Keithley 2700 issue - Weird behavior of U115 Comparator.

[petemate]

Hi,

I have an issue with a Keithley 2700 that I got. It doesn't work correct and won't show any meaningful readings in DC voltage mode, while DC curent works just fine (Haven't tested AC modes).

Note that I am unable to find a Keithley 2700 schematic, but for all intents and purposes it is identical to the Keithley 2000 schematic, which is found here: https://xdevs.com/doc/Keithley/2000/K2000.pdf. I will try to refer to everything as seen in the 2000 schematic.


Anyway, I traced the signal path throughout the device and ended up around the 4 JFETs at page 6 of the schematic. The LM339 that controls the JFETs had clear signs of being swapped out previously(hand soldered).  It seemed to me that they were not switching correctly, so swapped the LM339. Still nothing. I then removed all four JFETs(Q104,Q105,Q113 and Q108). I then shorted pins 1 and 2 of Q105, as this would feed the voltage straight into the DCV & Ohms X1 buffer, thus bypassing all the switching. And now the DC voltage works just fine(Havent tested with voltages above 2V, for obvious reasons).

I then turned my attention back to the U115, the LM339 that controls all the FETs. And here is the issue: It doesn't pull down to -15V when the switches are supposed to be off. This seems really weird. For the comparator switching Q104, I can clearly measure about 2.5V at pin 6 of U115 and 0V or 5V on pin 7(depending on the selected voltage range), but this doesn't transfer to the output(pin 1) of the LM339. Broken IC, I thought, so I switched it again. Still nothing. I kept probing around and by coincidence I noticed this *really* weird thing: When I probe pin 1 with Ch1 on my oscilloscope alone, it doesn't drop ti -15V. But if I *also* probe pin 7(with Ch2), then pin 1 will drop to -15 just as it is supposed to do. I don't get this. Ground clips of both probes is attached to TP102. I am thinking that it may be some sort of ground-loop kicking in and affecting the system, but that would have to pass through the 10Meg probes and I find that really unlikely. I also had the thought that it could be related to some incomplete voltage loop, but I don't see what would be missing from the circuit. Keep in mind that at this point in time, only R326(which is floating due to the Q104 having been removed) and R152 is present. So there should be nothing preventing U115 from pulling pin 1 down.

By the way, the U115 is supplied by +5V. Thats not indicated in the schematic, for some reason. But it makes sense, as the logic level signals controlling the comparator are also 5V.

I have of course checked all rails(they're fine) and all resistors between the U115 outputs and UNNAMEDNET146. They are fine(For some reason they are 100k on the 2700).

This is seriously getting on my nerves, so now I am posting here to try and find the solution. I know some of you guys have repaired quite a few 2000s before, so maybe you could tell me what to investigate further. Thanks!

Bonus question: What is the purpose of Q106 and Q107? Are they just acting as clamping devices in case of over-voltage?

[Kleinstein]

The main purpose of Q106 and Q107 should be clamping over-voltage. They also provide a kind of compensation to leakage currents.

The describes behaviour is a little odd. It sounds a little like something just ad the edge so that probing with the scope could bring it back to normal. Have you checked the 2.5 with a scope - just in case there is trouble. There could be a different 339 going crazy and effecting the 2.5 V level.

The other thing to check are the supposed to be logic signals - there have been cases of 4094 shift registers to fail too.

[petemate]

The main purpose of Q106 and Q107 should be clamping over-voltage. They also provide a kind of compensation to leakage currents.

The describes behaviour is a little odd. It sounds a little like something just ad the edge so that probing with the scope could bring it back to normal. Have you checked the 2.5 with a scope - just in case there is trouble. There could be a different 339 going crazy and effecting the 2.5 V level.

The other thing to check are the supposed to be logic signals - there have been cases of 4094 shift registers to fail too.

Thanks for your reply. Yes, its very odd. I have checked the logic signals and they are good. So is the 2.5V reference(its actualyl 2.7somethingV, but is stable. I am leaning towards some ground loop kind of thing, but I don't know where to look..

