Agilent 3458A repair

[franklin]

I’ve ordered the U8/U9 IC’s and sockets; supposed to be here on Monday 30th.
I’ll check the voltages you requested; still AUTOZERO OFF?

[franklin]

For DCI mode: U7A, p5 (to Q18): 0V (AUTOZERO ON or OFF)
For 100V DCV mode: U8C, p9 (to Q16): 5V (AUTOZERO ON or OFF)

I don’t have a desoldering station only a temp-controlled one and a handheld tin-pump. Thought I might cut the legs of the IC’s and solder them out one by one or is that a bad idea?

[Dr. Frank]

For DCI mode: U7A, p5 (to Q18): 0V (AUTOZERO ON or OFF)
For 100V DCV mode: U8C, p9 (to Q16): 5V (AUTOZERO ON or OFF)

I don’t have a desoldering station only a temp-controlled one and a handheld tin-pump. Thought I might cut the legs of the IC’s and solder them out one by one or is that a bad idea?

Sorry, obviously I accidentally  tested U7A in ACI mode, as DCI enters differently, via Q20 and Q22.
U8C level is fine, so U9 might be fully ok.

If you have no better tools, carefully cutting the leads is the least stressing way for the PCB.
Maybe you begin with U8 first.

It's a pity, though, because you could otherwise test these ICs on a breadboard, what has been damaged.. just for academic purposes.

I'm now looking forward to next week, if that's been the culprit.
Frank

[voltsandjolts]

I would recommend to cut the legs as close to the DIP package as possible.
Attack it with snips vertical.
Just to minimise risk of PCB damage.

..use nice small snips that get tips between legs, not electrician pliers!

[Dr. Frank]

@Frank

I have a question about these U7 / U8 (LM339): which one do you recommend?

Here is a list from Mouser:
https://www.mouser.de/Semiconductors/Amplifier-ICs/Analog-Comparators/_/N-cib1w?P=1z0z63x&Keyword=LM339&FS=True

Thanks.

These are uncritical parts, I would NOT recommend the ceramic package.
Mouser specifies the cheapest for 30V only, that might be a typo. The others are leadfree, obviously, so that should not have influence on the max. power supply.
Datasheets are linked incorrectly, so I can't check the parts from mouser.
T.I. has no wrecked the link themselves.

Anyhow, 30V max. is enough, 36V would be better.
Offset and bias don't play a role.
leadfree parts can of course be soldered with lead containing tin.

Frank

[BU508A]

Just for completeness, here is the link to the documentations available from Texas Instruments:

https://www.ti.com/product/LM339/technicaldocuments

[Dr. Frank]

yeah, also found it, no datasheets available either, and whole 339 family seems to to have 30V V_cc.
Really old stuff. 

[BU508A]

Found the datasheet, the one from Texas Instruments, added to this posting as PDF

[Kleinstein]

The normal Vcc is limited to 30 V, but AFAIK the outputs are open collector and can stand a voltage higher than VCC. The LM399 are kind of the standard part  to control JFETs switches and commonly used with more than 30 V at the outputs.

I would first exchange U8. Form the measured voltages it is essentially for sure that U8 is bad and chances are it could be the only fault. So far I see no problem with U9, it seems to work just as it is supposed to.

[franklin]

For Kleinstein

In one of your posts you wrote:

“One could even do a test with just an extra LM339 just put directly on top of (in parallel) the broken one. Chances are this could fix the error. With open collector outputs the 2 chips would not interfere with each other.  The comparator is only used as a kind of level shifter.”

I’ll hopefully receive the IC’s I ordered on Monday 30th. Piggy-backing an LM339 would be a quick test so I’m inclined to try that before doing something major.

[Dr. Frank]

I guess, paralleling another 339 is a bad idea.

The expected failure mode of U8 is that its + input of 8A draws excess current, therefore putting a 339 in parallel would not give an additional hint other than with the 470 Ohm resistor. 
There's danger to create a short, so I would not try this experiment at all.

Frank

[franklin]

Duly noted

[Kleinstein]

The measured voltages suggest that U8A has some positive current at the + input. To get some 150 mV at the 4094 output this would be a few mA at most. The signal level is still OK.
At the output U8A is likely not drawing any current. The observed 7.8 V are just what comes from the BOOT signal.  I very much doubt U8A would have a fixed 7.8 V level at the output, just the same as the reading. Drawing the pin to -20 V also works with the resistor - so not much current needed.

Anyway U8 is broken - so one could as well remove it carefully and have a socket for a new chip (just in case there is something odd that kills U8). It is very unlikely to do extra damage removing U8 - so the piggy pack test is not really needed and the old U8 has to go anyway.

[franklin]

Just received the parts ahead of time (nice). Removed U8 and U9, put in sockets and new ICs. It worked. I’m impressed, not of myself but of the lot of knowledgeable advice on this forum. I don’t think I had pulled it off myself. Now, the error I got is “ACAL required” so it will be interesting to see after letting the meter run for a few hours after performing ACAL.

[Dr. Frank]

franklin, hip hip hooray!! 

What a great news in difficult times!

It's always great fun when this community solves such problems remotely, especially on these famous 3458A.

So thanks also that you stayed persistent on failure finding.

After ACAL ALL, please repeat this leakage test, to briefly check whether the input path is ok.

As said, if you leave the instrument open in DCV 10V range, it should typically ramp downwards from zero with that given rate, i.e. about -2pA.
Another way is simply setting the 3458A to  FIXEDZ ON, that gives 10MOhm constant input impedance, 1V range, check zero reading with input short (should be about < 1µV), and then  let it settle with open jacks. -2pA then should give about -20µV reading.

Again, congratulation for your repair!

