Repair and experiments thread: HP 3458A U2

[TiN]

Some overnight data came thru. Looks like K2002-4 and HP3458B are very good friends now, running together down in the noise Also shows that 2002 is no much worse than 3458A, if we forget about ACAL and linearity of the latter for a second.



Green line is rusty 3458A.
Now having four "clocks" we can with good confidence say that 0.3ppm bump in first few hours did not come from meters, but from reference, as all four of them shown same data. You would be never be able to say this if use just one (even very good and just calibrated) meter.

Also there are often many concerns over shipping precision gear internationally, due to temperature, mechanical, pressure stresses, etc. In this case it all seem not to affect our homebrew DCV transfer (both 2002 and 3458B were calibrated in USA, while my lab is in half-globe away in Taiwan, for those who not frequent here) in significant way. Of course there is little chance that I just dont see the effect (e.g. all meters drifted same way due to shipping), but I found that very unprobable due to sample size (6 x LTZ1000A references, one of them being cross-shipped twice and retested 4 times, 3 x DMMs).

We also reaching the level on which calibration accuracy level (0.5-1ppm on 10VDC to be on safe side) cannot be improved much, no matter amount of money spent. Short of having own JJA system, surely. 

Having all this, resistance transfers likely be much more sensitive for shipping/stresses, as it's easier to damage sensitive elements than little semiconductor zener. 
We will see more once I get to resistance comparisons.

[Kleinstein]

It is no a surprise that the K2002 and 3458 are at about the same level for this test: it is expected that the main source for low frequency noise are the references and here both use the same type of LTZ1000, just like the DUT. So nearly half of the noise is expected to come from the DUT anyway.

The only possible problem for the K2002 could be the possibly not so good AZ implementation of the Keithley, as shown in the DMM7510 thread, but this might already be swamped by the references.

[TiN]

Cooked more data over weekend.
Three logs from LTZ modules and one from HP 3245A(modified with LTZ1000) set to 10.00000V.
Legend: HP3458A, HP3458B, K2002-4, K2002-6

[TiN]

Time has come...

I was thinking a while which meter to use for the wild experimentation, and after all things considered, decision was made to use my rusty 3458A as a patient. 3458U2 will have replaced front terminal block and missing inguard A1 shield, but remain unchanged for now.

Let the game begin, starting with easy part - DCI function. Why easy? Because it's simple component swap.

Current shunt modifications



I have broken A1 board, from which I removed original R213, R212, R211, R207, R209, R206 resistors and tested their TCRs to establish our baseline expectations.
Test results:
* R212 1 ohm - +8.3 ppm/K Datalog
* R211 9 ohm - +4.7 ppm/K Datalog
* R207 40K "reference" VHP101 - +1.4 ppm/K Datalog



I don't have much better replacement for 40K VHP101, so decided to leave that one alone for now, as we as 90 ohm (1mA range) and 4.53K shunts (10uA range), for but replaced rest with better resistors:

* R213 replaced with VPG foil shunt 319790 0.1 ohm 9W, measured TCR +0.55 ppm/K
* R212 replaced with VPG VPR221Z 1 ohm, measured TCR -0.86...-1.47ppm/K
* R211 replaced with AE foil resistor 9 ohm, measured TCR +1.5…+0.94 ppm/K
* R209 replaced with S102 foil resistor 634 ohm, measured TCR -1.2…1.25 ppm/K
* R206 replaced with CADDOCK USF340 resistor 500Kohm, measured TCR +0.78 ppm/K



After installing A1 board back, no SELF-TEST errors detected, ACAL ALL passed clean too.
Running GPIB app now collecting same 100mA running thru both meters from HP 3245A (LTZ-powered).

[ManateeMafia]

Looks like you are adding the undocumented pimp option 003. 

Will the alternating + and - TCR's have any impact on the readings or are the values too far apart?

[TiN]

There is still a long way before any celebrations can commit. 3458A holds many secrets within. One of which likely to be internal software corrections (unrelated to calibration constants) of components/resistors tempco. Why I think so. Well, here are some results. Test procedure was simple. DC current sourced thru both 3458X (modified) and stock 3458B and logged as usual. After getting somewhat stable current levels, data was corrected by computing offset and meters were covered with a blanket to raise internal temperature. Both meters stacked together, so temperature raise is expected similar.

Change of temperature (internal TEMP? sensor from A1's) was used as temperature delta and resulting tempco was calculated.

