Repair worklog : Old rusty HP 3458A

[TheSteve]

My new 'baby' arrived, all intact form the travels. Clean on the outside, under a ~2009 calibration seal, the inside looks like someone had smoked near it, lots of really fine black dust esp near the fan, unfortunately that is also close to the voltage reference. It had a known NVRAM failure (replacements arriving soon I hope) but also showed an ACAL AC required warning, (photo1) on first ACAL ALL , showed a new ?downslope error fault ( didn't photograph) but thought all that smoke wouldn't be good. After 2-3hours of cleaning with static protection, and re do ACAL AC it has resolved. Still need new NVRAM.

What programs do you guys use to 'talk' to your 3458, I have a Prologic GPIB-USB and Win XP PC? (PS I am not a good programmer!!!!!)
I want a good copy of the cal data before I change even the non CAL RAM.

That fine black dust has been inside every piece of test gear I have purchased that was used in any kind of industrial environment. I don't think it has anything to do with smoking(you'd smell it if it was). The stuff is generally pretty sticky, I've had some luck removing it with a 3M static safe vacuum but in extreme cases it requires compressed air which is not ESD safe.

[TiN]

ew ?downslope error fault ( didn't photograph)

Auch, hope it would not haunt your unit back later.

Keep us posted! Hope your repair would be easier than mine 

[bingo600]

What programs do you guys use to 'talk' to your 3458, I have a Prologic GPIB-USB and Win XP PC? (PS I am not a good programmer!!!!!)
I want a good copy of the cal data before I change even the non CAL RAM.

PHK uses a prologic with pylt
https://github.com/bsdphk/pylt

And i think that John Miles toolkit supports prologic also (Windows)
http://www.ke5fx.com/gpib/readme.htm

There is a 3458.exe in there , that ought to work once you have set up the toolkit

/Bingo

[VK5RC]

Thanks guys, my programming will have to get better. I think the KE5FX route looks the simplest (and safest) path for me. I have used other components of KE5FX's suite  before for screen print off of a Spec An. Again thanks to all esp TiN, appreciate your efforts. 

[plesa]

Does anybody want to follow TiN restoration?
http://www.ebay.com/itm/HP-Agilent-3458A-8-1-2-Digit-Digital-Multimeter-8-5-digit-DMM-/111824831441?

seems that reference boards, switching rods and transformer are popular components (missing or faulty)

[ManateeMafia]

I wouldn't doubt it was bought from same people. Bad transformer already adds $300+ to cost. Bad inguard supply is probably repairable.

The rest is probably faulty ...

[TheSteve]

I believe that is OldSchoolTechCorner's 3458A that he bought from the same source TiN did.

[TiN]

Yep, it is. I hope we will hear about it again..

[TiN]

Hmm, this project would not let me go...
I had DMM hooked to freshly modified 3245A with updated reference, monitoring 10VDC output, and reading was going down and down. In just a hour it was already 1.5 ppm's low. Ran ACAL, back to initial level, and then started drifing down again.  I suspected 3245A at first, so did next:

Connected 3245A output to K2001, using same wire as it was on 3458A.
Connected EDC MV106 10VDC to 3458A at same time.
Both meters sampled via simple LabView bodge, at NPLC10 both (NPLC100 and 200 shown same drift)

K2001 shown stable 10V output from 3245A, with noise ~1.2ppm (normal for 2001).
3458A drifted down again.

Then reversed sources (MV106 to K2001, 3245A to 3458A) , same stuff. After 5 hours it drifted down -6.5ppm's already. It's not room ambient temperature, it kept fairly constant ~27°C. I tried running colder (~22°C) and drift speed was bit faster, but value not return back when warmed up back to +27°C.

This morning switched sources back again, will see what happen when I get back home. 

There is a service note 18A, describing long-term and short-term drift issue. One of parts mention A9 reference returning to "pre-aged" state if left unpowered for long time, so I hope that's the case. I have meter for a week 24/7 already, note mentions 6 weeks to return to normal condition.
Other part of note suspects drifty A3 A/D board, with only solution to replace it. Oh noes, not A3 again 

I'll hook DMM to measure actual A9 LTZ's output to see what is happening with reference voltage.