Frank

 

[franklin]

I’ll do the pA test both ways. My previous message was a little premature though. After ACAL, I get ERROR 204 HARDWARE FAILURE – DC VOS DAC CONVERGENCE 64. However, the voltage (10V) is correct. I’ll have a look at what this message means or do you have any hints?

[franklin]

When instrument is switched on, I got ERROR 204 HARDWARE FAILURE – DC VOS DAC CONVERGENCE 64. When running ACAL, all is OK. Now I’m starting to get suspicious; what is happening? Switched the meter off and started again. No error; ACAL and all OK. Switched the meter off and started again. No error; ACAL and all OK. It seems everything is working fine but I didn’t like how it all started. I was even checking if I had forgotten something but could not discover anything suspicious. I’ll let it run for a while and see what happens.

[Dr. Frank]

maybe the ACAL constants have had a hickup only from the foregoing HW failure, which is healed now, after two ACAL.
Let's see, how it develops.
And maybe TiN can give his opinion, please, as he's got most experience

Frank

[franklin]

Tested temp; 35,2 ⁰C.
-   DCV, 10V (fixed range), AZERO OFF, FIXEDZ OFF (GOhm);  2 pA test: checks out to about -0,01 V/s
-   FIXEDZ ON, 1 V range (10 MOhm): Zero reading < 1 µV; raises to about 60 µV (short removed).

[Kleinstein]

The 60 µV at 10 M input resistance corresponds to some +6 pA. The drift rate is more like -1 pA, so a little confusing, but the configuration is not exactly the same, so there can be differences.  This are good values -  2pA is more like wishful thinking, best case, and the sign is likely not the same with all meters. AFAIK the specs are  < 20 pA  with no defined sign.

A second point would be to repeat the tests with AZERO on at some 10 PLC or more. AZ on is the normally used mode. Due to charge injection during AZ switching this may show a different input current.

The test using the drift rate can also be used at other voltages, not just near zero. So one could start with some + or - 9 V applied for some time and than watch the drift over a few minutes. The change in input current with input voltage defines the differential input resistance.

[Dr. Frank]

Tested temp; 35,2 ⁰C.
-   DCV, 10V (fixed range), AZERO OFF, FIXEDZ OFF (GOhm);  2 pA test: checks out to about -0,01 V/s
-   FIXEDZ ON, 1 V range (10 MOhm): Zero reading < 1 µV; raises to about 60 µV (short removed).

These values of -2pA and +6pA for the different methods are completely fine, so the DCV input path for the 10V, 1V and 100mV is obviously not damaged.

Remark: The AUTOZERO induces current spikes on the MUX, (which are only imperfectly compensated), therefore AZERO OFF gives the least bias current, which is reproducible and typical -2pA for different instruments.
These measurements require that the 3458A is warmed up for several hours, if I remember correctly I also found such different bias currents with the 10MOhm method when measuring 'cold' and 'warm'.

In our internal Metrology group, we compared 3458As and confirmed the findings of other authors, like Rado Lapuh in his book 'Sampling With The 3458A', where he also describes several methods for determining the bias current, typically -2pA again, and other parameters which are important for digitizing.

The bias current measurement around zero volt gives a mostly constant value, i.e. a linear voltage drift dU/dt for the capacitor method.
This voltage drift will saturate (stop) somewhere at several negative volts, in my case at about -5V.
In the end, you have to pre-charge the input capacitor to about that value, iterating from more negative or more positive values, otherwise this measurement will take forever.

This then gives you the impedance of the input, which is specified in the datasheet to be > 10GOhm.
The bias current on the other hand is only indirectly specified to be < 20pA.

This impedance is mostly determined by the impedance / leakage resistance of the input jacks, the input switch, the PCB material, whereas these -2pA are  given by the leakage of the FETs of the MUX and of the FET Q103A inside the pre - amplifier.

So in my case I determined an input impedance of -5V/-2pA ~ 2.5*10¹² Ohm, or about 2TOhm, which order of magnitude Lapuh also determined for several of his instruments.

Frank

[franklin]

Test AZ ON and 10 PLC; raises to about -55 µV (short removed). About the previous 60 µV; it should be -60 µV.

[Dr. Frank]

So the 10MOhm test is also done with AZERO ON?
Then both methods agree. Great!

You can also use the command APER 1, instead of NPLC 50, which would only do 1 AZ phase per measurement cycle, whereas NPLC 50 does 5 phases, as it's doing 5 * NPLC 10 in reality.
That may also reduce the charge injection.

Frank 

[Kleinstein]

That both cases with AZERO on and AZERO off agree is good. With the error message about DC offset DAC trouble I slightly feared that this could be one of the DAC used for the BOOT voltage offset of the charge injection compensation could have a fault. If both readings essentially agree this means that this part working really well.

For testing it would be more like using 1 PLC only, and maybe the 100 mV range. This would increase the switching effects even more.

[franklin]

Just started testing drift (inspired by TiN and Keysight Service Note 18A). I guess “CAL 2,1” displays the A9 reference calibration voltage and not the present value since the reading is fixed. I’ve been running it for only four days so far and it’s still running. This will be interesting and it’s something new to me just as the 3458A is.

I understand that the display should be kept OFF while the meter is ON for prolonged period to prevent wear and tear of the display. For this I’ve programmed one of the keys with “DISP MSG” and that does the trick for me but I guess this is old news.

TiN has some interesting views on oven temperature; lower it and get lower drift. When doing that I understand the reference voltage seem to raise when die temperature is lowered. Does this say something about oven temp when voltage at my reference A9 is 7,20741? Could it mean that there IS a lower oven temperature in my meter?

CAL? 2,1 (A9 voltage) = 7,20748898; since this doesn’t change with time, I suppose it is the reference voltage when it was calibrated. It matches closely to what I measured on A9 while instrument was open.