Older 3456A Repair

[matches]

Hi friends of old bench DMMs,

recently I got a older HP 3456A. I read several topics here and got some useful intel out of them. Also the reason why I was looking for a 3456A.
When I say "older" I mean it is an early model. It has the rather unusal fan installed. Look at the pictures attached. Fan_1


In the auction I only saw a very bad picture of the display. At least it was showing some numbers.
I have not powered it on yet. When it arrived, something was ratteling around.
It turned out to be the springs of the front panel buttons. So I guess it was good to not apply power.
See FP_Buttons.

I found all but one and put them back. Now the only button without spring is the SRQ button.
I'm hoping it will work without the spring, otherwise I will find a solution.

As I read in the other threads it would be a good Idea to recap.
I desoldered most electrolytics but it seams there is only one bad (reduced capacitanca).

For the bigger Sprague ones, I'm not shure.
C6, the 12 mF was measured to about D = 0.5 at 120 Hz (see C6_12000_uF).
ESR is beyond the capability of my DerEE (shows zero at 120 Hz, 1 kHz and above, it can't measure anything)

So is it OK or bad?
Does anyone have a datasheet?
Also what could I use as an replacement? I'd like to put something with similar physical dimension in (if fits electrically of course).

And for C5, the 4 mF Sprague, I only did an in-circuit measurement. I only would desolder it, if I have a replacement.
The terminals are big and I fear to stress it unduely during desoldering.

Also, does anyone have Servince Note 12?
I found 13, 14 and 15 on TIn's page.
Note 12 is about remmoving the fan. Maybe there are parts required which are not really avalable anymore.
But still, I'd like to have a look.

For pic's from a complete and cleaned unit, we will have to wait a little until I have all stuff together again.

Regards
Martin

[widlokm]

Hello Martin,

While I haven't seen this unit before I have some experience with old big Sparque capacitors. In my opinion the are very reliable, with parameters much better then typical cheap caps that You can buy today. I have a few of them (from some military scrap) and ESR or leakage is very low, and we are talking about >30years old caps. Of course some of them needed about an hour to "form" (not sure about English word - running capacitor near rated voltage from current limited power supply).
Dissipation factor of 0.5 seems large - can You check it on other meter (with DC bias preferably) or on scope to confirm?
Michael W. 

[MadTux]

From my experience, it's mostly the physically smaller capacitors (for +/-15 and 33V rail, if I remember correctly) in the inguard PSU in the  back that go bad. Sometimes,  the big blue 40000µF sprague cap for inguard digital is bad aswell, check that one if there is nothing on the display.
Never had outguard digital PSU fail on one of mine, so these probably don't go bad often (or not yet).

Fan looks like it's mounted the wrong way. I would rip it out completely, since the newer 3456As have it removed too. Makes it quieter and less dusty inside.

[lowimpedance]

That fan does look wrong, I have both versions of this meter at work so I can check on Monday if you want to be sure.
If removing the fan I think the pass transistor need's to be moved to the back panel as well.
Service notes are rather hard to find, (those you refer to are ones I found by accident when looking through some old handbooks at work when doing some repairs, no  12 though. ).
Frankly replacing all the caps in the PS at the rear is good insurance for the future as many errors are caused by dodgy caps.
 These are great meters, you will enjoy using it. !.

[dietert1]

Another recommendation: Unless you are absolutely sure what you are doing, don't try to calibrate that meter.
After fixing problems you may well arrive at a perfectly calibrated meter. Many years ago i bought two of them and also two Fluke 8502A. They also needed some repairs but all four still agree very well. I never felt any need for calibration (which would probably cost more than all four meters together).

Regards, Dieter

[jfphp]

I do not agree with the calibration : I have seen on a lot of 3456A that the drift over some years could be very important. Another tricky problem with the 3456A is the null with shorted inputs : there is no way to trim it and in most 3456A it is out of tolerance. Source are leaky fets in the input switches and it is a tediuous work to identify the faulty one.