10mADC range:
3458B : +3.36 ppm/K measured, 10ppm/K spec. Datalog
3458X : +10.92 ppm/K measured, 10ppm/K spec 



Next range 100mA

3458B : -10.6 ppm/K measured, 25ppm/K spec. Datalog
3458X : +61.6 ppm/K measured, 10ppm/K spec   



1A range

3458B : -4.65 ppm/K measured, 10ppm/K spec. Datalog
3458X : -1.88 ppm/K measured, 25ppm/K spec   



At first glance, 100mA range data does not make sense, right? We changed +8.3ppm/K original resistor to much better one +1ppm/K. That's 8 times more stable resistor. But if firmware actually uses +8ppm/K correction factor for DCI measurement calculation it apply +8ppm correction per each degree of temperature increase, right? TEMP? reading of 3458X increased +7.5C, which gives about +60ppm total correction. 

And that's about what we see in measurement data. However this brings more questions, how does meter reads it's temperature for compensation, as TEMP? sensor require switching, and only read by GPIB command in my datalog program each 50th sample, not every sample. Sounds like a conspiracy theory...

DCI 1A range however improved significantly, so perhaps correction factor not used for this range? Again more questions, than answers. Setting up my TCR setup again, to retest TCR of original shunts removed from A1 PCBA to see what was their original tempco, while 100uA datalog runs now.

[Kleinstein]

Right after soldering, there is a chance that the resistors can show drift. This can be due to a high temperature, but also due to relaxation of mechanical stress. So it may need a kind of burn in for the new resistors. Still the curves really look like correlated to temperature and not like aging / relaxation.

With the temperature there is also a change in the normal ADC calibration, ideally adjusted with ACAL if the temperature changes significant. Is this done in these curve too ? It might be worth a try to do the comparison for a voltage (e.g. 100 mV or 1 V) reading. Depending on how much is adjusted during such a normal ACAL run, they could also adjust for the TC of the shunts. And it is only the super stable 40 K resistor that might mater. So the TC and stability of the shunts might not be so important, except for self heating - so it would be mainly the 1 A range that could profit from a better shunt. 

For the resistors there could also be an effect of thermal EMF. Some resistors can show quite some thermal EMF if they get warm due to self heating, or external temperature gradients. So the TCR is not the only important property of the shunts.

[Tony_G]

This is a great thread but I've clearly missed something somewhere - Is it being called a 3458B because of the repair/modification work?

Don't mean to derail the thread, just wondering - I get the 3458X as the eXperimental one, just can't find the reasoning behind the 3458B.

TonyG

[TiN]

I just reference B as second unit, as I have two meters, and calling them both 3458A likely to confuse.

Kleinstein
ACAL does indeed correct all DCI shunts/ohm source errors by comparing to 40K reference resistor, but it takes 830 seconds to do so and not very convinient for transfers. So if tempco can be improved without ACAL, it can be useful for multiple comparisons/logs. I measured some temperatures before with nominal currents flowing thru shunts, and temperature raise was very small, less than 8K for 1A current.
ACAL was not used during data collection charts other but at the start on both meters. 

I might tap voltage output directly after shunt network, to see if it does actually change that much, or introduced tempco increase caused by meter's firmware math/processing.

[Dr. Frank]

TiN,
you measured the T.C.s on the resistors of this burnt board..
Are you sure it also had this VHP101, but not the older type with unknown (at least 1.3ppm/K) T.C. characteristics?

Your measurements were not very distinctive, but the T.C. you've measured were clearly above the original HP T.C. specification.
I'd assume T.C. ~ 0.3ppm for the real VHP 101.

To answer the question of Tony_G.. This improvement on the T.C.s of the shunt resistors might as well improve the DCI specification.
If you look closely at these, you'll come to the conclusion, that most of the uncertainty might arouse from the T.C. specification.

Also, the Ohm calibration stability might be improved, as several of these DCI shunt resistors were used in the Ohm ACAL routine.

Frank

[Kleinstein]

For the current measurements there is more than the TC of the shunts. There is also the drift of the 100 mV or 1 V DC range (I don't know if there might be an intermediate amplification used). So the drift of the shunts is only one part.
So it might be really worth comparing one of the DC ranges as well - maybe the 3458X was is not that good in this respect too.

The full ACAL takes a long time and is thus not that practical. Is there such a thing as a partial ACAL, e.g. only for the DC gain and ADC ?

[TiN]

Yes, one can run ACAL DCV (takes 130 seconds), ACAL OHM, ACAL ACV separately. I often use ACAL DCV when doing LTZ comparisons, to achieve transfer spec over longer times (hours-days).

