Repair : Yes, another little 3458A repair log.

[TiN]

Soon enough EEVBlog Metrology section will be all flooded with 3458A's...

Well, having sour taste of my own #2 3458A I decided than third one would have a better outcome, right!

No, did not rob the bank to buy yet another unit. Instead this unit came to me by it's own will. It was a destiny    .
Long story short, someone local (sorta, still another city 150km away) in Taiwan found my website, and decided to try luck with getting me to fix it.
Best part - is that neither I speak chinese, nor the owner speaks english, or ukrainian, or russian, which are the only languages I know  . With help of 3rd party we get things sorted out and I agreed to look on their sick box, in favour to have visit to their lab on some weekend to attempt a calibration of my 4920M. 

Let's cut the theory, and go right to the ER on this one. 

If you not familiar what is 3458A, I strongly recommend reading my unit one worklog
 
Initial photos and conditions



No cracks or dents, unlike my own 2nd unit this time. It's an old meter, still wearing old HP logo, so it's manufactured before HP T&M division separated into Agilent in 1999.



Just a bit dusty and dirty, but nothing too bad either. Label does not reveal any options, but actually unit itself reports Opt 001 present (extra memory). There is still non-damaged calibration sticker from Agilent on bezel, so it's a good sign. Meaning, good chance that nobody tried to fix sensitive boards without having proper qualification. It's obviously serviced by Agilent some time ago.



Serial is hidden for "just in case" reason, but it's 04xxx, much older than my first box. 

Day one: Power checks and initial diagnostics

After check of correct mains voltage selector switches (this is important, and can cause $337 USD minimum repair if unit plugged into 220VAC mains on 110VAC setting!  ), let's power unit up and check base measurement function.

 

Uh-oh. Test result shows it's 0.7% meter, not the 0.0008% one! Sick-sick poor meter.

Voltage rail	Expected	Measured	Error     Spec(1y)
7V +/-0.5ppm	-7.1637342 VDC	-7.2041502 VDC	+5641.7 ppm +/- 8 ppm
1.9 kOHM +/-8ppm	1899.972 kOHM	+1894.168 kOHM	-3054.8 ppm +/- 10 ppm
10 mADC, +/-10ppm	+9.999994 mADC	+10.073304 mADC	+7331.0 ppm +/- 20 ppm

Clear indication of failure, as even 20 year old unit with only factory calibration cannot drift so much off the spec. And it's indeed faulty, due to self-test failed as well on AC stage, giving error Flatness DAC Convergence : 198 which is test for 10V ACV range/function. Later after ACAL ALL attempt unit stopped operation with 202 HARDWARE FAILURE: SLAVE TEST - AC BOARD error message. It hints that A2 AC PCBA is faulty.

Additional confirmation is also incorrect reading on ACV, ~18.xxx VAC on 100VAC range and ~187.xxx VAC on 1kVAC range. There was no reaction or reading change on either short or open at input terminals.
Time to remove the covers and check boards for any obvious and visible component damages.



All boards visible on this side are original, with manufacturing date codes from year 1989-1990. It's even older than my own first HP 3458A. A9 DC reference module board is standard 8ppm/year model P/N 03458-66509, as expected.
A1 DC board is older Rev.C with weird and strange construction with two red 5mm LEDs hanging back to back on CR108 position. This position is supposed to be populated with 3V Zener diode, just like in newer A1s.



Perhaps HP factory ran out of 3V zeners and needed this unit built for customer, pronto? I'm sure Dr.Frank or Kleinstein know the correct answer to this!



Also there is no Q102 on Rev.C A1 board either. Reference 40 kOHM resistor is hermetic Vishay Precision Group part with custom part number.



This is old A5 outguard controller board, same as one in my first 3458A with EDIP-packaged NVRAMs. Firmware is stored in six UV-erasable EPROMs 27C512, revision 5.3 from week 20 year 1990. We can update it with help of UV-lamp and usual ROM-programmer, such as cheap TL866CS.



Hard to believe, but these 27-year old Dallas chips are still alive! Let's not play with time and replace them quickly
Also board can enjoy little bit of cleaning from dust and rust.



Schaffner dated 1984! Need to replace this EMI filter, if we want to ensure safe and long operation of the unit. These filters are prone to fail and burn if you get "lucky" with bad batch.

I will also replace those capacitors on the A6 power supply board, just to be on safe side. 27 years is not a small duration even for best top tier capacitors.

ADC board is old version as well, with different arrangement around oscillator package.



This is older 03458-66503 Rev.C board, with bit different layout than newer A/D boards. This is the board which matches original drawing and schematics of public HP 3458A Component-Level Information Package document with schematics. It also has U213 populated, but already late enough to feature two tantalum caps near ERC tag location.

Famous hybrid ASIC U180 marked as HPC5 USA 1808-0108. Let's hope it does not have drift or convergence issues, or this repair will quickly turn into expensive disaster.