[dietert1]

With respect to drift, would you give a number? Not everybody here is into calibration business.
Concerning the FET switches, isn't that a good example why calibration doesn't make sense until fixing all problems?

Regards, Dieter

[jfphp]

Several 10 E-4 in DC and Ohm. It is of good advice to make the performance test before the calibration (0 chek is not in the calibration because there is...no adjustement) and, from time to time, you escape a part of the cal process.
Tek (espacialy analog scopes) combined perf and adjust in the same procedure but very long (comparison between a Tek 465 and a HP 1740A : more than twice the time required by the second).

[matches]

Of course some of them needed about an hour to "form" (not sure about English word - running capacitor near rated voltage from current limited power supply).
Dissipation factor of 0.5 seems large - can You check it on other meter (with DC bias preferably) or on scope to confirm?

Hi Michael, good advice, I'll try that. C6 is in precondition with it's 16 rated volts. After that I do some measurements.
Here my Hameg 8142 with modulation capapilty comes in handy. I can apply AC with DC bias, current limited up to 1 A in 2 quadrants (source and sink limited to positive voltage).

Fan looks like it's mounted the wrong way. I would rip it out completely, since the newer 3456As have it removed too. Makes it quieter and less dusty inside.

Yes, it seemed a little odd to me as well. There is some kind of housing or duct where the fan sits in (not shown in the pictures).
From the backdating section and Change #10, which adresses the fan, I interpreted there are changes in some heatsinks.
For now I just had a rough look. But it's why I'd like to see this service note 12.
At least I want to hear the sound. But yes, I also prefere it without fan.
There was a lot dirt accumulated on the fan and on te filter. The filter was completly black. After a ultrasonic bath it is white again .

Another recommendation: Unless you are absolutely sure what you are doing, don't try to calibrate that meter.

Hello Dieter,
that is a concern of mine. I have a 34465A I could check it against. It is not sufficient for perfomance check or calibration but it's the best I have.
But I can take it into account and be aware of higher uncertainty for absolute measurements.
The big problem is the lag of sources. Only a couple of power supplys up to 30 V and a function generator.
It's a project or two for the future.

I do not agree with the calibration : I have seen on a lot of 3456A that the drift over some years could be very important. Another tricky problem with the 3456A is the null with shorted inputs : there is no way to trim it and in most 3456A it is out of tolerance. Source are leaky fets in the input switches and it is a tediuous work to identify the faulty one.

Thank you for the hint. At least the zero check is easy to do with a piece of copper wire.

Regards
Martin

[matches]

So I did measure the cap. It's a little better compared to the measurement of the DerEE.

I aplied 1 V DC bias and about 1 V peak peak sine wave modulated with 120 Hz, channel 4 on the scope, directly conected to the power supply output.
This one is just informative.

The voltage for calculations was grabbed as close as possible to the cap to remove errors of lead resistance.
The cap filtered it to about 234 mV (channel 1) with a current of 2.03 A (channel 3). All peak to peak values.
I also measured current and voltage with DMM's to 79.8 mV and 702 mA (AC-RMS, Z = 113.7 mOhm).
From the picture you can see an average phase of 68° giving a dissipation factor of 0.4 with 12.6 mF.

So... a little better than 0.5. But still...
Any recommendations?

[dietert1]

 I would just make a list and get this done, it's worth the effort.
A KEMET PEH200KC5150MB2 (15mF 40V) is specified with 16 mOhm at 100 Hz. You can safely assume all available axial caps are old stock and no better than yours...

Regards, Dieter

PS:
A) I now see that capacitor can be a 20V version.
B) Useful link: https://xdevs.com/fix/hp3456a/
C) I just connected one of our HP3456A to a AD587 based calibrator that gave 10.00002 V in 2009. After 30 minutes it is 10.00005 V. That means a 3 ppm change after 10 years.

