Keithley 197 - jitter/jumping last 2 digits

[trobbins]

My unit hasn't been serviced for decades afaik.  Display levels were jumping around at lower end quite noticeably, so recently had some time to open it up and with the unit vertical I could carefully flush the front panel slide switches using a contact cleaner and excercise the switches, and let it dry overnight.

Jitter/jumping is much reduced, but still consistently there between display refreshing on lowest two digits on all ranges and all modes (V-ohm-amps) including AC, with inputs shorted, and after a number of hours operation and with switches being excercised. Jitter reduces to just last digit typically on 200k and Megohm ranges.

Same jitter/jumping when testing a stable DCV like 5V.  As ranges increase to 1kV the 5.xx display is jittery on last 2 digits.

With a 10Meg resistor across the inputs, and 0.2VDC range, the display jumps around between 0.5xxmV and up to 2.xxxmV.  With a 2VDC range, the display jumps around between 0.0005xV and up to 0.0018xV.  That indicates an input leakage of circa 50-200pA, so perhaps not gross.

I guess that may suggest a slightly noisy common part of the circuitry from U102 onwards (ie. through Q120, Q121, U110, to U114 pin 17, and the phasing/clocking signals of the single slope A/D circuitry), as the jitter is common to all ranges and modes.

The closest relevant thread topic appears to be https://www.eevblog.com/forum/testgear/sick-keithley-197a/msg1139739/ .

Any advise appreciated on whether this type of fault has been troubleshooted to a particular region or parts.

Tim

[Kleinstein]

For an old Keithley meter one of the first points would be to check the filter caps. It may not be as urgent as in K2001 / K2002, but one may consider a change before they leak. At least I would measure the supply ripple.
The ADC uses the 5 V(A) as the reference signal, so this supply has to be really clean / low ripply.

I don't think the clock signal would be a problem - this is more the digital domain.

Some of the switches ( V , A ) can also effect all ranges.

[trobbins]

Thanks again Kleinstein for responding.

My unit has the GPIB accessory board, so I removed that assembly as the simplest first-in change to loading the power rails.  Luckily that pcb assembly is the root cause of the jitter/jumping, so I can easily troubleshoot that aspect (the main 2,200uF 16V bulk filter cap on that pcb is just showing a slight bulge on the X vent).

But like rocks in a river as the water level drops (or in Australian rivers its the 'snags'), I can now easily discern a small offset in ACV readings with the input shorted.  The 200mV range is showing 62uV; the 2V is showing 0.00011V; the 20V gives 0.0033V; 200V gives 0.020V; 1kV gives 0.11V.  There is no offset display for shorted DCV except 0.0003 on 20V range. 

So I guess that indicates some drift in operation related to the AC measurement circuitry around U104 and U106 (there is no change with AC/DC switch wiping). 

CR102 and CR104 are both 197-604 in the BOM.  It is a pain to visually identify the actual part number for those two diodes, but along the lines of a previous comment from you, a likely swap out would need a low leakage BAV199 (which I don't have to hand so can't easily test that scenario).

There is a 15uF polarised coupling cap between U104 and U106.  I could also add a temporary short on Q118 output to see if that is an influence.  I could also add a temporary short on pin 13 of U106 to local ground to see if that shows a non zero display, or confirms the issue is upstream.

Tim
 

[Kleinstein]

I doubt CR102/CR104 are the problem. those small diodes are normally rather robust. If needed for a test one could use the base to collector junctions of small transistors (e.g. 2N2222, 2N3904/6). The BAV199 would be OK, but it is a smaller case.

Excessive leakage with C113 could cause some offset problem as an DC offset of U104 would be visible. For a test one could use a smaller (e.g. 1 µF)  film cap to see if the offset disappears.
A small offset when in AC mode may be normal - there is noise and the RMS converters may not work that well at near zero. 

[trobbins]

I just shorted the U106 (AD637) input to gnd, and the display for 200mV range is 27uV; 2V range shows 0.00013V; 20V range shows 0.0014V; 200V range shows 0.013V; 1kV range shows 0.13V.   The AC Amp mode shows consistent last two digits as 28.  So it looks like that is down at the noise floor end.  I would have thought that offset was managed by calibration offset, although I don't see the 0V end being calibrated for ACV in the cal procedure, only near FS levels are calibrated.  I have Rev C, so have the calibration enable switch on the rear panel. 