[ManateeMafia]

TiN,

You may need more time. Attached is a spreadsheet for two 3458A's that I monitored for almost two weeks last year. They were both turned off for months and then left on for a couple of days before I started taking readings. Day 1 is after ~48 hrs warmup.

All data was recorded in Open Office. Hopefully, there is nothing that shows any serious issues with my 3458A's 

[TiN]

Can you save in Excel, plz?

[ManateeMafia]

Here you go. If this doesn't work, then I have more options.

[TiN]

Converted to PDF for openness  Thanks.
I'll do monitoring during this week for CAL 72, as SN 18 tells and make same looking data.
Have bad feeling for this... 

[ManateeMafia]

The results include all data. I think the service note only asks for 7 days or so for the results to be valid. I didn't like my numbers so I kept on recording them.

[Dr. Frank]

TiN,
You did not mention using ACAL in between? I mean after the first ACAL after one hour, and especially, when you changed room temperature.
It is very important to run this ACAL regularly, even if ambient does not change, but it's necessary if the RT changes by 1°C only!!

If ACAL DCV always brings reading back to the initial reading, the LTZ will not be the source of this excessive drift, and AN18 would not apply!!

The other components for DCV inside the 3458A have TCs of about 1ppm/K.
The only other possible source of drift in the DCV 10 range are the amplification resistors inside the A/D ASIC, which generate the +/-12V references for the multislope conversion. As said, an ACAL DCV will always cancel also their drift, if the LTZ itself is stable.

But if you did use ACAL more often than you described, and the 3458A did not display the original value afterwards, only then the LTZ itself may drift, due to the previous 95•C oven temp, i.e. it might show  hysteresis in form of slow creeping.

Maybe, in this case only, you directly check LTZ voltage, which you should know from initial assembly.

PS: Verify, that AZERO is ON! (annunciator in display NOT lit)

Frank

[acbern]

As Dr Frank points out, ACAL heals the usual drifts, and needs to be done frequently. More frequently than most users probably think, in my experience at least. You only find that out if you measure a source more stable than the 3458A, which may be rarely the case. So having one is nice. Other high accuracy meters which do not have a digital ACAL-like function may be inherently more stable (within certain limits). Thats probably the price you pay using a 3458A.
 
The AN 18A also describes a second cause of drift that may caue the unit to be out of spec, besides the LTZ1000, which is the hybrid used on the A/D. I have seen hybrids with the ceramic package (coincidence or not, at least reasonable due to hermeticity issues) be more stable, and the older types with the metal cover often seem to be at the edge or slighly above the SN 18A limits. Thats not too bad, actually, it only means you need to run ACAL more often (but thats good practice anyway).

Maybe what you see (likely amongs others, since 6.5ppm in 5 hours is quite a lot) is an excessive drift of that hybrid. What I would recommend is to do the SN 18A test (which takes a week, but after 2 days you can see a trend). I would recommend to read the related parameter on a daily basis, you then also see if it converges or just continues to drift away (may take you longer than 7 days to figure that out). I remember had that test done on one unit some time ago and the daily drift and the overal 7 day drift where in direct relationship (and only slightly above limit, no issue for use of the item once you know). As Keysight requires the test do be done for 7 days, they most likely assume that there are units that initially drift more and then converge, and meet spec in a 7 day window, but not after say 2 days. Doing a test once a day gives more data points with respect to this matter.

[TiN]

I got impression before that ACAL is targeted to cancel errors from temperature deviation. E.g. if meter was calibrated at +23C , but later ambient raised to +28C then ACAL would adjust gain and offsets to bring it back "in-cal", but now for 28C. Plus error of 40K VPG resistor tempco itself. So if temperature stays constant, ACAL should be doing almost nothing. This is also confirmed by Fluke's metrology papers I was reading before.

Also second to that, such large drift (more than 6 times of one by Keithley 2001, which have LM399), is not normal, even without any ACAL. So yes, I'll do 18A's and Ref measurement test and to see how it goes.