[matches]

OK, I give the original Spragues a try.
Thanks for the link. Here I got the service notes 13 to 15 from.
The smaller caps are going to be replaced by ones with higher capacitance as well.
Mainly because caps with equal ratings are smaller these days. I'd like to use the original PCS holes without having a cap flapping around.

The 3456A came with two different reference boards. Mine has the A24 with a referece "1902-0926" A25 Reference.
In this thread: https://www.eevblog.com/forum/testgear/teardown-hp-3455a-reference-module-11177b-01723/ one can see some measurements of it on the 3455A reference board.
Sadly nobody seems to know what device it actually is.
The other reference board (A25) has a LM299H.

Maybe reports of very stable and drifty ones are related to the different reference boards?

Regards
Martin

Correction: It has the A25 reference board with LM299H.

[dietert1]

So, i opened that HP3456A and found 03456-66525 Rev3 as reference. Board has a handwritten black "10". Chip is marked NS 6-0330 919. After 24 hours the result is 10.00007, so another 2 ppm.
Maybe a deviation of 10 ** -4 means the calibration expert was interrupted by a phone call the year before....

Regards, Dieter

[dietert1]

I just have some questions:

When i look at the HP3456A A25 reference circuit, there is a cap C501 39pF on the amplifier that could possibly have leakage. I was wondering what was the function of a small cap like this in a DC circuit? How to dimension that 39pF?
If the amplifier U501was unity-gain-stable, one could use a uF capacitor to avoid amplification of noise.

What is the function of R502? Isn't that just a noise generator? (There is no impedance match on U501 inputs)

Regards, Dieter

PS: I found the in-guard power supply has two voltage doublers where two diodes are missing to prevent reverse operation of C7 and C16 during power-up. Those diodes help for durability of the caps, if the caps aren't bipolar.

[Kleinstein]

A 39 pF capacitor is likely either PS or NP0 ceramics. Both types are not likely to develop leakage unless really mistreated (e.g. wrong solvent for cleaning could damage PS caps).

The capacitor may be needed for stability as the extra transistor in the loop can add some delay. It could likely be larger with no real problem. However it only filters about half the voltage anyway and most of the noise is from the reference, not the amplifiers. Higher frequency noise noise is not that important in the DMM. Though in a really low noise version some noise at around the modulation frequency could be a problem.

R507 is for input impedance match of the input of U501 - the inverting input has something around 2 x 50 K effectively in parallel.

[dietert1]

Thanks, now i saw the other 45.5 K in the resistor array in series with the 8.985K . Yes, that means R502 is for impedance matching.

Concerning C501 = 39pF of unknown quality: As far as i understand U501 is an OP-07, which is unity gain stable. Will try other values like some uF.  Also across the positive input.

When looking into the HP3456A reference circuit with a scope, i saw spikes of about 20 or 30 mV. Probably they originate from current switching in the ADC. The spikes are strong on emitter and collector of Q501. C506 is too far away, it doesn't help. The spikes enter into the reference circuit through the base of Q501. I could get rid of them with two 2u2 foil capacitors i put across CR502 and from its cathode to the collector of Q501.

By the way, in our service manual i found an EMail of 2000. And i remember getting the manual roughly at the same time as the two 3456A DVMs. Both units are from higher serial number ranges, yet one of them still has the fan. They were never calibrated in those 19 years and with the same AD587 calibrator one unit shows +7 ppm and the other one shows +10 ppm. So there is a 3 ppm difference after 19 years. I spent some hours doing maintainance on them. For example the older unit had once more lost some of those front button springs.

Regards, Dieter

[matches]

They were never calibrated in those 19 years and with the same AD587 calibrator one unit shows +7 ppm and the other one shows +10 ppm. So there is a 3 ppm difference after 19 years. I spent some hours doing maintainance on them. For example the older unit had once more lost some of those front button springs.

The have a difference of 3 ppm now. Do you have data from back than to compare with?
Still impressive!

Damn those front buttin springs! Do the buttuns still work?

I my have a blind spot. Where is R507?