The U104 (LF411CN) has a programmed gain of 5x for the 200mVAC range, so that seems to be the origin of the relatively large 62uV offset, compared to the other ranges being pretty much the noise floor of the AD637.  I quickly tried adding a 10k offset trimpot but it didn't reduce the 200mV range offset, so I think I will just have to enjoy having some offset (as I'm not keen on swapping out the 15uF tant cap, or any other surgery for now).

I had to replace the three main filter caps on the GPIB/Analog output board to eliminate any last digit jitter, although the biggest cap with the slight bulge was the dominant contributor.  All caps were Nichicon 85C, so perhaps could have done with the 105C range in hindsight.

PS. I can see now that the ACV specification is +/- 100 counts, so it seems the meter is within spec.

[Kleinstein]

If the coupling cap C113 is still Ok, a small DC offset of U104 would have no effect. The offset in the AC reading could be in part from the AD637 itself and just noise of amplifier. With 27 µV in the 200 mV range and 130 µV in the 2 V range, it looks like there is some of the background from the RMS converter and in the 200 mV and current ranges also some from the amplifier and the resistor R113.

With AC the noise contribution adds to the power and not to the voltage. So it is not about simply subtracting a constant voltage. In theory the computer part could remove the noise part correctly. However there is a nasty detail about the RMS converters: at low amplitude the BW of the RMS converter is reduced and thus only a part of the noise visible - so the "offset" is tricky.

[corehbola]

Hello @Trobbins.
I recently acquired (in Brasil) an Keithley 197 which has something in common to the behavior you noticed on yours.
I don't mean the jitter in particular, because mine has quite steady readings...but the slight input offset.
In the 200mV and 2V range (hi-z ranges), if left open, it develops a - always positive - voltage that varies from 250mV to 800mV depending on what excatly, I`m yet to know, it doesn't seem exactly related to temperature (but I can't rule it out altogether)...
And, like in yours, if connected to a 10Mhom resistance, it exhibits a voltage very similar to yours, between 1.2mV and 2.1mV.
It's curious that you mentioned exactly that value of 10MOhm in your post, because I had the same idea, and I was totally unaware of your post when I did my test.
When adding more resistance, the offset voltage adds up proportionally, and if I lower that resistance, if also lowers proportionally, so I did the math and, like in your case... it suggests an input leakage of - tops - 200pA.

So I ask you, did you ever chase this "quirk" further? Or did you ever get confirmation whether this is, or isn't, normal to this instrument.

The performance checks mentioned in the manual doesn't say anything about input bias current or leakage maximums,  and I didn't find much on the Internet in that regards.

I`d appreciate any hint you might be willing to share.

Kindest Regards,

Fabio

[Kleinstein]

200 pA sounds still reasonable. The input circuit is not made for the very lowest bias. This is also reflected in a >1 G specs - higher grade/more modern  meters tend to have > 10 GOhms and than something like < 50 pA of bias.
It is more that the actual input impedance of 500 mV/200 pA = 2.5 Gohms is relatively low. Still in spec, but already "relatively" close for a parameter that is often made by design an only spot checked.

It is not uncommon to have input impedance high enough so that even 20 pA of bias birngs the readings with an open input all the way to saturation / end of range.

edit: relatively high humidiy ( >60% RH) can contribute to more leakage from the PCB than one would see in dry climate.