[Kleinstein]

Having both probes ground at the same test point  should not make a big difference for the meter.  There is a ground loop for the scope, but this would only cause possible minor wiggles and RF pickup for what is seen on the scope.

The probe might also act just mechanical, bending the board and maybe help poor contracts / colder solder joints. It might be worth rechecking the solder joints related to the LM339.

[petemate]

Having both probes ground at the same test point  should not make a big difference for the meter.  There is a ground loop for the scope, but this would only cause possible minor wiggles and RF pickup for what is seen on the scope.

The probe might also act just mechanical, bending the board and maybe help poor contracts / colder solder joints. It might be worth rechecking the solder joints related to the LM339.

Its definitely something electrical.. Poking with a piece of plastic doesn't do anything. But poking with a piece of metal that isn't connected to something else(eg some tweezers) dies provoke a response that causes the comparator to operate correctly.

Edit: I just tried and its the same on all four comparator output pins of the LM339. If I add a probe to the non-inverting input(where data from the 4094s is), the output gets reliably pulled down, but if I use a tweezer, they toggle at exactly line frequency(picked up by me, probably). The fact that they do work seems to indicate that there is something wrong with the data from the 4094's, that perhaps isn't being leveled correctly, but is corrected as I measure it with a grounded oscilloscope probe..

Edit 2: Connecting a 1Meg from each of the non-inverting inputs of U115 to GND solves all issues with the DC input. It also drops the reference voltage from 2.8V to 2.5V Interesting.. Now I just need to find the actual source of the problem. May ideas?

[floobydust]

I would say you have an open-circuit between the comparator and the CD4094. You can check continuity.
Or the CD4094's output got killed by a previously blown LM339. The CD4094 should drive an LED probe if you load its output with one.

The Keithley 2.5V rail is a crappy design, it's high impedance 100k x2 from 5V with no capacitor across it, and it runs to multiple comparators. The 2.5V is probably oscillating nicely as
it can get also get loaded down (up) by any LM339, note that IC has PNP inputs and sources input bias current.
I think if any one comparator is differentially overloaded, it can knock the 2.5V rail around. It's like a big party line.

[petemate]

I would say you have an open-circuit between the comparator and the CD4094. You can check continuity.
Or the CD4094's output got killed by a previously blown LM339. The CD4094 should drive an LED probe if you load its output with one.

The Keithley 2.5V rail is a crappy design, it's high impedance 100k x2 from 5V with no capacitor across it, and it runs to multiple comparators. The 2.5V is probably oscillating nicely as
it can get also get loaded down (up) by any LM339, note that IC has PNP inputs and sources input bias current.
I think if any one comparator is differentially overloaded, it can knock the 2.5V rail around. It's like a big party line.

I will check if the 4094 can drive a LED, but I doubt that it is the issue as the comparator does indeed work correctly(when pulled to ground through a high impedance).

The weird part about the 2.5V design is that the voltage is slightly higher(about 2.8V) unless I add the 1Meg resistors. This means that some current is going *into* the 2.5V reference and to ground, causing the voltage across the resistor to rise slightly. Could be caused by an LM399 or one of the other comparators connected to the 2.5V.

Edit: I just checked and the 4094 can indeed drive a LED without any issues.

[Kleinstein]

It still sounds like a bad connection from the 4094 toward the lm339 - maybe just at the lm399. In a few cases micro cracks can develop just at the edge of a solder blob. I have not seen that with SMT parts, but with THT parts.

The LM339 have PNP transistors at the input. So the input currents are expected to shift up the 2.5 V level a little. However it should not case much trouble, as this is just used as a digital input toward a CMOS chip. So even 3 V would be still OK.

[petemate]

It still sounds like a bad connection from the 4094 toward the lm339 - maybe just at the lm399. In a few cases micro cracks can develop just at the edge of a solder blob. I have not seen that with SMT parts, but with THT parts.

The LM339 have PNP transistors at the input. So the input currents are expected to shift up the 2.5 V level a little. However it should not case much trouble, as this is just used as a digital input toward a CMOS chip. So even 3 V would be still OK.