One correction: my unit has the A25 Reference. If it is runnign fine again, I may try some mods. At least put some fome for thermal isolation and better stability of the heater.
I'm far from knowing what to do to improve a reference. First I have a project with a LM399 as a separete ref.

Also I took a high res picture of the A20 Board. These older designs are yust nice...
At the PSU, I replaced all electrolytics, except the 4000 µF Sprague. This one has 5 pins. If there are problems, i can replace it later.
And found another electrolytic in the logic section. It is placed at the power input. So I assume it's fine to replace it wth a film cap of higher value.
The spark gap looks like it had some, does it?

Regards
Martin

[Kleinstein]

The plan I have may not be that well readable so R502 and R507 are likely that same.

The way the ADC is build, it does draw quite some current pulses. Ideally I would expect quite some local buffering at the ADC. Local buffering makes the compensation more tricky - so the 39 pF may be a good compromise to make it stable and still fast enough to react to the load. Any change here could effect the ADC, possibly changing the gain a little.

If were to build a good LM399 based 12 V reference, I would use 2 x LM399 in series and divide down from 14 V to 12 V. This would result in less critical resistors, as only some 15 % of the output depends on the resistors. The buffering could improve with a faster OP than the OP07 (e.g. ADA4077).

[matches]

Ahh, OK. Yes, some numbers are beyond readable. Also some junction points leave room for interpretation.

I also thought about 2 LM399 in series. Some use multiple References to reduce noise.
Connectim them in series, I assume square root sums the noise of the references.
So maybe still better than amplified?

Regards
Martin

[Kleinstein]

Noise wise it is the same when using 2 references in series or average 2 references and than amplify by a factor of 2. Both ways are some 71% of the noise of a single reference amplified.

Using 2 in series would be mainly to start with 14 V and thus have less critical resistors. If starting from 7 V and than amplify to 12 V would need resistors with about 5 times better stability. At the level of stability useful for a 6 digit meter it may be cheaper to use a second reference instead of very good resistors. The lower noise is than more like an extra.

[dietert1]

The second unit has similar spikes and they originate from U402 in the ADC. There is a 100nF buffer cap C401, maybe that should increase. And i measure 140 uV (14 ppm) across the wire from the reference TP501 to U402 pin 7.

Yes, the front panel buttons do work again. I had to re-insert several springs and replace one lost spring. The other unit with a 19xxx serial doesn't have that problem.

The external resistor network of the reference may appear suspect, but reseating it and reseating the reference module did not result in a single ppm change. With my little mod i don't see noise anymore, before i had +/- 1 ppm changes per ADC reading. Whatever you do, a 3456A won't become that voltnuts thingy striving for ppb. Anyway, a precision of some ppm serves well for many labs.

What i would like to do is bring our AD587 geller reference to some calibrated 3458A in Germany and do a check. It is a USB device and its 37° thermostat is up within 10 Minutes. I remember trying that 10 years ago, but couldn't find anybody. We are near Karsruhe.

Regards, Dieter

PS: Just found an old comparison of our two HP 3456A. In 2008 the difference at 10V was 6.5 ppm. Same sign as the 3 ppm they show now.

[matches]

Very cool!
I'm not familiar with the AD587. How accurate would you expect a calibrtation of it to be?
Or what uncertainty would you consider acceptable?

So the 56 is up again. Not completley assabled due to lag of patience 
At first I thought its bad because the voltage input had zero resistance. It turns out it's not a good idea to plug the cable from the input selector switch at a test point anstead of the iinput pin 
You can see it in the picture of the A20 board I posted previously.

Measuring the breadboarded LM399 mentioned above, it is reading 15 to 18 counts low compared to my 34465A. 10.00045 V compared to 10.00061 V
And a nice surprice: An older 34401A differs by only 1 to 3 counts. 

Zero Offset is about 4 counts in the lowest range.
A Vishay Z201 100 Ohm was measured to 99.999 something Ohm's which is very fine.