[corehbola]

Thanks @Kleinstein,
So I think I will call it a winner and move on to having it calibrated and maybe try some harmless mod like adding a lateral LED lit backlight (or should I call it a sidelight?).
Usually, sidelighting a conventional LCD is more than enough to improve visibility in low light or under unfavourable lighting angles, and is mostly harmless as it just takes a few milliamps of plain D.C. current.
I`m really happy with this unit.
Another class of meter in comparison to what I used to have.
Rgrds,
Fabio

[corehbola]

Hello,
Suffering from some Obsessive Compulsive disorder, it still hurts me a little to see such fine meter developing something between 1 and 2mV when across a 10M Ohm input resistor, or yet, developing about 0.5V if left open (in Hi-Z ranges of 200mV and 2V).
Especially when my cheapo Radio Shack meter, when connected to the same 10M Ohm resistor develops 0.0mV (it's a 4000 count meter).
So, knowing it's an almost 40 years old meter that it could have sufffered some overload, I can't help thinking some of its input stage's components may be leaky (namely, the 4 input FET multiplexer) and, I alread found and read the - not so many - discussion threads there are on the Keithley meters that uses similar design, and learnt already those FETs are, for one, not very common nowadays (the PF5301 is, from Linear Systems, but it's expensive), and the other ones are not only unobtanium, but they it  seems they heve been even impossible to identify exactly (what are their specs, or what were their commercial p.n. equivalent).

This all leading me to think... But why not replace them by modern analog-switch ICs?
It seems there are quite a few from different brands displaying very low leakage of 1 to 2 pA (which puts them in the same league as the specs of the PF5301).

Is it a completely foolish idea, or is it something that may be worth pursuing?

Rgrds,

Fabio

[Kleinstein]

The JFETs for switching are one part of the leakage only.  I don't think the JFETs are that special - just TO92 fets are getting rare in general.
Replacing the FETs with a CMOS chips would really change the circuit - so you would essentially design you owns DMM. Usually JFET leakage can be quite good - compared to CMOS more like higher performance, but more tricky to use.

Another part of the leakage is from Q103 and Q127. These 2 would be prime candidates for leakage, both from the parts and also from damage (e.g. ESD).

A way to hunt the leakage current is to use local heat and see which parts react much to a little heat (e.g. +5 K). Most semiconductors double there leakage with about a 10 K temp rise. This also applies to many partially broken chips with excessive leakage.

[corehbola]

Thanks...
Hmm... Q103 and Q127... Those were the first ones I originally thought could have been damaged from prior abuse...
And after your comments, as a second thought now I might have ruled them out too soon.
I mean... since this DMM has a 10MOhm input resistor connected directly to the input, and switch-in the low side of input voltage divider, I thought - at first - that in the 200mV and 2V range, the signal path would always come through the 10M resistor (for DCV at least) (as it's the case with many Auto-Range designs).
But later I learnt (and was quite surprised) that in the 200mV and 2V range, signal goes directly (via K101, R107, C109, R106) and that's when I stopped thinking of Q103 and Q107 as a possible source altogether. (I even thought of Q114 and Q115, but I removed them altogether and it didn't change anything).

But Q103 and Q107, even though behind the 10M Ohm resistor, and out of the direct signal path, may be slowly charging  C109.
What puzzles me most is that, when the inputs are left open, the voltage slowly builds up (at about 10mV/sec), up to about 0.58V, and then it stops building up (suggesting there's a one diode voltage drop kicking in somewhere, limiting the voltage to increase any further...), but I can't see anywhere a diode in parallel to the signal path (apart of Q114 and Q115, but since they're bootstrapped, the voltage in them never reach anything near conduction).

I will follow your hint and take out Q103 (which is easy to remove and will be more conclusive).

Thanks... I`ll report back here.

[Kleinstein]

The input stage is different from mainyl modern handheld meters with a 10 M resistor in series for everything and than an inverting amplifier at the input.
10 M in series would add quite some additional noise. So for 5 digits and more this is no longer a good option.

The 0.58 V seen in case of an open input could be just coincidence - it is not unusual to have an open input to float all the way to overload.
If one has a way to record data, one could look at the speed a capacitor charges / discharge from a lower and higher voltage and see if the picture is more like a constant current plus resistance or more nonlinear with the bias current changing faster when above 0.6 V.


Other contributions to the input current can be U102 - what type of OP is this ?