Main purpose of getting meter like 3458A is to measure more stable sources, either short or long-term, and watching it drift away such as this kinda defeats the purpose, as it would surpass 1year specs in just few days!

Graph data:



Step ~8K seconds - changed EDC MV106 to 3245A outputs.
Hard to see, but just before end of graph - I did ACAL few minutes ago, and it does go back 9.99998 VDC (and already 9.999972 while I posted this).

[Dr. Frank]

I got impression before that ACAL is targeted to cancel errors from temperature deviation. E.g. if meter was calibrated at +23C , but later ambient raised to +28C then ACAL would adjust gain and offsets to bring it back "in-cal", but now for 28C. Plus error of 40K VPG resistor tempco itself. So if temperature stays constant, ACAL should be doing almost nothing. This is also confirmed by Fluke's metrology papers I was reading before.

Also second to that, such large drift (more than 6 times of one by Keithley 2001, which have LM399), is not normal, even without any ACAL. So yes, I'll do 18A's and Ref measurement test and to see how it goes.

Main purpose of getting meter like 3458A is to measure more stable sources, either short or long-term, and watching it drift away such as this kinda defeats the purpose, as it would surpass 1year specs in just few days!

Hello TiN,
you are basically completely right about the function of ACAL.

For a good working instrument , as the ones of ManateeMafia, acbern, and mine, you may achieve stabilities on the 3458A on the order of ~ 0.1ppm over >24h, but without ACAL in between (!), when ambient temperature is constant. I have measured exactly that, over 35h w/o ACAL!!

Anyhow, ACAL is an important function for the 3458A, so should not be underestimated to be used regularly.

You have initially drawn the wrong conclusion from that strong drift, i.e. that the LTZ reference is not ok.
AS ACAL brings reading back to origin, your LTZ ref is definitely ok.

The other possibility , i.e. a drifty ADC ASIC (also mentioned in the AN18, which I did not remember) is the other possibility, or maybe a problem with AZERO, which you might easily check with zero input.

And you're also correct, that drift you see, is much too high to be explained by the known bugs .. so I hope you will find a "reversible" error instead.

Frank

[Macbeth]

Time to swap ASIC's over between the 2 boards perhaps? (We didn't find out if the earlier boards ASIC is faulty after all).

This 3458A repair is a bit of a rollercoaster  No wonder OldSchool' got shut of his. I hope the new owner has the same tenacity as TiN  (come out, whoever you are! )

[Bill158]

Hello TiN,

It's no problem to blank the display completely including the annunciators, and without changing the firmware code.    I just tested that. 

Simply send "DISP OFF" plus a message text, that is a blank string inside two single quotation marks (!!), to the 3458A:  DISP OFF,'    '

Explanation - as always RTFM - in the user manual, chapter 6 - Command Reference, about "DISP", and the last line of the examples contains the crucial hint, p. 171/172 of the online pdf document!

You only may have to format these quotation marks correctly in your compiler, as these also were used for text string delimiters in many programming languages.
In Pascal the text string has to be formatted as:

     cmd:= 'DISP OFF,''    ''';

Frank

I haven't seen this method discussed before but pardon me if it has been presented.  If the 3458A is not under GPIB control and you want to turn the display completely off.  From the front panel press "BLUE" to get to the Command menu and then scroll to the "DISP" command.  Enter ">" and then scroll to the "DISP MSG" command and then press "ENTER".   The display will be completely blanked except for the "SMPL" which blinks every time a sample is completed.  To get the display back just press "ENTER" again.  I don't know if this saves the display from aging and going dim.  I am not that familiar with VFDs.  Maybe someone with more knowledge and experience on this subject can give us some more information.
Bill

[ManateeMafia]

This 3458A repair is a bit of a rollercoaster  No wonder OldSchool' got shut of his. I hope the new owner has the same tenacity as TiN  (come out, whoever you are! )

Agreed. I have been doing the beer and popcorn routine for the last week. Wonder why I haven't gotten my current project fixed yet?