I did not test everything yet, but so far it is nice. I think I'll let it run for a while, assemble it completley of course and do some more testing.

When I measure voltages (at the boards on test points etc.) How far off would be acceptable?
One percent? Ten Percent? As a roule of thumb...

Regards
Martin

[dietert1]

So your 3456A appears to be very stable, too. The internal voltages on our units are pretty precise. Both 5V are within 10 mV, +/- 15 V within 50 mV.

In one of our units the in-guard 5 V regulator gets extremely hot, because it is not sitting flat on its cooler (bad thermal contact). That's a pretty cheap error. Don't know yet how to fix that. Another one: The +33 to +46 V voltage doubler works on the in-guard 5 V output and produces voltage drops there, that are pretty clear on the scope. Maybe there should be a 10 Ohm resistor in series with CR 8. Also i am not sure if the 39 V zener CR 5 should be there or not. Our unit don't have it. Also the out-guard logic board (bottom) lacks some 100nF buffer caps, especially close to the isolation interface U21.

Our AD587 standard is based on a Geller reference board. The specifications say 5 ppm/°C, so i made a temperature controller for it running at about 37°. Everything very simple, no PT100 but NTC. It has a USB interface for power and control. There is a bistable HG-wetted relay inside that switches all 4 poles of the DVM (except guard) between an external input signal and the reference. So we can measure with the DVM with interleaved voltage reference checks. The reference runs very stable. From what i have seen i think it is at 8 +/- 2 ppm above normal. I will bring it to the metrology meeting in Stuttgart to have it checked against a certified Fluke 732A (1 ppm). Let's hope it won't be a hot day.
As far as i remember the specified uncertainty in the 10 V range of the 3456A is +/- 5 ppm . That depends a bit on the environment. What i have seen is better, maybe +/- 2 ppm - with all covers closed! Currently the thermometer displays 20.2°C.
The 3456A has several design problems/limitations that contribute up to10 or 20 ppm each. One test is reversing the 10 V standard and see what is the difference in displayed voltage. With our two units i see about 4 ppm changes. Also, when i operate the front panel switch "Guard = Lo", there is a change of 3 to 5 ppm. Another interesting experiment is the zero test with a direct short on the inputs. In autorange mode with autozero it can take several minutes until the displayed sign starts changing with noise. Maybe temperature changes on the input connectors.

Regards, Dieter

[matches]

When I came home from work I saw th 56 had "cought up" a little. It now shows only 3 to 5 counts less 

I was asking about the voltages because the 5 V for digital was only 4.8 something. And as a beginner in electronics I'm interested in a rule of thumb of course.
And sadly it takes a long time to undestand a circuit. So it is hard to follow your thoughts at times  yet every input helps

Regarding your temperature issue: maybe it's possible to gently convince the parts to come closer. Maybe a remount with a drop of thermal grease will do the job.

If you want a tracable calibration for your reference an 3458A alone may not be good enough for you. So the Idea with the 732A is good. Ideally you would bild some history where you can estimate the long term drift.
I was thinking of building a more versatile reference with 0.1, 1, 10 and 100 V, calibrate that with a 58, option 2 at work and then check my meters at home.
Or maybe send one "precise" Meter away for calibration and use it as a reference or transfer standard.

The 5 ppm Spec of the 56 is just the value for the range. You would have to add another 24 ppm of the reading.
And this is just the 90 day spec. I usually use the 1 year spec when having a 1 year calibration interval. For the 56 one would have to add some term per month over 90 days.

Regards
Martin

[dietert1]

Yes, i fixed that hot 5 V regulator by pulling it and inserting it in a better position, with grease. Now i found another problem in that unit with the front/back input selector switch.

If i had a 5 V line at 4.8 V i would check the head room. In our 3456As the raw supplies are between 8.8 and 10.5 V with less than 500 mV ripple. Or just pull and exchange that 5V regulator. New ones usually have lower tolerances. Calibration should not depend on that, because in those ancient designs the 5V was used only for logic.

Regards, Dieter