Previous tests of 1V DCV show max 0.6ppm change with 0.5C TEMP? variation over multiple hours. Perhaps worth to note, that 3458X is my older meter, which I use 24/7 since it's repairs finished February 2016. It's <1ppm stable for all DCV ranges. I can't tell same about newer unit, so that's why older one went for modifications.

Idea was to start with shunts and go down the signal path as it goes, just didn't expect 60ppm TC increase from only shunt change .

Dr.Frank
Could be. But they all look exactly same. I'm not going to replace 40K off my A1 just yet.

I can repeat 40K TCR measurement (of resistor from dead A1) once again with different meter, to confirm the value.

100uA range seem to be about same for both meters, below 1ppm/K.

[Dr. Frank]

TiN,
The earlier 40k resistors were different, and looked different.
They did not have that VHP101 marking, which was used from about 1995 / 96 onwards.
So you simply have to verify that, on your DUT, please.
Frank

[Kleinstein]

If on provoking the higher internal temperature there also is a significant change in air flow, this could cause additional drift / shifts. This is because this would change temperature gradients inside too. So changing the temperature due to reduced air flow is different from higher air temperature, even if the same internal temperature reading is reached. Some of the really hot parts (e.g. AC board) might even overheat.

It might be worth measuring the 1 Ohms resistor removed from 3458A - is this really so bad to show a TC in the 50 ppm range, so that old compensation would cause the now higher TC.

The resistance of the copper line between the 0.1 and 1 Ohms shunt could also contribute. So may the resistance of the E1 terminal of the 0.1 Ohms shunt, as thus is added to the 1 Ohms value. Also the 1 Ohms resistor looks like it is 4 wire, but used as 2 wire - so it's wire resistance also adds, while it is included in the simpler 2 wire resistor. Still this should not be enough to cause a 60 ppm/K effect.

[mimmus78]

TiN you must also check what happens with EMF.

Let the meter rest with zero current to "stabilise temperature", and remove leads from binding posts.
Do an ACAL. This should zero the current range to read zero with zero current (yes I know you know).

Than run your test and make the reading/temperature stabilise. After this happens remove stuff from binding posts, what is reading at ZERO now?
This should be EMF contribute to error.

On my unit at 1A range, after 10 minutes of running at 1A I have +4uA due to this "heating" effect.
Most of the heating/effect happens on fist minutes.
After removing current the meter returned to ZERO in a couple of minutes.

This effect seems not to influence much the 1A range as I swapped cables and got the "same" current reading, estimated to less than 1uA difference but maybe this EMF will produces more error on the lower current ranges.
Can you repeat this experiment with your BAD ranges?

EDIT: I tested also the other ranges and this thermal effect seems to be present only on the 1A range.

[mimmus78]

And anyway this remember me of someone ...

https://youtu.be/_CvNe-JhjD4

Inviato dal mio Nexus 6P utilizzando Tapatalk

[ManateeMafia]

Are you trying to convince TiN to repair some of these bad U180's?   

[mimmus78]

No I do not need to convince him, he already tried ... I never thought possible of repair stuff this way ...

[TiN]

Bit more data captured.

Retest 10mADC range:
3458B : +3.47 ppm/K measured, 10ppm/K spec. Datalog
3458X : +12.35 ppm/K measured, 10ppm/K spec 



1mA range

3458B : +0.55 ppm/K measured, 10ppm/K spec. Datalog
3458X : +5.48 ppm/K measured, 10ppm/K spec



100uA range

3458B : -1.42 ppm/K measured, 10ppm/K spec. Datalog
3458X : -0.45 ppm/K measured, 10ppm/K spec



No problem with 100uA range at all.

100nA range

3458B : <100 ppm/K measured, 210ppm/K spec. Datalog
3458X : <100 ppm/K measured, 210ppm/K spec 



Error/tempco of 100nA range mostly delivered from amplification stage, so this measurement have no meaning. It matches unmodified 3458, so replacement of 500K resistor was for nothing really.

DCV comparison, 100mV range

3458B : +1.72 ppm/K measured, 2.2ppm/K spec. Datalog
3458X : +0.63 ppm/K measured, 2.2ppm/K spec   



DCV comparison, 1000mV range

3458B : -0.2 ppm/K measured, 1.3ppm/K spec. Datalog
3458X : -0.128 ppm/K measured, 1.3ppm/K spec   



As we can see, DCV 100mV and 1V ranges are good and cause zero problems, and not cause of DCI excessive tempco. We can close A3 effect theory for now.

