meters and more often than not found the analog switching FETs to be troublesome (maybe due to an ESD event). One was quite a peculiar defect: every once in a while, the readings of the instrument gone totally havoc and the input relays went clicky-clacky. Turned out that the crystal oscillator that provides the clock signal to the isolated front-end circuitry (a/d converter, shift registers) somehow "decided" to run on a slightly different frequency than the supposed 1.2288MHz, but not enough offset to notice it directly. I only found this by coincidence when I observed the clock signal right at the moment when it "jumped" to the wrong frequency. I assumed a dodgy crystal but the real culprit turned out to be U121, a 4049 CMOS hex inverter, which measured completely normal externally but wouldn't work properly in the oscillator circuit.

Attempting a calibration on yours sounds reasonable... Fortunately, the ranges can be calibrated individually.

[giosif]

Coming back with an update that I performed calibration for the 20 mV range and, as suspected, I am now getting accurate readings in that range. 

I would like to thank Kleinstein, in particular, for his help and most informative explanations!
And thanks to you as well, TurboTom, for confirming the impedance measurements and overall observations!

Moving on, I want to understand this meter better, as it seems to be fairly straightforward in the way it is designed (in any case, it looks much simpler than something like the Fluke 8842A).