[corehbola]

Hi,
Well, first of all, Q103 and Q127 are not the culprits... I removed them altogether, and it didn't change the behavior whatsoever...
Spot heating components didn't bring up any conclusion, maybe because after it hits about 0.58V it stays there so, there's no room to see it worsening... This already smells like the source voltage of the leakage, whatever it is, is around that value (and not that it's being "clamped" at that level).
Spot cooling components, didn't help either, I tried cooling one by one all the multiplexer FETs, U102 (AD542 dated 8429), U101 (the bootstrap/ guadring buffer)... Of course that, at times, when I cooled them too much it would go all over the shop... but then I cannot rule-out some condensation. Everytime I cooled them carefully, I couldn't see anything outstanding.

But one thing did make A LOT of difference.
I lifted the input side of R107 and two things happened:
1. Leaving it opened, the build up voltage slowed down more than 20~30 times. Especially past 0.3V (granted, maybe this slowing down past 0.3V was already there, but since it was going relatively fast before, I didn't notice).
2. When I connect the 10Mohm resistor directly at the tip of R107 to ground, it still shows a predominantly positive voltage, but it visually averages at much lower levels, (most of the times the display is showing values around 100uV).

Both of those above can be very well from U102, but I would be very happy with that already.

With R107 disconnected as it is, it leaves very little (component wise) as possible culprits.
a. The P.C.B. (surface contamination under the switches because everywhere else is quite clean).
b. The Switches themselves - most likely -  .
c. Leakage from the relays coil to the contacts (I know, very unlikely, but easy to rule-out at least).
d. A handful of range switching FETs connected after the 10Mohm resistor. Again very unlikely because there's no positive voltage there for the leak to come from (only negative voltages at the gates, at least in DCV ranges).

Other than that, I can only think of RFI rectification, that is finding its way into C109... but from what is left of components, who would be the one rectifying? The switch contacts? The FETs?
On that hipothesis... I wonder if it would help to add a ferrite bead at R106 and R107 leads?

More to come... I`ll let all know.

Thanks for the insights so far.
 

[trobbins]

Hi Fabio, thanks for pursuing your 197 issue and discussing it.  I haven't as yet pursued further assessment of my own 197 issue, but may well get keen to do that.  I'm not keen on removing/swapping parts if I don't have to, but I would take on board Kleinstein's comment about warming individual parts to see if anything was noticed.  The use of cooling seems a bit more complicated if taken too far in step temperature, or contributing its own leakage.
Ciao, Tim

[corehbola]

Thank you,
And apologies for having hijacked your thread but, but your description of the offset voltage when applied to a 10M ohm resistor was (and is so far) exactly the same as mine, so I really thought there could be a common root cause.
I`m not a collector and I only want a meter of this league, working as well as it possibly can so, I`m not much worried about swapping parts (if necessary) for the sake of troubleshooting, or lifting some component leads to rule out possible causes.
And also, because the guy who I purchased this meter from, has quite a decent lab at home, with a reference voltage generator and a good and recently calibrated 7 1/2 digits HP multimeter, and he told me already, should I chose to chase the minor flaws this meter has, that after I`m done, I`m welcome to take the meter there again and he will re-calibrate it for me.
So, as it stands, I`ll take my chances a little bit further at least, because the worse it can happen is losing its calibration and I can have it calibrated again.
If, in the process, I find some flaw that can be useful for your meter... it will save you from the troubleshooting.
This meter, it's in general good shape and doesn't have signs of extensive rework. The soldering of pretty much all electronics is its original wave soldering (or was until I start touching it).
But it does show it has had its switches replaced (because some pads are in bad shape) and there's one broken trace underneath the switches (like if it has been fused, suggesting some pretty nasty overload) and this broken trace has been bodged wired on the bottom side of the PCB.
From the looks of it, whoever did it, was some real service center (Keithley or else, I can't tell) because they used teflon insulated wire to do the bodge and not some random piece of wire, and the switches are original replacement parts for what I can see from pictures of other 197s.
I`m convinced that it's a good opportunity to try to improve it, and in the process gain some experience in servicing this kind of meters. Some components may be difficult to source, but it has not a single proprietary component, and this is a very good thing.
I`ll let you know.

[trobbins]

Fabio, I have no concerns at all about this thread being used for any similar 197 faultfinding - that is the benefit of threads as a means to collate info and make it easier to find

[corehbola]

Thank you TRobbins,

I didn't find the culprit yet, but matters have advanced somewhat.

