HP 3456a odd wander

[gmac34]

I have an HP 3456a that I have been repairing (bad capacitors and some oxidation), and that is now “under observation” to see if it performs correctly.
I do not have a calibrator so my checks are based on comparing the 3456 to a keithly 2000 using an HP 3245 as voltage source.

Overall the calibration on the 3456 seems agreeing well with the 2000 (within 5 counts) but I have noticed an odd behaviour specifically in the top of the 10v range:
Let’s say I start in the 1v range, feeding in 1v, the two meters agree very well (2-3 counts difference) and are stable compared to each other.
If I then switch to the 10v range (keeping 1v in, but also true to up to 7-8v ) they also agree and are stable
If I then put in 9-10v they start very much aligned (less than 5 counts) but shortly after the 3456 starts drifting down and seems to settle to -20 counts from the 2000 within a couple of minutes.

If I lower the input voltage for a while (even keeping them in the 10v range), and then back to 10v input the reading goes up a bit (let’s say within 10 counts), and then starts climbing down again.

I would also add that even though the 2000 is not recently calibrated it is stable in it reedings so I would conclude that the 3456 is at fault and not the 2000

It seems like something is warming up with the “higher” input voltage.
I have been staring at the schematics and I’m not sure what could be the culprit.
I cannot test with voltages toward the top of the 100V or 1000V range and I didn't notice anything similar in the 1v or 100mv ranges.

I wish to collect some, all the participants are connected with GPIB so I can prepare some data logging to document the odd behaviour, so I'm open to suggestions on how to best set that up.

[Kleinstein]

It may be a good idea to do a similar test in the 1 V range, so near its upper range.
If the effect does not happen in the 1V range this would indicate that the ADCs is not responsible.

A thermal effect should no happen just near the end as if there is a threshold. Thermal effects should result in INL contributions that look like U² or U³ and thus already starting at lower votlages.
One could also check the opposite polarity.

For the 10 V range there should not be a gain stage before the ADC. So there are no other resistors involved to set the gain, except at the ADC.
There is a slight chance to get a possible error from the input bias going up at a high voltage, e.g. from a JFET approaching it break-down voltage. 10 counts = 100 µV would be 1 nA with some 100 K resistance used for the protection. Such a current should be detectable.
Added current could also happen in a more pulse form from AZ switching. This would become more obvious with shorter integration and than averaging.

[gmac34]

Hi Kleinstein, I cannot reproduce the effect in the 1v or 100mv ranges, they appear stable.
I wish to setup some data collection to see exactly how this non linearity appears, and if the threshold that I seem to noticed is actually there.

Another thing that would suggest that it might not be a thermal issue is that when I lower the voltage the reading is immediately stable.

what do you mean by "AZ switching"?
thanks

[bdunham7]

Could you try it at 9, 10 and 11V and try to quantify the amount and rate of the settling?

AZ = Auto Zero.  You can try turning it off.

[Wallace Gasiewicz]

I have repaired my 3456 last year. A question: Are the resistance measurements OK at different values? Especially at high values?Your problem sounds like a leaky fet.

[gmac34]

I made some tests:
to read the plots you can see on the top the voltage measured by the keithley 2000 and on the bottom the difference in voltage between the 3456 and the 2000. the hp 3245a is used as voltage source, can only do up to 10v but it's quite stable. All the participants were warmed up for at least 1h

1st test - 8V (10V range) for 20min:
They start quite close but quickly the 3456 drops, the rate of change diminishes over time and they seem to stabilise after 15min (-50uV in the first 5 minutes and then -20uV in the following 10minutes)

2nd test 1V ( 1V range)
With a 1v input there is no drift, they stay within 5uV from each other (20min test)

3rd test 5 to 10v (1v step on the 10v range) 10min per voltage step
there is drift on all ranges, but it seems to jump up at every increment. even if the previous step had time to stabilise the following higher voltage takes a few minutes to drift down some more.

4th test -10v,
I cannot see any drift here

5th test 10v on the 10v range, the 2000 measures tp 303, input of the ad converter, this should be 10v on full scale for each range. This one is measured with auto zero off

voltage on tp 303 drifts down with time but it's quite stable compared to the reading on the 3456, this would mean that the drift originates in the input amplifier

the resistance measure seems to work fine, but I can only test with 0.1% 25ppm/C resistors, so cannot really say if it's ok

[Wallace Gasiewicz]

Look at section 8 C 41 of the operating and service manual. The section deals with problems on certain ranges. It may help you to locate the problem. It could be a FET that is leaky. If you locate the FET that is in the 10 V range and short it out the range may work indicating the FET needs to be replaced. The FETs are just used as switches, not amps. The Resistors in the circuits are used for feedback and can also be the problem. 
I suspect the FET in the 10 and 1000 volt range. Please read the service manual at this section. 

Anyway start there.
wally

[Kleinstein]

I dont't think the measurement at TP really points to drift originating at the amplifier. Some drift of the amplifier in AZ mode is normal and would be suppressed in the AZ mode. The relatively stable reading more points to a problem with the switching part, like too slow settling at the AZ steps or maybe too much input current in AZ mode. It may help to measure the input current (e.g. 10  M input resistance handheld voltmeter in 200 mV to get a 20 nA current range. Another point could be comparing the 1 PLC and 10 PLC mode of the 3456 - so possibly more switching effect with 1 PLC.