[TiN]

Bill158, your way does not remove annunciators, basically it's same as sending DISP MSG,'    '; GPIB.

There are two main reasons why our lovely VFD screens die out:
1. Segments which are being turned on for long time are loosing their active layer, causing brightness drop. Only way to help this - don't let pixels be lit for long time, if you don't need them. You could mitigate this a bit by raising driving current, but that would just speed whole dimming process even more.
2. Filament getting oxidized from constant operation, reducing overall efficiency of tube emission. This can be improved by temporary driving filament with much higher current till it red glows to burn oxides off. I covered this in detail on Keithley 2000 VFD rejuvenation task.

Having fun, guys? That's okay, one day truck with golden bars will crash on my street too 

Attached coax cable to LTZ A9 output, hooked to K2001.



Guess what, it's not doing it now! Murphy gotcha. LTZ output stays within 1.8ppm window, likely much much less, K2001 just have more noise (I have LED lights around on, soldering iron on, Tek LA on, far from quiet setup now). 3458A and 2002 connected via same cable to EDC MV106 10V, both NPLC 50, both MATH nulled at beginning of measurements.

3458A shows 7.5uV (0.75ppm), 2002 shows 12.3uV (1.23ppm).

 

[Bill158]

Bill158, your way does not remove annunciators, basically it's same as sending DISP MSG,'    '; GPIB.

There are two main reasons why our lovely VFD screens die out:
1. Segments which are being turned on for long time are loosing their active layer, causing brightness drop. Only way to help this - don't let pixels be lit for long time, if you don't need them. You could mitigate this a bit by raising driving current, but that would just speed whole dimming process even more.
2. Filament getting oxidized from constant operation, reducing overall efficiency of tube emission. This can be improved by temporary driving filament with much higher current till it red glows to burn oxides off. I covered this in detail on Keithley 2000 VFD rejuvenation task.

All I was trying to do was turn off the pixels so that #1 wouldn't be happening as fast as leaving them on.  The fact that the "SMPL" flashes once in a while isn't going to cause much degradation if any.  I figured that these VFDs would be very much like a CRT in an oscilloscope.  Even if you turn the intensity all the way down you are still causing oxidation of the cathode.  Look at many 576/577 curve tracers where the intensity was left up but just a dot at the lower left corner of the CRT.  Burned a hole in the phosphor so that it is dim right at that spot.  But at least you are not losing the active layer on the pixels if you don't have the display showing anything.
Thanks for the answer on VFDs.
Bill

[VK5RC]

Just got off the phone from Keysight Australia, 
Standard Calibration (including altering cal ram if needed)  $1331 (plus gst)  7-10 working days,
NATA calibration (includes error bars)  AUD $1730(plus gst)
One time repair (if not grossly damaged) AUD$3343 (plus gst).
Ooooh ooooh!

[Macbeth]

I don't understand why hobbyists are really arsed about NIST traceable calibration, etc. Even with it the instrument could pass and be grossly out of what we really want (which is perfection!). The manufacturers stated cal figures are the extreme ends of the bell curve. Yes, as engineers we have to treat all devices as if they are at that extreme. But just not MY device, mmmkay!   

Now I guess the cal lab will simply pass anything even at the ends of the curve rather than bother with actually adjusting the device to meet spec. When you pay extra so they do adjust they don't bother perfectly zeroing the adjustment, just make sure it is within the min-max deviation.

All of this is great for traceable calibration performed by and for grunts who act like trained monkeys. It allows a way of tracing legal blame for when a space shuttle explodes, a drone misses target, or a nuclear power plant goes spreading it's radioactive goodness around the planet...

I just want all my instruments to agree with each other. The "Macbeth volt" and the "Macbeth second" are more important to me. Of course I want them to be exactly 1V and 1Hz in the rest of the natural universe as well as just in my lab, but I'm not so disappointed if my 1V is 10ppm out... ok maybe I am, 1ppm isn't so bad...

If we can get the "EEVBlog volt" which is basically all the voltnuts accuracy averaged then I reckon it will be a better standard than NIST anyway