As I mentioned earlier, this unit has signs of rework in the gang switches (desoldering pump marks on several of the PCB traces and pads) and, as it also had residues of solder flux cleaning (or a failed cleaning attempt as the underside of the board was gross), I thought that, although risky, I would never be sure of the state of the switches and the level of contamination on the PCB, under the switches.
And, as it goes, where some tens of pA is enough to produce errors in hi-impedance readings... I decided to remove the whole switch assembly and clean the board and clean the switches thoroughly (by taking them all apart and, physically cleaning them inside and out and re-lubing).

While I'm at it, I took some pictures of the board, under the switches, and post here for future reference.



Studying the manual, I think I have a solid plan, that will allow me to spot what is the origin of the input leakage current that is causing the 1~2mV readings when I ground the leads through a 10Mohm resistor (just like in your unit).
Using machined round socket pins, I`ll socket the four MUX FETs (Q110 to Q113) and R116 (the 200k resistor that goes into the Input Buffer AD542) so that I can isolate any of them, together, or one-by-one.
Then, using the troubleshooting mode and selecting between modes 0u2, 1u2, 2u2 and 3u2, and connecting an external hi impedance meter to the 197's input (where the voltage is building up) I will be able to identify the source of the leakage current and, with some luck, source the parts that are needed to amend it.
Finding a replacement to the AD542 nowadays is not at all difficult and there are many alternatives that can match or out-spec this old part.
The MUX FETs... Those may take a while to arrive and are not exactly cheap, but they're still orderable from DigiKey so I think I will order a few of them even if it's for the sake of it (or for the future).
As for the attenuator range switching FETs, those ones are not yet so clear to me if they can be obtained and which current part number they would be but, on the other hand... there's no way they can be the culprits (at this point at least, where I`m only testing on 200mV and 2V range), because they can't possibily be the source of ANY positive (outgoing) current, as they're not connected to any positive voltage source....

And, by the way, I`m doing all these testing now with all the switches removed. I soldered jumper wires to emulate the switches in the positions Power ON, DC, Volts, 2V range to elimintate any extraneous source of leakage.

I'll keep you posted.

[Kleinstein]

The switching JFETs should not be that critical. Likely most of them are some PN4393 / J113 or similar as the typical switching JFET. Many others should be OK, as the voltage is still relatively low and this is only a 5 digit meter.

[corehbola]

Thank you Kleinstein...
I`ll keep those FET P/N in mind... As you well said, TO92 FETs are not that common nowadays and it's always good to know some suggestions.
The MUX FETs on the 197 are of type PF5301 and are still available from Linear Systems and for the likes of it, their main "selling point" is the low leakage, at max. 5pA.
I don't have enough experience with FETs and with this level of low current to be able to even tell what is the typical leakage of a "vanilla" switching FET, but I assume that it's probably bigger than that 5pA, otherwise they wouldn't bother to highlight this particular spec.
They don't have particularly low channel resistance. I measure the ones in my 197 at anything between 7kOhm and 12KOhm.
The datasheet of the PF5301 suggests even worse than that, with Idss from 30 to 500uA (@Vds = 10V and Vgs = 0V) suggesting an Rds ON between 20k and 300k (not very "on" to my taste, but in these hi-Z application, it shouldn't matter anyway).

I wonder if you own yourself a 197 (or 197A). I noticed mine is probably from a very early revision.
Besides many small (and probably insignificant differences) I noticed one particular difference between mine and the only available User/Service manual there is, which is about the Guard Rings around sensitive components.

The generally available manual mentions basically 2 major guard rings... one around the range switching FETs, connected to the bootstrap buffer U101A output (pin7), and the other around the MUX FETs + Input Buffer OpAmp, the AD542 U102 and its gain switching FET, connected to the other bootstrap buffer, U101B (pin 1).

It happens that on my 197 in (and I don't know if in newer ones it's like mine or it's like it's shown in the newer documentaiton), the guard ring around the MUX FETs is not connected to U101B pin 1, but it's actually connected to ground (so it's not really like a bootstrapped guard ring as the schematics suggests, but it's more like a ground shield).