This board was clearly replaced by Agilent service center before, as it has much newer date codes on packages, around 1998-1999 year, and has Agilent service calibration tag. Also it's likely a refurbished before board, there is still solder flux residue around parts F701, C701, C709, C710, Q701, C406, Q908 (J304), Q905, Q904, CR901, Q901 (2N3094).



No visible damages observed, but resistance measurement on fuses revealed blown F701 which is post-regulator fuse on +15V rail. AHA!

Day next: Repair on A2 PCBA

Probing around revealed blown fuse F701.



After removal of bad F701 fuse and suspected Q701 2N4401 NPN transistor power supply line +14va was powered from external SMU, set to +14.000VDC with current compliance 100mADC for safety reasons.



This allowed unit to pass full self-test and successfully complete ACAL self-calibration. Current consumption was stable around +35 mADC, sometimes peaking at +39 mADC during AC range relay switching/operation. At power on during meter init current peaks at 56 mADC.



Yes, I know, replacing fuse with jump-wire is bad bad bad, but I had no fuse in this package for a quick test. Guilty! 

After 200 OHM resistor test, other parts that got replacement are C701, C710, and Q701 with new NPN. This measure resolved issue with +14va rail and provided stable +14.1..+14.2 VDC on C710. This also fixed error offsets on the DCV/OHMF/DCI functions as well! Too easy to believe!



Let's do same test as before, with stable LTZ from fairy-tale thread...   

Voltage rail	Expected	Measured	Error	_. Spec(1y)
7V +/-0.5ppm	+7.1637424 VDC	+7.1637519 VDC	+1.326 ppm	+/- 2.55 ppm
Extended test for 7V	+7.1637466 VDC, data	+7.1637324 VDC, data	<2.2 ppm	+/- 2.55 ppm

Can it be any better? Within specs even after 24 year old calibration?! I even powered on one of my Keithley 2002's to check that it's indeed correct voltage coming out of the reference.

Easy stuff, A5 board updates

Now since analog inguard stuff works, rest of repair is piece of cake, really.



Firmware is Rev 5. Time to update to latest Rev.9. Here is collection of firmware ROM dumps for reference.
Firmware ROM dumps for HP/Agilent/Keysight 3458A, read by general purpose ROM programmer. From information we have, firmware dumps are compatible with all hardware versions.

As it's already 1:35 am here, I couldn't be bothered to convert textile formatting from my site to BB-codes for forum, so for firmware you just go there, alright? 

After reassembly unit powered on just fine, and completed full Self-Test and Acal ALL without issues. 

 

CAL? 1,1 and CAL? 2,1 values presented as well. Not that this has any use for anyone here, but just nice to have good details in article, and I like taking photos

 

CALSTR? which stores calibration string returns: Date = Wed Nov  4 07:43:04 1992,  Temp('C) = 38.1. This could be deceiving, as CALSTR does not need to be updated after calibration process complete, so it could be just left original from factory calibration. Doesn't mean much. Given test results, meter calibration is likely was happen after 1992, otherwise it would be extremely good result over all these years, which is hard to believe.

Here's unit after 10VDC calibration, reading output of my modified pimped up HP 3245A. You can see data before calibration around 3pm and after. 0.3ppm difference, I'd call close enough!



I'll have this box running another week, to make some comparisons to my gear to ensure everything stable and well, and then travel to visit happy owner, together with Wavetek 4920M, to run some of tests using their lab gear. I heard they have some Fluke toys there....

Summary

This went very well, and a good difference to both my rusty units from eBay, so there still is a little chance that on lucky day, under lucky moon, in lucky place one of us can buy 3458A in condition and with minor issues like covered here for some reasonable amount of cash. Just keep looking on what are you getting, before trigger evil Buy it now button.

Thanks for reading. 

As usual, as end result, I get another article on my site, yey!

[Kleinstein]

2.7 V zeners are generally rather poor performance. So the two LEDs could be an upgrade.

Not having the fuse may not be that bad - the 78L15 usually is current limiting (e.g. 140 mA typical peak). But that opens the question how this fuse could blow in the first place.  I would check the 14 volts rail for tantalum caps, that might gone bad. It might be worth to look for the source of this - seems to be not the first repair in this area. Don't know if it is worth changing the 78L15 - maybe current limiting does not work so well on that one or there are glitches with it.

A cause for a temporary / rare overload on the 14 V rail would be a latch-up in one of the chips powered from there. But it might be hard to find a culprit here.

Looks like there are spare fuses (F001-F003) on the board - just below the switch.

[TiN]

Thanks. Actually I used one of those fuses already (1/8w one) and it worked, but once I tried ACAL it blew again. That was before bad transistor replacement. I checked tantalum caps and replaced one, but they are fine.

Also during 100V test today seems new issue discovered.
https://xdevs.com/mark9_100v/
On this range reading slowly creeps up. I would suspect A3 if not previous test when 10V range just fine. Something with 100v/1kV divider on A1?

ACAL DCV restores reading back to spot on value. It's not the source, as my meter doesn't do this stuff.