I agree that this is what it sounds like, but it is just very improbable that all four comparators have this bad connection. There is perfect continuity between the pins.

I have no idea what to investigate next..

[duak]

Does the LM339 have its Vcc bypassed properly?  Comparators have a fair bit of gain and four of them in one package is an invitation to oscillate especially if there is no designed in hysteresis in the circuit.  The capacitance of a probe or even a 1M0 resistor can be enough to stabilize it.

Cheers,

[petemate]

Does the LM339 have its Vcc bypassed properly?  Comparators have a fair bit of gain and four of them in one package is an invitation to oscillate especially if there is no designed in hysteresis in the circuit.  The capacitance of a probe or even a 1M0 resistor can be enough to stabilize it.

Cheers,

No, its not bypassed according to the schematic. Im not sure if there is a bypass cap on the other side of the PCB, though. But nothing in the schematic.

I don't see any signs of oscillations. The rails are stable(<10mVpp) and I see no signs of oscillation on my scope on either inputs or the rails.

[floobydust]

The CD4094 output should be, at all times low impedance. It can output a 1 or 0 as 5V or 0V.
Connecting a 1MEG resistor to its output, is saying this is not true.

Is the output enable pin 15 high on the CD4094?

Otherwise, I'm wildly guessing you have an open ground between AGND and DGND. They cannot be too far apart, weird the CD4094 is connected to AGND on the schematic, I would have used DGND.

edit: Never mind, I see the AGND is optically isolated from the DGND side.

Power for the LM339 is +5V, -15V and the CD4094 is +5V, AGND so I would ensure these rails are all correct;
The BAV199 CR115 on the +/-15V Vregs is odd, I expect +15.6V and -15.6V

[petemate]

The CD4094 output should be, at all times low impedance. It can output a 1 or 0 as 5V or 0V.
Connecting a 1MEG resistor to its output, is saying this is not true.

Is the output enable pin 15 high on the CD4094?

Otherwise, I'm wildly guessing you have an open ground between AGND and DGND. They cannot be too far apart, weird the CD4094 is connected to AGND on the schematic, I would have used DGND.

edit: Never mind, I see the AGND is optically isolated from the DGND side.

Power for the LM339 is +5V, -15V and the CD4094 is +5V, AGND so I would ensure these rails are all correct;
The BAV199 CR115 on the +/-15V Vregs is odd, I expect +15.6V and -15.6V

Just a few minutes ago I tried, in disperation, to replace the LM399(U115) yet again. This time something new happened. It worked, sort of, for a little bit, then gave up again. This time the result was different. The outputs of the 4094 were held high by what I guess is some sort of internal fault of the LM339 pulling up everything. So was the 2.5V reference. Removing the LM339 again brings the signals from the 4094 back to normal again. Now the question is what causes the LM339 to fail. I'll try replacing all the JFETs around the circuit once more.. Just to be on the safe side.

I agree that those BAV199's are odd but they do work. And yes, the voltage is +/- 15.6V.

[duak]

I don't have a schematic for the 2700 but I have a reverse engineered one for the 2000 that I think came from the EEV site.  I looked at the datasheet for the LM339 and see that there's a couple of possible failure modes to consider:
o  too much current in output -- not likely as the outputs seem to drive only FETs.
o  too much current in input if brought below Vee  -- two of the LM339 inputs are driven by signal 'OHMA' but I can't see where it comes from.  Vee is apparently tied to AGND so if 'OHMA' is brought below -0.3 V relative to AGND and enough current flows, the LM339 could be damaged.

Supposing the digital and analog grounds are supposed to be isolated, maybe there's a shorted cap or other component applying a potential difference between the two grounds.  Look in the power supply section for caps to ground on the transformer secondary and regulator chip insulators.  I recall one instrument where an SMD part fell off the PCB and became a jim-dandy short in another part of the circuit.

Best o' luck

[floobydust]

I would hold off changing IC's, there's something else going on. Clean the PCB with IPA and q-tips, look for solder troubles first with a magnifier.