Original A1's resistors TCR tests:

Test results:
* R211 9R ohm from A1 - -1.38 ppm/K Datalog - measured directly by K2002
* R209 634R ohm from A1 - +4.65 ppm/K Datalog - measured directly by K2002
* R206 500 Kohm from A1 - +1.86 ppm/K Datalog - measured directly by K2002
* R212 1 ohm shunt from A1 - +1.7 ppm/K Datalog - measured directly by K2002
* R213 0.1 ohm HP 03458-82501-1 shunt from A1 - -6.2...-13 ppm/K Datalog - measured directly by K2002
* R213 0.1 ohm HP 03458-82501-1 shunt from A1 - 7.5 ppm/K from 25C to 35C, 1000mA test current Datalog - measured by Keithley 182-M

Summary table:

Resistor	Original A1	New resistor	Range	Test result
R213 0.1R	+7.50 ppm/K	+0.55 ppm/K	1 A	Better
R212 1R	+1.70 ppm/K	-0.86...-1.47 ppm/K	100 mA	Out of spec
R211 9R	-1.38 ppm/K	+1.5...+0.94 ppm/K	10 mA	Out of spec
R210 90R	Unchanged	unchanged	1 mA	Out of spec
R209 634R	+4.65 ppm/K	+1.2...+1.25 ppm/K	100 uA	Better
R208 4.53K	Unchanged	unchanged	10 uA	Not tested
R207 40K	Unchanged	unchanged	1 uA	Not tested
R206 500KR	+1.86 ppm/K	+0.78 ppm/K	100 nA	Same

Perhaps I'll swap 9R for 10mA range back to see if that "fixes" the 10mA TC.

Dr. Frank
40K resistor from dead A1 is Vishay 301031 hermetical



40K on my 3458A's A1 is Vishay VHP101 (photos posted above). I think same is in my second unit, as it's 5 years younger.

[Kleinstein]

The 1 Ohms resistor is also used in the higher ranges. So the 60 ppm / K drift of that resistor would also cause something like 6 ppm/K for the next higher range.
How was the PC of the new 1 Ohms shunt measured: As a 4 wire resistor, or as a 2 Wire resistor like use in the circuit and 4 wires only from than on ?

If the contacts to the resistor have a rather high resistance, this could cause the problems, not only for the 100mA range, but also the 10 mA range. Wire resistance at the other side would add another 6 ppm/K and thus would end up at 12 ppm/K as observed. So chances are the 1 Ohms resistor is 4 wire mode only and may thus not be suitable here.

[TiN]

That thought crossed my mind today. TCR was measured as 4-wire resistor, with each pin connected individually to K2002.
I connected both current and sense pins together in meter, so it's definitely different that TCR measurement was done.
Also test current is different, K2002 source 7.2mA during measurement, instead of 100mA used in actual meter.

Here's graph during TCR test, 1 ohm resistor is rosewood color. Simple box calculation gives -1.9ppm/K (ramp from +20°C to +40°C).


As bonus for thread, I got cover off 3458B for some checks, took few pics. A1 R207 confirmed to be indeed VHP101.
A3's logic is now in ALTERA MAX V CPLD (1270 ALE).



Those unpopulated test points on the corner are JTAG. Curious to see if PLD is actually secured  . 



CPLD Reads just fine Backed up just in case. CRC matches the sticker label.



Love these Tek grabbers, always handy 



Reference A9 module. 15K VAR installed (prior to calibration in January) to reduce LTZ temperature  .

[mimmus78]

"Curious to see if PLD is actually secured ..." and few post later someone got fired 

Anyway TiN why you used this "not even encapsulated" audio resistor?
OK it's 100K over 15K so any change in this resistor is attenuated ... but isn't it too much risky like this?

Warranty is expired on my unit, so time to void this stickers, have a check inside and get some photo for your collection.
I will tear down again in next two weeks ... cause I want to install some less nosier fan inside.
So if you are curios of something else just let me know.

[mimmus78]

More photos ...

[TiN]

Anyway TiN why you used this "not even encapsulated" audio resistor?
OK it's 100K over 15K so any change in this resistor is attenuated ... but isn't it too much risky like this?

Long story, but she'll be alright. It's not really naked, there is still thick enamel coating over it.

... cause I want to install some less nosier fan inside.

New A3 too . However I'd strongly suggest against replacing fan, it's airflow speed and force are important and I'd keep them as designed.
After all it's not that noisy (compared to Keithley 2304 or Keithley 2002 series gear it's barely audible).

[TheSteve]

I find the fans in my 3458A's beyond annoying. With two of them running the frequencies from the fans seem to beat together. I can actually hear it throughout the entire house. On another floor isn't isn't loud, but I still know it is them. I am using the exact fans Keysight sells for the 3458A.