[HighVoltage]

You are getting good at 3458A troubleshooting, thanks for sharing your thoughts or repair.
Hmm, why would Fuse 701 be bad if nothing else would be broken down the line ....

[TheSteve]

You are getting good at 3458A troubleshooting, thanks for sharing your thoughts or repair.
Hmm, why would Fuse 701 be bad if nothing else would be broken down the line ....

I believe Q701 was bad.

[Kleinstein]

Even if Q701 is blown, I don't see a reason why this should blow the fuse. It would give 15 instead of about 14 V to the circuits. Most part should be Ok with this. I would suspect that Q701 was blown together with the fuse.  With an other event of blowing the fuse one should really check the reason behind that. This could be a more serious problem, like spikes on the supply. I doubt that replacing Q701 fixed it.

The drift in 100 V range looks suspicious. Maybe time to check the 10 V range again, just in case.

With only 100 V there should not be significant heating of the divider.

[Twoflower]

Not an expert here. But the sparc-gap looks a bit strange. Could it be that it had already fired?

[quarks]

I wonder why anyone would use Dallas 1230 instead of normal SRAM chips

[TiN]

For me it's simple, I had 1230's, but I have no SRAMs for opt.001 Main two RAMs should be NVRAM, unless you want to reprogram all meter settings every reset/power cycle .

[VK5RC]

Great work TiN, Thanks for sharing.
Robert

[quarks]

For me it's simple, I had 1230's, but I have no SRAMs for opt.001 Main two RAMs should be NVRAM, unless you want to reprogram all meter settings every reset/power cycle .

Hello TiN,

the price for a 1230 is 10 to 20 times higher then a normal SRAM and you do not even need the option 1 for any of your tests right now.
Also if you use a 1230 the very first time the internal battery will be triggered to start to drain and even if you replace them later with the normal SRAM, your 1230s will not have the shelflife of an unused one anymore. I would not have done it, because most likely a 1230 will be harder and harder to get in the future.

bye
quarks

[TiN]

That is quite alright. I got unit with NVRAMs and I return it with nothing less. Also schedule does not permit to wait another week or two for SRAMs to arrive, unit planned to be back to owner next weekend, unless something goes wrong.
I'm aware of DS1230's operation, thanks.

Today I tested DCV noise using methods discussed before in article and this EEVBlog thread.











Green line is my own meter, red line is unit from this thread. You may notice that noise after NPLC 100 on user's box goes down, as one would expect, but on mine it creeps up by a lot.
I even retested my meter, using very same short, same setup, and same EZGPIB code. Hmm...

Perhaps it's all because user's meter is upside down and electrons don't fall down from ADC U180? 

[ManateeMafia]

Didn't that 3458A #3 go through some strong earthquakes? Some settling of contents and NPLC's may have occurred.   

[TiN]

Who knows.

I did retest again for DCV.



No issue on 10VDC, both boxes are within ±0.2ppm variation over 11 hour span, which is down in noise of used external LTZ1000 module during measurement.
This time I used 1kV range on both meters. Same slow ramp as on 100VDC range, ~+2.8ppm/hour on DUT. Front or rear terminals - same thing.

Seems A1 has some problemo.

Btw, this is good example to show that "SELF TEST PASS" message CANNOT be used as measure to judge if unit working properly. Test procedure during self-test is simply too fast to catch any of these pesky drifts. This applies to any DMM, not just HP in topic. Buyers beware  And owners, I can't stress more, must have GPIB logging setups running from time to time to even see these failures.

[plesa]

Btw, this is good example to show that "SELF TEST PASS" message CANNOT be used as measure to judge if unit working properly. Test procedure during self-test is simply too fast to catch any of these pesky drifts. This applies to any DMM, not just HP in topic. Buyers beware  And owners, I can't stress more, must have GPIB logging setups running from time to time to even see these failures.

This confirmed my worry with 3458A on eBay and I'm curios if this will be obvious during calibration process.
Same for convergence errors, when you needs to run meter for few hours on higher NPLC.
Are you going to send it to official calibration?

[Kleinstein]

One point worth checking might be the input bias current. If the input amplifier has excessive bias, that is drifting, this could cause the drift in the 100 and 1000 V ranges, as the divider is a relatively high impedance source compared to the protection circuit in the 10 V range.
Testing the input current is a good idea anyway, as dirt of leaking protective diodes can cause excess current.

[TiN]

What would be the best procedure?

I probed TP11 with regard to input LO and multiplied output with *100.007 math gain. Here's the result:



Interesting enough, 100VAC range using 70VAC input does not have drift issue . Using different input signal path for AC?
Graph below is AC 1kHz.



As side bonus, I captured ACAL DCV procedure, using my 3458A as sampler at NPLC1 speed



CSV-datalog of ACAL DCV, measured TP11 point on A1

Took photo of A9. Bottom cap side has funky arm to act as spacer.



Custom HP marking on LTZ1000A