[gmac34]

The measurement at the test point was with the auto zero disabled, if that makes any difference.

I measured the input current using a uCurrent in series with the input of the 3456a and I see less than 3nA going in the meter with a 10v input.

Regarding the FETs, I did go trough the manual and I tried to bypass q308, which switches the 10v range, but that did not make a difference.

[bdunham7]

I measured the input current using a uCurrent in series with the input of the 3456a and I see less than 3nA going in the meter with a 10v input.

That actually seems high.  Can you measure it at 5V and then -10V?

[gmac34]

I think I am at the limit of what I can measure, I seem to get 2-3nA at 1 v and between 3-4nA for +5 -10 and +10
I also tried +-50V and I get 5uA then
I'm zeroing the meter using the short position on the uCurrent

[Kleinstein]

The µCurrent is not really good for measuring such small current. Just a 10 M input resistance Handheld DMM is more suitable 1 nA would be 10 mV at 10 M and thus easy to measure. Cable isolation can be tricky and also cause some leakage in the pA range.

2-5 nA of input current could easily cause the observed error as internally there are 100 K resistors in the protection.

[gmac34]

I digged out my old fluke 8060 and with hat I can barely measure 0.5-0.2mv across it's 10Mohm input and this is coherent with the drop in the reading on the 3456a.
I don't have anything that would be capable of measuring the input impedance of the 3456a but it is at least above the 100Mohm my 2000 can measure.

[gmac34]

The last suspect before going to sleep is q119 it is a N-channel Jfet  connected as a diode clamp to 13v on the input (up to the 10v range, 100 and 1000v are routed to an attenuator), there is also a q118 clamping to -13v I think it would be worth a try to remove it and see if that could be the culprit.

[bdunham7]

Wouldn't Q118 be more likely?  If they are accessible, just popping them out might be easier than thinking about it...

[gmac34]

off course, it could only be the one connected to -13... I said it was late

[Wallace Gasiewicz]

I understand that you can only generate 10 V for test.It would be nice if you could generate 50 but...Does the unit drift at all when you use 10 volts in and you use the 100 volt manual scale using 6 digit resolution??

[gmac34]

I can generate more with a power supply but there will be lot more noise, I didn’t notice any drift in the 100v range while I was feeding around 95v from a bunch of power supplies in series but the last few digits where jumping around a lot. I also didn’t see any drift putting 10v in the 100v range. A better test would be the 1000v range as it uses the same gain configuration as the 10v but there is no way I can make anything close to that.

Q118 was not the culprit. Next I will check the two FETs that are in series with the input 116 and 103.

[Kleinstein]

I digged out my old fluke 8060 and with hat I can barely measure 0.5-0.2mv across it's 10Mohm input and this is coherent with the drop in the reading on the 3456a.
I don't have anything that would be capable of measuring the input impedance of the 3456a but it is at least above the 100Mohm my 2000 can measure.

0.2-0.5 mV corresponds to 20-50 pA, which is about the usualy spec limit for a high end DMM. Unless the current goes considerably higher on warm up this should not be the problem.
The low leakage current implies a high input impedance in the >100 Gohm range.

If looking for leakage from semiconductors a quite effective technique is local heating of the suspect parts, as the semiconductor leakage (usually also the one due to defects) goes up the temperature.


Possible culprits could also be the LM339 comparators used to control the JFET switches. There are reports of these chips failing in different old meters.  If there is leakage to the negative side some JFETs may not turn on fully. For the 10 V range this would be pin 13 of U102 that should not be much lower than the input in the non AZ mode.
A point to check is if the measurements are done with the analog fitler enabled or not. The filter may have leakage current to load down the signal in some cases.

[gmac34]

13 of U102  measures just a bit under 10v and should be ok, now I managed to trace 10v to the end of the input switching, I can measure a stable 10v after r103

then it goes into q310-11 U307 and U308, and from there it is incorrect (measured on both side of q308).
The analog filter was always disabled and doesn't seem to add any problem if enabled.

I also made another run with plc of 1 with similar drift

[gmac34]

I tried to make a test with 100v but it's inconclusive, if you squint there might be a drift as well but it's quite small and with such an unstable source there can be other factors

[Kleinstein]

At the amplifier the 30  V supply could be an issue (e.g. ripple or too low). Another possible point at the amplifier could be too much current for the JFET amplifier, e.g. visible as too much voltage at R308

[gmac34]

30v supply is stable and measures 29.8V did not find a tolerance on the SM.
for r308 I get -18v on one side and -15 on the other

[Kleinstein]

The 30  should be well good enough. Similar the current through R308 looks like the expected value and thus not too much voltage headroom lost.

[Wallace Gasiewicz]

I think you are very close.Just to be sure you said the signal is low on both sides of Q 308 ? Not both sides of  U308?
HA1 is the signal that turns on the 10 volt and 1000 Volt ranges. HA are on P 155 of manualIt turns Q 308 ON. thru U 304.
Make sure HA 1 is stable and should be 5 V     pin 7 of U304. It should be present in 10 Volt and 1000 Volt ranges.
If this is OK  then the problem is either in U 304b or Q 308  P 188   

Kleinstein Mentioned that a LM 399 can be the culprit, He was referring to a different LM 399, Hopefully he is correct.  U 304 is a LM 399, if it is leaky the voltage to the gate of Q 308 will change causing your problem.

This is the Unity Amplifier and the other amps are not involved. This fits with your observation the the other ranges are OK,