I would like to know how is it actually in the 197 newer models... if it's connected to ground (and the schematic then is just plain wrong) or if they really changed it on the PCB and made it follow the guard ring of the input buffer (instead of ground as it is in mine).

Rgrds,

[Kleinstein]

The PF5301 is indeed a very high impedance JFET, about equivalent to the 2N4117 / 4118.  This may be useful at some positions, but not that well suited at other positions in series with tha divider resistor, due to the high on resistance.  Chancs are they should used different fets for switching. The exact type should not matter that much, unless the leakage currents get too high. There is an PF5103, that is quite similar to the PN4393 / J113.
The fets  Q105, Q106 , Q109 and Q116 should be more lower resistance types. Q109 may even be more something like PM4392 for even lower resistance at the cost of more votlage needed to turn off.

With the leakage of the FETs, the actual leakage is often quite a bit better than the specs - which are more like test limits. Testing for very low currents takes time and is thus expensive. So with many types the spec limits are not very tight and typical parts can be way better. This is one reason why selcted parts may be specified to get parts screened for tighter limits.

The on resistance is not equal to Vds / IDSS. If not directly given a better estimate is  1/gfs. So for a gfs of 70 -500 µs this would be 2-14 Kohms.
The measurement should be some with a relatively low testvoltage, a higher test voltage can effect the result.

[MathWizard]

I have a 197 and used MG chemicals contact cleaner (legit brand), and the performance was horrible after, it left a resin in there, I had it take it all apart and clean it, which was a nightmare.
 
Does anyone know a good replacement relay for these? In the schem. it's the K102, custom, Coto 3500-0021 instrumentation relay. I've searched a lot online, and I can't find info on the custom model#, or replacements.

It's some Low Thermal EMF, High voltage, w/ Hg option available. So IDK what the contact materiel was either. It's single coil, normally open, shielded/gnded. It uses 5V w/470R coil. I'm not worried about the driver logic, I just need a recommendation for the spec's, or any similar relay or suitable replacement for a 5.5 or 6.5d meter.

I haven't looked at the schematic in a long time, but I made a sim of most of it. I never understand the slope-integrator section good enough to get it working. But the rest of it worked ok.

[Kleinstein]

It should be relatively rare that the relay fails. If leakage gets a problem, a good bake out (some 120 C ? for quite some time) could help, by removing water. It would at least be a chance.
Low thermal, high voltage reed relays are tricky. To some degree a 2 pole relay with both contacts in series helps: it gets the higher voltage rating and the termal gradients tend to be symmetric to give some compensation of the thermal EMF. With Hg wetted contrcts the choice of contract material is very limited - it is effectively Hg. Many of the normal contact materials react with mercury.

The ADC differs a bit from the normal multi-slope ADC in that the run-up part is combined with the input before the switch to the integrator and the rundown part is with a single slow slope only. So it is multi-slope run-up and just 1 slow slope rundown. The supply to U117A is the main reference.

[MathWizard]

I burned out the coil somehow switching the diagnostic modes, and then I really broke it trying to get it out of the can, I should have heated it up, or cut the top of the can off.

I've only looked at ideal op-amp integrators, next time I get deep back into this thing I'll know a bit more. Hopefully I haven't damaged anything else I can't easily replace.
here's a pic of the NI of U108, when the DMM input shorted, the yellow trace

[Swainster]

Just to add, my K197A suffered from a significant offset in AC volts (didn't check AC current but was probably the same). This offset caused the 2V range to read something like a volt, and the millivolt range to overload even with a shorting plug fitted I.e. it was pretty easy to spot. I traced this to the output of U104, where it's DC level went further and further from AGND as the ranges were switched, hitting the rail at the millivolt range (5x gain mode). In my case, at least, this did indeed turn out to be caused by faulty CR102/CR104. According to another K197 thread on this forum, these are just 1N4149. All I had to hand was some off-brand 1n4148, but these fixed the issue and got the lower AC ranges working again. Having said this, I've since stocked up with some quality 1N4148, 1N4149 and low leakage BAS45A. Is it worth me going back in and swapping out the dodgy 4148s for the BAS45As?