I thought it might be latch-up as there are three rails (+5V, +15V, -15V) involved and if the +5V is late coming up, you might get the LM339 latching up through the CD4094?
Is U120, the second LM339 working OK? No IC's running hot?

I can't see a JFET causing any of this, most have gate resistors. It's good to check V across for a leaky/shorted gate.

[petemate]

Hi guys, thanks for your replies.

Here are the updates:

I tried removing everthing back to the 4094s. Probing each output revealed that one of the 4094s was bad. It would reflect pulses off the input stream directly to the output and have a rather high low-level voltage of about 2.5 V. Changed that immediately. Now all logic levels were correct. I then mounted an LM339(at U115), still without JFETs Q113,115,104,108 and probed its outputs. It worked! For a little bit, that is. I then tried lifting its legs, but it was gone. I then removed that one and mounted a whole new LM339, with all output legs lifted. Same thing, it worked but only a little bit. confuses me even more, cause it means that there is something going on with the digital signals or the power, that caused the IC to die. Simply can't explain it. Probing back, it looks like one of the 4094 outputs is dead as well now. Edit: They all appear to work.

>o  too much current in output -- not likely as the outputs seem to drive only FETs.

The LM339s are able to withstand a short to ground(in this case -15)V for an unlimited amount of time. So, in theory at least, that shouldn't be the problem.

> o  too much current in input if brought below Vee  -- two of the LM339 inputs are driven by signal 'OHMA' but I can't see where it comes from.  Vee is apparently tied to AGND so if 'OHMA' is brought below -0.3 V relative to AGND and enough current flows, the LM339 could be damaged.

I haven't even gotten to the OHMA opamp yet. Im basically only focusing on U115, since its outputs are the only ones that do not working according to the switching tables in the service manual.

> Supposing the digital and analog grounds are supposed to be isolated, maybe there's a shorted cap or other component applying a potential difference between the two grounds.  Look in the power supply section for caps to ground on the transformer secondary and regulator chip insulators.  I recall one instrument where an SMD part fell off the PCB and became a jim-dandy short in another part of the circuit.

There are no power issues that I can see. Everything is fine and stable, no overshoot at startup, no shorts or anything.


> I would hold off changing IC's, there's something else going on. Clean the PCB with IPA and q-tips, look for solder troubles first with a magnifier.

Yes, there is definitely something else going on. Its very weird.

I thought it might be latch-up as there are three rails (+5V, +15V, -15V) involved and if the +5V is late coming up, you might get the LM339 latching up through the CD4094?

Could be, but there must be some coupling effect here. The CD4094 works just fine when U115/LM399 isn't mounted, but as soon as it gets mounted, it starts acting up, possibly taking CD4094's with it.

> Is U120, the second LM339 working OK? No IC's running hot?

Nope. I don't have a thermal camera and I ain't too happy about poking around in there(due to contaminating the sensitive areas), but the digital stuff and the control stuff for switching the JFETs doesn't seem to get hot.

> I can't see a JFET causing any of this, most have gate resistors. It's good to check V across for a leaky/shorted gate.

I completely agree. I have not seen any shorted FETs or similar. At the moment they are all removed and there is little continuity between their pins.

Edit: I dont know what happened, but it seems that its going the right way now. I got Q104 and Q108 working now. I tried swapping out the  LM339 *agan* and was probing around, and suddenly it was working. Now, Q105 and Q113 still aren't working, but I discovered that I can get the output that controls Q105 into a pretty severe osillation by changing around the ranges on the multimeter. But I can't tell if the 4094 is dead again, now..

I am starting to suspect that having those JFETs in actually helps the circuit stabilize or something(I know its really floating without, but that shouldn't affect the comparator..). Anyway, I'll do some more swapping tomorrow and see where it takes me.

[duak]

D'oh!  My bad, I was misremembering which part was acting up.  I looked at the schematic I had and chased down the signals out of the 4094 and into the LM339s and couldn't see any way spurious voltages could be generated without a short or open as it seems they're all tied to AGND.

This is a fairly recent model, isn't it?  Does it use SMD parts on a four layer board?  What I'm thinking of is an open in one or more of the power or gnd pins, maybe from a bad via.  CMOS will still kind of work because the protection circuitry on the inputs and intrinsic diodes on the ouputs will divert current to or from the circuits the chip is connected to.  I'd bet if an LM339 had a poor gnd, it'll source current from its inputs by way of the intrinisic diodes and maybe even sort of work.

Best o' luck

[petemate]

Hi duak, thanks for replying!

Status at the moment is that I got everything but Q108 and Q105  working. The outputs from U115 controlling those transistors just won't go negative, but stays around 0V. I have no idea why. And yes, this is with the inverting output at 2.5V and the inputs at 0V. I have changed the U115 countless times and they all act alike. I am beginning to suggest that certain outputs of the 4094's are dead, because replacing one of those(I121) fixed other issues for me. But why would their voltages then correctly go from 0 to 5V, when toggling different ranges? Makes no sense to me. But I ordered a bunch of HEF4094's(couldn't get the MC14094B) and then we'll see whats going on.

The result is that DCV range now works at 100mV, 1V and 10V ranges(and is *very* spot on), but not at 100V and 1000V ranges.

[floobydust]

Be aware CD4094's vary between manufacturers, some 15V, 18V, 20V rated.
NXP HEF4094B 15V
MC14094B 18V

[petemate]

Be aware CD4094's vary between manufacturers, some 15V, 18V, 20V rated.
NXP HEF4094B 15V
MC14094B 18V

They are both 18V max rating. But the in-circuit is 5V so everything should be fine

[petemate]

Almost everything worked when I replaced the 4094's. I then managed to screw up elsewhere, by trying to exchange parts to get my 2000 working, so now I am back again to the debugging stage. I have never had more trouble repairing a board. It is extremely sensitive.

[TizianoHV]

A few days ago, my trusty Keithley 2790 (as 2700) broke after being close to some big inductors (without being connected) and started showing similar symptoms:
AC volts, AC current, DC current and Frequency were working ok but DC voltage and resistance (2W and 4W) were showing garbage and on the 0,1-1-10Vdc ranges the input impedance was 24kohm (should be Gohms).
The self-test showed a lot of errors: 301.1, 301.2, 302.1, 302.2, 303.1, 303.2, 304.1, 306.2, 400.2, 401.2, 402.2, 403.2, 600.1.
After opening it I found
-Q203 shorted gate to source (I think it's needed only when using scanner cards).
-U131 (current source) had one of the negative input dead short to ground (pulling down the 2.5V reference).
-U209 was pulling down the 2.5Vref too.
-U115 (DC / resistance input switching) couldn't pull down the outputs to -15V and letting all the mosfets on.
While waiting for replacements I used some LED and pulldown resistors (Image1) to verify that the mosfets were ok.
I noticed a weird thing: during powerup Q104, Q105, Q113 turns on together. If I'm correct this will "connect" SENSE+ to INPUT+. Someone else, with a Keithley 2700 or 2000, could verify? Without opening the meter, you can use a second multimeter and measure the resistance between Sense+ and Input+ during powerup (if it's as mine the resistance should be around 192kohm for a few seconds).

(K2000 schematic-> https://xdevs.com/doc/Keithley/2000/K2000.pdf)

[petemate]

Hi TizianoHV,

I don't have time to measure it tonight, but I'll check the resistance tomorrow

[petemate]

When measuring the resistance between Sense+ and In+, for a short second I measure approx. 110-something during startup. This is measured with my K2000 on the K2700. It isn't completely consistent, though, so I suspect it has something to do with the setting of U121 from the previous power cycle.

Also, the reading was hard to capture since it only exists for one single (fast) reading. I used my phone on slow-motion and got the following still-framaes from it. As you see, the display hasn't updated all the way, but it seems reasonable to assume its around 100kR.

[TizianoHV]

Replacements just arrived!

After replacing U115, U131, U209 the Keithley is working again! Self-test passes without errors, calibration seems ok for all functions.


At first I had a "false bad" causing one of the mosfets to stay always on, luckily it was just a cold solder joint.

(I still need to find a replacement for Q203 but it should be needed only for scanner card operation)